SMARTPHONE APPS AND VIRTUAL REALITY AS ROAD SAFETY I … · smartphone safe-driving app as an intervention tool. The off-the-shelf safe-driving app "Flo - driving insights" (Flo)

SMARTPHONE APPS AND VIRTUAL

REALITY AS ROAD SAFETY

INTERVENTIONS: EXAMINING THEIR

REAL-WORLD EFFECTS FOR YOUNG

DRIVERS

Daniel Lyubomirov Vankov Master of Business Administration

Master of Finance Bachelor of Finance

Submitted in fulfilment of the requirements for the degree of Doctor of Philosophy

Centre for Accident Research & Road Safety – Queensland

School of Psychology and Counselling

Faculty of Health

Queensland University of Technology

2019

Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers i

Keywords

Apps, Consumer-Oriented Technology, Drink-driving, Driving Under the Influence, Drug-driving, Interventions, Risky Driving Behaviour, Road Safety, Smartphones, Speeding, Theory of Planned Behaviour, Virtual Reality, Young Drivers.

ii Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers

Executive summary

1.35 million people lost their lives on the road in 2016 with young drivers being

at higher risk. Calling for innovation, there is evidence that current global road safety

efforts do not yield the desired results. In recent years, novel Consumer-Oriented

Technologies (COT) have been introduced to reduce road trauma. For the current

research, COTs are defined as technologies that answer consumer (or individual

driver) wants or needs. As such, they are not necessarily designed from a road safety

perspective. While COTs may be quickly adopted, the evidence is lacking about their

impact on safety. Thus, this research sought to answer the fundamental question of

how does a COT intervention influence young drivers' safety.

This PhD program of research evaluated the effects of two COT-based

interventions over a period of three months. One of the interventions used a

smartphone safe-driving app to reduce speeding (480 participants). The other used

Virtual Reality (VR) simulations of risky driving to influence driving under the

influence of alcohol or drugs (DUI) (329 participants). The participants were assigned

to an intervention condition or a control condition (no intervention) to allow for robust

evaluation.

The evaluation framework was underpinned by an extended Theory of Planned

Behaviour (TPB). Self-report questionnaires were administered before exposure to the

interventions. Follow-up surveys were conducted approximately three months after the

initial surveys. The findings did not provide evidence for the safety benefits of using

the two deployed COTs beyond some limited secondary effects. This suggested that

the positive impact of some COTs may be overstated. However, the findings should

be interpreted in the light of the encountered limitations, such as high drop-out rates,

lack of information on participants' pre-intervention familiarity with COT, lack of

naturalistic data for comparison or data from control participants also being potentially

impacted by the experiments. Nevertheless, this research contributed new and valuable

knowledge towards using COTs in prevention efforts, with practical considerations in

road safety interventions' design, implementation and evaluation.

Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers iii

Abstract

Transport, particularly road transport, is part of people's everyday lives,

powering economies and growth around the world. However, it comes at a cost, and

in many cases, its impact has considerable negative consequences to people's health

and well-being. In 2016 alone, a record 1.35 million people lost their lives on the road.

Young drivers are globally disproportionately affected. The international community

engages a vast amount of effort to minimise road trauma with concrete targets set both

globally, through the United Nations Decade of Action for Road Safety and the

Millennium Development Goals, and locally, with countries establishing their policies

and programs to support road safety. Unfortunately, there is evidence that those global

efforts to reduce deaths and injuries caused by road crashes do not yield the expected

results.

The rapid development and integration of technology in recent years are

regarded as a potential route to reduce road trauma. There is evidence in the literature

that technology has both positive and negative impacts on drivers' performance. For

example, a class of COTs that makes its way into the lives of drivers, such as

smartphones and many of the apps delivered through them, answer consumer (or

individual driver) wants or needs. However, they do not contribute to executing the

driving task.

Some COTs predominantly introduce risks for drivers but, nevertheless, are

expected to expand their presence in the young drivers' ecosystem. The literature

highlights smartphones as a significant source of unwanted and unintended distraction,

thus a big threat in terms of safety. However, unlike for other COTs, the literature also

suggests that smartphones may have the potential to positively influence young drivers

in the context of carefully designed safe-driving apps. A different type of software

application, such as VR simulations, is making its way in prevention efforts outside

the cars. As VR simulations are not used while real driving takes place, they are seen

as a safer option to raise awareness on driving-related risks in conditions that sit

between the laboratory and the real-life experience. Both academia and business

embrace the power of such COTs and offer solutions in an attempt to address road

iv Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers

safety problems in general, and young drivers' overrepresentation in road crashes, in

particular.

COTs solutions may be quickly adopted and integrated into awareness-raising

initiatives. However, limited research has been undertaken to investigate if they offer

real safety benefits. The newer the technology, the less knowledge there is around its

impact. This PhD program of research used a multidisciplinary approach to address

this research gap, sitting at the intersection of road safety, psychology and human-

computer interaction fields.

The research process drew upon a combination of data collection methods to

address the research questions appropriately (e.g., qualitative methods in systematic

reviews and a focus group, and quantitative methods in the evaluation of

interventions). Initially, the problems of young drivers were investigated, and the

behaviours of interest were identified. Then behavioural theories were examined to

determine the most suitable framework to be applied. As a result, this PhD program of

research was underpinned by an extended TPB as a theoretical framework. The

studies’ implementation of the adopted framework also addressed limitations,

observed in prior road safety studies underpinned by TPB, such as not implementing

interventions as part of the research or not collecting data before and after the

interventions to evaluate their effect. The TPB constructs of most interest were the

participants' intention and the participants' self-reported behaviour. These constructs

and their predictors were assessed before and after the two implemented interventions.

Before the two COT-based interventions were implemented, two systematic

literature reviews of COTs investigated the application of smartphone safe-driving

apps and VR simulations of risky driving in road safety research. In the first systematic

review, positive safety benefits for young drivers in naturalistic settings were reported

as a result of a deployed smartphone safe-driving-app intervention in only three out of

80 papers that had been selected for full-text review (22 papers included in the

qualitative synthesis, 13 of them involved young drivers aged 18-25). No safety

benefits for young drivers from using VR simulations of risky driving were found

during the second systematic review with a qualitative synthesis of 6 papers.

Nevertheless, the systematically-explored body of evidence informed the two

intervention studies' evaluations. The systematic reviews provided insights on how two

examples of contemporary COTs, smartphone safe-driving apps and VR, were

Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers v

previously used within a road safety context, similar to the current program of

research.

Preliminary work was implemented to establish criteria and to select a

smartphone safe-driving app as an intervention tool. The off-the-shelf safe-driving app

"Flo - driving insights" (Flo) was selected as a tool for the smartphone safe-driving

app intervention. Due to a lack of the same variety and availability of apps to choose

from in the VR domain, the software "3D Tripping", for which real safety benefits had

not yet been investigated, was selected and used as an intervention tool in the VR

intervention.

The first intervention explored whether Flo as an intervention tool can positively

influence the young drivers' (aged 18 to 25) intention not to speed as well as their

subsequent self-reported behaviour of not speeding during the three months of the

intervention. Self-report questionnaires were administered before (n=480) and after

(n=210) the intervention period. A Control group (n=126 after the intervention period)

was established so that any general shifts over this period could be identified and not

assigned as a result of the intervention. A Flo leaderboard was created in which the

study participants could observe each other's driving scores, i.e. driving performance

and achievements. Periodically those driving scores were recorded so that it was

possible for the research team to follow individual driver's progress (n=62).

The collected data were assessed at several levels. Pre-intervention data were

analysed through a 3-step hierarchical multiple regression analysis on intention not to

speed. It was found that a significant variation in intention not to speed was explained

by demographic variables (gender, age and driving license) (6%), by TPB constructs

(instrumental attitude, affective attitude, subjective norm, descriptive norm, self-

efficacy and perceived controllability) over and above the demographics (48%) and by

additional predictors (past behaviour of not speeding, perceived risk, moral norm,

peers' norm, and impulsivity) over and above TPB (19%). Past behaviour of not

speeding was both the strongest individual (69%) and the strongest unique predictor

(18%). Including sensitivity to punishment and sensitivity to reward at the third step of

the model increased the explained variance and decreased the individual and the

unique contribution of past behaviour of not speeding by 2% in each case.

The effect of the intervention was analysed through a series of one-way and two-

way analysis of covariance (ANCOVA) tests. No statistically significant effects of the

vi Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers

intervention were observed. No consistent effects were revealed by exploring the

leaderboard data, either.

Post-intervention data were analysed by a 3-step hierarchical multiple regression

to investigate whether behaviour of not speeding during the three months of the

intervention could have been predicted by the participants’ baseline data. A significant

variation was explained by demographic variables (gender, age and driving license)

(10%), by TPB constructs (intention not to DUI, self-efficacy and perceived

controllability) over and above the demographics (40%) and by additional predictors

(past speeding behaviour, perceived risk, moral norm, peers' norm, impulsivity,

sensitivity to reward and sensitivity to punishment) over and above TPB (14%). Once

again, past behaviour of not speeding was both the strongest individual (55%) and the

strongest unique predictor (10%).

Finally, one-way and two-way ANCOVA tests were performed to investigate

for any potential negative side effects, namely an increased smartphone engagement

amongst the study participants as a result of the intervention. Effects with statistical

significance were found only in the interaction between condition and gender, which

suggested that the intervention facilitated a decreased level of smartphone engagement

amongst the Intervention group male participants.

The second intervention aimed to examine the effect of VR simulations of risky

driving on DUI intention and self-reported behaviour amongst young people (aged 18

to 25) with "3D Tripping" as an intervention tool. Similar to the first intervention

study, self-report questionnaires were administered before (n=329) and three months

after (n=138) the VR intervention was implemented. A convenience Control group

(n=39 three months after the intervention) was established to help identify general

shifts over time.

The collected data was assessed at three levels. Initially, in a 3-step logistic

regression analysis of pre-intervention data, the demographic variables (gender, age

and driving experience) explained a significant 6.3% to 11.1% of the variance in

intention not to DUI. Adding the TPB constructs (instrumental attitude, affective

attitude, subjective norm, descriptive norm, self-efficacy and perceived controllability)

explained 21.8% to 38.3% of the variance in intention not to DUI. Adding additional

predictors (past behaviour of not DUI, perceived risk, moral norm, peers' norm, and

impulsivity) explained 30.1% to 52.9% of the variance in intention not to DUI (36.5%

Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers vii

to 73.3% when sensitivity to punishment and sensitivity to reward were also added).

Past behaviour of not DUI was the strongest statistically significant unique predictor

(p = .006, odds ratio = 43.08).

A series of chi-square tests for independence, McNemar's tests and Wilcoxon

Signed Ranks tests were performed to evaluate the effect of the intervention. No

statistically significant effects were found.

A second 3-step logistic regression analysis was performed after the intervention

to investigate whether participants’ behaviour of not DUI during the three months after

the intervention could have been predicted with their baseline data. The demographic

variables (gender, age and driving experience) explained between 7.7% and 13.1% of

the variance in behaviour of not DUI during the three months after the intervention.

Adding the TPB constructs (intention not to DUI, self-efficacy and perceived

controllability) explained between 11.3% and 19.3% of the variance in behaviour of

not DUI during the three months after the intervention. At the third step, the model,

including additional predictors (past behaviour of not DUI, perceived risk, moral

norm, peers' norm, impulsivity, sensitivity to punishment and sensitivity to reward),

explained between 24.3% and 41.5% of the variance in behaviour of not DUI during

the three months after the intervention. The strongest statistically significant unique

contributor was peers' norm (p = .005, odds ratio = 1.63).

Besides the evidence for both the predictive power of TPB and the positive

effects in regards to smartphone engagement, i.e. distraction, a behaviour of secondary

interest in the smartphone safe-driving app study, findings from both intervention

studies did not provide statistically significant evidence of safety benefits in regards to

the main behaviours of interest, speeding and DUI. These findings suggest that the

positive impact of some technologies, in general, and of the two deployed COTs, in

particular, may be overstated. They also point to the need for robust evaluations to be

undertaken before technology applications roll-out to the general public. Overall, this

program of research highlighted the usefulness of theoretically-grounded evaluation

concerning emerging solutions offered by COTs. Although bringing to market is what

developers are usually interested in, the premature release of COTs may lead to

unsustained claims.

This research contributed new and valuable knowledge towards using COT in

prevention efforts targeting the public, in general, and young drivers, in particular.

viii Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers

Such information may provide an evidence-based foundation for road safety

stakeholders, such as road safety social entrepreneurs and researchers, in their search

for new and better instruments to address existing risks on the road. The PhD thesis

describes practical considerations in the design, implementation and evaluation of

interventions, including involvement of a large number of young people, using social

media for recruitment, evaluation of self-reports and using off-the-shelf technology

applications for prevention purposes. Furthermore, by examining not only young

drivers' cognitive determinants but also their demographic and personality

characteristics, this research acknowledged that people are different by nature and

there is no "one size fits all" solution when it comes to promoting safe driving

behaviour on the road.

Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers ix

Table of contents

Keywords .................................................................................................................................. i

Executive summary .................................................................................................................. ii

Abstract ................................................................................................................................... iii

Table of contents ..................................................................................................................... ix

List of figures ........................................................................................................................ xiii

List of tables ............................................................................................................................ xv

Glossary of terms ................................................................................................................ xviii

List of abbreviations ............................................................................................................... xx

Funding and awards ............................................................................................................. xxii

Statement of original authorship ......................................................................................... xxiii

Acknowledgement .............................................................................................................. xxiv

Chapter 1: Introduction ...................................................................................... 1

1.1 Research problem ........................................................................................................... 3

1.2 Research aim and objectives ........................................................................................... 5

1.3 Significance of the research ............................................................................................ 6

1.4 Outcomes ........................................................................................................................ 7

1.5 Document outline ........................................................................................................... 8

Chapter 2: Literature review .............................................................................. 9

2.1 Young drivers' risky driving behaviours ......................................................................... 9

2.2 Contributing factors ...................................................................................................... 11 2.2.1 Driving experience ............................................................................................. 11 2.2.2 Optimism bias ..................................................................................................... 13 2.2.3 Gender ................................................................................................................ 13

2.3 Interventions ................................................................................................................. 14 2.3.1 Driver training .................................................................................................... 14 2.3.2 Media campaigns ................................................................................................ 16 2.3.3 Law enforcement ................................................................................................ 17

2.4 Consumer-oriented technologies .................................................................................. 17 2.4.1 Context ............................................................................................................... 18 2.4.2 Safe-driving apps ................................................................................................ 19 2.4.3 Virtual reality ..................................................................................................... 21

2.5 Conclusion and identification of a research gap ........................................................... 22

Chapter 3: Theoretical considerations informing intervention evaluation framework .......................................................................................................... 25

3.1 Introduction .................................................................................................................. 25

3.2 Transtheoretical Model of Health Behavior Change .................................................... 26

3.3 Health Belief Model ..................................................................................................... 29

x Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers

3.4 Social Cognitive Theory .............................................................................................. 31

3.5 Theory of Planned Behaviour ...................................................................................... 34 3.5.1 Criticisms and limitations of TPB ..................................................................... 36

3.6 Extending TPB ............................................................................................................. 36 3.6.1 Additional normative influences ........................................................................ 37 3.6.2 Risk perception .................................................................................................. 38 3.6.3 Personality characteristics .................................................................................. 39

3.7 Conclusion ................................................................................................................... 40

Chapter 4: Research design ............................................................................... 42

4.1 Research questions ....................................................................................................... 42

4.2 The extended TPB and selecting intervention tools ..................................................... 45

4.3 Methodology ................................................................................................................ 48

4.4 Intervention studies’ designs ........................................................................................ 50 4.4.1 Participants ........................................................................................................ 50 4.4.2 Surveys .............................................................................................................. 51 4.4.3 Variables ............................................................................................................ 55 4.4.4 Analyses ............................................................................................................. 58 4.4.5 Ethics ................................................................................................................. 60

Chapter 5: Study 1 - Systematic review of safe-driving apps ........................ 61

5.1 Rationale for conducting a systematic review ............................................................. 61

5.2 Method ......................................................................................................................... 62 5.2.1 Search databases ................................................................................................ 62 5.2.2 Literature search criteria .................................................................................... 62 5.2.3 Search term ........................................................................................................ 63

5.3 Search and screening results ........................................................................................ 63

5.4 Findings ........................................................................................................................ 65 5.4.1 Studies’ designs and samples ............................................................................. 75 5.4.2 Sensors and measures ........................................................................................ 76 5.4.3 Benefits .............................................................................................................. 77

5.5 Summary ...................................................................................................................... 78

5.6 Discussion .................................................................................................................... 79

Chapter 6: Selecting a safe-driving app ........................................................... 82

6.1 Focus group design ...................................................................................................... 82 6.1.1 Participants ........................................................................................................ 82 6.1.2 Procedure and materials ..................................................................................... 83 6.1.3 Data analysis ...................................................................................................... 84

6.2 Findings from the Focus group .................................................................................... 84 6.2.1 Vision on young people's use of technologies in the car ................................... 85 6.2.2 Discussing smartphone safe-driving apps .......................................................... 86

6.3 Synthesis of Focus group’s findings ............................................................................ 88

6.4 Selecting a safe-driving app for an evaluation ............................................................. 90

6.5 Conclusion ................................................................................................................... 96

Chapter 7: Study 2 - Intervention with an off-the-shelf smartphone safe-driving app .......................................................................................................... 98

Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers xi

7.1 Introduction .................................................................................................................. 98

7.2 Method ........................................................................................................................ 101 7.2.1 Study design ..................................................................................................... 101 7.2.2 Recruitment ...................................................................................................... 101 7.2.3 Intervention tool ............................................................................................... 102 7.2.4 Procedure .......................................................................................................... 103 7.2.5 Participants ....................................................................................................... 105 7.2.6 Intervention ...................................................................................................... 106

7.3 Hypotheses .................................................................................................................. 107

7.4 Preliminary analysis ................................................................................................... 109 7.4.1 Missing data ..................................................................................................... 110 7.4.2 Data transformation .......................................................................................... 110 7.4.3 Dropouts ........................................................................................................... 112 7.4.4 Assumptions checks ......................................................................................... 112 7.4.5 Personality characteristics ................................................................................ 114

7.5 Results ........................................................................................................................ 114 7.5.1 Participants' intention not to speed before the intervention (RQ2.1, H.1 -

H.3) 114 7.5.2 Changes in salient beliefs (RQ2.2, H.4 - H.7) .................................................. 119 7.5.3 Predictors of behaviour of not speeding during the intervention (RQ 2.3,

H.8 - H.10)........................................................................................................ 124 7.5.4 Potential negative effects: Self-reported smartphone engagement (RQ2.4,

H.11) ................................................................................................................. 126

7.6 Discussion ................................................................................................................... 131 7.6.1 Findings ............................................................................................................ 131 7.6.2 Strengths ........................................................................................................... 135 7.6.3 Limitations........................................................................................................ 136

7.7 Conclusion .................................................................................................................. 138

Chapter 8: Study 3 - Systematic review of VR simulations of risky driving .... 140

8.1 Rationale for conducting a systematic review ............................................................ 140

8.2 Method ........................................................................................................................ 141 8.2.1 Search databases ............................................................................................... 141 8.2.2 Literature search criteria ................................................................................... 141 8.2.3 Search term ....................................................................................................... 141

8.3 Search and screening results ....................................................................................... 142

8.4 Findings ...................................................................................................................... 143 8.4.1 Studies’ samples ............................................................................................... 146 8.4.2 Measures ........................................................................................................... 146 8.4.3 Benefits ............................................................................................................. 146

8.5 Summary ..................................................................................................................... 147

8.6 Discussion ................................................................................................................... 147

Chapter 9: Study 4 - Intervention with VR simulations of risky driving ... 149

9.1 Introduction ................................................................................................................ 149

9.2 Method ........................................................................................................................ 150 9.2.1 VR tool and intervention .................................................................................. 150 9.2.2 Recruitment ...................................................................................................... 155

xii Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers

9.2.3 Data collection procedure ................................................................................ 157 9.2.4 Participants ...................................................................................................... 158

9.3 Hypotheses ................................................................................................................. 158

9.4 Preliminary analysis ................................................................................................... 160 9.4.1 Missing data ..................................................................................................... 160 9.4.2 Dropouts .......................................................................................................... 160 9.4.3 Assumption checks and data transformation ................................................... 161 9.4.4 Personality characteristics ................................................................................ 163

9.5 Results ........................................................................................................................ 163 9.5.1 Participants' intention not to DUI before the intervention (RQ4.1, H.12 -

H.14) ................................................................................................................ 163 9.5.2 Changes in salient beliefs (RQ4.2, H.15 - H.17) ............................................. 168 9.5.3 Predictors of behaviour of not driving under the influence of drugs or

alcohol after the intervention (RQ4.3, H.18 - H.20) ........................................ 172

9.6 Discussion .................................................................................................................. 175 9.6.1 Findings ........................................................................................................... 175 9.6.2 Strengths .......................................................................................................... 177 9.6.3 Limitations ....................................................................................................... 178

9.7 Conclusion ................................................................................................................. 180

Chapter 10: General discussion ........................................................................ 182

10.1 Overall contribution ................................................................................................... 182

10.2 Integration of findings, strengths and limitations ...................................................... 185 10.2.1 Theoretical considerations ............................................................................... 185 10.2.2 Practical considerations ................................................................................... 187 10.2.3 Methodological considerations ........................................................................ 189

10.3 Future research directions .......................................................................................... 190 10.3.1 A researcher's wish list to COT developers ..................................................... 194

10.4 Chapter summary ....................................................................................................... 197

References ............................................................................................................... 201

Appendices .............................................................................................................. 219

Appendix A Smartphone safe-driving apps on Google Play and iTunes .......... 221

Appendix B Study 2 Questionnaire ...................................................................... 226

Appendix C Study 4 Questionnaire ...................................................................... 233

Appendix D Additional Study 2 Effects of the intervention models .................. 246

Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers xiii

List of figures

Figure 1.1. Number and rate of road traffic death per 100,000 population: 2000–2016 (WHO, 2018) ............................................................................... 1

Figure 1.2. Widening gap between the actual and desired progress towards the EU 2020 target (Adminaite et al., 2018) ........................................................ 2

Figure 1.3. Australian progress until 2017 towards fatality target (BITRE, 2018) .............................................................................................................. 3

Figure 2.1. Percentage of South Australia drivers involved in a crash five years after licensing (Austroads, 2008). ................................................................ 15

Figure 3.1. TTM stages of change ............................................................................. 27

Figure 3.2. The Health Belief Model ......................................................................... 30

Figure 3.3. Social Cognitive Theory Model ............................................................... 32

Figure 3.4. Theory of Planned Behaviour (Ajzen, 1991) ........................................... 34

Figure 3.5. Extension of the Theory of Planned Behaviour in the current program of research. ................................................................................... 41

Figure 4.1. Outline of the overall thesis methodology ............................................... 49

Figure 5.1. Data extraction flowchart based on the PRISMA statement. .................. 65

Figure 6.1. Focus group visual brainstorming tools ................................................. 84

Figure 7.1. Time on screen as reported by Flo for each trip ................................... 100

Figure 7.2. Smartphone with Flo, providing real-time feedback while driving. ..... 102

Figure 7.3. Safe-driving app intervention design .................................................... 104

Figure 7.4. Example screenshot of Flo GoOz leaderboard ..................................... 104

Figure 8.1. Data extraction flowchart based on the PRISMA statement. ................ 142

Figure 9.1. A user is getting used to managing the VR driving simulator. .............. 151

Figure 9.2. The VR software visualises parking of the vehicle before entering the night club. ............................................................................................. 152

Figure 9.3. A choice to experience impaired driving as a result of ecstasy, cannabis or magic mushrooms influence is given to users. ....................... 152

Figure 9.4. A user driving under the VR-simulated influence of magic mushrooms. ................................................................................................ 153

Figure 9.5. A participant, operating the VR driving simulator in front of their peers. .......................................................................................................... 155

Figure 9.6. A participant is operating the VR software on a fully adjusted VR driving simulator. ....................................................................................... 155

Figure 10.1. Outline of the thesis studies and findings ............................................ 184

Figure 10.2. Example of future safe-driving app intervention design ..................... 193

xiv Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers

Figure 10.3. Example of a future DUI VR intervention design ................................ 194

Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers xv

List of tables

Table 1.1. Definitions of key terms ......................................................................... xviii

Table 4.1. COTs' selection criteria, derived from the extended TPB framework. ..... 47

Table 4.2. Constructs and time of measurement. ....................................................... 54

Table 4.3. Items, adapted from Elliott and Thomson (2010). .................................... 56

Table 4.4. Items, adapted from Gannon et al. (2014). ............................................... 58

Table 5.1. Impact and effect of apps, games and gamification on young drivers' road safety. ................................................................................................... 66

Table 6.1. Country of origin and gender of participants in the Focus group. ........... 83

Table 6.2. Criteria for smartphone safe-driving apps, synthesised from Focus group’s findings. .......................................................................................... 89

Table 6.3. Smartphone safe-driving apps (yes=1, no=0, double points for SC3). ..... 91

Table 7.1. Means, standard deviations and bivariate correlations for the TPB variables at Time 1 (n=480). ..................................................................... 115

Table 7.2. 3-step hierarchical multiple regression analysis, predicting intention not to speed for all participants at Time 1, with demographic factors, TPB variables and additional variables as predictors (n=480). ............... 116

Table 7.3. Linear multiple regression analysis predicting Intention not to speed at Time 1 with demographic factors, TPB variables and additional variables, including sensitivity, as predictors (n=210). ............................ 118

Table 7.4. Means, standard deviations and bivariate correlations for the standard TPB variables at Time 2 (n=157). .............................................. 119

Table 7.5. Means and standard deviations for the Control and the Intervention groups' intention not to speed at Time 1 and Time 2 (n=157). .................. 120

Table 7.6. Interaction effects between Condition and personality characteristics, intention not to speed, adjusted for Time 1 values (n=157). ..................................................................................................... 120

Table 7.7. Means and standard deviations for the Control and the Intervention groups' behaviour of not speeding at Time 1 and Time 2 (n=157). .......... 121

Table 7.8. Interaction effects between Condition and personality characteristics, past behaviour of not speeding during the three months of the intervention, adjusted for Time 1 values (n=157). .............. 122

Table 7.9. Effect of the intervention on TPB variables, adjusted for Time 1 values (n=157). .......................................................................................... 122

Table 7.10. 3-step hierarchical multiple regression analysis, predicting behaviour of not speeding during the three months of the intervention for all participants at Time 2, with demographic factors, TPB variables and additional variables as predictors (n=210). ....................... 125

xvi Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers

Table 7.11. Mean, standard deviation and bivariate correlations for the phone interaction variables at Time 1 (n=157). ................................................... 126

Table 7.12. Self-reported phone interaction at Time 1 and Time 2 (n=157). .......... 127

Table 7.13. Mean, standard deviation and bivariate correlations for the phone interaction variables at Time 2 (n=157). ................................................... 128

Table 7.14. Phone interactions' means and standard deviations for the Control (n=126) and the Intervention group (n=31) at Time 1 and Time 2. .......... 128

Table 7.15. Effect of the intervention on phone interaction variables, adjusted for Time 1 values, with Condition as a fixed factor (n=157). .................... 128

Table 7.16. Interaction effects between Condition and personality characteristics, phone interaction variables, adjusted for Time 1 values (n=157). .......................................................................................... 129

Table 8.1. Virtual reality in the road safety literature ............................................. 144

Table 9.1. Frequencies, means, standard deviations and bivariate correlations for the TPB variables at Time 1 (n=329). .................................................. 164

Table 9.2. Logistic regression analysis, predicting Intention not to DUI for all participants at Time 1, with demographic factors as predictors (n=329). ..................................................................................................... 165

Table 9.3. Logistic regression analysis predicting Intention not to DUI for all participants at Time 1 (n=329) with demographic factors and TPB variables as predictors. .............................................................................. 165

Table 9.4. Logistic regression analysis predicting Intention not to DUI for all participants at Time 1 (n=329) with demographic factors, TPB variables and additional variables as predictors. ..................................... 166

Table 9.5. Logistic regression analysis, predicting Intention not to DUI for all participants at Time 1, with demographic factors, TPB variables and all additional variables as predictors (n=138). ......................................... 167

Table 9.6. Frequencies, means, standard deviations and bivariate Spearman correlations for the TPB variables at Time 2 (n=138). ............................. 168

Table 9.7. Dichotomised DVs' frequencies per group condition (n=138) ............... 169

Table 9.8. Logistic regression analysis predicting behaviour of not DUI during the three months after the intervention for participants at Time 2 (n=138) with demographic factors as predictors. ..................................... 173

Table 9.9. Logistic regression analysis, predicting Behaviour of not DUI during the three months after intervention for participants at Time 2, with demographic factors and TPB from Time 1 as predictors (n=138). ..................................................................................................... 173

Table 9.10. Logistic regression analysis, predicting Behaviour of not DUI during the three months after intervention for participants at Time 2, with demographic factors, TPB and additional predictors from Time 1 as predictors (n=138). ............................................................................... 174

Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers xvii

Table 10.1. Interaction effects between Condition and personality characteristics, intention not to speed, adjusted for Time 1 values (n=144). ..................................................................................................... 246



10.4. Interaction effects between Condition and personality characteristics, intention not to speed, adjusted for Time 1 values (n=210). ..................... 249



Table 10.7. Effect of the intervention on TPB variables, adjusted for Time 1 values (n=62). ............................................................................................ 252

xviii Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers

Glossary of terms

The following Table 1.1 provides definitions of key terms that are used in this

document.

Table 1.1. Definitions of key terms

Consumer-

oriented

Relating to the needs and interests of individual consumers,

rather than businesses (Cambridge Business English Dictionary).

Distraction Anything that diverts the driver’s attention away from the

primary tasks of navigating the vehicle and responding to critical

events (NHTSA, 2010).

Driving under the

influence of

alcohol

Illegal behaviour in which the driver operates a vehicle with

blood alcohol concentration (BAC) level above the limits set by

the law.

Driving under the

influence of drugs

Illegal behaviour in which the driver operates a vehicle after

consuming illegal drugs.

Gamification Using game design in a non-game context in an attempt to

boost drivers' motivation and commitment to use a new in-

vehicle system (Diewald, Möller, Roalter, Stockinger, & Kranz,

2013).

Immersive (for

VR)

Keeps the immersion alive and engaging, rather than being

merely impressive, by enabling interaction with nearly

everything in the virtual world. It also offers good content or

gameplay that’s independent of technology, making VR

interactions core to the experience, and easing the user quickly

and smoothly into the virtual world (Intel)1.

Incentives A thing that motivates or encourages someone to do something

(Oxford Dictionary of English).

Leaderboard A leaderboard is a rank list of the people involved. Its purpose is

to show them where they rank in a gamified system. Those at the

1 https://software.intel.com/en-us/articles/guidelines-for-immersive-virtual-reality-experiences - Accessed on 04/03/2019

Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers xix

top enjoy the fame of being seen by all. To the rest, the

leaderboard shows where they stand relative to their peers

(Duggan & Shoup, 2013).

Persuade To make someone do or believe something by giving them a

good reason to do it or by talking to that person and making

them believe it (Cambridge Dictionary of English).

Realistic (for VR) Makes the virtual world seem real by providing smooth 3D

video, realistic sound, intuitive controls for manipulating the

environment, and natural responses to the user’s actions in the

virtual world (Intel)1.

Risky Driving

Behaviour (in the

case of young

drivers)

Any risky driving undertaken by the young driver which

increases the likelihood of the young driver being involved in a

car crash and may harm or fatally injure the young driver

themselves, their passenger(s), and other road users such as

pedestrians, cyclists, drivers and passengers in other vehicles

(Scott-Parker, 2012).

Simulation An imitation over time of real-world processes or system

operations (Banks, 1998).

Speeding Illegal behaviour in which the driver operates a vehicle at a

speed which is above the set speed limit for the respective

section of the road.

Virtual reality or

VR

A medium composed of interactive computer simulations that

sense the participant's position and actions and replace or

augment the feedback to one or more senses, giving the feeling

of being mentally immersed or present in the simulation (a

virtual world) (Sherman & Craig, 2018).

xx Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers

List of abbreviations

ADAS Advanced Driver Assistance Systems

AIHW Australian Institute of Health and Welfare

AONSW Audit Office of New South Wales

ATC Australian Transport Council

BAC Blood alcohol concentration

BIS-11 Barratt Impulsiveness Scale Version 11

BITRE Bureau of Infrastructure Transport and Regional Economics

CDCP Centers for Disease Control and Prevention

COT Consumer-oriented technologies

DUI Driving under the influence of alcohol or drugs

EC European Commission

GDL Graduated Driver Licensing

GPS Global Positioning System

HBM Health Belief Model

ICT Information and Communication Technology

ITF International Transport Forum

NHTSA National Highway Traffic Safety Administration

OECD Organisation for Economic Co-operation and Development

PBC Perceived Behavioural Control

PRISMA Preferred Reporting Items for Systematic Reviews and Meta-Analyses

SET Self-Efficacy Theory

SCT Social Cognitive Theory

TPB Theory of Planned Behaviour

TRA Theory of Reasoned Action

Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers xxi

TTM Transtheoretical model of health behavior change

VR Virtual reality

SPSRQ Sensitivity to Punishment and Sensitivity to Reward Questionnaire

WHO World Health Organisation

xxii Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers

Funding and awards

The program of research constituted a PhD project with a focus on young

drivers, funded by the Australian Government Department of Education through a

2015 Endeavour Postgraduate Scholarship. The project sought to build on Daniel

Vankov's already established prior experience as a road safety social entrepreneur in

relation to the implementation of road safety education programs for young drivers (18

to 25 years). These programs had incorporated COT tools (e.g. mobile driving

simulators and VR) and communication strategies to reach out to young drivers in

Europe, Asia and Latin America.

The Queensland University of Technology awarded Daniel Vankov with a

2018 Student Leadership Award for his work and contributions.

Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers xxiii

Statement of original authorship

The work contained in this thesis has not been previously submitted to meet

requirements for an award at this or any other higher education institution. To the best

of my knowledge and belief, the thesis contains no material previously published or

written by another person except where due reference is made.

Signature:

Date:

QUT Verified Signature

xxiv Smartphone Apps and Virtual Reality as Road Safety Interventions: Examining their Real-World Effects for Young Drivers

Acknowledgement

I wish to thank my supervisory team. Without them, this thesis would not have

happened. To my principal supervisor, Dr Ronald Schroeter – thank you for your

patience, support and guidance, which ultimately made it possible to complete this

program of research. To my associate supervisor, Professor Andry Rakotonirainy –

thank you for making time in your busy schedule to review. To my second associate

supervisor, Dr Melanie White - thank you for introducing me to the depths of

psychology, a field I did not know so much about before embarking on my PhD

journey. It was both a privilege and an honour to work with all of you. I am grateful

for all I have learned as a consequence of that opportunity.

I would also like to express my special gratitude to Professor Divera Twisk. Your

support came when I most needed it. This support made it possible for me to

understand the full value of the quality work we were doing.

I wish to thank the Australian Government Department of Education for

awarding me a 2015 Endeavour Postgraduate Scholarship. This scholarship enabled

me to embark on my four-year-long research journey with the Centre for Accident

Research and Road Safety – Queensland (CARRS-Q), Queensland University of

Technology. I cannot imagine a more supportive and welcoming environment for both

my research project and my extracurricular activities. Thus, I would like to thank

explicitly the CARRS-Q Directors, Professor Narelle Haworth and Professor Andry

Rakotonirainy, who supported my ideas for open-access participants' recruitment,

open public interventions, vocational visits, invited talks, and fundraisers.

"The hardest part of any journey is taking the first step". Thank you for helping

me take mine!

Chapter 1: Introduction 1

Chapter 1: Introduction

Global efforts to reduce deaths and injuries caused by road crashes are starting

to wear off. The Global Status Report on Road Safety (WHO, 2018) shows that despite

all the efforts of the international community to address the problem, road fatalities

continue to climb (see Figure 1.1). 1.35 million people lost their lives on the roads in

2016 (WHO, 2018). In different parts of the world, initiatives, such as the Decade of

Action for Road Safety 2011-2020, bring different results with the latest statistics

calling for novel complementary approaches in the prevention efforts. A report,

exploring road safety from a slightly different perspective, suggests that two-thirds of

the reduction in road fatalities in the years between 2008 and 2014 were not due to

targeted efforts but were a result of the 2008 Global Financial Crisis (ITF, 2015). The

same report points out that unemployment and the accompanying less affordable

travel, especially when it comes to young drivers, as well as generally fewer kilometres

driven, were amongst the contributors for improved road safety within this period. The

established causality suggests an explanation of the recent upward trend in road

fatalities, i.e. global economic recovery, may have reversed the positive trends of

trauma reduction in road safety. Statistics from developed economies around the world

support the suggestion.

Figure 1.1. Number and rate of road traffic death per 100,000 population: 2000–2016 (WHO, 2018)

The U.S. National Highway Traffic Safety Administration (NHTSA) recorded

8.4% increase of fatalities in 2015 in comparison with 2014, the highest rate on U.S.

2 Chapter 1: Introduction

roads since 1964, followed by a 5.4% increase in 2016 (NHTSA, 2017) before a

reduction of 1.8% in 2017 (NHTSA, 2018b). The population was affected in all

segments (male/female, day/night, drivers/pedestrians).

The European Union (EU), which has been one of the best examples of achieving

results with targeted actions, also shows mixed results with a widening gap between

actual and desired progress towards targets (Adminaite, Calinescu, Jost, Stipdonk, &

Ward, 2018). Figure 1.2 visualises the recorded reduction in road fatalities between

2010 and 2017 (light blue line) and the desired trend projected target line from 2010

onwards (deep blue line). The European Transport Safety Council recorded an increase

of fatalities in 2014 and 2015 before reverses in 2016 and 2017 (Adminaite et al.,

2018). Thus, the achieved average annual reduction in road deaths since 2010 in EU

equals 3.1% while a 6.7% average was needed to reach the EU 2020 targets (Adminaite

et al., 2018).

Figure 1.2. Widening gap between the actual and desired progress towards the EU 2020 target (Adminaite et al., 2018)

Australia is globally regarded as a high-achiever in road safety, but recent

statistics reveal negative trends. The number of fatalities increased by 5% from 1,150

in 2014 to 1205 in 2015 and by further 7.3% to 1293 in 2016 before a reduction of

5.2% to 1226 was recorded in 2017 (BITRE, 2018). The figures are a serious concern,

given that Australia was consistently ahead of its fatalities' reduction targets. The

achievements may further deteriorate if the economy continues to do well. Figure 1.3


illustrates the Australian fatalities trends since 2008. The solid blue line shows the

targeted number of fatalities; the dotted grey line shows the current trend.

Figure 1.3. Australian progress until 2017 towards fatality target (BITRE, 2018)

1.1 RESEARCH PROBLEM

The target population of this PhD program of research is young drivers, aged 18

to 25 with a valid driver's license, who drive a car. Research findings suggest that

young drivers have a 2 to 3 times higher crash risk than experienced drivers (SafetyNet,

2009). However, such an estimate does not reveal the gravity of the full picture. Young

drivers drive much less than their experienced colleagues. If miles driven is taken as a

basis, their risk factor to be involved in crashes is estimated at ten times higher than

the risk factor of experienced drivers (McKnight & McKnight, 2003).

Statistics across jurisdictions evidences overrepresentation of young drivers in

road crashes. In Australia, the 17-25 age group accounts for 19% of the fatalities with

the rate remaining above the national average (BITRE, 2018). In the US, in 2016, the

15 to 20-year-old age group represented only 5.4% of all licensed drivers but was

involved in 9% of the fatal crashes (NHTSA, 2018a). In 2017, in the EU, the 18 to 24-

year-old represented 8% of the population but 14% of the road fatalities (EC, 2018).

In the 30 member states of the Organisation for Economic Co-operation and

Development (OECD), the proportion of the young drivers under the age of 25 in the

population is 10.1% while in the fatalities it is 26.7% (OECD, 2006). Not surprisingly


globally young people, aged 15 to 24 years, are overrepresented in road crashes with

traffic injury being the leading cause of death in the 5-29 age group (WHO, 2018).

There are numerous reasons for young drivers to be exposed to increased crash

risk. Major reasons are poor control of the vehicle (Patten et al., 2006), inability to

identify hazards on the road (Pollatsek, Fisher, & Pradhan, 2006) and sensation

seeking (Hatfield, Fernandes, & Job, 2014; Schroeter, Oxtoby, & Johnson, 2014;

Scott-Parker, Watson, King, & Hyde, 2013). Young drivers are also more likely to

crash as a result of distraction (McEvoy, Stevenson, & Woodward, 2006). Helping

them to understand the consequences of reckless behaviour as well as the impact on

safety of each decision behind the wheel is a serious multidisciplinary challenge to

which COTs may be an answer.

Young drivers are early technology adopters (Lee, 2007). Current research

explores COTs in road safety aiming not only to transform driving into a constantly

engaging and fun activity but also to cultivate constructive driving behaviour

(Schroeter, Rakotonirainy, & Foth, 2012; Steinberger, Schroeter, Foth, & Johnson,

2017). Some COT applications, such as Fiat eco:drive or Nokia's "Routine Driving"

and "Driving Miss Daisy", have attracted attention and have already been reviewed

(Bellotti, Berta, & De Gloria, 2014). Such COT applications were made available to

the general public and explored serious gaming concepts, i.e. they were designed with

a primary goal different than mere entertainment. However, there is evidence showing

that the positive effects of using them are not guaranteed and can be compromised by

undesired driving behaviour (Ecker, Holzer, Broy, & Butz, 2011). COTs are not

necessarily designed from a road safety perspective. COTs may be responding to

consumers' (or individual drivers') wants or needs, unrelated to the primary driving

task.

While COTs are often seen as a potential threat in road safety research, they offer

the unique characteristic of being able to provide more acceptable suggestions from

the perspective of young drivers (Lee, 2007). Therefore, COTs (see Section 2.4)

represent opportunities, which the current research leveraged, to positively influence

young drivers. An extended TPB (Ajzen, 1988) was operationalised to evaluate the

effects of these opportunities (see Section 3.7).

Leveraging evidence-based opportunities to deliver safety benefits for young

drivers, such as the ones the current program of research evaluated, may be


mainstreamed through the young drivers’ ecosystem. The young drivers’ behaviour

and safety on the road are influenced by their ecosystem. This ecosystem is complex

and involves a multitude of stakeholders (Scott-Parker, Goode, & Salmon, 2015).

These stakeholders have different types of relationships with the young drivers, e.g.

authoritarian (government, police), commercial (car manufacturers, insurance

companies, driving instructors), personal (parents, peers) or not-for-profit (design

researchers, not-for-profit organisations (NFPs)). Tackling major social problems,

such as young drivers' road trauma, requires joint efforts. Yet, in practice, they are

typically assigned to governments or NFPs (Porter & Kramer, 2011).

NFPs are unique in that they are managed by visionary social entrepreneurs

(SEs). The commercial side of their endeavours is a necessary means to deliver good

(Kotler, Hessekiel, & Lee, 2012). SEs are believed to be able to mainstream research

findings into practice. Their contributions have been researched in multiple fields such

as civic engagement (Levinson, 2012), social transformation (Mair & Noboa, 2006),

sustainable development (Seelos & Mair, 2004) as well as in health promotion

(Catford, 1998). SEs are often carriers of social innovation (Lettice & Parekh, 2010),

having unique skills to a) take a new perspective on the problems, b) create new

ecosystems, and c) appeal to the customer base.

In the context of this PhD thesis, it should be noted that the author has established

considerable background experience as a SE in relation to the implementation of road

safety education programs for young drivers. These skills contributed towards this PhD

project to be supported by a 2015 Endeavour Postgraduate Scholarship (see Funding

and awards). As such, the author's personal drive to contribute towards reducing road

trauma amongst young drivers by using NFPs as a vehicle forms the backdrop of this

thesis.

1.2 RESEARCH AIM AND OBJECTIVES

The PhD aimed to examine the effects of two COT-based interventions to reduce

risky driving behaviour among young drivers. The first one transformed an existing

risk source (smartphones) into one for motivating safe driving behaviour. The second

one deployed VR for the same purpose. Although it is unlikely that any chosen

example of COTs would represent their full and true potential, it would have been out

of scope for the current PhD program of research to deploy the designed evaluation


framework (see Section 3.7) to investigate several different smartphone safe-driving

apps and VR simulations of risky driving. Therefore findings are reported about the

two used COTs only, and shall not be generalised to COTs in general.

The evaluations of both interventions were grounded in TPB. As such, both of

them assessed the impact of novel technology-based practice-oriented approaches to

raise awareness on driving-related risks outside the laboratory environment, i.e. in the

real world, without incurring substantial development and deployment costs as part of

the studies themselves. The following three key objectives were pursued as part of the

research investigation:

1. Understand to what extent the use of smartphone safe-driving apps and VR

simulations of risky driving are associated with safety benefits for young

drivers in the literature;

2. Identify smartphone safe-driving apps and VR simulations of risky driving

that could potentially persuade young drivers to adopt safer on-road

behaviour as a result of road safety interventions; and

3. Investigate the extent to which the use of an example safe-driving app and a

VR simulation of risky driving was associated with positive changes in

participating young drivers' self-reported intentions and self-reported

behaviour on the road over three months.

1.3 SIGNIFICANCE OF THE RESEARCH

International efforts continue to focus on road crash prevention, but their

potential to sustain the positive reduction trends is undermined by the global economic

recovery (ITF, 2015). Existing strategies have their place, but new ones are needed to

support further international efforts to reduce road trauma. In-vehicle COTs, and

particularly the smartphone, are very often regarded as a major cause for crashes,

especially in relation to young drivers (WHO, 2011). At the same time, a number of

road safety stakeholders embrace their positive potential in an effort to reduce existing

risks and help young drivers improve their behaviour. Outside the vehicles, VR

applications attempt to create environments close to the real world while eliminating

the negative consequences of reckless decisions (in the case of road safety). As such,

road safety stakeholders have started to embrace this opportunity towards raising


awareness through VR simulations of risky driving (Lang, Liang, Xu, Zhao, & Yu,

2018). However, the safety benefits of such initiatives have not been researched.

The significance of the current research is that it followed a practice-oriented

methodology that used psychological and social constructs applied to COT

applications that aim to motivate safer driving behaviour to reduce road trauma

amongst young drivers in the real world. The PhD program of research used two

currently available COT applications, a smartphone safe-driving app and VR

simulations of risky driving, to facilitate the acquisition of safer driving practices and

the cessation of risky driving behaviours amongst the target group, focusing on:

- Operationalising of two COT-based interventions which did not appear to

increase risks, such as distraction, for the participating drivers;

- Conducting studies in the participants' free-living environment;

- Quantifying the changes in driving intention and behaviour through the

evaluation of self-reported data.

1.4 OUTCOMES

The research delivered new knowledge and insights about the effects on young

drivers' intention and behaviour from the rapidly expanding use of 1) smartphone safe-

driving apps and 2) VR simulations of risky driving.

The outcomes of this research are twofold:

1. Contributing to a better understanding of the real safety benefits from using

two examples of COT applications, a smartphone safe-driving app (see

Chapter 7) and VR simulations of risky driving (see Chapter 9), for risk

prevention purposes in a road safety context; and

2. Informing the future evaluation of interventions, supported by COTs and

more specifically by smartphone safe-driving apps and VR simulations of

risky driving, that aim to reduce risky driving behaviours in young drivers

(see Subsection 10.3).

In the broader context of the current program of research, the author of the

current thesis believes that those research outcomes stand a higher chance to be later

operationalised in the real world through the involvement of SEs. Given SEs’

experience in ameliorating a diverse set of problems, they were consulted along with


road safety researchers as an expert reference group within the framework of the

current program of research (see Chapter 6). Thus, the current project pursued not

simply innovation but rather innovation that might possibly be applied in the real world

on a larger scale using SEs as a vehicle. This innovation was pursued in regards to

addressing:

1. A knowledge gap by investigating what might be the actual safety benefits

in the real world of the growing quantity of COTs, available to drivers, by

assessing the impact of two COT examples, a safe-driving app and a VR

simulation of risky driving; and

2. A number of limitations, identified in the literature, such as examining self-

reported scores over time to inform more robust conclusions.

1.5 DOCUMENT OUTLINE

Following this Introduction in Chapter 1, a Literature review is presented in

Chapter 2. It reviews and discusses the young drivers' problem and what COTs may

influence drivers’ behaviour. Applicable theories are reviewed in detail in Chapter 3.

Chapter 3 also focuses on how the TPB was operationalised in the Research design,

details on which are given as Chapter 4. Chapter 5 reports on Study 1, systematic

review, and explores in depth the application and utility of smartphone apps in road

safety research with a focus on young drivers. Chapter 6 establishes criteria for

selecting a safe-driving app as an intervention tool and searches and evaluates the

available apps on the app stores to inform the final selection. Chapter 7 (Study 2)

evaluates an operationalised intervention with an off-the-shelf smartphone safe-

driving app, which was selected following a systematic selection process that is

outlined in the previous Chapter 6. It examines the intervention impact on the young

drivers’ self-reported intention not to speed and their behaviour of not speeding during

the 3-month intervention period in their free-living environment. Chapter 8 (Study 3,

systematic review) investigates the actual and potential application and utility of VR

simulations of risky driving in road safety research and practice, with a focus on young

drivers. Chapter 9 reports on Study 4, which evaluates an operationalised intervention

with a VR simulation of risky driving in which participants were driving a virtual car,

simulating the DUI of their choice (alcohol or drugs). It provides information on the

effect of the intervention on the young drivers’ self-reported intention not to DUI and

behaviour of not DUI in the three months after the intervention. Chapter 10 concludes

the document with an overall discussion.

Chapter 2: Literature review 9

Chapter 2: Literature review

Chapter 2 reviews the young drivers' problem in the road safety domain. It begins

with exploring the fatal five risky behaviours that are associated with the

overrepresentation of young people in road crashes. It continues with a review of

contributing factors to those crashes that were considered when designing the current

program of research. An examination of interventions deployed as countermeasures

follows. The Chapter continues to describe evidence in the literature on addressing

young drivers' risky road behaviour through different types and implementations of

COT-based road safety interventions. COTs use in road safety is discussed, resulting

in an overview of the additional risks they introduce to the drivers. The chapter

concludes with the main considerations, which this PhD program of research

subsequently focuses on and the articulation of the Research Gap.

2.1 YOUNG DRIVERS' RISKY DRIVING BEHAVIOURS

The global problem of young drivers’ safety remains a focus of risk prevention

interventions due to unsatisfactory results of past implementations (Scott-Parker,

King, & Watson, 2015). Traffic crashes do not only lead to young people being

overrepresented in road fatalities (BITRE, 2018), but they are also the leading external

cause for hospitalisation due to injury (AIHW, 2008). Several risky behaviours,

classified as the fatal five (speeding, DUI, not wearing a seatbelt, fatigue and

distraction), are often reported by young drivers (Scott-Parker & Oviedo-Trespalacios,

2017).

Speeding is ruled to be the cause of 43% of young drivers' fatal crashes in

comparison to 23% for older drivers involved in fatal crashes (AONSW, 2011).

Almost 50% of young people admit speeding at least once by 10 to 25 km/h during

their last ten trips (D. Smart et al., 2005). Despite the consequences of speeding being

widely-known, young drivers are very likely to engage in such behaviour, without

being ashamed by the fact, and are more likely to report it (Fleiter, Watson, Lennon,

& Lewis, 2006; Horvath, Lewis, & Watson, 2012).

DUI impairs driving performance and, as a result, increases the risk of crashes

(Hingson, Heeren, Levenson, Jamanka, & Voas, 2002). The risk increases when the

10 Chapter 2: Literature review

driver is young (Peck, Gebers, Voas, & Romano, 2008), which is why Graduated

Driver Licensing (GDL) systems adopt zero alcohol tolerance (see Subsection 2.3.1).

Still, a recent US survey shows staggering statistics in youth drink driving. In the 30

days before the survey, 7.8% of the young driver respondents had driven after drinking

alcohol, while 20% rode with a driver who had been drinking (CDCP, 2016). In

Australia, drink-driving is identified as a primary contributor to 30% of the fatal and

9% of the non-fatal injuries for all drivers (ATC, 2011). Drug-driving is a main

behavioural factor in 7% of the fatal and 2% of the non-fatal crashes for all drivers in

Australia (ATC, 2011) with younger drivers (18–29 years) reporting higher than

expected engagement in such behaviour (Ward, Schell, Kelley-Baker, Otto, & Finley,

2018). ATC (2011) does not provide the respective crash statistics stratified by age

groups. While numbers by age groups in the case of DUI cannot be found in reports

such as the one of ATC (2011), there is evidence that DUI-related risks increase five

times for young drivers, aged under 21, in comparison with older drivers (Peck et al.,

2008).

Fatigue is a primary contributor to 6% of all crashes and 15% of the fatal crashes

(Legislative Assembly of Queensland: Parliamentary Travelsafe Committee, 2005).

Yet, 80% of drivers report driving while fatigued (Obst, Armstrong, Smith, & Banks,

2011). Research identifies fatigue as the psychological state to most commonly impair

young drivers (Wundersitz, 2012). Similar to other risky behaviours on the road, young

people are more likely to engage in driving when fatigued than their more experienced

colleagues (McGwin Jr & Brown, 1999; Rhodes & Pivik, 2011).

Not wearing a seatbelt is a contributing behavioural factor in 20% of the fatal

and 4% of the non-fatal crashes for all Australian drivers (ATC, 2011). There is

evidence that the use of seatbelts reduces the number of fatalities on the road (Dinh-

Zarr et al., 2001). In recent years seatbelt use by young drivers reached up to 85% and

is comparable with the trends in the general population (Pickrell & Liu, 2015).

Nevertheless, restraint non-use during short trips is still reported by 20% of young

Australian drivers (Scott-Parker & Oviedo-Trespalacios, 2017), which may further

contribute to their overrepresentation in fatal crashes.

Last, but not least, distraction is the most commonly observed risky behaviour

in young people with 41.5% admitting sending an SMS or an e-mail at least once in

the past month (CDCP, 2016). Currently, in the US, 10% of fatal crashes and 18% of


injury crashes are reported as a result of distraction (NHTSA, 2015). More than 50%

of young drivers are identified as distraction-prone (Schroeder, Meyers, & Kostyniuk,

2013). As it was the case with the other fatal four behaviours on the road, as a result

of being distracted young drivers are more vulnerable while driving than their older

colleagues (McEvoy et al., 2006). The problem is likely to deepen as a result of the

increase in the number of electronic devices in the cars (Parliament of Victoria Road

Safety Committee, 2006; WHO, 2011), a forecast largely confirmed by both reviews

of the literature (Oviedo-Trespalacios, Haque, King, & Washington, 2016) and crash

statistics (NHTSA, 2017).

The reviewed literature highlighted speeding and DUI as the deadliest of the fatal

five risky behaviours on the road with young drivers being at a higher risk of crash

involvement as a result of both of them. As a consequence, the potential safety benefits

of reducing speeding and DUI amongst young drivers can be perceived as

comparatively higher than the potential safety benefits, related to the other three risky

behaviours. This makes speeding and DUI reduction a suitable target of road safety

efforts. It is worth acknowledging that examples in the literature can be found with

speeding defined as "driving at an illegal speed over the limit" or "driving at an

inappropriate speed" or both. Not all sources define their understanding of "speeding"

and make the distinction explicit. For the current PhD program of research, speeding

is defined as "driving at an illegal speed over the legal limit".

2.2 CONTRIBUTING FACTORS

Young drivers’ crash involvement is influenced by a multitude of characteristics

(road, vehicle, personality, etc.) as well as broader social factors (family, friends)

(Scott-Parker, 2012). Variables related to those factors and characteristics are likely to

interact and influence each other at any given point of time, ultimately influencing the

young people’s behaviour on the road. Following is a review of the literature exploring

some of those factors that were considered relevant for the current PhD program of

research.

2.2.1 Driving experience

Driving experience, or more likely the lack of it, can be considered an objective

constraint for young drivers to fully understand the potential implications of their

behaviour on the road. Limited driving experience results in a diminished capability


to both recognise and respond to road hazards (Deery, 2000; Scialfa et al., 2011). In

contrast to their own beliefs of being better than experienced drivers (Gosselin,

Gagnon, Stinchcombe, & Joanisse, 2010), there is evidence that young drivers

recognise fewer risks than their experienced colleagues (Fisher, Pradhan, Pollatsek, &

Knodler Jr, 2007). Failure to comprehensively understand the situation on the road in

any given moment may put young drivers at a disadvantage when it comes to

anticipating risks and adequately responding to them when they materialise.

Anticipating and responding to risks is likely to improve with driving practice.

So more practice is recommended for the risk of crash involvement to be reduced, but

the more young drivers drive, the higher their crash risk (Prato, Toledo, Lotan, &

Taubman - Ben-Ari, 2010). GDL programs are designed as a response. They require

experienced drivers to monitor novice drivers, thus reducing the risks of exposure

while the much-needed practice is being acquired. However, such supervision is

usually limited during the first year of driver licensing. Extending it may not be

practical or possible. For example, when behaviour is illegal and by definition should

not take place on the road, extended supervision might not have the chance to tackle

them due to their rare occurrence. Nevertheless, such behaviours contribute to causing

crashes and acquiring experience in relation to them in a safe (laboratory) environment

may save lives in the real world.

The present PhD program of research made an effort to compensate for the

young drivers' inexperience by providing them with an opportunity to gain experience

in two separate interventions:

1) In a rational and understandable way, a smartphone safe-driving app, able to

interpret driving events and communicate feedback, simulated constant supervision in

the young drivers' free-living environment while they were operating their vehicles in

their daily routine. The feedback was coupled with impartial and unemotional advice

on how safety could have been improved.

2) In safe experimental conditions, a VR simulation of risky driving allowed

young drivers to let their emotions off-leash and experience the dangers of DUI. The

participants were able to see how their driving abilities were affected as a result of the

simulated intoxication in a safe environment.


2.2.2 Optimism bias

Lack of driving experience may not be a standalone contributing factor to young

drivers’ increased driving risks, particularly in critical situations, although it is

associated with increased risk of crash involvement (McCartt, Mayhew, Braitman,

Ferguson, & Simpson, 2009). "This will not happen to me." often flashes through

people's mind when they are exposed to negative information. This kind of optimism

bias makes people believe that negative things are more likely to happen to others than

to themselves (Weinstein, 1980). Young drivers are prone to optimism bias related to

their driving (Gosselin et al., 2010; Harré, Foster, & O'Neill, 2005; Horswill, Waylen,

& Tofield, 2004; White, Cunningham, & Titchener, 2011). Researchers report that

young drivers see themselves as being better drivers than their peers and, thus, they

think they are less likely to be involved in crashes (Harré et al., 2005; Horswill et al.,

2004; White et al., 2011). Other research reports that young drivers have the same self-

perception when compared with more experienced drivers, too (Gosselin et al., 2010).

This, as a consequence, may lead to increased risk-taking and subsequent crash

involvement.

Young drivers' overestimation of their driving skills, combined with their

tendency to underestimate potential risks, increases their overall crash risk. Revealing

an accurate picture of one’s driving performance may help address young drivers’

optimism bias and subsequently improve their driving behaviour. The current PhD

program of research explored that assumption through COTs interventions that

provided immediate personalised feedback to the involved participants.

2.2.3 Gender

Males are consistently overrepresented in crash fatalities in comparison to

females (BITRE, 2018). Young males are more likely to engage in risky behaviour on

the road (Rhodes & Pivik, 2011) which is why the gender effect is often assessed as

part of road safety studies, e.g. investigating speeding (Fernandes, Hatfield, & Soames

Job, 2010; Horvath et al., 2012) or DUI (Fernandes et al., 2010; Peck et al., 2008).

Thus, gender differences are at least controlled for (Horvath et al., 2012; Parr et al.,

2016) or investigated in detail (Obst et al., 2011; Rhodes & Pivik, 2011).

There is evidence that males and females are impacted differently by road safety

interventions (Lewis, Watson, White, & Elliott, 2013). In general, males are more


tolerant than females to take risks while driving (Redshaw, 2006). Males also report

violating traffic regulations more often than females (Castellà & Pérez, 2004;

Constantinou, Panayiotou, Konstantinou, Loutsiou-Ladd, & Kapardis, 2011). Lewis et

al. (2013) found that it was important for male drivers to feel in control while, at the

same time, they perceived to have little control over their speeding. Lewis et al. (2013)

suggested that mass media message content should be developed with young males as

a primary target as speeding-related beliefs were particularly relevant for them.

Nevertheless, evidence suggests that such campaigns had less effect on male than on

female drivers (Lewis, Watson, & White, 2010; Lewis, Watson, & Tay, 2007; Lewis,

Watson, Tay, & White, 2007).

In the current program of research, gender, together with driving experience and

age, was controlled for as a demographic characteristic in the regression models to

allow for a more accurate assessment on the predictive contribution made by

theoretical constructs in regards to the dependent variables (DVs) of interest. When

allowed by the specific statistical test and by the respective sample size, gender was

also investigated as a moderator, which allowed for a more in-depth exploration of the

intervention effects.

2.3 INTERVENTIONS

The literature reviewed above provides insights and evidence on why young

drivers may be taking unnecessary risks on the roads and what are the potential

consequences for them. Another set of knowledge can be utilised in road safety

intervention design, implementation and evaluation that originates in previous

prevention efforts. Road safety stakeholders address young drivers’ vulnerability on

the road at multiple levels through a variety of interventions and initiatives, some of

which are further discussed in turn.

2.3.1 Driver training

Driver training aims to provide young people with the necessary minimum

experience for them to be able to be licenced as drivers. It is regulated by the

government and delivered to learner drivers by more experienced drivers, driving

instructors and, to a lesser extent, by road-safety-related businesses. Countries, such

as Australia and the USA, have introduced GDL systems, which vary in structure and

complexity. It should be noted that the current program of research involved


participants, who were licenced drivers under the Australian GDL system (Walker,

2014). In Australia, novice drivers start with a learner's (L) license. After passing a

learner driver’s test, drivers are issued with a provisional driver’s licence, initially a

provisional one (P1) license, followed by a provisional two (P2) license. Finally, after

holding a provisional licence for between two to four years, depending on the state or

territory, novice drivers are eligible for an open driver’s licence. The process of going

through the different license stages is characterised with decreasing limitations. For

example, learner drivers can drive only under the supervision of an openly-licensed

driver. This requirement ceases when the driver obtains a provisional license.

However, at that time, a restriction on driving with multiple passengers starts applying.

The most relevant license limitations for the current research are the ban on using

mobile phones and the zero BAC requirement during the entire provisional period

(Walker, 2014), which are also recommended by the Australian GDL policy

framework.

In an effort to improve driver training there has been a trend towards more

supervision hours and delayed licensing, as well as more diverse supervision,

involving as many stakeholders (e.g. parents and schools) as much as possible

(Senserrick, 2007). As a result, learner drivers under supervision are the safest in the

world, but this fact changes after their provisional license is granted (see Figure 2.1)

(Mayhew, Simpson, & Pak, 2003).

Figure 2.1. Percentage of South Australia drivers involved in a crash five years after licensing (Austroads, 2008).


The tenets that better safety on the road is achieved by more intensive driver

training is widely accepted (Watson, 1997). However, the literature does not suggest

that the conventional driver training necessarily leads to safer driving habits (Vernick

et al., 1999; Watson, 1997). Neither Vernick et al. (1999) nor Watson (1997) provides

an explicit answer as to “why” increased driver training does not result in a reduced

crash rate. Both studies found no evidence across jurisdictions in support of instilling

more driver training upon young drivers. However, neither of them challenges the fact

that driver training does improve driving skills. The problem is that this improvement

is not translated into improved safety levels. A more recent review of the literature

confirms that training improves skills, but it also confirms that evidence for crash

reduction, as a result, remains questionable (Beanland, Goode, Salmon, & Lenné,

2013).

2.3.2 Media campaigns

While it remains ambiguous that drivers’ training improves road safety, other

educational tools are put to work with that purpose. Fear-based media campaigns, such

as, for example, the Northern Ireland Department of Education Road Safety Anti Drink

Driving Ad2, try to reveal the full scope of possible consequences of reckless driving

in order to change driver attitudes and in turn facilitate safer driving practices. Such

campaigns are usually initiated by governments or road-safety-related businesses. At

the same time, a growing body of evidence suggests that they have little effect on the

target group, especially on the riskier male drivers (Lewis, Watson, & White, 2010;

Lewis, Watson, & Tay, 2007; Lewis, Watson, Tay, & White, 2007).

Other studies provide mixed results. For example, Tay (2005) provides evidence

of the anti-speeding media campaigns in Victoria independent effect on male drivers'

crash rates. However, merely visualising risks may not be sufficient to change

behaviour (Shope & Bingham, 2008). As discussed earlier (see Subsection 2.2.2)

young people are susceptible to optimism bias. They underestimate the likelihood of

being involved in a crash as a consequence of the risks they are taking, compared to

the likelihood of it happening to others (Gosselin et al., 2010; Harré et al., 2005;

Horswill et al., 2004; White et al., 2011). Thus, the overall unsatisfactory results of

2 https://youtu.be/0x4Qrjyf4lQ


media campaigns may be rooted in the inherently optimistic way the young people

think.

2.3.3 Law enforcement

Other measures take place in cases where mass education produces

unsatisfactory results. Law enforcement is the major government tool, channelled

through the Police, to tackle unlawful driving behaviour. It is believed to have

complementary effects with media campaigns (Tay, 2005). The author provides

evidence of its independent effect on drink-driving crashes and interactive effect with

the respective media campaign on speeding-related crashes. Its effectiveness builds on

people’s perceptions about the risk of apprehension in combination with the

probability of sanctions (Tay, 2005). Tay (2005) sees its success as a function of the

police presence level and the hit rate of the respective intervention.

Law enforcement increasingly utilises innovative technologies to improve

success. For example, alcohol ignition interlocks do not allow the engine to start if the

driver is above the legal blood alcohol concentration (BAC) limit. There is evidence

that alcohol ignition interlocks devices are feasible in both commercial and public

settings (Silverans, Alvarez, Assum, Evers, & Mathijssen, 2007). They are an example

of a context-based personalised enforcement when the right of decision, i.e. starting

the vehicle, is taken away from the driver and delegated to the technology to avoid the

possibility for an unfavourable outcome, i.e. crash. Other technologies systematically

provide similar decision support to drivers in the context of other detected potential

risks such as distraction or drowsiness (Kashevnik & Lashkov, 2018), which may

indirectly support law enforcement.

2.4 CONSUMER-ORIENTED TECHNOLOGIES

The use of technologies in the road safety domain is regarded as posing a mix of

potential benefits and threats to the drivers, in general, and young drivers, in particular.

There is evidence that using contemporary technologies in the car can make safe

driving an engaging, challenging and enjoyable task (Schroeter et al., 2012;

Steinberger et al., 2015). However, not all evidence provides support for their positive

impact on safety.

The presence of multiple technologies competing for the drivers' attention, such

as connecting smartphones, in-car video, audio and other electronic devices, is very


likely to deepen existing problems, such as distraction (Parliament of Victoria Road

Safety Committee, 2006). On the one hand, there is evidence that the use of technology

devices while driving distracts young drivers from their primary task: driving (Blanco,

Biever, Gallagher, & Dingus, 2006; Ferguson, 2003; Lee, 2007; WHO, 2011), which

may lead to tragic consequences. On the other hand, more and larger screens, in

combination with innovative interface concepts and possibly large-scale augmented

reality experiences, will be introduced in the cars to maximise car manufacturers'

revenue (McKinsey&Company, 2014). These are often guided by marketing and

consumer demands, rather than by safety considerations. The general sentiment is that

the problem is likely to both increase and evolve together with the number and

complexity of the available technologies (WHO, 2011). The way drivers use and

interact with COTs should be holistically examined, leaving no room for

underestimation of potential unintended safety counter-effects. Ignoring safety

considerations when COTs' design takes place may lead to unexpected consequences.

2.4.1 Context

COTs can provide additional experiences that may or may not be related to the

driving tasks and are not necessarily designed considering the safety of the drivers. For

example, they allow drivers and passengers to listen to music, watch movies, make or

receive calls, send and read text messages, browse the Internet, etc. In the literature,

COTs are often regarded as a source of distraction, which increases driving associated

risks (Parliament of Victoria Road Safety Committee, 2006). With the evolution of the

smartphone, currently, most COTs, including GPS (Global Positioning System), can

be found in a single device in the car (Regan, Williamson, Friswell, Hatfield, &

Grzebieta, 2012). This makes smartphones perhaps the most widely recognised

example of a COT that has infiltrated the car. However, this does not change the nature

of their adverse effects on driving (Rowden & Watson, 2013), and they continue to

cause major road safety concerns by distracting drivers. Klauer, Dingus, Neale,

Sudweeks, and Ramsey (2006) found that 65% of near-crashes and 78% of all crashes

are due to reduced drivers’ attention to their primary task.

COTs have been shown to drain important cognitive and operational capabilities

of the driver (Kircher, Ahlström, & Patten, 2011). A body of evidence in the literature

reveals a number of common behaviours observed in drivers when distracted by using

COTs. These include reduction of speed, lower control of the vehicle speed, often


changing position on the road, reduction of the ability to detect hazards, and an overall

decline in reaction time (Regan, Lee, & Young, 2009).

For example, looking away from the road, which is often a requirement to

operate a COT, reduces the drivers' abilities to brake in a case of an event or to keep

the vehicle in the lane (Lamble, Laakso, & Summala, 1999). The authors exposed 12

drivers, aged 19 to 27 years, to attention-demanding tasks in the car while approaching

a decelerating vehicle on the road. Haque and Washington (2013) provide further

evidence that young drivers' reaction time increased in general in the presence of a

distraction. In a simulator study in three conditions of phone conversations (with a

handheld device, with hands-free and no conversation), Haque and Washington (2013)

recorded slower reactions when a hands-free device was used, and there was no

looking away from the road involved to operate it. Recarte and Nunes (2000) provide

clues about why that might be happening. The authors found changed scanning

patterns in mobile phones users while driving. By observing eye movements while

participants performed verbal and scanning tasks, Recarte and Nunes (2000) recorded

less attention paid to the mirrors, the dashboard or the road, which are vital to detecting

and responding to hazards. Using a phone while driving introduces a second cognitive

task, competing for the driver's attention that is unrelated to the core one of operating

the vehicle. As a result, dual-task drivers make more errors and detect fewer hazards

(Briggs, Hole, & Land, 2016).

The reviewed literature suggests that technology applications that are not

directly related to the driving tasks, which is the case of COTs, increase the risk of

distraction. However, COTs can potentially persuade, i.e. by giving user drivers a good

reason to drive safer, and may influence their behaviours as a result (Fogg, 2009).

When it is designed to persuade drivers positively, COTs can foster safer driving

behaviour (Schroeter et al., 2012; Steinberger et al., 2015), an implication assessed by

the current PhD program of research.

2.4.2 Safe-driving apps

Mobile phone use while driving increases the chances of crash four times (White,

Hyde, Walsh, & Watson, 2010). The drivers themselves are well aware of the problem,

with 14% of them confirming that distraction contributed to their crash (McEvoy &

Stevenson, 2007). Still, a very large number report using their mobile phone in the car:


43% receive calls, 36% make calls, 27% read and 18% send text messages (White et

al., 2010).

Mobile phones, including smartphones, although identified as a major source of

distraction while driving (WHO, 2011), also offer various opportunities in terms of

novel technology-driven road safety approaches. Apps coming out of academia, such

as the CarSafe app (You et al., 2013), have been developed to monitor driver behaviour

and warn drivers about identified risks. However, they have only had limited uptake

to date. Insurance companies (e.g. AAMI in Australia, State Farm in the US,

Telefonica in Spain, and AVIVA in the UK) also embrace the opportunity through

proprietary smartphone safe-driving apps in an effort to promote safer driving

practices (also sell insurance and collect driving data). Other companies and start-ups

have created driving apps such as Rookie Dongle, Flo or Automatic. Some of those

apps not only monitor speed through GPS but can also block incoming calls and

messages on the mobile phone. Others essentially serve the purpose of a driving coach

with feedback, aiming to help drivers to learn and apply the road rules.

Safe-driving apps often leverage gameful designs or gamification (Diewald et

al., 2013) to boost motivation and commitment to use. The gamification of driving

may lead to an increased drivers’ engagement with their driving tasks (Steinberger et

al., 2017). Safe-driving apps use a number of tools to trigger such a positive outcome,

such as challenges, feedback, social approval, and rewards (Markey, 2014). For

example, while using such apps, the drivers may earn points for driving "safely". Those

points can then potentially be redeemed for rewards or shared as achievements on

social media.

There is a variety of safe-driving apps that claim to be able to improve the

drivers’ skills and performance. For example, You et al. (2013) developed and tested

a smartphone app (CarSafe) to alert drowsy and distracted drivers. They tested the app

on the road with 12 drivers aged 23-53. Six of them drove in a controlled environment

with a co-pilot that gave them instructions to perform dangerous manoeuvres when the

situation was safe. The other six were monitored while driving in their usual daily

routine. While there was real driving performance being observed, the data collection

was related to assessing the CarSafe accuracy rather than to evaluate its influence on

the drivers. No data was collected about changes in the drivers' behaviour as a result

of using the app, which is the focus of the current thesis.


Another study evaluating a smartphone safe-driving app intervention focused on

the usage of smartphone apps rather than on safety benefits. Musicant and Lotan

(2015) deployed a smartphone app (RefuelMe) that at the point of the intervention was

available for iPhone users for free. Their target group was composed of 21 soldiers

aged 18-19 and their friends, also 21 people. The authors collected data through three

sources: the app (naturalistic: 29,335 recorded trips, scored based on recorded G-force

events, such as speeding, hard acceleration, hard braking and fast cornering),

surveys/phone interviews (self-reports) and the soldiers’ Facebook page (qualitative

data). A limitation of the study was that it did not establish a self-reported baseline for

comparison and could not evaluate behavioural changes. The authors inferred driving

behaviour from the recorded trip scores and observed a decrease in scores with the

progress of the study. Similar to the intervention implemented in the current research,

the app was monitoring and recording the real driving behaviour as well as providing

feedback (real-time, at the end of the trip and weekly). Although driving behaviour

was observed, the focus of the study was on app usage and what could motivate

adoption. Since the study was focused on adoption and usage, driving performance

feedback was very forgiving.

Musicant and Lotan (2015) found that once all incentives were obtained by the

participants, they stopped using the app. They attributed this to the fact that obtaining

the rewards was more related to mere participation than to performing safer. Besides

the incentives, another possible explanation of Musicant and Lotan (2015) result was

that the app had to be manually started and stopped. While the current program of

research also utilised incentives, all participants were eligible to receive the same

incentive at the end of their participation, allowing for an investigation of the influence

of a safe-driving app on participants’ driving behaviour, regardless of the reasons for

using it. Long-term adoption was not a targeted result. A self-starting capability of the

smartphone safe-driving app was a desired feature when selecting an intervention tool.

2.4.3 Virtual reality

Moving away from the possibility to introduce additional risks in the young

drivers’ experience on the road, a second opportunity to use technology for improving

young drivers’ safety was utilised as a part of this PhD program of research. Young

drivers' use VR in their daily lives (Lang et al., 2018). An increasing number of

immersive virtual experiences is currently becoming available for households as


consumer electronics. HTC Vive, Oculus Rift, PlayStation VR, Google Daydream

View and Samsung Gear are commercially available consumer-grade devices that

bring the experience on demand.

Designated tasks, delivered through VR and experienced by users, were found

to potentially lead to prosocial behaviour (van Loon, Bailenson, Zaki, Bostick, &

Willer, 2018). Blascovich et al. (2002) argue that VR is a well-suited methodological

tool for experimental social psychology that helps overcome limitations such as lack

of replication or control-mundane realism trade-off. Its main advantage is the

possibility for individual perspective-taking, which is arguably more successful than

traditional role-playing exercises (van Loon et al., 2018). As a result, there is evidence

that VR can increase empathy ((Garner, 2017; Ingram et al., 2019; van Loon et al.,

2018). Nevertheless, Ahn, Bailenson, and Park (2014), Morina, Ijntema, Meyerbröker,

and Emmelkamp (2015), Schwebel, McClure, and Porter (2017), Theng, Lee,

Patinadan, and Foo (2015) and van Loon et al. (2018) provide mixed evidence for

behavioural change success in their studies. Thus, whether VR interventions ultimately

lead to a behavioural change is not certain.

The situation in road safety research is not different, signifying that VR is just

making its way into it. There is a limited number of road safety studies leveraging this

technology, despite it providing an opportunity to simulate life-threatening situations

in safety. Most of the VR research has been in the domain of pedestrian safety

(Bhagavathula, Williams, Owens, & Gibbons, 2018; Morrongiello, Corbett, Foster, &

Koutsoulianos, 2018; Schwebel, Combs, Rodriguez, Severson, & Sisiopiku, 2016;

Schwebel et al., 2017) but little evidence was found on the VR potential to deliver

safety benefits. This may or may not be due to the novelty of the technology but makes

a systematic investigation of their real effect, both as evidenced in the literature and as

part of a real-world intervention, a gap that needed to be addressed.

2.5 CONCLUSION AND IDENTIFICATION OF A RESEARCH GAP

The presented literature review showed that existing efforts might not have the

potential to reduce crashes further. Existing strategies, delivered through driver

training, media and law enforcement, are working, but their effect seems to plateau.

Although many of those strategies are focused on young drivers, young drivers

continue to be overrepresented in crashes (BITRE, 2018; NHTSA, 2018a; EC, 2018;


WHO, 2018), with evidence of their increased risk exposure. The overall situation calls

for novel complementary approaches to persuade young drivers to reduce their risk-

taking, which, in turn, could help lead reversing the crashes' trend, bringing numbers

below the plateau.

The current program of research utilised novel interventions as complementary

approaches. It focused those interventions on risky behaviours where arguably the

potential benefit is highest, i.e. behaviours, which are significant contributors to both

the number and the severity of road crashes. As evidenced by the literature, the risky

behaviours of speeding and DUI bear high potential severity of consequences that may

follow if young drivers engage in them. Thus those were targeted by the two

intervention studies, evaluated in Chapter 7 and Chapter 9.

As a contemporary response to the call for novel complementary approaches,

instead of trying to train the young drivers out of their bad habits, the current research

evaluated the potential of technology to deliver additional experience to the young

drivers and, as a consequence, to improve their driving behaviour in regards to the two

targeted behaviours, speeding and DUI. Fogg (2009) suggested that technologies may

change users' behaviours through persuasion. Schroeter et al. (2012) and Steinberger

et al. (2015) supported that view further suggesting that COTs, designed to persuade

drivers, could help young drivers adopt safer driving behaviour.

Available COTs, namely smartphone safe-driving apps and VR simulations of

risky driving, provide opportunities to help address speeding and DUI as risky

behaviours. These opportunities are currently being explored by both academia and

businesses. However, the reviewed literature did not provide extensive evidence for

safety benefits for the young drivers, stemming from smartphone safe-driving apps

and VR simulations of risky driving. Such safety benefits seemed to have been

predominantly measured in conditions that do not resemble the use of the two example

COTs in the real world, e.g. through simulator tests with small samples and lack of

investigation of long-term effects. Thus, little is known about those two COTs' safety

impact in real-world road safety interventions, although both are readily available for

the general public and, more so, they are targeting young drivers.

At the same time, it is acknowledged that traditional literature reviews may

suffer from bias, originating in the reviewer's impressions (Mulrow, 1994). Systematic

literature reviews are regarded as a means to address such bias, which shall deliver


improved reliability of findings as well as an increased confidence in the conclusion

(Fineout-Overholt, Melnyk, Stillwell, & Williamson, 2010). As systematic reviews

represent a scientific investigation on their own (Mulrow, 1994), the two systematic

reviews, part of the current program of research, are presented as separate Chapter 5,

focusing on smartphone safe-driving apps, and Chapter 8, focusing on VR simulations

of risky driving.

Overall, the missing evidence of real-world effects from the use of smartphone

safe-driving apps and VR simulations of risky driving represented a significant

research gap, which the current program of research aims to address. This gap

represents the basis for the current project to systematically evaluate the effect of

smartphone safe-driving apps and VR simulations of risky driving. Thus, the

investigation looks at those effects both as evidenced in the literature and as part of

intervention studies. The two implemented intervention studies focus on behavioural

change the two COTs can trigger in the ordinary young driver, in the longer term, in

their free-living environment. Their findings are presented in Chapter 7, for an

example smartphone safe-driving app COT, and in Chapter 9, for an example VR

simulations of risky driving COT.

Chapter 3: Theoretical considerations informing intervention evaluation framework 25

Chapter 3: Theoretical considerations informing intervention evaluation framework

Chapter 3 reviews theories that were applied in domains similar to the ones

explored by the current PhD program of research: health risks prevention, road safety,

human-computer interaction and young people. It presents the considerations taken

into account when choosing the most suitable framework to guide the evaluations of

the implemented interventions. Those considerations include whether the theory was

considered to be well suited:

1) to evaluate behaviour and its underlying constructs;

2) for the nature of the targeted behaviours (speeding and DUI);

3) to be applied at an interpersonal level, as the implemented interventions

regard young drivers as part of a dynamic system involving other people;

and

4) to account for the interrelation between its constructs.

The final choice was made depending on how well the respective theory fit the

needs of the planned interventions.

3.1 INTRODUCTION

Behaviour change is considered achieved when there is evidence for new

behaviour adoption on the basis of new knowledge acquisition (Bandura, 1986). Thus,

for the current PhD project's interventions to be considered successful, their evaluation

has to find significant evidence that a desired behavioural change on the individual

level or, at least, change in the intention to perform the behaviour of interest, was

achieved, as a result of the respective intervention. Appropriate theoretical grounding

is not only a sound basis for evaluating the intervention achievements, but it also

provides insights on how behavioural change could be motivated amongst the

participants (Kohler, Grimley, & Reynolds, 1999).

26 Chapter 3: Theoretical considerations informing intervention evaluation framework

Researchers are extensively using behaviour change theories to underpin

interventions' design, development and evaluation. The current PhD program of

research targeted behavioural change in regards to achieving positive changes in young

people's intention and behaviour on the road. Both personal and social factors are

found to influence the behaviour of young drivers (Shope & Bingham, 2008). Keeping

in mind that young drivers' behaviour is influenced by many factors, this chapter

discusses the choice of a theory that informed this PhD project, both theoretically and

methodologically.

Theories with a history of being applied in fields relevant for the current research

such as health promotion, human-computer interaction, young people, road safety and

combinations of them were considered when choosing an appropriate one to guide the

current research evaluation framework. In their systematic review covering 256

articles (out of 8680 articles initially retrieved), Davis, Campbell, Hildon, Hobbs, and

Michie (2015) identified 82 theories in the social and behavioural sciences literature.

Three of them accounted for 165 (60%) of the reviewed articles across behaviour

change: Transtheoretical model of health behavior change (TTM) (Prochaska &

Velicer, 1997) (33%), TPB (Ajzen, 1988) (13%), Social Cognitive Theory (SCT)

(Bandura, 1986) (11%) and Health Belief Model (HBM) (Hochbaum, Rosenstock, &

Kegels, 1952) (3%). The Theory of Reasoned Action (TRA) (Ajzen & Fishbein, 1980)

and the Self-Efficacy Theory (SET) (Bandura, 1977) were also included in the analysis

as they are the antecedents of TPB and SCT respectively. These were the theories

considered, and following the detailed considerations are provided.

3.2 TRANSTHEORETICAL MODEL OF HEALTH BEHAVIOR CHANGE

TTM (Prochaska & Velicer, 1997) argues that each individual is in a different

stage of behavioural change in relation to a specific health-related behaviour. The

theory is widely used to explain the way people engage in activities to change their

behaviour, how they progress through the changes, and what efforts are put in place to

maintain new and better behaviour. According to the model, there are six stages of

change (Precontemplation (Not Ready); Contemplation (Getting Ready); Preparation

(Ready); Action; Maintenance and Termination) with the first five being measurable

(see Figure 3.1). TTM suggests strategies, based on identified individual readiness to

change, which can help a person move from one stage to the next one. The opposite

process (relapse) is also possible at any stage.


Figure 3.1. TTM stages of change

TTM is applied in various health-related interventions, such as stress

management (Prochaska et al., 2012), reduction of smoking (Prochaska, DiClemente,

Velicer, & Rossi, 1993), improving energy balance (Van Duyn et al., 1998; Velicer et

al., 2013) or obesity reduction (Mauriello et al., 2010). In many cases, TTM is used to

assess interventions' impact on more than one aspect (e.g. technology, young people

and risky behaviour), relevant to the current PhD program of research. For example,

Prochaska et al. (2012) used a telephone and an online program coaching to improve

wellbeing in terms of improved life evaluation, healthy behaviour, emotional and

physical health. Aveyard et al. (1999) used three TTM-based computer sessions and

three class lessons over a year to reduce smoking in young people and their peers.

Velicer et al. (2013) implemented a computer-based intervention to prevent substance

abuse. Thus TTM is identified as a useful model in multidisciplinary research

including human-computer interaction (Aveyard et al., 1999; Di Noia, Contento, &

Prochaska, 2008; Gold et al., 2016) and persuasive technologies that change users'

behaviours and attitudes through social influence and persuasion (Fogg, 2009).

In contrast with other health domains, the literature review did not reveal many

TTM-based applications in the road safety field. Five studies in three fields were

identified in which the model is applied: recidivist drink drivers (Freeman et al., 2005;

Polacsek et al., 2001), work-related safety (Banks, 2008; Murray, White, & Ison, 2012)

and more recently, senior drivers (Kowalski, Jeznach, & Tuokko, 2014).

Polacsek et al. (2001) implemented an intervention for drink drivers, attending a

“Driving while intoxicated” school, to investigate if the Victim Impact Panels (VIPs)

of Mothers Against Drunk Driving had an additional effect on recidivism or on

progressing individuals through the stages of change towards not drinking while

driving. VIPs are designed to reveal the full scope of the negative consequences of


substance-impaired driving, helping offenders recognize and internalize the offence's

long-term effects. In the same field, Freeman et al. (2005) investigated the TTM

constructs of stages of change and self-efficacy in recidivist drink drivers. The authors

used TTM-based scales to measure self-efficacy levels (Drinking/Driving Efficacy

Scale (Wells-Parker, Burnett, Dill, & Williams, 1997)) and motivation to change drink

driving behaviour (Readiness to Change Questionnaire (Rollnick, Heather, Gold, &

Hall, 1992)), Stages of Change for Drink Driving Questionnaire (Wells-Parker,

Williams, Dill, & Kenne, 1998)). This shows that available scales can be successfully

combined in a comprehensive questionnaire in an effort to build a more holistic picture

of an intervention's effect.

Kowalski et al. (2014) used the TTM framework to develop specific questions

addressing constructs such as stages of change, decisional balance, change processes

and self-efficacy. Their study demonstrated how the TTM helps to understand at what

stages of change the participants were, and whether participants were aware of the

need to change their behaviour. The latter is a prerequisite of going through

behavioural change.

Linking TTM to the more narrow settings of occupational road safety, Banks

(2008) explored the correlation to determine relations between the individual stages of

change and crash involvement, fatigue and distraction. The author concluded that the

stage of change was a significant independent predictor of crash involvement.

Predicting potential crash involvement through the participants' stage of change can

be useful for the current research. Such link suggests that young drivers' crash

involvement can potentially be reduced if an intervention manages to help young

people progress from one stage of change to another.

Like most models, TTM has attracted criticism (Littell & Girvin, 2002; West,

2005). Littell and Girvin (2002) see the biggest problem in trying to oversimplify

complex behavioural change processes into stages. Other researchers found instability

of the stages themselves (De Nooijer, Van Assema, De Vet, & Brug, 2005; Hughes,

Keely, Fagerstrom, & Callas, 2005), challenging the model presumption that people

make coherent and stable plans. The model also neglects the fact that many health

problems arise from semi-automated unhealthy habits, which are not easy to change

(West, 2005). For example, Velicer et al. (2013) reported limited results in directly

addressing smoking and alcohol. The authors suggested that reactivity and


defensiveness towards behavioural change may have been due to the behaviours’

addictive nature. So, for some behaviours, the model can propose an incorrect

intervention strategy (West, 2005).

TTM could have been considered relevant to the current research. This PhD

program of research aimed to improve young drivers’ behaviour on the road, i.e. push

them from one stage to another in respect of their risky behaviour. TTM explores five

different stages. The advantage of TTM is that aids understanding at which stage of

change the study participants are, in particular when dealing with groups that are not

homogenous. Dividing a participant pool into subgroups, based on their stage, further

allows understanding how effective a respective intervention is in the case of each

different stage, i.e. by assessing whether and how far the participants from the

respective stage of change progressed, as a result of the intervention. Thus the model

could provide valuable information for what type of participants such interventions

could deliver maximum benefit.

In relation to the current program of research, a critical aspect in determining the

TTM as a suitable model to address the research gap is that the two interventions to be

evaluated (a smartphone safe-driving app, see Subsection 7.2.6, and VR simulations

of risky driving, see Subsection 9.2.1) were not stage-tailored. Tailoring them to the

needs of a group of participants carried the risk of the chosen target stage not being

identified correctly due to oversimplification of the behavioural processes in driving.

The current program of research also looked into problematic driving habits, which in

many cases may be semi-automated and, thus, might not be accounted for by the

model. In addition, TTM focuses on the individual and not on the interpersonal level.

Accounting for interpersonal relations was a needed model's characteristic so that

normative influences for achieving a positive change can be assessed. Thus, TTM was

considered not fully suiting the needs of the current project.

3.3 HEALTH BELIEF MODEL

HBM (Hochbaum et al., 1952) addresses the TTM limitation of dividing the

process of behavioural change into stages by looking at the process as a whole. HBM

also does not assume that people make plans, and necessarily follow them. The model

argues that a person has to perceive a direct threat, with serious and potentially fatal

outcome, to consider changing behaviour. Additionally, such change has to be directly


related to a benefit that is within reach, i.e. the effort to obtain the benefit is less than

the benefit itself (see Figure 3.2).

Figure 3.2. The Health Belief Model

The HBM predictive power finds varying support in the literature. Janz and

Becker (1984) investigated 46 studies and found the HBM construct to be statistically

significant in predicting behaviour, with ratios of 81% for susceptibility, 65% for

seriousness, 78% for benefits, and 89% for barriers. However, other authors raise the

question of whether, despite good results, other theories are a better fit. For example,

Quine, Rutter and Arnold (1998) examined the predictive power of HBM through path

analysis in a longitudinal study. They looked at its ability to offer an explanation of

the factors that determine helmet use by school-aged cyclists. The study compared the

HBM results with other results, coming from comparable studies underpinned by TPB.

Quine, Rutter and Arnold (1998) found that TPB offered greater predictive ability.

Şimşekoğlu and Lajunen (2008) reported similar results. They compared the predictive

power of HBM and TPB as well as the fit of the two theories to collected self-reported

data on seatbelt use among students, front-seat passengers. Şimşekoğlu and Lajunen

(2008) reported that HBM was not a good fit for the data, while TPB was.

HBM is known to suffer from conceptual difficulties. For example, there is no

indication of how the different beliefs influence each other or how they influence the

behaviour when combined (Quine, Rutter, & Arnold, 2000). Furthermore, the model

does not consider tailoring strategies to encourage healthier behaviour as well as the

accuracy of the information about the behaviour (Mackenzie, 2016); it does not


address the TTM limitation of focusing mainly on the individual; and it does not

consider other factors that may have an influence on one's health behaviour, such as

norms.

HBM has been applied in various fields, relevant to the current project, such as

health prevention (Janz & Becker, 1984), young people (Gillibrand & Stevenson,

2006) and human-computer interaction (Ng, Kankanhalli, & Xu, 2009). With evidence

of being successfully used in road safety studies, investigating young people's safe

behaviours, HBM could be considered relevant for the current research, which

investigated safety benefits from using two examples of COT on the young drivers'

speeding and DUI behaviour. For example, Ghavami, Harandy and Kabir (2016) used

HBM constructs in a questionnaire to assess the effect an HBM-designed intervention

had on primary school students in regards to obeying traffic rules. HBM could inform

what the likelihood is that each young driver would cease risky driving, which of the

constructs is the most relevant in predicting the behaviour, and what health promotion

advice is best suited for an intervention. However, the HBM's primary focus on the

individual makes it difficult to integrate interpersonal factors. Thus, HBM was

considered as not being the most suitable to inform the current research interventions'

evaluation framework.

3.4 SOCIAL COGNITIVE THEORY

With a history of being applied for evaluation purposes, the Bandura (1977) Self-

Efficacy Theory (SET) was considered as an overarching framework. SET aims to

explain and predict behavioural changes as a result of interventions. As the name

suggests, SET is focused on self-efficacy, i.e. on how much an individual believes they

are able to achieve specific goals. However, similar to TTM and HBM, SET alone

does not account for interpersonal factors (e.g. social norms).

SCT (Bandura, 1986) extends SET with many of its constructs being similar to

HBM constructs (Donovan, 2011). SCT suggests that humans function as a result of

interactions between their environment, personality and behaviour (see Figure 3.3).

SCT argues that personality is developed through observational learning and social

experience. The model goes beyond the SET's limitation of not accounting for

interpersonal factors, which are accounted for by the “environment” construct in SCT.


The environment constitutes the external influences on one's behaviour while the

personality is one's own motivational factors to perform the behaviour.

The model's personality dimension includes key constructs which influence

performing the desired behaviour: 1) self-efficacy or how much an individual believes

they are able to achieve specific goals, outcome expectations or the individual's

expectations, in case they perform a behaviour; 2) self-control or how much an

individual is able to autonomously regulate their own intentions and behaviours,

reinforcements or internal or external responses to an individual's behaviour, affecting

their likelihood to continue it; and 3) observational learning or the ability of an

individual to reproduce a behaviour after observing it in others. Within these

constructs, SCT considers both self-reflection, i.e. whether an individual is able to

analyse their own behaviour critically, and potential influences of personality

characteristics.

Figure 3.3. Social Cognitive Theory Model

SCT has a long track record of being applied in fields relevant to the current

project. It was used to inform studies, concerned with health risks prevention (Miller,

Shoda, & Hurley, 1996; Schwarzer & Renner, 2000; Wallace, Buckworth, Kirby, &

Sherman, 2000), human-computer interaction (Compeau & Higgins, 1995; Ifinedo,

2016), young people (Ifinedo, 2016; Wallace et al., 2000) and their peers (Compeau

& Higgins, 1995; Ifinedo, 2016; Rana & Dwivedi, 2015; Wallace et al., 2000).

The theory has also been used to inform road safety research. For example,

Tranter and Warn (2008) used an SCT-based questionnaire to investigate the attitudes

of mature drivers towards speeding and traffic rules violations. The study found

support that personality characteristics, such as higher interest in motorsports,

associate with a higher propensity to engage in speeding behaviour and can predict


speeding violations. Providing further evidence around speeding, Yıldırım-Yenier,

Vingilis, Wiesenthal, Mann, and Seeley (2016) used SCT to explore the relationships

between thrill-seeking, attitudes towards speeding and driving violations. Their study

found that driving offences can be directly predicted by thrill-seeking, a personality

influence in SCT. Investigating into the nature and mechanisms of influence between

SCT core constructs (personality, behaviour and environment), Scott-Parker (2012)

used the model as an overarching framework to explore young people’s risky driving

behaviour. The author found that all three factors (personality, behaviour and

environment) were associated with young drivers’ risky behaviour on the road.

While recognising its relative utility, some researchers see SCT as a collection

of logical statements that are difficult to empirically test (Smedslund, 1978). Other

researchers see its constructs as based on variables that are not well defined, and that

cannot be observed and assessed (Lee, 1989). For example, the instruments to measure

self-efficacy may not be carefully developed and validated (Frei, Svarin, Steurer-Stey,

& Puhan, 2009). In addition, Mackenzie (2016) questioned SCT's ability to account

for motivation at the moment of executing the behaviour.

The SCT could have offered a suitable theoretical grounding for this PhD

program of research, e.g. its use to explain risky behaviours in young drivers. The

model can inform interventions' evaluation design and can provide insights into the

participants' self-efficacy (How to support personal confidence in achieving behaviour

change results? What information to be provided to increase self-reflection?), and

normative influences (How to shape the intervention environment to encourage safe

behaviour on the road?). However, the lack of carefully developed and validated

measurement instruments is a notable limitation. Developing and validating

questionnaires was outside of the scope of this PhD program of research. In addition,

through its real-time feedback feature (see Subsection 7.2.3), the smartphone safe-

driving app in the Study 2 intervention could influence participants when they perform

risky driving. The SCT's questioned ability to account for motivation at the moment

of executing the behaviour would challenge its suitability to evaluate the effects of

such interactions, making the limitation a valid concern for the current program of

research. As a result, SCT was not considered a good fit for the present research

evaluation purpose.


3.5 THEORY OF PLANNED BEHAVIOUR

The current PhD program of research envisaged interventions that can

potentially not only improve driving behaviour but also change young people's

intention to perform specific behaviours. Ajzen and Fishbein's (1980) TRA makes an

effort to predict behaviour and regards intention as the best predictor. According to

TRA, intention, in turn, is influenced by the individual’s attitudes and norms.

However, a recognised limitation of the theory is that TRA does not account for how

the individual sees the effort (easy or difficult) that has to be put in place to achieve

the desired behaviour, or, at least, to form an intention to perform it. As a response for

the need of improving the theory, Ajzen (1988) added perceived behavioural control

(PBC) to TRA, as an additional third factor, influencing both intention and behaviour,

which saw TRA evolve into TPB (see Figure 3.4).

Figure 3.4. Theory of Planned Behaviour (Ajzen, 1991)

According to TPB (Ajzen, 1991), intention to perform a behaviour is the best

predictor of future behaviour, as its immediate antecedent. In turn, intention is

predicted by three interrelated factors: 1) how favourable, or unfavourable, the

behaviour is perceived to be (attitude), 2) whether important others are perceived as

approving or disapproving the behaviour of interest (subjective norm), and 3) how

easy, or difficult, performing the behaviour is perceived to be (perceived behavioural

control, PBC) (Ajzen, 1991). PBC is also considered as a direct predictor of behaviour.

The TPB components have been used to explain various risk-related behaviours in the

health domain, e.g. safer sex, uptake of vitamin C or cycle helmet use (Rutter & Quine,

2002). Rutter and Quine (2002) provided evidence that around 40%, on average, of the

variance in both health behaviour and intention can be explained through TPB.


Supporting the evidence for TPB's predictive power in the health domain in

general (Rutter & Quine, 2002), a body of literature reports on the theory's suitability

to evaluate constructs in relation to risky behaviours on the road, such as speeding and

DUI, which are of primary interest for this program of research.

Complex behaviours such as speeding and DUI can be difficult to evaluate.

However, the TPB has been shown to be useful in this context, e.g. by Stead, Tagg,

MacKintosh, and Eadie (2005). The authors found that the TPB constructs predicted

between 47% and 53% of the variance in the participants' speeding intention and

between 33% and 40% of the variance in the participants' speeding behaviour. Others,

such as Warner and Åberg (2008) found that TPB constructs account for up to 73% of

the variance in intention to speed; and Elliott and Thomson (2010) found that TPB

constructs predicted 55% of the variance in the participants' speeding intention and

47% of the variance in the participants' speeding behaviour. In a study focused on

young adults' DUI, Chan, Wu, and Hung (2010) found that the TPB explained 79% of

the variance in intention to drink and drive. In another study, Potard, Kubiszewski,

Camus, Courtois, and Gaymard (2018) found that the standard TPB (attitude,

subjective norm, and PBC) explained 44%, while the extended TPB (including past

behaviour with the standard constructs) explained 52% of the variance in DUI

intention. Although speeding and DUI are complex behaviours to investigate, these

studies demonstrate the TBP's suitability as an evaluation framework.

In addition, TPB was also found suitable to predict distraction in drivers, the

behaviour of secondary interest for the current program of research. Chen, Donmez,

Hoekstra-Atwood and Marulanda (2016) used the TPB framework to assess attitudes,

social norms, and PBC in drivers, engaging in distracting tasks that were not relevant

to the driving task. The authors found TPB to explain 45.2% of the variance in

distraction. Comparable findings were reported by Bazargan-Hejazi et al. (2017), who

found that TPB explained 47% of the variance in intention to engage in phone

distraction while driving.

TPB addresses some of the suitability issues identified for the TTM, HBM and

SCT in the context of this program of research. TPB is not stage-tailored and accounts

for interpersonal relations through its subjective norm, addressing the discussed TTM

limitations of bring stage-tailored and focusing on the individual. Despite not being

stage-tailored, TPB allows for tailoring strategies to encourage specific behaviour, for


example, through interventions as in Quine, Rutter, and Arnold (2001). It also explores

different beliefs, and their influences through its attitude construct, addressing the

discussed HBM limitation of not accounting how different beliefs interact and

influence behaviour together. There are also validated questionnaires in the TPB

toolset, as in Lennon, Oviedo-Trespalacios, and Matthews (2017), Chen et al. (2016),

Haque et al. (2012), Elliott and Thomson (2010) and Quine et al. (2001), addressing

the discussed SCT limitation of lack of carefully developed and validated instruments.

3.5.1 Criticisms and limitations of TPB

A criticised TPB assumption is that it sees the person as having all the resources

and skills to enact the behaviour of interest (Mackenzie, 2016). Unconscious

influences on behaviour (Sheeran, Gollwitzer, & Bargh, 2013) are also pointed out as

a major limitation, as TPB exclusively focuses on rational reasoning. Other researchers

underline the TPB static explanatory nature as a limitation (Sniehotta, Presseau, &

Araújo-Soares, 2014). Sniehotta et al. (2014) also suggest that the main problem of

TPB is in the validity of its predictions, as the sequence of influences, it explores, is in

conflict with the available evidence. For example, shifts in behaviour, as a

consequence of intervention, are not always moderated directly through the TPB

constructs and, sometimes, when the behaviour is driven by a habit, reverse causation

is possible (Webb & Sheeran, 2006). In such cases, the intention has little influence on

the behaviour and past behaviour is a much stronger predictor to both intention and

future behaviour.

Some of those limitations can be addressed through research design. For

example, the static nature limitation could be addressed by a longitudinal design of the

studies, exploring shifts in the TPB constructs over time, as in Quine et al. (2001) and

in Stead et al. (2005). The assumption that people have the needed resources and skills

to enact a behaviour could be addressed by providing research participants with

additional resources and skills that could enable them to perform the behaviour of

interest. Other limitations require an extension of TPB, as discussed in turn.

3.6 EXTENDING TPB

Conner (2015) suggests that a more constructive approach of capitalising on the

vast body of knowledge surrounding TPB shall be adopted by extending the theory,

instead of following Sniehotta et al. (2014) proposal to retire it. For example, to


establish causality effects, an intervention's evaluation could compare self-reported

results against shifts in the underlying TPB constructs as a result of the intervention,

as in Stead et al. (2005), or could explore past behaviour in the analysis, as in Elliott

and Thomson (2010). A deeper investigation of causality effects within TPB could

also be achieved by splitting the original TPB constructs into components and

assessing them separately (Conner & Sparks, 2005; Elliott & Thomson, 2010).

Unconscious influences can be addressed by exploring personality characteristics in

implemented studies (Sheeran et al., 2013) or demographic factors, as in Horvath et al.

(2012) (see Section 2.2. for a discussion on gender and driving experience). Conner

(2015) suggests in addition to personality characteristics, other potential predictors can

also be explored to provide further explanation on why an intervention may have

triggered a behavioural change if such is found in the first place.

The literature offered evidence for potentially useful constructs that can be used

to extend TPB to account for additional influences, i.e. 1) the normative influences,

not accounted for by TPB, of moral norm and peers' norm, 2) risk perception, and 3)

the personality characteristics impulsivity, sensitivity to punishment and sensitivity to

reward. Those are discussed in the following sections.

3.6.1 Additional normative influences

Intention is, in general, weakly predicted by the TPB subjective norm (Armitage

& Conner, 2001). At the same time, conformism influences the behaviour of young

drivers in their early stages of learning to drive (Falk et al., 2014). Acquiring the

driving license usually happens when young people are trying to find their own

identity, pursuing independence from their parents (Engström, Gregersen,

Hernetkoski, Keskinen, & Nyberg, 2003; Laapotti, Keskinen, Hatakka, & Katila,

2001) while being influenced by social norms or influencing each other through

establishing norms within their inner circle of friends. Forming driving habits, that not

only lead to behaviour in compliance with the traffic legislation but also lead to safe

traffic participation, may be critical at this time. Such an effort can potentially reduce

future efforts and costs. To extend the TPB predictive validity, Armitage and Conner

(2001) suggest the expansion of its normative component.

Friends or peers may reinforce the perception of a behaviour being right or

wrong (Conner & Sparks, 2005). Drivers may be much more influenced by their peer

drivers than they realise (Chen et al., 2016). Young drivers are very susceptible to peer


pressure, especially when COT, such as a mobile phone (Chen et al., 2016), is involved

(Lee, 2007).

Peers have a significant impact, usually directed towards risky, unsafe and illegal

driving behaviours (Engström et al., 2003; Falk et al., 2014), and are considered as a

major source of increased crash risk for young adults (Falk et al., 2014). For example,

young drivers are more likely to engage in speeding when approved by their friends

(Fleiter et al., 2006; Horvath et al., 2012). Similarly, in the case of DUI, Sela‐Shayovitz

(2008) showed that perceived peer pressure had a significant impact on young drivers,

as well as on their involvement in DUI-related road crashes. Other researchers suggest

that the potential of peer influence can also have a positive effect, i.e. protect against

risky behaviour (Kaye, 2014; Otto, Ward, Swinford, & Linkenbach, 2014; Weston &

Hellier, 2018).

Thus, as part of interventions' evaluation, the present PhD program of research

could assess the contribution of peers' norm., i.e. whether the participants' friends are

seen as disapproving or approving of the respective participant engaging in the

behaviour of interest. For the purpose of the current PhD program of research, peers

were defined as other young people, at the same, or nearly the same age as the

respective participants, who may, or may not, be their friends, and may, or may not,

be participating as participants in the current program of research.

Regardless of the nature of or the reason for the risky behaviour, it may also be

negatively influenced due to the fact that behaviour is perceived as "normal" in the

first place (Ward et al., 2017). The predictive role of moral norm is argued to be distinct

from the standard TPB constructs (Conner & Sparks, 2005; Manstead, 2000). The

individual's moral norm would see engaging in a behaviour perceived as correct or

incorrect from a personal perspective and, thus, assessing it may complement

evaluations of other norms such as the subjective norm (Ajzen, 1991). Building on

that, as part of the interventions' evaluation, the present PhD program of research could

assess the contribution of moral norm.

3.6.2 Risk perception

Tay (2005) established that avoiding negative consequences is the motivation

behind specific behaviour. He indicated the importance not only of the probability of

punishment but also the individual risk tolerance (of getting punished, for example).


People decide whether taking a risk is justified based on a number of factors, including

the perception of vulnerability, the severity of the threat and potential benefits

(Hochbaum et al., 1952). A better understanding of risk perception in a specific

population may provide additional insights when assessing the impact of an

implemented intervention on the involved participants. Risk perception was shown to

have a strong influence on risky driving intention (Ward et al., 2017) and behaviour

(Rhodes & Pivik, 2011), including speeding and DUI (Fernandes et al., 2010), the

behaviours of current inquiry. Thus, the participants’ perceived risk of being involved

in a crash or of being caught by the police while performing specific behaviour could

be assessed within the current PhD program of research.

3.6.3 Personality characteristics

Personality characteristics are influential and, as such, have been explored

widely in previous research in an effort to explain young people’s risk-taking (Scott-

Parker, 2012). Young drivers’ personality characteristics have been studied in relation

to each of the fatal five risky behaviours on the road: speeding (Tao, Zhang, & Qu,

2017), DUI (Fernandes et al., 2010; Luk et al., 2017), not wearing a seatbelt (Fernandes

et al., 2010), fatigue (Fernandes et al., 2010) and distraction (Parr et al., 2016).

The impact of personality on young drivers’ behaviour such as speeding and DUI

can, therefore, not be dismissed. In regards to those behaviours, the literature provides

evidence for the influence of the following personality characteristics: impulsivity,

sensitivity to punishment and sensitivity to reward.

Impulsivity was previously studied in relation to young drivers (Scott-Parker,

2012). It is seen as the young drivers' risky driving most robust predictor (Luk et al.,

2017). A number of studies have shown a positive relationship between self-reported

risky driving and impulsivity (Constantinou et al., 2011; Pearson, Murphy, & Doane,

2013). Thus, it may be critical to assess its contribution when evaluating the impact of

an intervention.

Sensitivity to punishment and sensitivity to reward were also found to predict

risky driving (Constantinou et al., 2011). Castellà and Pérez (2004) established that

sensitivity to reward was positively correlated, while sensitivity to punishment was

negatively correlated with traffic rules violations. In a recent literature review,

Sensitivity to reward was further shown to have a high negative impact on the young


drivers' risky driving behaviours (Scott-Parker & Weston, 2017). With evidence

whether young drivers are more influenced by the perspective of being punished or by

the chance of being rewarded, an intervention can take completely different forms. The

intervention designers can choose whether to stress more on its negative (punishment)

or on its positive (rewards) aspects, depending on their expectations for the target

group composition.

In this PhD program, impulsivity, sensitivity to punishment and sensitivity to

reward were therefore considered to provide an understanding regarding the effect of

the implemented interventions.

3.7 CONCLUSION

Evidence for new behaviour adoption is needed to confirm successful behaviour

change (Bandura, 1986). In the case of interventions, evaluating their achievements

through an appropriate theoretical framework would provide the necessary insights

(Kohler et al., 1999). Researchers use different behavioural theories to underpin

evaluations, depending on the specifics of the interventions, they investigate. The

presented review explored some of the most widely applied theories in the social and

behavioural sciences literature. However, given the characteristics of the current PhD

program of research, not all of them were found suitable to guide the present research

evaluation framework.

Building on the wealth of existing TPB literature, an extended TPB (Ajzen,

1988) framework was considered most suitable to inform the evaluation of the two

interventions (see Figure 3.5), part of this PhD program of research, where the

influence of two COTs on young drivers' intention and behaviour was investigated in

regards to reducing speeding and DUI. Both interventions provided an opportunity to

collect, analyse and compare self-report (through surveys) and limited observational

(through a chosen smartphone app leaderboard) data. Analysing the collected data

through the extended TPB framework could both address known TPB limitations and

provide insights on changes in the participants' driving intention and behaviour.

Examining TPB constructs and additional predictors could build a complete picture of

which construct accounted for changes in the participants’ intentions and self-reported

behaviours, and where the intervention had a greater effect.


Figure 3.5. Extension of the Theory of Planned Behaviour in the current program of research.

Overall, the extended TPB (Figure 3.5) was considered a potential good fit for

the present PhD program of research with respect to a better understanding of young

drivers' salient beliefs towards speeding and DUI. The following research design

chapter provides details on how this extended TPB was operationalised in terms of

surveys, variables and applied analysis.

42 Chapter 4: Research design

Chapter 4: Research design

Chapter 4 describes the research design methodology that was adopted to

achieve the aim and objectives of this program of research. The current thesis aimed

to close the identified research gap (see Section 2.5) by examining the effects of two

examples of COTs, a smartphone safe-driving app and VR simulations of risky

driving. As discussed earlier, the research investigation pursued three key objectives

to accomplish the aim:

1. Understand to what extent the use of those two COTs is associated in the

literature with safety benefits for young drivers;

2. Identify an example of each of two COTs that could potentially persuade

young drivers to adopt safer on-road behaviour; and

3. Investigate the extent to which such behavioural change happened, as a result

of two real-world interventions with the two example COTs as intervention tools.

First, the Chapter defines the research questions, answered later in the thesis to

address those three objectives. Then, it provides details on establishing theory-based

selection criteria. Finally, a general overview of the adopted research methodology

within the current program of research is presented before discussing in more detail

the two intervention studies’ design.

4.1 RESEARCH QUESTIONS

The identified research gap (see Section 2.5) called for an investigation of the

real-world impact of COTs that are readily available to young drivers to experience,

such as smartphones safe-driving apps and VR simulations of risky driving. The

limited knowledge about their safety benefits required a systematic approach to start

closing the gap.

Building on the generated knowledge from the literature, focusing on using safe-

driving apps in research (see Subsection 2.4.2), a need for a more in-depth systematic

review was identified to explore research question one:

RQ1. What is the state of the art evidence of the safety benefits of smartphone

safe-driving apps for young drivers?

Chapter 4: Research design 43

A large number of smartphone safe-driving apps is available online today. A

larger number of smartphone safe-driving apps exist when taking into consideration

those available outside app stores, e.g. when developed for research purposes only.

The wide availability of smartphone safe-driving apps is contrasted against limited

scientific evidence for a positive impact on their users in real life. Unfortunately,

despite this lack of evidence, smartphone safe-driving apps’ developers may be

tempted to claim that such positive impact exists as part of their marketing efforts, e.g.

based on monitoring driving behaviour through the app itself, which can include

detection of risky events, such as speeding. Overall, the fact that speeding plays such

a significant role in young drivers' safety, the rapid increase in the number of available

safe driving apps, and the limited evidence about these apps' safety impact in a free-

living environment, led to research question two:

RQ2. How do young drivers’ self-reported behaviour of not speeding and

intention not to speed alter in their free-living environment, as a result of exposure to

a smartphone safe-driving app intervention?

Additionally, the adopted extended TPB framework (see Section 3.7) allowed

exploring in-depth the two constructs of interest, i.e. self-reported behaviour of not

speeding and intention not to speed. Such a focus could potentially supply more

detailed information on the effect the smartphone safe-driving app intervention had,

e.g. the influence of which constructs could be explored further to increase the impact.

Assessing self-reported behaviour of not speeding and intention not to speed would

allow determining 1) what the young drivers planned to do in the absence of an

intervention, 2) how the intervention impacted them, and 3) was it possible to predict,

with the information, available for the drivers before they were subjected to an

intervention, how they actually behaved after receiving the intervention. Thus, the

following secondary questions were addressed in the case of the smartphone safe-

driving app intervention:

RQ2.1. What did we know about the participants before the intervention, and

to what extent could the extended TPB framework predict their intention not

to speed?

RQ2.2. Did the intervention change the participants' salient beliefs, as

depicted by the TPB constructs?


RQ2.3. Using the extended TPB framework, to what extent could the data

available before the intervention predict the participants' behaviour of not

speeding during the intervention?

RQ2.4. How did the intervention influence the participants’ engagement with

their smartphones?

Investigating the above research questions further our understanding to what

extent the use of one smartphone safe-driving app may be associated with safety

benefits for young drivers. This understanding can potentially provide support for

using such apps in the framework of novel technology-based approaches to reduce

road trauma.

As part of this research's aim to examine COTs more broadly, another COT was

examined, Virtual Reality (VR). VR is less widely spread, compared to smartphones,

but technology advancements and sinking prices are leading to continued proliferation.

In the road safety context, VR (see Subsection 2.4.3) provides a unique opportunity to

simulate potentially life-threatening situations in a safe environment. While

experiencing VR, users can perform designated tasks, which may assist them in taking

a different perspective through the simulated experience. As discussed earlier, such

shifts in the individual perspective-taking may lead to behavioural change. Due to the

novelty of the technology, however, there is limited knowledge about VR's potential

to deliver safety benefits in parallel with such behavioural change, which led to

research question three:

RQ3. How is VR applied in road safety research to motivate behavioural change

in young drivers?

In the light of the limited knowledge about safety benefits from using VR

technology applications, ongoing global interventions utilise VR to raise awareness on

the risks of DUI, another behaviour with a high negative impact on young drivers (see

Section 2.1). At the same time, the added value of such VR simulations of risky driving

as a tool in those real-world interventions had not yet been evaluated. Therefore,

research question four was:

RQ4. How do young drivers’ self-reported behaviour of not DUI and intention

not to DUI alter in their free-living environment as a result of a VR intervention?


Similar to the case with the smartphone safe-driving app intervention, the

adopted extended TPB framework allowed exploring in-depth self-reported behaviour

of not DUI and intention not to DUI through the following secondary questions:

RQ4.1. What did we know about the participants before the intervention, and

to what extent could the extended TPB framework predict their intention not

to DUI?

RQ4.2. Did the intervention change the participants' salient beliefs, as

depicted by the TPB constructs?

RQ4.3. Using the extended TPB framework, to what extent could the data

available before the intervention predict the participants' behaviour of not

DUI after the intervention?

Overall, it was argued that, grounded in the extended TPB evaluation

framework, the two COTs-based interventions, leveraging a safe-driving app and VR

simulations of risky driving, could reduce young drivers’ speeding or DUI,

respectively. Therefore, by separately answering the research questions formulated

above, the following overarching research question was assessed:

How do COTs-based interventions influence young drivers?

This overarching research question is aligned with the overall aim of this PhD

thesis. The research questions, defined above, align with individual studies and the

overall adopted methodology, as discussed in the following sections.

4.2 THE EXTENDED TPB AND SELECTING INTERVENTION TOOLS

To be able to influence people, i.e. change behaviour, Ajzen (2006) suggests that

interventions be designed towards influencing one or more of its predictors. However,

Ajzen (2006) does not specify what interventions (e.g. face-to-face, media campaign

or others) may deliver that influence. Fife‐Schaw, Sheeran, and Norman (2007) agree

that TPB is silent in guiding appropriate strategies to influence its basic constructs.

The intervention intended to influence the young participating drivers by using

two COTs as tools, a smartphone safe-driving app in Study 2 and a VR simulation of

DUI in Study 4. Even though there is a lack of specific strategic guidance, the adopted

extended TPB evaluation framework (see Section 3.7) can still serve as a starting point

in establishing criteria for selecting examples of such COTs. Such criteria will


explicitly identify which of the extended TPB underlying constructs could potentially

be influenced by the two implemented COTs-based interventions.

An intervention cannot change all extended TPB constructs. Thus, the current

section focuses on 1) what variables in the extended TPB model can be influenced, 2)

how a decision can be made whether the respective COT can influence them, and 3)

which of them have been shown to be most influential in predicting the targeted driver

behaviours.

Regarding the first point (what), it is noted that no intervention can change

demographic variables. Furthermore, Personality characteristics, such as impulsivity

and sensitivity, are perceived as relatively stable (Scott-Parker, 2012). Thus, they are

not seen as potentially changeable in the short term, either. Ajzen (2006) suggests that

interventions should focus on salient beliefs as they are readily accessible and might

be influenced. Thus, the 6 potentially changeable constructs (attitude, norms, PBC,

moral norm, peers' norm and perceived risk) can be used to provide guidance on the

selection of COTs within the current extended TPB framework.

Regarding the second point (how), the lack of TPB insights on what could

potentially be successful in influencing its constructs (Ajzen, 2006; Fife‐Schaw et al.,

2007) leave the adopted constructs' definitions themselves (see Section 3.5 and Section

3.6) as a starting point of establishing selection criteria for COTs (see Table 4.1). Those

selection criteria were formed as binary questions, directly pointing at the respective

construct's definition. A positive (yes) answer carries 1 point. A negative (no) answer

does not carry points. The number of positive answers determines the relevant COT

ranking. The selection criteria were applied in Chapter 6 and Chapter 9.

Elaborating on the third point (most influential), the selection criteria might

consider not only the constructs but also their potential to mediate the desired changes.

The literature agrees that intention is a stronger predictor of behaviour than PBC

(Armitage & Conner, 2001; Fife‐Schaw et al., 2007; Hagger & Chatzisarantis, 2005).

However, the evidence around predicting intention is mixed. Norms are generally seen

as having lower predictive power in TPB than attitude and PBC (Armitage & Conner,

2001; Fife‐Schaw et al., 2007; Hagger & Chatzisarantis, 2005). Fife‐Schaw et al.

(2007) point out attitude as typically having the strongest predictive value. However,

assumptions on predictive value should better be made in regards to the specific


behaviour of interest. Those behaviours in the case of the current program of research

are speeding and DUI.

Table 4.1. COTs' selection criteria, derived from the extended TPB framework.

Selection

criteria (SC) N Construct Definition Question

SC1 Attitude

How the driver sees the

behaviour, favourable or

unfavourable.

Does the COT help the driver better

see whether their behaviour on the

road is favourable or unfavourable?

SC2 Norms

Whether the driver's important

referents would approve or

disapprove their engagement in a

particular behaviour.

Does the COT provide information on

how the driver's important referents

see their behaviour?

SC3 PBC

How easy, or difficult, the driver

perceives performing the

behaviour.

Does the COT improve the driver's

ability to perform the behaviour?

SC4 Moral norm Whether the driver perceives the

behaviour as "normal".

Does the COT help the driver

understand the morality of their

behaviour?

SC5 Peers' norm

Whether the driver's peers are

seen as disapproving or approving

of the respective participant

engaging in the behaviour.


how the driver's peers perform the

behaviour?

SC6 Perceived

risk

Whether the driver perceives a

risk of being involved in a crash

or being caught by the police

while performing a specific

behaviour.


the possibility the driver to be

involved in a crash or to be caught by

the police while performing the

behaviour?

A review of TPB studies focused on speeding indicates that PBC, and not

attitude, is the strongest predictor of intention (Elliott and Thomson, 2010; Stead et

al., 2005; Warner & Åberg, 2008). While Stead et al. (2005) and Elliott and Thomson

(2010) provide support for attitude having the second strongest value and confirm

norms as having the lowest contribution, Warner and Åberg (2008) found the reverse.

In DUI studies, PBC was found as the strongest predictor by Moan and Rise (2011)

and Potard et al. (2018), followed by norms and then attitude. Chan, Wu, and Hung

(2010) provide evidence that attitude has the strongest influence on intention, followed

by PBC and norms. Given that typically PBC is the strongest predictor of intention in


the case of speeding and DUI, interventions may focus on using COTs that are

expected to influence PBC. Thus, a positive answer to the PBC-related question should

be weighted more highly when using the questions as criteria for selecting the most

promising COT.

4.3 METHODOLOGY

This PhD program of research used a mixed-methods design, consisting of both

qualitative and quantitative approaches (Figure 4.1). Similar research tracks were used

to expand the knowledge around the real-world effects of using both COTs,

smartphone safe-driving apps and VR simulations of risky driving, as intervention

tools. They were structured as follows:

As the first step within each track, Study 1 and Study 3 followed PRISMA,

evidence-based guidelines on a minimum set of items for systematic reviews reporting.

The two studies investigated the two COTs’ characteristics and their effects on young

drivers' behaviour and safety, as evidenced by the available literature. Both systematic

reviews confirmed the research gap, identified in Section 2.5. The systematic reviews

identified limited knowledge about safety benefits in the real world, delivered to young

drivers by smartphone safe-driving apps in respect of reducing speeding, and by VR

simulations of risky driving in respect of reducing DUI.

The second step focused on selecting suitable examples of the two COTs, a

smartphone safe-driving app and VR simulations of risky driving, to be used as

intervention tools. The established theory-based selection criteria (see Section 4.2)

were used to evaluate potential candidates. Chapter 6 enriched the knowledge,

generated by Study 1, with insights from a Focus group and a systematic review of

smartphones apps stores, to inform the choice of a safe-driving app. As a result, Flo

was selected as a good fit to be assessed within this specific program of research. At

the same time, VR apps stores are not as richly stocked as the smartphones apps stores.

Road safety VR software for ordinary consumers is still not available. Thus, selecting

VR software as an intervention tool was not performed systematically. Rather, based

on convenience, the software "3D Tripping" was obtained from its developers to be

deployed as an intervention tool in the framework of the current program of research.


RESEARCH TRACKS

SAFE-DRIVING APPS VIRTUAL REALITY

Step 1 Study 1 (Chapter 5) Study 3 (Chapter 8)

Systematic Review of Safe-driving Apps Systematic Review of VR

RQ1: What is the state of the art evidence of

the safety benefits of smartphone safe-driving

apps for young drivers?

RQ3: How is VR applied in road safety

research to motivate behavioural change in

young drivers?

METHOD: Systematic Review (adhering to

the PRISMA guidelines)

METHOD: Systematic Review (adhering to

the PRISMA guidelines)

ANALYSIS: Narrative ANALYSIS: Narrative

Step 2 Chapter 6

Selecting a safe-driving app

Step 3 Study 2 (Chapter 7) Study 4 (Chapter 9)

Intervention with an off-the-shelf smartphone

safe-driving app

Intervention with VR simulations of risky

driving

RQ2: How do young drivers’ self-reported

behaviour of not speeding and intention not to

speed alter in their free-living environment, as

a result of exposure to a smartphone safe-

driving app intervention?

RQ4: How do young drivers’ self-reported

behaviour of not DUI and intention not to DUI

alter in their free-living environment as a result

of a VR intervention?

METHOD: Intervention with cross-sectional

pre- and post-surveys, complemented by

driving scores from safe-driving app

leaderboard.

METHOD: Intervention with cross-sectional

pre- and post-surveys.

ANALYSIS: Hierarchical multiple regression,

Analysis of covariance.

ANALYSIS: Logistic regression, McNemar's

test, Chi-square test for independence,

Wilcoxon Signed Ranks Test.

Figure 4.1. Outline of the overall thesis methodology

The third and final step was to investigate how the selected two examples of

COTs influenced young drivers in their free-living environment as part of two

interventions. Each intervention leveraged one of the two COTs. Surveys were

administered before the interventions took place to establish a baseline on all repeated

measures. A second self-reported set of data was collected from the participants


approximately three months after the first survey. Collecting data at two time points

allowed assessing the long-term effects of the two interventions, as depicted by those

repeated measures. As a result, the two intervention studies were designed to follow a

similar process. The section below provides details on this design, which was

developed before the actual interventions took place, i.e. at the point of seeking ethical

clearance.

4.4 INTERVENTION STUDIES’ DESIGNS

The common features of the studies at step three of the two research tracks are

described in the following subsections of this chapter. Details on their actual

implementation as well as on the obtained results that are specific for each separate

study, such as design, participants and procedures, are later described in separate

sections in the respective chapters. This separated reporting avoids repetition in the

respective chapters due to the common features.

4.4.1 Participants

Study 2 and Study 4 involved participants. A short overview is provided below

with more details included in Chapter 7 and Chapter 9.

Study 2 and Study 4 tested the predictability of the TPB constructs and the

additional variables through regression. A higher number than advised by statistical

texts was targeted in anticipation for possible dropout during the second data

collection. Tabachnick and Fidell (2007) specify that n=>104 + m, where m is the

number of predictors, is needed so that separate regressions can be conducted in each

setting.

In order to participate in Study 2 or in Study 4, a participant had to be aged 18

to 25 with a valid driver's license. In Study 2, an additional criterion was that the young

drivers had to drive a car a minimum of 100 kilometres per month, and use a car as the

only means of transport (to avoid collecting data that originates in travelling by means

other than a car). The target number of participants in Study 2 was 140. However, as

the study methodology allowed for a larger number of participants, which would have

increased the statistical significance of the obtained results, more participants were

expected to be recruited. Thus the potential number of participants was capped at

1,000. In Study 4, the additional criteria "Have no history of seizures or epilepsy" was

used when recruiting the Intervention group due to a number of potential risks


associated with the use of the VR headset while seated in a static position. Initially,

the target number in Study 4 was set for a minimum of 200 participants per condition

to anticipate potential dropouts.

Participants in both Study 2 and Study 4 were expected to perform self-screening

if they meet the inclusion criteria to participate before they consent and complete a

survey. Consent was required for all participants. It was specific to the respective study

and was implied, i.e. such was considered obtained after a participant went through the

study information sheet, generated their anonymous identifier, according to a

predefined formula, and started completing the survey. An anonymous identifier was

generated by the respective participants themselves, as per a predefined formula, and

included: day of birth, first letter of first name, first letter of family name and last two

digits of mobile number (example 24DL08). Those anonymous identifiers were used

in both interventions to connect data sets for the same participants from the two times

they completed the respective surveys.

The participants' e-mail addresses were kept so that they can be contacted to

complete the second survey, approximately three months after the first one. However,

the records linking their anonymous identifier with their e-mails were destroyed after

the first time they completed a survey. After destroying the link, the anonymous

identifiers were the only means to connect datasets that originated from the same

person. Only through those datasets, upon destruction of the link, it was impossible for

the person to be identified.

Participants in Study 2 and in the Control group of Study 4 were recruited

through Facebook. Participants in the Intervention group of Study 4 needed to be

present in person to experience the VR simulation of DUI. Thus, convenience

sampling was used for their recruitment through live events. Two tablets with Internet

connection were available for them to complete the survey before they were admitted

to operate the VR driving simulator. A detailed description of the Study 2 recruitment

process is provided in Section 7.2. Section 9.2 provides details on the Study 4

recruitment process.

4.4.2 Surveys

The current project interventions were evaluated through an extended TPB

theoretical framework (see Figure 3.5) to overcome some TPB limitations (see Section


3.5) relevant to the current program of research. Following Elliott and Thomson's

(2010) work, the current PhD program of research assessed potential separate effects

of dichotomised standard TPB constructs (attitude, subjective norm, and PBC, which

were defined in Section 3.5). For the purpose of the current PhD program of research,

attitude was dichotomised into instrumental attitude (cognitive), determined by how

the driver sees the behaviour, e.g. good or bad, and affective attitude (emotional),

determined by how they feel about the behaviour, e.g. will they enjoy the behaviour or

choose to not perform the behaviour, etc. Subjective norm, referring to whether

individual's important referents would improve or disapprove their engagement in a

particular behaviour, and descriptive norm, or whether those important referents are

thought to perform the behaviour themselves, were explored separately. Self-efficacy,

i.e. the individual's ability to perform the behaviour, and perceived controllability, i.e.

whether the environment would constraint or provide an opportunity for the behaviour

to be performed, were assessed as separate predictors, too.

A review of the literature provided support for such a decision. For example,

affective attitude (Lawton, Parker, Manstead, & Stradling, 1997; Rhodes & Pivik,

2011) and instrumental attitude (Elliott & Thomson, 2010) can be distinguished as

separate predictors of intention. A meta-analysis of twenty-one hypotheses (total

sample N = 8097) showed both descriptive norm and subjective norm as statistically

significant independent predictors, too (Rivis & Sheeran, 2003). Self-efficacy and

perceived controllability are empirically separable (Ajzen, 2002), and they are used in

such a manner in road safety research, e.g. in Horvath et al. (2012). These reports

corroborate the approach of using split components of the standard TPB measures,

which in turn allowed the establishment of readily distinguishable causal effects.

The evaluation of the interventions within this PhD program of research utilised

additional predictors in addition to TPB while assessing for a behavioural change that

might have been triggered by the intervention. Changeable in the short-term

constructs, past behaviour (Conner & Sparks, 2005; Elliott & Thomson, 2010; Gauld,

Lewis, White, & Watson, 2016; Haque et al., 2012), moral norm (Conner & Sparks,

2005; Elliott & Thomson, 2010; Manstead, 2000), peers' norm (Conner & Sparks,

2005; Fleiter et al., 2006) and perceived risk (Gannon, Rosta, Reeve, Hyde, & Lewis,

2014; Haque et al., 2012; Rhodes & Pivik, 2011), previously being utilised to extend

TPB, were explored together with more stable personality characteristics (impulsivity,


sensitivity to punishment and sensitivity to reward), previously found to be relevant to

young drivers (Scott-Parker, 2012). Thus the rational reasoning, on which TPB is

criticised to be exclusively focused (Sniehotta et al., 2014), was complemented by

measuring potential unconscious influences (Sheeran et al., 2013), which originated in

people's distinctive personality.

Both interventions employed online surveys for the participants. The surveys

consisted of three parts (see Appendices C and D for the complete surveys) appearing

in fixed order both at Time 1 (before the intervention) and Time 2 (after the

intervention) (see Table 4.2). The surveys at Time 1 and Time 2 of each intervention

were offered approximately three months apart in both interventions. Three months

were considered enough for the participants to have had the opportunity to perform the

risky behaviour of interest (Bingham et al., 2011). Each time the survey was expected

to take approximately 10 minutes to complete.

Part 1 of the surveys, demographic data, contained 4 items in Study 2 and 3 items

in Study 4: age, gender, type of driver's license (in Study 2) or driving experience (in

years, in Study 4) and state of residence (measured in Study 2 only). Demographic

data in Study 4 were collected at Time 1 only. At Time 2 in Study 4, the Intervention

group participants, only, were asked what did they choose to experience when driving

the driving simulator with "3D Tripping" VR software: alcohol, ecstasy, magic

mushrooms or cannabis.

Part 2 of the survey contained 16 repeated-measure items in Study 2 and 13 in

Study 4. Single items were used to measure theory components to maximize response

rate and minimise fatigue biases (Hart, Rennison, & Gibson, 2005) in the domains of

speeding (Study 2) and DUI (Study 4). 11 items were adapted from Elliott and

Thomson (2010) to measure TPB variables as well as the additional predictors: past

behaviour, moral norm and peers' norm. Detailed information on the level of

adaptation is provided in Table 4.3 (see Subsection 4.4.3). Elliott and Thomson (2010)

developed and validated their scales through a study involving 1403 participants. The

participants were English drivers, within the age range 18 to 91, caught for speeding,

up to four months before the study. Internal consistency (Cronbach’s α) ranged from

0.84 to 0.97 for the different scales. The other additional predictors were two items

adapted from Gannon et al. (2014) to measure perceived risk and three items borrowed

from Gauld et al. (2016), and used only in Study 2, to measure smartphone use to


examine whether the smartphone safe-driving app did not increase distraction amongst

the participants.

Table 4.2. Constructs and time of measurement.

Construct No. of

items

Study 2

Measured at...

Study 4

Measured at...

Source

Time

1

Time

2

Time

1

Time

2

Demographic data 4 / 3 ✓ ✓ ✓

N.A. VR experience 1 ✓

Intention 2 ✓ ✓ ✓ ✓

Adapted from Elliott and

Thomson (2010)

Attitudes 2 ✓ ✓ ✓ ✓

Norms 4 ✓ ✓ ✓ ✓

PBC 2 ✓ ✓ ✓ ✓

Past behaviour 1 ✓ ✓ ✓ ✓

Risk perception 2 ✓ ✓ ✓ ✓ Adapted from Gannon et al.

(2014)

Smartphone use 3 ✓ ✓ Gauld et al. (2016)

Impulsivity 30 ✓ ✓ Patton and Stanford (1995)

Sensitivity to

reward

24 ✓ ✓

Torrubia, Ávila, Moltó, and

Caseras (2001) Sensitivity to

punishment

24 ✓ ✓

Part 3 of the survey was different at Time 1 and Time 2. It explored the

participants' personality characteristics. As those were not expected to change in the

short term, the related questionnaires were administered only once. At Time 1, data

was collected on the participants' impulsivity, using the 30-item Barratt Impulsiveness

Scale Version 11 (BIS-11) on 4-point Likert scales (Patton & Stanford, 1995). Internal


consistency (Cronbach's α) of BIS-11 varies between .79 and .83 (Patton & Stanford,

1995). At Time 2, data was collected on the participants' sensitivity to punishment and

sensitivity to reward, using the 48-item Sensitivity to Punishment and Sensitivity to

Reward Questionnaire (SPSRQ) in yes/no format (Torrubia et al., 2001). SPSRQ

returns two scores for each individual as a result. One represents their sensitivity to

punishment and the other their sensitivity to reward. The higher the score, the more

sensitive the person is. Torrubia et al. (2001) reported a sensitivity to punishment

Cronbach's α of .82 for females and .83 for males, and a sensitivity to reward

Cronbach's α of .75 for females and .78 for males.

4.4.3 Variables

Participants' speeding (Study 2) or DUI (Study 4) intention and past behaviour

were used both as outcome variables (dependent variables, DVs) and as predictors

(independent variables, IVs). Thus, they are included in the following description of

the predictor variables.

TPB items, adapted from Elliott and Thomson (2010) (see Table 4.3 for the

adaptations with the changes being underlined), were used to measure the TPB

constructs using 9-point scales (scored 1–9). Intention to perform the behaviour of

interest, speeding or DUI, (2 items) focused on 1) what the driver believes will happen

and 2) what the driver believes is possible to happen. Attitudes towards the behaviour

of interest (2 items) focused on drivers' 1) cognitive (instrumental) attitude and 2)

emotional (affective) attitude. Norms (2 items) explored 1) what would people, who

are important to the driver, think of them violating the behaviour of interest (subjective

norm) and 2) what those people will do themselves according to the driver (descriptive

norm). PBC (2 items) explored 1) the drivers' own view of their abilities (internal

factors, self-efficacy) and 2) whether it would be possible for them to use those abilities

in real life (external factors, perceived controllability). Example question: To what

extent do you intend to drive faster than the speed limit over the next 3 months?

(Possible answers: No extent at all (1) to A great extent (9)).

Some of the additional predictors used in the surveys were also adapted from

Elliott and Thomson (2010) (see Table 4.3 for the adaptations with the changes being

underlined) and measured on 9-point Likert scales (scored 1–9). Past behaviour,

speeding (in Study 2) or DUI (in Study 4), explored the drivers' recall of their actual

behaviour in the recent three months. Two items were used to measure additional


norms: 1) what would drivers' peers, in particular, think of the driver behaviour (peers'

norm); and 2) the drivers' understanding towards their own behaviour (moral norm).

Example question: Would your friends disapprove or approve of you driving under the

influence of alcohol or drugs over the next 3 months? (Possible answers: Definitely

disapprove (1) to Definitely approve (9)).

Table 4.3. Items, adapted from Elliott and Thomson (2010).

Construct Original item Item used in Study 2 Item used in Study 4

Intention

To what extent do you

intend to drive faster than

the speed limit over the next

6 months?


intend to drive faster than


3 months?


intend to drive under the

influence of alcohol or drugs

over the next 3 months?

Intention

How often do you think you

will drive faster than the

speed limit in the next 6

months?


will drive faster than the

speed limit in the next 3

months?


will drive under the


in the next 3 months?

Instrumental

attitude

How bad or good would it

be for you personally if you

drove faster than the speed

limit over the next 6

months?



drove faster than the speed


months?



drove under the influence of

alcohol or drugs over the

next 3 months?

Affective

attitude

How unenjoyable or

enjoyable would it be for

you personally if you drove

faster than the speed limit


How unenjoyable or





How unenjoyable or



under the influence of


next 3 months?

Subjective

norm

Would the people who are

important to you disapprove

or approve of you driving













next 3 months?

Descriptive

norm

How often do you think the

people who are important to

you will drive faster than


6 months?



you will drive faster than


3 months?



you will drive under the



Self-efficacy How confident are you that

you will be able to avoid

How confident are you that


How confident are you that



driving faster than the speed


months?

driving faster than the speed


months?

driving under the influence

of alcohol or drugs over the

next 3 months?

Perceived

controllability

Over the next 6 months,

how much do you feel that

avoiding driving faster than

the speed limit is under

your control?

Over the next 3 months,

how much do you feel that

avoiding driving faster than

the speed limit is under

your control?

Over the next 3 months, how

much do you feel that

avoiding driving under the


is under your control?

Moral norm

How wrong do you think it

would be for you to drive











next 3 months?

Past speeding

behaviour

How often did you drive


over the last 6 months?



over the last 3 months?



alcohol or drugs over the last

3 months?

Peers' norm






Would your friends

disapprove or approve of

you driving over the speed


months?

Would your friends

disapprove or approve of

you driving under the



Perceived risk (2 items), adapted from Gannon et al. (2014) (see Table 4.4 for

the adaptations with the changes being underlined), was measured on 9-point scales

(scored 1–9) focusing on 1) how the drivers' perceived the risk of being involved in a

road crash and 2) whether they worry about being caught by the Police. Example

question: If you were to drive over the speed limit over the next 3 months, how much

would you worry about being involved in a road crash? (Not at all worried to worried

very much).

Smartphone use (3 items from Gauld et al. (2016) used only in Study 2) was

explored with a focus on the drivers' behaviour with respect to 1) initiating (less)

communication; 2) monitoring/reading (less) communication; and 3) responding (less)

to communication, on a 7-point scale where each score was related to a certain

frequency of use (more than once per day; daily; 1–2 times per week; 1–2 times per

month; 1–2 times per 3 months; once a year; never). Example question: How often do

you do the following on your smartphone while driving: Initiate communication on


social interactive technology? (Starting a communication) (Possible answers: More

than once per day; Daily; 1–2 times per week; 1–2 times per month; 1–2 times per 3

months; Once a year; Never).

Table 4.4. Items, adapted from Gannon et al. (2014).

Construct Original item Item used in Study 2 Item used in Study 4

Perceived

risk

If you were to drink

walk, how much

would you worry

about being involved

in a road crash?

If you were to drive over the

speed limit over the next 3

months, how much would

you worry about being

involved in a road crash?

If you were to drive over the next

3 months under the influence of

alcohol or drugs, how much

would you worry about being

involved in a road crash?

If you were to drink

walk, how much

would you worry

about being involved

in a road crash?

If you were to drive over the

speed limit over the next 3

months, how much would

you worry about being

caught by the Police?

If you were to drive over the next

3 months under the influence of

alcohol or drugs, how much

would you worry about being

caught by the Police?

All items described above were repeated-measures, thus used both at Time 1 and

Time 2. Data on personality characteristics, used as additional predictors in the

analysis, were collected only one time per intervention.

BIS-11 (Patton & Stanford, 1995) was used at Time 1 of the interventions to

measure impulsivity as a construct. BIS-11 is a 30 item self-report tool using 4-point

Likert scales (1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost

Always/Always)). Example question: I plan tasks carefully. (Possible answers: 1

(Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)).

SPSRQ (Torrubia et al., 2001) replaced BIS-11 at Time 2 of the interventions as

part 3 of the survey to measure the drivers' sensitivity to punishment and sensitivity to

reward. SPSRQ is a 48 item self-report tool, using a "yes/no" format, with 24 items

focusing on each of the two constructs. Example question: Do you often refrain from

doing something because you are afraid of it being illegal? (Possible answers:

Yes/No).

4.4.4 Analyses

Data from study participants was collected through Google forms in Study 2 and

QUT Key Survey software in Study 4. The data from the driver surveys were checked,


coded and entered into Statistical Package for the Social Sciences (SPSS Statistics 25).

A record was kept on data coding and scales recoding. Due to the predominant use of

closed questions, there was limited missing or invalid data. Descriptive statistics for

all variables were examined. The data were checked for outliers. Outliers did not

appear to have an overly influential impact on the analyses.

The data were checked to test for assumptions of parametric tests, including

normality, linearity, homoscedasticity, homogeneity of variance and multicollinearity.

Inspection of standardised residual scores, scatterplots, skewness and kurtosis values,

95% trimmed means, visual inspections of histograms and Shapiro-Wilk statistics

revealed that assumptions were sufficiently met in Study 2. Thus, parametric tests were

used (one-way between-groups multivariate analysis of variance (MANOVA), 3-step

hierarchical multiple linear regression, one-way and two-way analyses of covariance

(ANCOVA)) to analyse the data in Study 2. Initially, the analysis of Study 4 was

intended to follow the same process as Study 2. One-way between-groups MANOVA

was used to preliminary assess the data. The Study 4 DVs' data had significant

deviations from normality and was, therefore, dichotomised. Thus, non-parametric

tests were used to analyse the data (3-step multiple logistic regression, Chi-square test

for independence, McNemar's test, Wilcoxon Signed Ranks Test). Further details

about the performed analyses are provided in the relevant sections of Chapter 7 and

Chapter 9.

Despite the different set of tests, used in Study 2 (parametric) and Study 4 (non-

parametric), regression analysis was used to analyse the data from both studies. Since

data collection used measures, developed by Elliott and Thomson (2010), data analysis

was guided by their methodology, too. Elliott and Thomson (2010) not only

reconceptualised the standard TPB by dichotomizing the original constructs but also

retained their separately defined components when assessing their predictive strength.

Thus instrumental attitude and affective attitude were used in the models instead of a

single attitude measure. Separate subjective norm and descriptive norm were used

instead of a subjective norm only. Finally, self-efficacy and perceived controllability

were explored separately as part of PBC. More details on the undertaken approach are

provided in Section 3.7 above.

The difference between the current analysis and the approach of Elliott and

Thomson (2010) is that demographic factors were included in the analysis before TPB


at Step 1. That was done to account for the influences of gender, age and driving

experience (denoted by driving license in Study 2), an approach guided by Horvath et

al. (2012). This allowed for a more accurate reflection of the standard TPB constructs'

predictive ability, assessed at Step 2, over and above the demographics. The predictive

ability of the additional predictors (past behaviour, perceived risk, moral norm, peers'

norm, impulsivity, sensitivity to reward and sensitivity to punishment), over and above

the TPB variables, was investigated at Step 3.

4.4.5 Ethics

Study 2 and Study 4 involved the participation of consenting adult participants.

Ethical clearance was obtained from the QUT Human Research Ethics Committee. In

Study 2, the participants completed two online surveys and used a safe-driving app

while driving (QUT Ethics Approval Number 1700000846). In Study 4, the

participants completed two online surveys and drove a VR driving simulator (QUT

Ethics Approval Number 1800000214).

Chapter 5: Study 1 - Systematic review of safe-driving apps 61

Chapter 5: Study 1 - Systematic review of safe-driving apps

This Chapter 5 presents the first study of the program of research. The study

investigated to what extent smartphone safe-driving apps were previously explored in

research and whether any safety benefits were reported. It addressed RQ1:

What is the state of the art evidence of the safety benefits of smartphone safe-

driving apps for young drivers?

Following a PRISMA design (Liberati et al., 2009), Study 1 reviewed the

available literature, in terms of safe-driving apps' characteristics and effects on young

drivers' behaviour and safety. First, the search results from the two used databases are

presented, followed by key findings and a discussion.

5.1 RATIONALE FOR CONDUCTING A SYSTEMATIC REVIEW

The risk of road crashes appears to be higher for young drivers than the risk for

older drivers (McKnight & McKnight, 2003; SafetyNet, 2009). The increased risk of

crash involvement manifests itself into young drivers being overrepresented in road

fatalities. For example, young Australian drivers represent 12% of the population, but

19% of the driver fatalities (BITRE, 2018). Five risky behaviours (speeding, DUI, not

wearing a seatbelt, fatigue and distraction) as pointed out as main contributors to that

statistics (Scott-Parker & Oviedo-Trespalacios, 2017).

At the same time, young drivers embrace technology, and researchers explore

paths to use COTs to reduce the young drivers' risky behaviours (Lee, 2007; Schroeter

et al., 2012; Steinberger et al., 2017). Safe-driving apps are an example of such COTs

that are regarded as having a potential to positively influence the young drivers

(Castignani, Derrmann, Frank, & Engel, 2017; Musicant & Lotan, 2015; You et al.,

2013). However, a small proportion of the existing literature provides knowledge

about the effects of safe-driving apps in the real world. There are also differences

between how safe-driving apps are operationalised for research purposes, i.e. different

study designs, different measurement, and different outcomes. These inconsistencies

made it difficult to make conclusions about the effects, stemming from using safe-

62 Chapter 5: Study 1 - Systematic review of safe-driving apps

driving apps to reduce young drivers' risky behaviours. Furthermore, no other

systematic review looks systematically into the literature for evidence for such effects.

5.2 METHOD

The PRISMA guidelines were used as a framework for the current systematic

review. A review protocol was developed following the guidelines with the following

steps:

1. Development of the research question.

2. Identification of search databases.

3. Definition of scope, inclusion, and exclusion criteria.

4. Definition of a search term.

5. A systematic search for information.

6. Screening and selection of studies (see PRISMA flowchart as Figure 5.1).

7. Review of selected articles.

8. Summarising of findings.

5.2.1 Search databases

Relevant papers were identified through searches in Transport Research

International Documentation (TRID, https://trid.trb.org) and Scopus

(https://www.scopus.com). Both databases are widely used for PRISMA-guided

systematic reviews in the context of road safety (Oviedo-Trespalacios, Truelove,

Watson, & Hinton, 2019; Shekari Soleimanloo, 2016; Staton et al., 2016).

5.2.2 Literature search criteria

Papers published from 2010 onwards, and written in English, were considered

for inclusion in the review. The search results were limited to the year 2010 as a

starting point to cover the time contemporary smartphone apps started to appear on the

consumer market. This trend is indicated by the sharp increase of public interest in

apps (see Google Trends) and the word "app" being named "Word of 2010"3. The

actual and potential application and utility of smartphone apps, games and

gamification in (a) road safety research more generally, (b) road safety practice more

generally, (c) young drivers road safety research specifically, and (d) young drivers

3 https://www.americandialect.org/app-voted-2010-word-of-the-year-by-the-american-dialect-society-updated


road safety practice specifically, was put in focus. Games and gamification are

regarded in the literature as potential routes to boost drivers’ engagement (see

Subsection 2.4.2). Thus, the two terms were used as complementary search terms to

smartphone apps to avoid potential omission of studies that focus on the games and

gamification aspects, rather than on the fact that those aspects are delivered through a

smartphone app. As a result, papers exploring active use of apps, games and

gamification on the driver's smartphone, with or without feedback influencing their

driving behaviour, were explored. Papers not relevant to driving, mobile phone use

literature reviews, evaluation of traffic data, phone use surveys, medical, technical

solutions, detection and assessment of driver distraction or fatigue, traffic modelling,

crash prediction, road condition detection, theoretical discussions, and ones with focus

on vulnerable road users with no connection to drivers were excluded.

5.2.3 Search term

The search term deployed in TRID was (road OR driver) safety (app OR apps

OR "smartphone application" OR game OR games OR gamification) ("mobile phone"

OR smartphone). The search term deployed in Scopus was ( ALL ( road OR driver )

AND ALL ( safety ) AND ALL ( app OR apps OR "smartphone application" OR

game OR games OR gamification ) AND ALL ( "mobile phone" OR smartphone

) ).

5.3 SEARCH AND SCREENING RESULTS

The search in TRID, executed on September 12, 2017, returned 864 records.

Those results were limited to the years 2010 – 2017, which reduced the records to 377

records. By excluding languages other than English, the number of records became

365. The titles of those 365 records were screened for relevance. After the ones

considered to be potentially relevant were marked, 80 remained and were downloaded

for abstract screening. After screening of the abstracts, 35 records were marked for

further processing.

The search in Scopus returned 1246 records. Similar to above, those results were

limited to the years 2010 – 2017, which reduced the number of records to 1178. By

excluding languages other than English the number became 1164. Screening for

relevance in this specific case was initially performed by groups of publications by the

author. Scopus has functions on the left side of the screen that allows for quick search


within the current search results based on different criteria, such as the authors. Using

this function, authors were ranked by the number of identified papers and explored

top-down. Within the authors' work, their field of interest could then be more easily

recognised by screening a group of titles as a whole. This allowed the exclusion of

several papers at the same time as not being relevant, such as medical research or

crowdsensing, also reducing the number of authors to be looked at subsequently. This

approach of grouping papers was applied to authors with three or more listed papers.

25 authors were excluded using this approach, which reduced the result to 1088

records. The remaining authors had 2 or fewer publications. Those and all other

remaining titles were individually screened for relevance irrespective of the authors'

other title. After screening titles, 120 selected records were downloaded to screen their

abstract. Out of those, 55 were marked for further processing after the abstracts'

screening.

All papers marked for further processing after abstracts' screening from both

databases were imported into an Excel sheet to allow removing of duplicates. 80 papers

remained after removing duplicate records. 68 full texts were downloaded. 12 could

not be found online and needed to be found elsewhere. Nevertheless, one referred to

commercialising the outcome of a paper that was already found. Another record was a

presentation. Two referred to projects that had recently started, but there were no

results published, yet. This left eight documents to be found. Their authors were

emailed and asked to provide the respective article. Five of the contacted authors sent

their articles. Three remained missing from the initial selection of 80. One of those

was recognised as already existing in the current selection. After working through the

full texts of the documents, 31 articles were included in the detailed full-text analysis.

Four main themes were identified across the 31 papers: usage (3 studies); eco-driving

(6 studies), safety in addition to eco-driving (4 studies) and safety-only (18 studies).

As the current PhD program of research focused on evidence for safety benefits as a

result delivered by smartphone safe-driving apps, the 3 studies that focused on usage

and the 6 studies that focused on eco-driving were excluded as irrelevant. Thus, 22

papers were retained to be included in the qualitative synthesis (see Figure 5.1).


Figure 5.1. Data extraction flowchart based on the PRISMA statement.

5.4 FINDINGS

Once the final selection of articles to be analysed in detail was completed, the

elements of interest to the current study were summarised. Type of study, type and

number of participants, sensors used, measures taken, and focus of the studies, as well

as their findings, were considered in the analysis of the 22 papers, included in the

qualitative synthesis (see Table 5.1).


Table 5.1. Impact and effect of apps, games and gamification on young drivers' road safety.

Authors Intervention Type of study

(design details)

Type of

participants

Number of

participants

Sensors used Measures

taken

Focus Summary of findings

Zhang et al.

(2014)

The app warned drivers before an

accident blackspot. It also advised

them to take a break from driving

after a certain time. Both were

provided in real-time while driving

on expressways through an icon and

voice. No interaction was necessary.

The app is started and stopped

manually.

The app assessed the driving, based

on smoothness, speed, acceleration

and deceleration, as well as drivers'

recall on their driving safety. It

provided scores, driving history and

ranking based on the assessment.

Naturalistic

(each driver

driving on five

different routes

of expressways)

University

students

5 GPS Speeding,

acceleration,

deceleration,

driving

smoothness

Safety Participants obeyed

speed limits

represented by very

high scores on

speeding (over 80).

Birrell,

Fowkes, and

Jennings

(2013)

The app provided driving safety

and fuel efficiency feedback to the

drivers in real-time. Warnings were

issued in relation to lane departure

and headway distance. Advice was

Naturalistic (a

50 minute

mixed route

driving

scenario)

30 males

(mean age

42.33) and

10 females

40 LDW camera,

OBD2,

accelerometer

GPS

Time

headway, lane

changes,

glance

Safety

and

eco

Three times

reduction of

tailgating and

improvement of


given in relation to gear changes,

acceleration and braking. All advice

was identical and followed a

predefined script as the driving

route was predetermined.

(mean age

40.6)

frequencies

and duration

4.1% in fuel

efficiency.

Creaser et al.

(2015)

The app was recording phone use

while driving of a control and two

intervention groups. The app was

starting automatically, and there

was no need for participants to

interact with it. However, the

participants were reminded to

mount the phone before driving. In

one of the intervention groups, the

phone usage was blocked. In the

second intervention group, the app

additionally was sending

notifications to parents when risky

behaviour was detected.

Naturalistic (12

months free

driving)

Novice

teen drivers

(mean age

16.03, 130

males, 144

females)

274 Phone blocking

app

Phone use Safety Compared to the

control group, the

intervention groups

used their

smartphones

significantly less for

calling and texting

while driving.

Rodrigues,

Macedo,

Serpa, and

Serpa (2015)

The 3D game was introducing

traffic regulations to promote safe

driving through. While driving in

the game, participants were

Simulator

(educational 3D

traffic rules

game)

13 males

and 2

females (19

15 N/A N/A Safety The game stimulated

motivation and

learning while being

fun.


provided with feedback about their

traffic rules violations, when

detected.

to 36

years).

Steinberger,

Proppe,

Schroeter, and

Alt (2016)

In a driving game environment, the

app was re-engaging the drivers in

the driving task when speed changes

were necessary due to speed limits. It

was also providing feedback on the

participants’ driving behaviour while

driving.

Simulator (90

minutes

sessions)

19 male

drivers (18

to 25), 5

researchers

(26 to 36)

24 N/A Speed, eye

glances

Safety Improved driver

engagement,

decrease in speed

violations, increased

visual distraction.

Riener and

Reder (2014)

The app was gathering speed, gear-

shifting and braking force data. It

was providing visual and auditory

recommendations to the drivers in

real-time. The drivers were driving

an equipped car on a predefined

route. Their driving performance

was ranked.

Naturalistic

(19.4 km long

straight

commuter track

with no sharp

bends)

Males (23

to 26 years)

9 OBD2, GPS,

accelerometer,

Open street map

Speed, gear-

shifting, brake

force.

Safety

and

eco

No evidence of

improvement due to

the app steering

recommendations.

Rahman,

Qiao, Li, and

Yu (2016)

The app was alerting the

participants while driving for traffic

hazards (headway, speed, and

acceleration) through sound, visual,

and voice warnings.

Simulator (20

minutes

sessions)

12 males,

12 females

24 N/A Headway

distance,

headway time,

speed, and

Safety Both worker fatalities

and vehicle collisions

were reduced.


acceleration/d

eceleration

He et al.

(2017)

The app was issuing curve-speed

warnings in real-time to enable the

driver to improve the longitudinal

control of the vehicle.

Naturalistic (a

selected road

with curves)

N/A N/A GPS, compass Digital maps,

vehicle speed,

vehicle height

and curve

radius

Safety Decreased lateral

acceleration on a

dangerous curve with

the warning system

enabled.

Fitz-Walter,

Johnson,

Wyeth,

Tjondronegor

o, and Scott-

Parker (2017)

The app was collecting trip data to

allow easier and more accurate

recording of learner drivers’ driving

practices.

The app was starting automatically.

However, it had to be stopped

manually.

Naturalistic (1-

month free

driving)

Learner

drivers

25 N/A Weather, start

and end time

and location.

Safety No significant

change in the

behaviour despite the

gamified version

being seen as more

enjoyable and

motivating.

Botzer,

Musicant, and

Perry (2017)

The app was issuing collision

warnings in real-time. The app had

to be turned on and off manually.

Naturalistic (one

to two weeks)

Age from

24 to 60

26 Smartphone

camera, GPS,

motion sensors

Time-stamped

acceleration

and warnings

Safety Warnings of

imminent collisions

triggered

decelerating, thus,

safer driving. Fewer

warnings were issued

with time. However,

21/26 drivers stopped

using the app after


the experiment

ended.

Louveton,

Mccall,

Koenig,

Avanesov, and

Engel (2016)

The app was giving different tasks

to the drivers while they were

driving on a straight road after a

lead vehicle. The tasks were timed

according to speed change events of

the lead vehicle. The tasks were

both voiced and visualised.

Simulator 15 female

and 14

male (22 to

49)

29 N/A Task

completion

time and error

rate

Safety Increased cognitive

load and poorer

performance,

generated visual

distraction.

Birrell,

Fowkes, and

Jennings

(2014)











Naturalistic (50-

minute mixed

route driving

scenario)

30 males

(mean age

42.33) and

10 females

(mean age

40.6)

40 LDW camera,

OBD2,

accelerometer

GPS

Time headway Safety

and

eco

Three times

reduction of

tailgating and

improvement of

4.1% in fuel

efficiency.

Jiang, Zhang,

Chikaraishi,

Seya, and

The app was introducing different

information to the drivers during the

intervention period at 5 stages:

1. Only collected data.

Naturalistic (3

months free

driving)

Drivers,

who used

expressway

for at least

100 GPS Speeding,

acceleration,

deceleration,

Safety The app influences

driving safety

significantly when

underpinned by a


Fujiwara

(2017)

2. Information on rest areas and

blackspots was introduced.

3. Service and parking area

information was added.

4. Drivers were given the opportunity

to self-evaluate their driving. On that

basis, scoring was introduced. This

was followed by introducing driving

advice.

5. The last function of the app was

the “Drive & Love” safety

education campaign.

4 times per

month

driving

smoothness

careful combination

of provided

information

depending on the

driver's stage of

change.

Schartmüller

and Riener

(2015)

The app was measuring speed, lane

position and distance to the front

vehicle. It was issuing visual and

auditory warnings to the drivers in

real-time. The system was

preconfigured and run fully

automated.

Naturalistic

(every

participant had

to drive 10,4 km

long track

twice)

Age range

18 to 60

17 Build-in camera Vehicle

detection and

tracking, lane

detection and

tracking,

vehicle

distance

estimation

Safety Enhanced perception

of minimal distance

required.

Birrell and

Fowkes

(2014)




Naturalistic

(fixed driving

route)

10 males

and 5

females

15 (out of

40)

Adapted LDW

camera, OBD2,

Headway, lane

departure, gear

change,

Safety

and

eco

No visual distraction

caused by the in-

vehicle smart driving









(over 21

years)

accelerometer,

GPS

acceleration

and braking

system providing

feedback.

Li, Qiao,

Qiao, and Yu

(2016)

The app was calculating headway

distance and time to collision in a

pre-equipped vehicle. It was

providing a real-time warning

message to the driver when

predetermined thresholds were met.

The warning message was

encouraging the driver to keep a

safer distance.

Naturalistic

(limited local

community

route)

N/A N/A OBD2, GPS Braking

distances,

deceleration

and speed.

Safety Improved speed

compliance and

deceleration

performance, keeping

safer distances to

intersections and

other vehicles.

Ryder, Gahr,

Egolf,

Dahlinger, and

Wortmann

(2017)

The app was providing visual

hotspots warning to the drivers in

real-time. The warnings issued,

based on analysing historical

accident data.

Naturalistic

(four weeks)

Professiona

l drivers

57 OBD2 Location,

driver’s

personality,

dangerous

braking

events, vehicle

speed

Safety Driver behaviour

improvement through

in-vehicle accident

hotspots warnings

influenced by the

individual’s

personality. No


immediate effect on

driver behaviour

outside lab

experiments.

Hu et al.

(2015)

The app was calculating a mood-

fatigue profile of the driver in real-

time. Based on the specific

calculation, it was proposing

suitable mood-calcified music.

Simulator 32 males

and 16

females

48 Front camera Mood, fatigue Safety Decreased fatigue

and negative mood

compared to a

traditional

smartphone-based

music player.

Creaser et al.

(2015)

As a result of in-vehicle

monitoring, the app was providing

real-time warnings when an unsafe

driving behaviour was detected, e.g.

missing seatbelt, speeding or phone

use. It was sending messages to the

parents of some of the drivers if

they did not comply with the

warning.

Naturalistic (12

months free

driving)

Newly

licensed

teens

300 Accelerometers,

GPS, in-vehicle

Arduino

microprocessor,

seatbelt sensor,

passenger

sensors

Speeding, hard

turning,

braking,

accelerations,

seatbelt

Safety Reduced risky

driving behaviours.

Musicant and

Botzer (2016)

The app was issuing sound and

visual collision warnings in real-

time.

Naturalistic (2-3

weeks)

8 females

and 18

males (24

to 60 years)

26 GPS, camera

and smartphone

dynamic

sensors.

Time-stamped

warnings and

speed.

Safety Lower speed when

issued warnings,

safer distance.


Williams,

Peters, and

Breazeal

(2013)

The app was using voice and facial

expressions to communicate with

the driver while they were

travelling in a simulated

environment. It was detecting

phone communication events (e.g.

getting late for a calendar

appointment, sending or receiving

messages), it was suggesting

solutions on how to respond.

Simulator 20 males

and 24

females

(mean age

28.6 years)

44 N/A Navigation,

collision

warnings,

Internet,

entertainment

systems and

messaging

services

Safety Less interaction

stress, more often

safety precautions

and increased

companionship with

the assistant in

comparison to

smartphone users.

Birrell,

Young,

Stanton, and

Jennings

(2017)



drivers in real-time. During high

driver cognitive load, e.g. driving in

a city, the presented information

was limited. More information was

presented during driving with lower

cognitive demands, e.g. on a

highway.

Simulator (5-min

simulated

scenarios) and

Naturalistic (a

mixed 0.1 miles

driving route)

N/A 25

(simulator),

40

(naturalistic)

OBD2, GPS,

camera

Headway, lane

departures,

acceleration,

gear changing

Safety Modulated driving

workload towards

manageable levels

depending on current

driving task

demands.


5.4.1 Studies’ designs and samples

The analysis of the 22 papers revealed that the studies were implemented in

two types of settings, naturalistic (16 papers) and simulator (7 papers). One study was

implemented both in naturalistic and simulator settings (Birrell et al., 2017), hence the

sum of studies implemented in one or the other setting is more than the number of

analysed papers.

While simulator studies tend to be similar in that they are implemented in

laboratory conditions, the analysed naturalistic studies offered greater design

variability. Nine of the naturalistic studies were characterised by predefined routes (He

et al., 2017; Riener & Reder, 2014; Zhang et al., 2014). The other seven studies focused

on the drivers' behaviour in their free-living environment (Botzer et al., 2017; Jiang et

al., 2017; Ryder et al., 2017). The free-living driving studies were characterised by

time limits, ranging from one week (Botzer et al., 2017) to 12 months (Creaser et al.,

2015).

The samples' design offered a similar variability. Two naturalistic studies ((He

et al., 2017; Li et al., 2016) did not provide any information for their participants.

Birrell et al. (2017) supplied information about the number of participants without

other details. The other 19 studies provided more comprehensive information. For

example, the number of reported involved participants in the different studies ranged

from 5 (Zhang et al., 2014) to 300 (Creaser et al., 2015). Some studies reported the

participants' occupation, e.g. university students (Zhang et al., 2014) or professional

drivers (Ryder et al., 2017), without providing information on gender for example.

Other studies were interested in other characteristics, such as frequency of using a

certain road (Jiang et al., 2017) or driving experience (Fitz-Walter et al., 2017). Some

studies focused only on male drivers (Riener & Reder, 2014; Steinberger et al., 2016),

while a majority had both males and females as participants (Birrell et al., 2013;

Creaser et al., 2015; Rahman et al., 2016). The reported age range of the participant

had a large spread, from a mean age of 16.03 (Creaser et al., 2015) to an upper limit

of 60 years (Botzer et al., 2017; Musicant & Bolzer, 2016; Schartmüller & Riener,

2015).


5.4.2 Sensors and measures

The contemporary smartphones possess a number of sensors that safe-driving

apps can leverage. Those are the internal clock, cameras, GPS, accelerometer,

gyroscope and magnetometer. Additional information can be generated by monitoring

the software, operated on the smartphone. Such information may include what apps

are being used, whether phone calls are being initiated or answered, or whether

messages are being read or written.

One study reported on the use of a phone-blocking app to monitor the

participants' smartphone use (Creaser et al., 2015). Six studies did not provide

information on any sensors used, five of which reported on simulator studies

(Louveton et al., 2016; Rahman et al., 2016; Rodrigues et al., 2015; Steinberger et al.,

2016; Williams et al., 2013) and one on a naturalistic study (Fitz-Walter et al., 2017).

Smartphone sensors were used in 14 of the analysed studies.

The most widely used smartphone sensor was the GPS, 12 studies, e.g. Zhang

et al. (2014), Birrell, Fowkes, and Jennings (2014), and Riener and Reder (2014).

Another sensor, used in the analysed studies, was the accelerometer, 7 studies, e.g.

Creaser et al. (2015), Birrell and Fowkes (2014) and Botzer et al. (2017). The

smartphone cameras were the third sensor used in more than half of the studies, 7

studies, e.g. Birrell et al. (2017), Musicant and Botzer (2016) and Hu et al. (2015).

The different sensors provide opportunity measures to be taken about different

risky behaviours. For example, the GPS and the smartphone accelerometer allow

assessment of speed, acceleration, deceleration and driving smoothness (Botzer et al.,

2017; Riener & Reder, 2014; Zhang et al., 2014). The smartphone cameras allow for

vehicle detection, vehicle tracking, lane detection, lane tracking, and headway distance

estimation (Birrell et al., 2013; Birrell et al., 2014; Schartmüller & Riener, 2015).

In addition to smartphone sensors, some of the studies used OBD2 to collect

additional data. The OBD2 connects directly to the car systems. Thus, the collected

data might be expected to be more accurate than the data generated by smartphone

sensors. The use of OBD2 was reported by seven of the reviewed studies. OBD2 was

most often used in conjunction with the smartphone GPS and accelerometer to collect

additional data about speed, acceleration and braking, e.g. in Li et al. (2016), Birrell et

al. (2014) and Riener and Reder (2014). Different than the smartphone GPS and


accelerometer, the OBD2 can provide data about gear-shifting, which is useful

information when eco-driving is analysed together with safety (Birrell & Fowkes,

2014; Riener & Reder, 2014).

5.4.3 Benefits

Out of the 22 analysed papers, 18 papers focused entirely on safety, while 4

papers explored safety together with eco-driving. Three of the eco-driving papers

reported results from one study, in which a 4.1% improvement in fuel efficiency

together with 3 times reduction of tailgating was observed (Birrell et al., 2013; Birrell

et al., 2014) while no visual distraction was caused by the in-vehicle smart driving

system (S. A. Birrell & Fowkes, 2014). Those findings did not find support in the

fourth eco-driving paper, which concluded with no evidence of improvement due to

the app steering recommendations (Riener & Reder, 2014). Both eco-driving studies

were implemented in naturalistic settings with fixed driving routes.

The studies, focused only on safety, offered a greater variability of both

implementations and results. Implementations were not only in naturalistic settings (12

studies), as in the case of eco-driving but also in simulated environments (7 studies),

with one study using both.

The simulator studies reported mixed results. For example, Rodrigues et al.

(2015) found their smartphone game implementation to stimulate motivation and

learning while being fun. More directly related to safety, Steinberger et al. (2016)

found improved driver engagement and decrease in speed violations. Reduction of

both vehicle-to-vehicle crashes and worker fatalities was reported by Rahman et al.

(2016). Hu et al. (2015) found a decrease in fatigue and negative mood. Birrell et al.

(2017) reported on modulated driving workload towards manageable levels, a result

which finds support in Williams et al. (2013). Williams et al. (2013) found less

interaction stress, more often safety precautions and increased companionship.

However, not all studies found a positive impact on safety. Louveton et al. (2016)

reported on a generated visual distraction, leading to an increased cognitive load and

poorer performance. The increased visual distraction was also reported by Steinberger

et al. (2016), although it did not degrade lane-keeping performance.

The diversity of findings was greater in naturalistic settings implementations.

For example, Creaser et al. (2015) reported less distraction of their Intervention groups


in comparison to the Control group. With regards to speeding, Zhang et al. (2014)

reported increased compliance with speed limits. Li et al. (2016) also found that

participants improved their speeding profiles. In addition, Li et al. (2016) observed

improved deceleration rates, extended braking distances to the leading vehicle or

intersection stop line. Musicant and Botzer (2016) reported on speed and headway

distance improvements, as a result of issued warnings. In-car warning systems were

used to trigger a variety of results, such as a decreased lateral acceleration on a

dangerous curve (He et al., 2017), decelerating to avoid collisions (Botzer et al., 2017),

enhanced perception of minimal distance required (Schartmüller & Riener, 2015) or

reduced frequency of risky driving behaviours in general (Creaser et al., 2015).

However, despite the evidence of effects on driver behaviour, Ryder et al.

(2017) found it challenging to trigger driving behaviour improvements in naturalistic

settings. The authors showed that the drivers’ personality influences their likelihood

to improve behaviour as a result of an intervention. In support, Jiang et al. (2017) found

a significant influence on driving safety when the information provided to the drivers

is customised to their personality. For example, safety diagnosis and blackspot

information were found beneficial for careless and irritable drivers, while feedback,

self-diagnosis and drivers' ranking was beneficial to drivers who wanted to decrease

their speeding (Jiang et al., 2017).

5.5 SUMMARY

Only 16 of the 22 reviewed papers were concerned with driving studies in the

real world; 6 referred to studies in simulated conditions only. In addition, 3 of the

papers referred to the same study, leaving 14 studies, exploring effects in naturalistic

settings, thus, being somewhat similar and relevant for the current program of research.

Out of those 14 papers, only 4 explicitly reported as being focused on young drivers.

One of those 4 papers reported no significant change in the behaviour despite the

gamified version being seen as more enjoyable and motivating (Fitz-Walter et al.,

2017). Only three studies, focused on young drivers, reported positive safety benefits

from the deployed interventions (Creaser et al., 2015; Creaser et al., 2015; Zhang et

al., 2014).

Zhang et al. (2014) focused on speeding. In a naturalistic driving study, they

used the smartphone GPS to monitor the speed of five university students. Each young


driver drove on five different predefined routes of expressways. The authors'

conclusions were that the participants obeyed the speed limits, which was depicted by

the smartphone app awarding them very high scores on speeding (over 80).

The other two studies described results from a 12-month field operational test of

a Teen Driver Support System (TDSS) on the roads of Minnesota (Creaser et al.,

2015). The test involved 300 newly licensed teens who were provided through a

smartphone with in-vehicle real-time feedback about their risky behaviours. Those

behaviours were reported to the parents of a subset of the teens through text messages.

Additional data was collected through the vehicle outside the smartphones. Depending

on the type of feedback received, the participants were divided into three conditions:

control with no feedback, a first treatment group with TDSS only and a second

treatment group with TDSS and parental notifications. The participants were

remunerated with USD 300 upon completion and could keep the smartphone and its

accessories after the end of the study. As a result of the test, it was found that in

comparison to the control group, the two intervention groups called and texted

significantly less per mile driven (Creaser et al., 2015). Another observed result was

that the driver alerts, generated through the in-vehicle monitoring, lead to reduced

frequency of risky driving behaviours (Creaser et al., 2015).

5.6 DISCUSSION

Smartphone safe-driving apps can be used for low-cost safety interventions that

take advantage of the built-in smartphone sensors' capabilities. However, while

answering research RQ1 (What is the state of the art evidence of the safety benefits of

smartphone safe-driving apps for young drivers?), it was found that a safe-driving app

can deliver a variety of safety benefits. Benefits include increased work safety

(Rahman et al., 2016), improved drivers’ mood (Hu et al., 2015) or enhances distance

perception (Schartmüller & Riener, 2015).

Some researchers used the same smartphone apps they designed, developed or

obtained to investigate different behaviours in the involved drivers. For example, the

same app modulated driving workload towards manageable levels (Birrell et al., 2017)

and reduced tailgating (Birrell et al., 2014). Another app improved speed limit

compliance on a predefined route (Zhang et al., 2014) but also increased safety over

three months (Jiang et al., 2017). A third app improved deceleration patterns (Botzer


et al., 2017) and promoted lower speeds in general (Musicant & Botzer, 2016). A

fourth app reduced both drivers’ phone interactions (Creaser et al., 2015) and the risky

driving behaviour frequency in general (Creaser et al., 2015). Thus, the systematic

review provided evidence for the potential of smartphone safe-driving apps to deliver

safety benefits in regards to different behaviours or several behaviours at the same

time.

Despite the evidence for safety benefits, delivered by safe-driving apps in

general, only three studies in the current systematic review reported clear safety

benefits for the involved young drivers in naturalistic settings. None of those three

studies would easily qualify as a scalable real-world intervention. One assessed only

five people, who drove on a predefined stretch of a highway (Zhang et al., 2014). The

other two were part of a multimillion-dollar investigation with substantial incentives

used to recruit and retain participants (Creaser et al., 2015; Creaser et al., 2015).

The reviewed literature provided support not only for the need for a further

investigation into the safety benefits delivered by smartphone safe-driving apps but

also for the adopted study design. The fact that a number of past studies focused on

speeding in naturalistic settings (Li et al., 2016; Musicant & Botzer, 2016; Zhang et

al., 2014) suggests that speeding is a behaviour that is likely to be influenced by a

smartphone safe-driving app intervention. Increased compliance with speed limits was

observed in both simulated (Steinberger et al., 2016) and naturalistic conditions

(Zhang et al., 2014). However, the assessed time period varied from 5 minutes (Birrell

et al., 2017) to one year (Creaser et al., 2015). Thus, drawing comparisons and

clustering benefits presents a challenge due to the differences in the settings in which

the studies were implemented.

The feedback, issued by a safe-driving app, was found useful in a number of

settings, such as lowering speed in naturalistic driving (Musicant & Botzer, 2016),

decreasing fatigue and reducing negative mood in simulator driving (He et al., 2017),

preventing collisions in naturalistic driving (Botzer et al., 2017), improving distance

perception in naturalistic driving (Schartmüller & Riener, 2015), and reducing young

drivers’ risk-taking in naturalistic driving (Creaser et al., 2015). Furthermore,

increased distraction generated by smartphone safe-driving apps (Louveton et al.,

2016; Steinberger et al., 2016) could be possible unwanted side effects of a safe-

driving app intervention. Such evidence provided support for including distraction-


related questions in the Study 2 surveys. These questions had the objective to help

assess whether participants’ smartphone interactions while driving did not increase as

a result of the safe-driving app intervention.

From a more theoretical perspective in safe-driving app interventions, evidence

was found on the impact of drivers' personality on their behaviour (Jiang et al., 2017;

Ryder et al., 2017). This evidence supported an earlier decision the adopted TPB

theoretical framework to be extended with constructs that measure personality

characteristics (see Subsection 3.6.3).

Additionally, despite a large number of safe-driving smartphone apps readily

available online, and more developed for research purposes, the literature provided

limited information on the considerations made when choosing one as an intervention

tool. Thus, the next chapter explores the topic of selecting a safe-driving app from

these currently available, which can motivate safe driving behaviour among young

drivers as part of an intervention.

82 Chapter 6: Selecting a safe-driving app

Chapter 6: Selecting a safe-driving app

This Chapter 6 provides details on the work done in the process of investigating

available safe-driving apps and selecting one for deployment as part of an intervention.

It begins with describing the implemented Focus group design, participants, findings

and limitations. The Focus group is followed by a description of the investigation of

the Google Play and iTunes app stores to identify available safe-driving apps.

Subsequently, the pros and cons, as per the Focus group recommendation, of a

selection of some more popular apps are investigated. Three apps are tested in the real

world before finally selecting one app to be deployed as an intervention tool.

6.1 FOCUS GROUP DESIGN

The first step in choosing the most appropriate app was to consult an expert

reference group, which was implemented via a Focus group. Substantial previous

experience in road safety at an executive level or in road safety research, an ongoing

commitment for cooperation, and a professional record of successful impact in the

field were considered when selecting the participants. The design of the Focus group

purposefully involved road safety entrepreneurs (SEs). This involvement was guided

by the author's underlying personal interest and experience working as a SE, as well

as the motivation to establish grounds for using the research outcomes in the real world

through NFPs (see Section 1.1).

The Focus group with the road safety experts took place as part of their existing

international meeting in Shanghai, China. The Focus group lasted for two hours. It was

fully audio- and video-recorded, to capture all visual elements of the discussion, and

subsequently transcribed. This structure of the process helped to capture how the

discussion evolved over the course of the Focus Group while the participants interacted

with the visuals and changed them.

6.1.1 Participants

The participants in the Focus group represented a convenience sample of

academia (3 people), applied research (1 person) and project leaders (6 people) in the

domain of awareness-raising road safety interventions for young drivers. The Focus

group sample was stratified to be both gender (6 males and 4 females) and culturally

Chapter 6: Selecting a safe-driving app 83

(6 countries) balanced (see Table 6.1). All invited participants had experience in risk

prevention programs for young drivers and were involved in road safety projects on

an international level.

Table 6.1. Country of origin and gender of participants in the Focus group.

Country of origin Gender

Male Female Argentina 1Austria 2 1Belgium 1China 3Hungary 1Romania 1

Total 6 4

A QUT ethical clearance (Approval Number 1600000340) was obtained before

participants were involved in the Focus group. Participants were anonymised by

assigning an identifier to each of them. Identifiers were assigned randomly depending

on the sequence the participants' consent forms were collected and stored. Identifiers

range from P1 to P10. Those identifiers are referred to in the analysis when a quote is

cited.

6.1.2 Procedure and materials

The Focus group procedure followed a two-step approach, discussing the

following:

Step 1. Personal views about how young people use in-car information and

communication technologies (ICT) and what innovative ICT can

influence them positively.

Step 2. Discussing smartphone safe-driving apps.

Participants were encouraged to share their experience in ICT and in methods of

ICT implementation with high potential to positively influence young drivers’

behaviour. During the discussion, the road safety experts' attitudes towards ICT, its

evolution, and what challenges it brings in the vehicles, were explored. They were able

to elaborate by supporting their statements with information about what worked, how

it worked, and why did they think it worked in their cultural context. Furthermore, they

discussed what could work better, and why did they think it could work better. Visual

brainstorming tools (flipchart drawings and post-it notes), coordinate plane and SWOT


analysis (see Figure 6.1) and a short video of the DriveScribe app review4 further

facilitated the discussion.

Figure 6.1. Focus group visual brainstorming tools

6.1.3 Data analysis

The Focus group audio- and video-recorded session was transcribed and coded

in NVivo 11. A thematic analysis was performed to identify patterns in the participants'

responses (Braun & Clarke, 2006). The subsequent interpretation of the data facilitated

the generation of codes, which were aggregated into themes. The main identified

themes were 1) vision on young people's use of technologies in the car, and 2)

smartphone safe-driving apps. Three subthemes, describing considerations to facilitate

the adoption of smartphone safe-driving apps, were identified under theme 2. Those

were 2.a. young drivers' needs and their safety, 2.b. young drivers' ecosystem, and 2.c.

road safety stakeholders' needs. Selected quotes, relevant to each theme and subtheme,

are integrated into and discussed as part of the following Section 6.2.

6.2 FINDINGS FROM THE FOCUS GROUP

The sample of 10 participants produced 110 statements (mean rate of 11

statements per participant, ranging from 1 statement to 21 statements per participant).

A statement is defined as a participant's input, from the moment they start speaking to

the moment they stop. Those statements represented the respective participant's

opinion, perception or attitude in relation to one or more identified theme or subtheme.

4 https://www.youtube.com/watch?v=pbqqUpO6qXI


As discussed above, two main themes and three subthemes emerged from the focus

group discussion. Details in relation to each theme and subtheme are presented in the

subsequent subsections.

6.2.1 Vision on young people's use of technologies in the car

The first theme in the Focus group revolved around the use of technologies in

the car and how young people interact with them. The participants shared that,

according to them, young people use technology in cars because it is there and they

like it. The discussion quickly centred on ICTs in the car that have no added value to

the driving tasks, yet appear attractive and engaging to young drivers. Those

technologies are usually not “safety-enhancing equipment" [P10]. Distracting devices

came more into focus, and a new division of technology users appeared: the ones who

have money and the ones who do not have money. "Those who can afford to have all

of the equipment they want all of it." [P6] The ones that cannot afford to have the

expensive equipment were referred to by P6 as "the normal guys, [who] are driving

carefully." However, after a short reflection, P6 added "But actually the mobile phone

is still there. It's really a bad habit, and they are using it everywhere." This reflection

was supported by the most pessimistic participant in relation to ICT, [P3], who stated

"According to me using the new technology in car in ... is not so popular", but later

added, "besides the mobile and social media." Thus, the Focus group participants

viewed technology as being out there and widely available to the drivers. In their

opinion, drivers with financial means are likely to possess more sophisticated

technology. However, according to them, the drivers’ financial means have no effect

on the level of smartphone penetration in cars.

The mobile phone emerged as a ubiquitous problem, carrying a low benefit and

a high risk. Participants highlighted that this was irrespective of the drivers’ social

status. A discussion on why drivers use mobile phones, although they know it is

dangerous and illegal, emerged. The moderator raised the question of what

technologies could influence young people positively, given that the ones they like are

usually related to the increase of risks, associated with driving. The participants agreed

that mobile phones, driving under the influence of alcohol and speed (the behaviours

of interest for the current program of research) are the three most common reasons for

increased risks among young drivers. They also agreed that they have to be tackled

differently and that there is available ICT to prevent risks, related to each of the three.


For example, Alcolock was identified as most likely to prevent driving under the

influence of alcohol, Intelligent Speed Adaptation - to control the speed, and

"everything that prevents mobile devices to be used, provided they are in the car, is

also good for reducing accidents and getting safer roads" [P7]. The participants shared

a common view that, since young people are attracted to risk-enhancing devices

through their characteristics, a similar approach should be used when designing

prevention tools.

6.2.2 Discussing smartphone safe-driving apps

The initial Focus group discussion around young people and in-car ICT was

summarised by P6:

"If it is to make it attractive to the young [drivers], it has to be in their own

language. I mean, yeah, if it is smartphones and social media, you have to be on there,

and try to get their attraction there. [sic]"

P6 pointed a direction to follow if young drivers’ are to be reached. Regardless

of the technology, it has to be attractive and to speak to the young drivers in a way to

be easily understood. Since smartphones and social media are believed to be both

attractive and understandable than they should be used to reach out to young drivers.

To visualise an implemented potential transformation of a smartphone from a

risk-enhancing to a safety-enhancing device, the DriveScribe review video was shown

to the participants. DriveScribe leverages smartphones and social media in the form of

a well-intentioned safe driving app, which might be perceived as speaking in the young

drivers’ own language. The video review triggered the question if safe-driving apps,

in general, have an added value to the driver's tasks, or they simply increase the

distraction. P2 shared that, from what was seen, the app is a little different than a GPS

with a speed warning. The only difference was earning points. P4 supported the view

that the app shows what is already known by the driver with the only difference that it

is supported by numbers. P4 wondered if this could change the attitude. P4 argued that

the data can still be useful for third parties such as insurance companies, for example,

when they calculate premiums. P7 generalised the situation with new technology, and

narrowed down how helpful application should be framed:


"We have to move forward with the technology as the technology moves forward.

So, I think, it's the important step when a new problem arises when you look for

solutions to overcome the new problems."

The subsequent discussion uncovered the features that may make young drivers

embrace safe-driving apps. The Focus group participants provided their considerations

for research design efforts that aim at deploying safe-driving apps for research

purposes. As the Focus group included SEs, accounting for such consideration may

provide benefit in the long term, beyond the scope of the current program of research

(see Section 1.1). For example, using SEs’ insights as guidelines here may help

mainstream adoption of safe-driving apps in future implementations, outside road

safety research (see Section 1.4). Thus, as per the summarised opinion of the Focus

group participants, a safe-driving app intervention is likely to succeed when the

following is met:

1. Young drivers' needs and safety (subtheme 2.a) should be accounted for by

designing an attractive solution ("if it is not mandatory by law, you have to be

attractive.” [P8]) that is interesting, “sexy” and speaks to young people in their

own language so that it does not “scare” them ("Connected to insurance, that

would scare and nobody will use it." [P8]). The solution should embed

incentives (“You have to have incentives." [P8]), incorporate gamification or

social network sharing ("share it in their social networks" [P8]), to support

long-term adoption. It should also be fixed, hands-free to reduce any potential

for increasing driving risks (“It should be fixed in the car so that you cannot

use it with your hands while driving." [P10]). Optional blocking of incoming

calls, texts, notifications or social media, can be a good function to enhance

safety, too ("Other functions should be blocked." [P10]). Audio feedback can

help keep eyes and attention on the road, which is essential in reducing road

hazards ("The feedback should be auditory as well." [P10]).

2. Dynamics and interrelations in the young drivers' ecosystem (subtheme 2.b)

should be considered when aiming for widespread use. The involvement of as

many stakeholders as possible can support the joint efforts of road safety

researchers. Possible connection with GDL can provide a whole new level of

motivation for using a safe-driving app (“driving schools to start to use it [...]”

[P6]; “connect it [apps] to [...] driving license probation” [P1]).


3. Road safety stakeholders' needs (subtheme 2.c) should also be accounted for

when designing implementations to be used in both laboratory and in real-

world conditions. Thus their experience and support can be used when to

understand how driving behaviour can be influenced more efficiently (“You

can use [apps...] like a kind of simulator when you do an education campaign”

[P4]; “[apps...] be used in real life or real driving.” [P7]). The opportunity to

replace expensive equipment with other quality means is important in every

endeavour (“If it can replace the very high-cost simulators. [That means] Low

cost!” [P4]). Low cost will undoubtedly appeal and, thus, boost adoption. Last,

but not least, the implementation should trigger discussions, especially in the

open public ("the data should be available to the public." [P10]), for example

through a leaderboard. This will make the researchers' efforts more visible and

will open the door for more feedback that could help adoption and

improvements.

6.3 SYNTHESIS OF FOCUS GROUP’S FINDINGS

The purpose of the implemented Focus group was to provide understanding

about the road safety experts' attitudes towards ICT, its evolution, and what challenges

it brings in the vehicles. More specifically, the Focus group sought to gain a

comprehensive understanding of potential real-world limitations that might discourage

safe-driving apps' adoption. The young drivers' specific needs and motivations were

explored in the process of identifying such limitations. The Focus group was

exploratory in nature and added to the present program of research a better

understanding of potential criteria which may boost safe-driving apps adoption if met.

Those criteria aligned with some of the theory-based criteria for selecting COTs.

Alignment was found with the criteria, relevant for influencing norms and PBC, the

expected strongest predictor of intention (see Section 4.2). More specifically, SC2

(relevant for norms) is consistent with two focus group's recommendations. First,

involving as many stakeholders as possible in supporting the joint efforts of road safety

researchers would potentially include important referents of the participants. Those

important referents can both express their attitude towards the participants' behaviour

or share information about their own behaviour. And second, such exchange of

information can serve as a trigger for open public discussions together with the

interventions, which was the other SC2 relevant recommendation of the focus group.


Discussions in public can additionally reinforce the establishment of norms. SC3

(relevant for PBC) is somewhat consistent with the recommendations not to distract

drivers. Not distracting the drivers would keep them engaged in their primary driving

task, which, in turn, would enable them to perform the behaviour of interest better. To

achieve this SC3-relevant outcome, the focus group recommended 1) devices to be

fixed and hands-free, 2), incoming communication to be blocked, and 3) the app to

provide audio feedback. Finally, SC5 (relevant for peers' norm) aligns with

recommendations such as to include leaderboards and enable social network sharing.

A leaderboard would allow the young drivers to see how their peer young drivers rank.

This information would potentially motivate them to improve their behaviour to enjoy

the fame of being at the top (Duggan & Shoup, 2013). Social network sharing would

allow them to reinforce the fame, thus, reinforcing the peer pressure.

The focus-group-based criteria, which are complementary to the theory-based

ones, are summarised in Table 6.2. Those criteria should be carefully applied in other

research projects and should be considered together with the Focus group limitations.

The main limitation of the Focus group was related to the DriveScribe app. Its review

facilitated the discussion by providing a real-life example. However, it also biased the

discussion as the Focus group participants kept referring to it while the objective was

to discuss smartphone safe-driving apps in general. This potentially reduced the scope

of the findings. Having a more diverse number of safe-driving apps reviews could

potentially add value to future discussions. Thus results should not be generalised.

Table 6.2. Criteria for smartphone safe-driving apps, synthesised from Focus group’s findings.

Criteria Findings

Low-cost In order to boost adoption, it would be best if the app is free for users and

does not require hardware, such as a dongle, in addition to the smartphone

itself.

Safety The app should be able to provide live feedback to the driver as well as run

invisibly in the background, thus, not causing additional distraction. It also

shall have a self-starting capability.

Availability The app shall not be geographically restricted.

Information The app shall provide various types of information. It shall have a web

interface. It shall provide detailed after-trip feedback on a Google map. It


should also feature user groups (i.e. proprietary leaderboards) with the

achievements being possible to share on social media.

6.4 SELECTING A SAFE-DRIVING APP FOR AN EVALUATION

The next step, following the Focus group, was to browse the app stores (Google

Play and iTunes) to identify safe-driving apps. Based on findings from the Focus

group, apps were searched and studied. The used terms were "road app", "smart

driving", "safe driving" and "OBD game". Sixty-six apps from the search results in

Google Play and 20 from iTunes were selected for detailed investigation (see

Appendix B). The established list of safe-driving apps was narrowed down to three

apps, shortlisted for user testing. Finally, one was selected to be evaluated as an

intervention tool.

The online reviews, left by users in the smartphone safe-driving apps’ marketing

profiles, were used to narrow down the number of apps to be investigated in more

detail. The Focus group recommendations, discussed in the previous section, were

compared against the pros and cons of a selection of some more popular apps. In Table

6.3, those apps were scored using the established theoretical selection criteria (see

Section 4.2). Double points were assigned to a positive answer on the PBC-related

question because PBC is seen as typically the strongest predictor of speeding.


Table 6.3. Smartphone safe-driving apps (yes=1, no=0, double points for SC3).

Name Pros Cons Scoring

Flo - driving

insights

1. Can provide live feedback to the driver or can run invisibly in the

background. The life feedback can help the driver understand their real-

time behaviour and potentially inform conclusions whether it is

favourable or unfavourable (SC1).

2. Provides detailed after-trip feedback on a Google map. This detailed

information can potentially help the driver understand the implications of

any changes in their behaviour, thus enabling them to improve it (SC3).

3. Offers leaderboard. The leaderboard shows to the driver how their peers

are doing (SC5). Comparing with their peers can help the driver assign a

moral value to their behaviour (SC4).

4. Free for users.

5. No geographic restriction.

6. Has a web interface.

7. Does not need a dongle.

1. Has problems with

synchronising with the GPS

signal when on autostart.

2. Drivers with short trips may

in general score lower than

drivers with long ones.

3. Does not have all car

manufacturers and models.

SC1 SC2 SC3 SC4 SC5 SC6

Yes No Yes Yes Yes No

1 0 2 1 1 0

Total points: 5


AAMI Safe Driver

1. Monitors for exact speed limits, not the general ones. The speed limits

feedback can help the driver understand whether it is favourable or

unfavourable (SC1).




3. Good mix of gamified elements (scores, badges). Receiving scores and

badges can represent a higher morality of the behaviour (SC4).

4. Runs in the background.


6. Free for users.

1. Cannot provide real-time

feedback.

2. Fails to record and analyse

long journeys.



signal when on autostart.

4. Very wide thresholds are

set for recording an offence.

5. Designed to sell an

insurance product.


calculating the overall

score.


Yes No Yes Yes No No

1 0 2 1 0 0

Total points: 4


Hellas Direct (one

of the DriveWell

implementations)




2. Uses all common gamification elements (scores, leaderboards, badges,

etc.). Receiving scores and badges can represent a higher morality of the

behaviour (SC4).

3. Offers a leaderboard. The leaderboard shows to the driver how their peers

are doing (SC5). Comparing with their peers can help the driver assign a

moral value to their behaviour (SC4).


5. Runs in the background.

6. Free for users.

1. Available only in selected

jurisdictions, i.e. not being

tested in Australia.

2. Always attached to

insurance products.

3. Looks like over-gamified –

may benefit from a "lite"

version.

4. Difficult to advise when the

user is not the driver.

5. Does not have a user-start

option.


No No Yes Yes Yes No

0 0 2 1 1 0

Total points: 4


SafeDrive

1. Blocks calls and texts. Thus, it helps the driver perform the behaviour

(SC3).

2. Free and not geographically restricted.

3. Tries to connect gamification (earning points) with the real world

(receiving rewards). Received points can represent a level of morality of

the behaviour (SC4).

4. Works on auto-start.

1. Does not monitor other data

than mobile phone usage

data.

2. No real rewards besides

discounts.

3. Has problems with setting

user info.



signal.


No No Yes Yes No No

0 0 2 1 0 0

Total points: 3

Automatic 1. Uses accurate vehicle data. The accurate vehicle data can help the driver

understand whether it is favourable or unfavourable (SC1).

2. Provides detailed feedback. This detailed information can potentially help

the driver understand the implications of any changes in their behaviour,

thus enabling them to improve it (SC3).

3. Assist with crash alert (SC6).

4. Diagnoses the car.

1. Costs 99.99 AUD.

2. Requires a dongle.

3. Restricted to the US.

4. Does not support all cars.


Yes No Yes No No Yes

1 0 2 0 0 1

Total points: 4


Rookie Dongle

1. Uses accurate vehicle data. The accurate vehicle data can help the driver

understand whether it is favourable or unfavourable (SC1).

2. Notifies parents or guardians about reckless driving. Thus, it may trigger

information on how driver's important referents see their behaviour

(SC2).

3. Provides detailed feedback. This detailed information can potentially help

the driver understand the implications of any changes in their behaviour,

thus enabling them to improve it (SC3).

4. Does not require a smartphone as it has a built-in GPS and mobile

connection.

1. Costs 338.99 EUR with a

one-year subscription.

2. May discourage adoption

because of notification to

third parties.


Yes Yes Yes No No No

1 1 2 0 0 0

Total points: 4


The final step was to install three apps on the candidate's smartphone for real

driving testing. The initial criteria, used for that specific purpose, were derived from

the focus group's recommendations. Those were:

- Availability. Not all apps had permissions to be used in Australia, i.e. their

usage was geographically restricted.

- Affordability. The apps had to be free of charge to use, which, if deployed

in an intervention, might increase the likelihood of a widespread adoption

and, thus, the potential for recruiting a larger sample.

As per the focus group's recommendations, the compliant apps were Flo, "AAMI

Safe Driver" and "SafeDrive". All three apps were available in Australia and were free

to use. Data were collected for more than 2,000 kilometres of driving. It is worth noting

that common problems were encountered in all three apps during the tests. For

example, there were problems with the apps synchronising with GPS signal, when on

autostart, i.e. when the apps were not started manually by pressing a designated button.

There were also app-specific problems. For example, failure to record and analyse long

journeys and problems with calculating the overall score were encountered in "AAMI

Safe Driver". "SafeDrive" had problems with registering user information. Although

no major problems were encountered with Flo, shortcomings were not missing. For

example, short trips, in general, generated lower scores than long trips, which can

potentially put at disadvantage drivers that drive on shorter distances. Shortcomings

with "AAMI Safe Driver" and "SafeDrive" were encountered, too. For example

"AAMI Safe Driver" did not provide real-time feedback, while "SafeDrive" was

focusing on mobile phone usage.

6.5 CONCLUSION

The Focus group participants seemed willing to explore the positive potential

of technologies in their work. This presented an opportunity for the COTs evaluated

as part of the current thesis to be potentially applied on a larger scale in the real world.

With the current program of research providing evidence around using COTs in

interventions that go into the research participants’ free-living environment, the

findings discussed in the thesis may well add value beyond the scope of the thesis

itself.


The undertaken real driving review explored existing apps, i.e. whether they

are suitable to be deployed as an intervention tool in the framework of the current PhD

program of research. Real-time feedback and collecting driving data capabilities of the

smartphone safe-driving apps were essential for the current program of research. Thus,

Flo was better suited than "AAMI Safe Driver" and "SafeDrive". This decision was

supported by 1) the real driving review, 2) the smartphone safe-driving apps systematic

literature review (see Chapter 5), 3) the Focus group participants’ opinions and

recommendations (see Section 6.3), and 4) the theoretically-grounded apps' scores (see

Table 6.3), where Flo scored 5, the highest of the three tested apps. These

considerations lead to the selection of Flo to be evaluated as an intervention tool going

forward.

98 Chapter 7: Study 2 - Intervention with an off-the-shelf smartphone safe-driving app

Chapter 7: Study 2 - Intervention with an off-the-shelf smartphone safe-driving app

Chapter 7 presents the second study of this program of research. Study 2 assessed

whether a smartphone safe-driving app intervention could influence young drivers'

self-reported behaviour of not speeding and intention not to speed. First, a brief

introduction of the study is presented, before outlining its key aims, method and

hypotheses, followed by the study results, and a discussion.

7.1 INTRODUCTION

Risk prevention efforts yield unsatisfactory results, leaving the safety of young

drivers in focus (Scott-Parker et al., 2015). Young people continue to be

overrepresented in road fatalities (BITRE, 2018), with speeding causing 43% of them

(AONSW, 2011). At the same time, available in-car COTs, such as infotainment

systems or smartphones, compete with the primary driving task for drivers' attention,

threatening for existing road safety problems, e.g. driver distraction, to further

deteriorate (Parliament of Victoria Road Safety Committee, 2006). In parallel with

increasing concerns that problems related to technologies are likely to both increase

and evolve (WHO, 2011), both academia and businesses suggest that the power of

COTs, in general, and smartphones, in particular, provide untapped opportunities to

reduce risks, and, more importantly, to reduce risks amongst young drivers (see

Chapter 5 and Chapter 6).

Study 2 examined the effects of a safe-driving app intervention to reduce risky

driving behaviour, more specifically speeding, amongst young drivers. It evaluated an

intervention that deployed a safe-driving app, Flo (see Chapter 6 for the app selection

process), to transform existing risk sources (smartphones) into ones for motivating safe

driving behaviour. The study operationalised an approach, suggested by Schroeter et

al. (2012), to persuade young drivers to behave safer on the road without the need to

reduce the fun from driving or to incur substantial deployment costs.

Even though smartphones allow the monitoring and assessment of other risky

driving manoeuvres beyond speeding, such as hard acceleration, hard braking and fast

Chapter 7: Study 2 - Intervention with an off-the-shelf smartphone safe-driving app 99

cornering, those were not targeted. The focus of the current research was to understand

whether, and to what extent, the intervention with an off-the-shelf safe-driving app as

a tool positively influenced the young drivers' intention not to speed as well as their

subsequent self-reported behaviour of not speeding within the TPB evaluation

framework (see Figure 3.5 in Section 3.7). In Study 2, a quantitative evaluation of the

implemented safe-driving app intervention was conducted based on surveys-collected

data. The choice of TPB as an overarching framework to guide this quantitative

evaluation was carefully determined after a review of the most widely used theories in

the social and behavioural sciences (see Chapter 3).

Study 2 aimed to answer RQ2: "How do young drivers’ self-reported behaviour

of not speeding and intention not to speed alter in their free-living environment, as a

result of exposure to a smartphone safe-driving app intervention?"

The intervention had a span of three months. The variables identified as being of

most interest were:

1. Intention not to speed, measured before the intervention, as it related to

what the drivers planned to do during the following three months without

being influenced;

2. Behaviour of not speeding during the three months of the intervention,

measured after the intervention, as it reflected what the drivers had been

doing during the intervention.

In the literature, speeding is defined as "driving at an illegal speed over the limit"

or "driving at an inappropriate speed" or both. However, "driving at an inappropriate

speed" can be regarded as a less specific definition. Thus, for the current purpose,

speeding is defined as illegal behaviour, which sets a defined threshold for a behaviour

to be considered as speeding.

It was expected that as a result of the intervention participants in the Intervention

group would report significantly greater behaviour of not speeding during the three

months of the intervention as well as greater intention not to speed in comparison with

the Control group. To achieve the expected result, the smartphone safe-driving app Flo

was deployed to be used by the Intervention group participants over three months. It

served the purpose of a driving coach with feedback on the targeted behaviour in the

framework of this project (speeding), as well as other risky manoeuvres (hard


acceleration, hard braking and fast cornering). A driving coach with feedback was

expected to potentially influence the intervention participants' perceived behavioural

control (PBC, see Figure 3.5 in Section 3.7) by showing them that whatever they do

on the road had a consequence that could be measured and reported. Such an

experience may show the participants they are in control of what is reported through

their behaviour. If successful, through the established relations within the TPB, such

an influence would impact a) the dichotomised TPB constructs (see Subsection 4.4.2),

and b) the young drivers’ self-reported behaviour of not speeding and intention not to

speed.

Notwithstanding, the app of choice was not expected to increase distraction

significantly, thus, to negatively influence the participants' behaviour in regards to

initiating, monitoring/reading, and responding to social interactive technology on their

smartphones while driving. As a potential clue to the participant, the app provided

insights into the time each participant had their phone screen active while driving (see

Figure 7.1).

Figure 7.1. Time on screen as reported by Flo for each trip

Study 2 was analysed in four parts to address the overarching RQ2. The first part

looked at the whole sample of participants before the intervention was applied to part

of them (RQ2.1). It explored their intention not to speed and its predictors. The second

part focused on the changes the intervention might or might not have triggered in

regards to speeding (RQ2.2). The third part assessed how much of the self-reported

behaviour of not speeding during the three months of the intervention could have been

predicted with the data available before the intervention took place (RQ2.3). Finally,


the fourth part assessed whether an additional risk, namely increased distraction, was

introduced with the intervention (RQ2.4).

7.2 METHOD

7.2.1 Study design

Study 2 was designed as a randomised controlled experiment in the

participants' free-living environment. The intervention was implemented as part of

their normal daily routine with no intention that routine to be impacted by the

intervention. All participants completed the same questionnaire at the point of

recruitment. After recruitment, the list of participants was split by gender, to ensure

gender balance in both subsequent conditions. Without any other consideration, half

of the male participants and half of the female participants were randomly assigned to

the Intervention group. The remaining halves were assigned to the Control group. The

gender stratification was performed to ensure that both conditions reflect the original

gender balance of the recruited sample.

The Intervention group was asked to install and use for three months the

intervention tool, the Flo smartphone safe-driving app. The Control group was

instructed that they would be contacted after three months and that they are not

expected to do anything else before that. After the three-month intervention period

expired, both groups were contacted with a request to complete the identical second

questionnaire (see Appendix B). The design of the questionnaire is discussed in

Subsection 4.4.2 of this thesis.

7.2.2 Recruitment

Recruitment was conducted between the 01st and the 30th of April 2018.

Otherwise, potential participants were not time-constrained to consider participation.

As personal presence was not required, participants were allowed to complete the

survey at any time of the day and at any device, when and where it was convenient for

them.

Participants were initially approached online through QUT students' e-mail lists

and social media. A Facebook ad campaign was implemented as part of the social

media outreach. The campaign was set to target people who were aged 18 to 25;

resided in Australia; spoke English; had interests in smartphones, driving or motor

vehicles. The ad showed a static image of hands on a wheel and a look over a car


dashboard with a clear highway as a perspective. The accompanying text was

communicating to the viewer the additional requirement of driving 100 kilometres per

month, which was impossible to set as a targeting parameter of the ad campaign, and

was inviting them to complete the online survey. The text was also informing them

that the survey would take 10 minutes to complete. The campaign reached more than

50,000 people. The click-through rate was approximately 5%.

There was no cost related to the participants' involvement in the study. However,

the safe-driving app uses smartphone sensors. Continued use of those sensors, such as

GPS for example, running in the background can decrease battery life. The driving

app, we used, was built to use minimal power, but power consumption in the

smartphone was considered an indirect cost to be borne by the participants.

Participation in a random draw of gift vouchers was offered as an incentive to

participate. A participant was eligible for incentive after completing both surveys (at

Time 1 and at Time 2). The study had a total incentive fund of 1,500 AUD divided

into 10 Coles/Myer vouchers of 150 AUD each.

7.2.3 Intervention tool

The smartphone safe-driving app, Flo (see Figure 7.2 and Chapter 6 for

information about the selection process of Flo) was the intervention tool in Study 2.

Flo essentially served the purpose of a driving coach with feedback, aiming to help the

participating drivers learn to drive safer. It was installed on the intervention group

participants' smartphones.

Figure 7.2. Smartphone with Flo, providing real-time feedback while driving5.

5 Screenshot source: https://www.youtube.com/watch?v=A9GLHohkqbo


In general, Flo provides insights to the driver in real-time by tracking

movements, using GPS, and calculating values for cornering, acceleration and braking.

It can also silently work in the background, without providing real-time feedback, to

minimise distraction. Based on the gathered data, it assigns scores for every trip. The

app saves all trip data, including driving behaviour, which the driver can review after

driving either through a web profile or through the app itself. For users that join a

leaderboard, a score of their driving behaviour ranks them with respect to other users.

Driving scores and rankings were additional data that were collected through Flo for

some of the participants.

Flo automatically detects and records the driving trips, i.e. drivers do not need

to remember to start the app before every trip. However, it should be noted that during

a trip, Flo cannot distinguish between being a driver or just being a passenger. This

means that if the smartphone owner is just the passenger, the app will self-start and log

data as if they were the driver. This limitation led to the inclusion requirement of

"Drive a car as the only means of transport" for a participant to be eligible (as noted in

the participant information sheet).

7.2.4 Procedure

A participants' information sheet was provided to the participants online, as a

cover sheet of the first survey, before they started completing the survey. At Time 1

(April 2018), the self-completion questionnaires were used to collect data on

demographics, TPB constructs (self-efficacy, perceived controllability, instrumental

attitude, affective attitude, subjective norm, descriptive norm, intention not to speed

and behaviour of not speeding) and additional predictors (past speeding behaviour,

perceived risk, moral norm, peers' norm, impulsivity) (see Figure 3.5). Detailed

information on the items is provided in Subsection 4.4.3.

Upon completion of the Time 1 survey, the participants were randomly divided,

nevertheless gender-stratified as noted earlier, into a Control group with no

intervention (n=241, 117 females) and an Intervention group with app deployment

(n=243, 123 males) (see Figure 7.3). The participants in the Control group were not

required to do anything in relation to the study during the three-month intervention

period.


Figure 7.3. Safe-driving app intervention design

The Intervention group was instructed to install Flo from either Google Play or

iTunes, depending on the operating system of their smartphone. Subsequently,

invitations to join the in-app GoOz leaderboard group (see Figure 7.4) were issued

through the app web interface. The group was specifically created for the current study.

As some participants did not receive those invitations, a separate e-mail was sent to

them with instructions on how to initiate joining the group from their end. The group

allowed the research team to observe and periodically record how many trips each

participant made during the last 30 days, and what their current driving score was.

Figure 7.4. Example screenshot of Flo GoOz leaderboard

At Time 2 (three months after Time 1, August 2018), an invitation to complete

the second survey was sent to all participants. Two reminders were sent to participants


who did not complete the survey. The ones that completed it were excluded from the

reminders by matching their anonymous identifiers. The self-generated participants'

anonymous identifiers were used to link datasets from Time 1 and Time 2, related to

the same person.

7.2.5 Participants

Initially, 504 young drivers completed the first survey. Partially completed

surveys were not considered. After removing 23 duplicates, as well as one entry of

participants who explicitly requested in writing that they wanted to opt-out of the study

after they had completed the first survey, 480 cases (245 male; Mage = 20.88 years,

SD = 2.10) were retained for analysis. This number was well above the initially

targeted minimum desired number of 140 participants (see Subsection 4.4.1).

Of the 480 participants, 217 (45.2%) reported to have an open driver's licence,

193 (40.2%) reported having a provisional (Year 2) licence, 67 (14.0%) – a provisional

(Year 1) licence, and 3 (.6%) – a learner licence (see Subsection 2.3.1 for details on

the Australian GDL framework).

The biggest group of participants reported living in Victoria (141, 29.4%),

followed by New South Wales (132, 27.5%), Queensland (99, 20.6%), Western

Australia (42, 8.8%), South Australia (39, 8.1%), Tasmania (18, 3.8%), Australian

Capital Territory (8, 1.7%) and Northern Territory (1, 0.2%). The sample distribution

was aligned with the general distribution of the population in Australia6. The

Australian Bureau of Statistics reports 32.0% of the population as living in New South

Wales, 25.8% in Victoria, 20.0% in Queensland, 10.4% in Western Australia, 7.0% in

South Australia, 2.1% in Tasmania, 1.7% in the Australian Capital Territory and 1%

in the Northern Territory. Data by states and territories was not explored further in the

analysis due to low numbers in some of the cases and because no evidence could be

found in the literature that pointed towards a potential effect of the participants'

geographic location within Australia.

At Time 2, 210 young drivers (109 male; Mage = 21.01 years, SD = 2.12)

completed the second questionnaire. The dropout rate of 56.25% exceeded the initially

6 http://www.abs.gov.au/ausstats/[email protected]/latestProducts/3101.0Media%20Release1Mar%202018 (Accessed on 18/12/18)


expected 14.29%, as encountered by Musicant and Lotan (2015). Nevertheless, due to

the larger initial sample, the collected 210 cases were more than the 120 initially

sought. The Intervention group remained with 84 participants, while the Control group

remained with 126. Each of the two groups had more than the initially targeted min

n=60, surpassing two and three times, respectively, the n=42 reported by Musicant and

Lotan (2015).

A look at the Flo leaderboard revealed that only 62 participants joined it. Thus a

decision was taken to revise the number of participants, which were considered an

Intervention group, to make sure that the analysed data belonged to people who

actively participated in the intervention. Twelve of the 62 participants in the

leaderboard never generated a score, suggesting that they did not use the app after

signup, and were therefore removed. As a result, 50 participants remained as being

considered part of the Intervention group. The data of 19 of those 50 participants could

not be reliably linked between the leaderboard and the survey data through the

anonymous identifiers. Thus 31 entries remained for the main analysis in this thesis.

It has to be acknowledged that for the purpose of analysis, the Intervention group

can be constituted differently, depending on the analyst’s preferences. For example, it

can be argued that all Intervention group participants, who completed the second

questionnaire, are part of the Intervention group as there is no evidence that they did

not use the app. Another option would be to analyse data from participants who are

considered highly engaged in the study, e.g. generated a score during more than half

of the intervention period. A third option could be to have a demographic match

between the Intervention group and the Control group. Those additional options might

potentially provide evidence for different impacts of the implemented intervention.

Should the reader prefer the analysis to be based on one of the three presented

variations of Intervention group constitution, these alternative analyses are presented

in Appendix D. However, in this chapter, a participant is considered to be part of the

Intervention group only when data could be reliably linked across the two surveys, and

the Flo leaderboard as this is sufficient evidence that they used the smartphone app

and therefore underwent the intervention as defined in this thesis.

7.2.6 Intervention

Between completing the survey at Time 1 (April 2018, before the intervention)

and Time 2 (three months after Time 1, August 2018), the participants from the


Intervention group were subjected to Flo, the off-the-shelf safe-driving app. Only those

participants were expected to use Flo for three months, while they were driving. They

were not required to complete any tasks outside their normal daily routine so that their

free-living environment (see RQ2) remains intact by the current study. Using the app

was expected to persuade them to adopt a safer driving behaviour as evidenced by their

self-reports at Time 2 with self-reports at Time 1 as a baseline. The participants from

the Control group were not expected to do anything during that same period.

7.3 HYPOTHESES

The deployed extended TPB framework (Ajzen, 1988) (see Section 3.7)

constitutes a powerful model for evaluating interventions. It identifies determinants of

behaviour that can potentially be influenced, thus, modified. It further identifies other

determinants that cannot be influenced but still determine behaviour. In that respect,

initially looking at the predictors of intention not to speed at baseline to address RQ2.1,

it was hypothesised that:

H.1. Demographic variables (gender, age and driving license) would account

for a significant variation in intention not to speed. Speed is a major contributor

to crashes (AONSW, 2011), and the increased crash risk is shown to associate

with both inexperience and age (McCartt et al., 2009). Also, gender is shown

to play a role in the young drivers' risky behaviours (Scott-Parker, 2012).

H.2. TPB constructs (instrumental attitude, affective attitude, subjective norm,

descriptive norm, self-efficacy and perceived controllability) would account

for a significant variation in intention not to speed, over and above the

demographic variables. According to Sniehotta et al. (2014), TPB may lack

sufficient predictive power in longitudinal studies, or in studies with samples,

coming outside the university campuses, a view not shared by Conner (2015),

and not supported by findings of studies the current one leverages on, e.g.

Elliott and Thomson (2010).

As suggested by Conner (2015), an extended theoretical framework may address

widely noted criticisms and limitations of the TPB, as well as provide additional

insights on the effect of the deployed intervention. Additional predictors may also

explain additional variation in the DVs, after controlling for TPB variables. A detailed


discussion about the additional predictors, used to extend TPB, is presented in Section

3.6. Thus, it was hypothesised that:

H.3. Additional predictors (past behaviour of not speeding, perceived risk,

moral norm, peers' norm, impulsivity, sensitivity to reward and sensitivity to

punishment) would account for a significant variation in intention not to speed,

over and above the TPB variables.

New technology designed to persuade drivers can help young drivers adopt safer

driving behaviour in naturalistic settings as a result of safe-driving apps interventions,

focused on speeding (Li et al., 2016; Musicant & Botzer, 2016; Zhang et al., 2014)

(see Chapter 5). Thus, the implemented intervention provided a basis to investigate

RQ2.2 or the actual safety benefits of a safe-driving app. For the effects of the

intervention, it was hypothesised that after the intervention:

H.4. Participants in the Intervention group would report significantly greater

intention not to speed in the future than the Control group participants.


behaviour of not speeding during the three months of the intervention than the

Control group participants.

H.6. Due to the expectation that the app can influence attitudes, PBC, moral

norm and peers' norm (see Section 6.4), the safe-driving app intervention

would have positively influenced the Intervention group participants'

instrumental attitude, affective attitude, self-efficacy and perceived

controllability, moral norm and peers' norm directly. Through the correlations

of those constructs within the framework, we also expected subjective norm,

descriptive norm and perceived risk to be influenced indirectly.

H.7. The intervention would improve the participants' driving, as represented

by the observed driving scores in the safe-driving app leaderboard.

In intervention design, knowing what your participants would do in the future

could be very useful information. Such knowledge could help intervention designers

calibrate interventions better, and in advance, to address potential risks. Thus, to

address RQ2.3, predictors of behaviour of not speeding during the three months of the

intervention were investigated to assess how much of the behaviour during the

intervention, reported at Time 2, could have been predicted with the information


available before the intervention, as reported at Time 1. In that respect, it was

hypothesised that:

H.8. Demographic variables (gender, age and driving license) would account

for a significant variation in behaviour of not speeding during the three months

of the intervention.

H.9. TPB constructs (intention not to speed, self-efficacy and perceived

controllability) would account for a significant variation in behaviour of not

speeding during the three months of the intervention, over and above the

demographic variables.

H.10. Additional predictors (past behaviour of not speeding, perceived risk,

moral norm, peers' norm, impulsivity, sensitivity to reward and sensitivity to

punishment) would account for a significant variation in behaviour of not

speeding during the three months of the intervention, over and above the TPB

variables.

Mobile phones are identified as a major source of distraction while driving

(WHO, 2011). Thus, the current intervention provided an opportunity to investigate

RQ2.4, or if despite the good intentions of implementing the intervention, the safe-

driving app did not increase the drivers' distraction. In that respect, potential negative

effects were investigated. For the self-reported smartphone interaction behavioural

measures, it was hypothesised that:

H.11. The intervention would not significantly increase the Intervention group

drivers' distraction in terms of initiating (less), monitoring/reading (less) or

responding (less) to communication in comparison to the Control group

participants.

7.4 PRELIMINARY ANALYSIS

Before analysing the results in detail, a preliminary data analysis was performed,

to deal with missing data, to transform data, to decide how to deal with dropouts, and

to establish the participants' profiles in regards to their personality characteristics.

These preliminary analyses are discussed in turn.


7.4.1 Missing data

There was no missing data. The setup of the data collection required all questions

to be compulsorily answered with a limited number of answer options to choose from,

i.e. successful submission of the online questionnaire could only occur if once all

questions were completed.

7.4.2 Data transformation

Following data collection, measures for intentions, attitudes, norms (with the

exclusion of moral norm) and behaviour were recoded (transformed), so that higher

scores indicated greater agreement with the construct (perceived negatively-geared

answer to the left of the scale, smaller value, and perceived positively-geared answer

to the right of the scale, higher value). The reported Boolean answers on smartphone

use (initiated (less) communication, monitored/read (less) communication, and

responded (less) to communication) were assigned numerical values and, thus, were

converted into a scale from 1 (More than once per day) to 7 (Never), i.e. again,

perceived negatively-geared answer to the left of the scale, smaller value, and

perceived positively-geared answer to the right of the scale, higher value.

The two separate intention questions "To what extent do you intend to drive

faster than the speed limit over the next 3 months?" (A great extent to no extent at all

after recoding) and "How often do you think you will drive faster than the speed limit

in the next 3 months?" (All the time to never after recoding) for the construct intention

were strongly and significantly correlated (Pearson's r = .79, p < .001). Thus, they were

combined (through finding an average) into a single measure intention not to speed.

The other TPB measures were retained as separate in the analysis in accordance with

the Elliott and Thomson's (2010) model, an approach discussed in further detail in

Subsection 4.4.2.

The questions "If you were to drive over the speed limit over the next 3 months,

how much would you worry about being involved in a road crash?" (Not at all worried

to worried very much) and "If you were to drive over the speed limit over the next 3

months, how much would you worry about being caught by the Police?" (Not at all

worried to worried very much) for the construct perceived risk were strongly and

significantly correlated (Pearson's r = .53, p < .001). Thus, they were combined


(through finding an average) into a single measure perceived risk, to be used as an

additional predictor.

Gender was recoded, from a string, into a numeric variable, with "0" denoting

males and "1" denoting females.

For the regression analysis, driving licence was recoded from a string into a

numeric variable with "0" denoting a learner license, "1" denoting a provisional (year

1) license, "2" denoting a provisional (year 2) license, and "3" denoting an open

license.

For the ANCOVA analysis, driving licence was recoded from a string into a

numeric variable with "1" denoting both a provisional (year 1) license and a

provisional (year 2) license, and "2" denoting an open license. The different recoding

was implemented because none of the learner drivers and only a low number of

provisional (year 1) drivers completed the second survey.

For the ANCOVA analysis, participants were grouped according to the

following scores:

- BIS-11 scores for impulsivity, as per Stanford et al. (2009):

o Denoted with "1", participants with scores lower than 52 were

part of the low impulsivity group.

o Denoted with "2", participants with scores between 52 and 71

were part of the "normal" impulsivity group.

o Denoted with "3", participants with scores above 72 were part of

the high impulsivity group.

- Sensitivity to punishment score, with a cut-off the mean scores of the

distribution, or 13.51 (SD=5.60):

o Denoted with "1", participants with scores lower than 13,

inclusive, were part of the low sensitivity to punishment group.

o Denoted with "2", participants with scores above 14, inclusive,

were part of the high sensitivity to punishment group.

- Sensitivity to reward score, with a cut-off the mean scores of the

distribution, or 11.50 (SD=4.60):

o Denoted with "1", participants with scores lower than 11,

inclusive, were part of the low sensitivity to reward group.


o Denoted with "2", participants with scores above 12, inclusive,

were part of the high sensitivity to reward group.

7.4.3 Dropouts

A preliminary one-way MANOVA was performed in regards to the provided

answers on the extended TPB variables (11 DVs) at Time 1. The IV was whether Time

2 questionnaire was completed or not. Homogeneity assumption was met, with a Box's

M p = .34. No significant difference was found for the DVs (Wilks' Lambda = .99, F

(11, 468) = .63, p = .80, ηp2 = .015) between the participants who completed only the

Time 1 questionnaire and those who completed both questionnaires. A further look

into each DV with a Bonferroni-adjusted α level benchmark set at .004 (standard value

of .05 divided by 11, the number of DVs), did not reveal any significant difference

between the two groups on any of the measures. Thus, the data collected from the

participants at Time 2 could be considered representative for the participants of the

whole sample. Nevertheless, the full data set was retained in the analysis at Time 1.

7.4.4 Assumptions checks

To determine whether random assignment to the Intervention and the Control

group was successful, a preliminary one-way between-groups MANOVA was

performed on answers about demographics (so variables gender, age and driving

experience as depicted by driving license) at Time 1. The IV was the condition, so

Intervention or Control. No significant difference between the two groups of

participants was found (Wilks' Lambda = 1.00, F (3, 476) = .09, p = .97, ηp2 = .001).

Thus, the random assignment to the two conditions was considered successful.

Normality for the DVs, intention not to speed, measured at Time 1, and

behaviour of not speeding during the three months of the intervention, measured at

Time 2, was assessed statistically, via skewness and kurtosis, and visually, via

histograms and Q-Q plots. Intention not to speed was negatively skewed at Time 1 (-

.68, std. error = .17, z = -6.09). The value for kurtosis was -.65 (std. error = .22, z = -

2.91). Although the absolute values fell within the generally accepted range of -2:2,

the calculated z-value for skewness suggested a departure from normality (Kim, 2013).

However, Kim (2013) suggests that z-values be ignored in samples of more than 300

participants. Histograms, with imposed normal curves, and Q-Q plots examination

suggested normal distribution. The Shapiro-Wilk test suggested a non-normal


distribution (p < .001). However, it is considered very sensitive and potentially

unreliable in samples > 50 (Elliott & Woodward, 2007).

Behaviour of not speeding during the intervention was positively skewed (.04,

std. error = .17, z = 0.24) with negative kurtosis (-1.21, std. error = .33, z = 3.63). Both

values fell within the generally accepted range of -2:2. Similar to the case of intention

not to speed, the calculated z-values suggested a departure from normality, but could

safely be ignored (Kim, 2013). The histogram, with an imposed normal curve, and a

Q-Q plot examination, suggested normal distribution while Shapiro-Wilk did not

suggest one (p < .001).

An examination of the Boxplots in each of the two DVs' case did not reveal

outliers, and also suggested normal distribution. Given that despite negative kurtosis

underestimation of variance disappears in samples > 200 (Tabachnick & Fidell, 2007),

the present study assumed normality was sufficient to explore the data with parametric

tests.

Additional normality checks were performed together with investigating each

regression model. Visual inspections of the regressions standardised residual P-P plots

and Scatterplots suggested no major deviations from normality. Results showed a

Mahalanobis distance above the suggested values (Tabachnick & Fidell, 2007).

However, subsequent inspections of the data files revealed only one or two cases

exceeding the suggested values for different tests. The maximum Cook’s distances

were negligible, indicating that the identified outliers did not have a major influence

on the data analysis.

Multicollinearity was investigated as part of the regression analysis due to high

correlation coefficients between some variables. This was most noticeable between

past behaviour of not speeding and intention not to speed (r=0.83, p < .001), which

were both entered as IVs at Step 3 of the linear regression model to predict behaviour

of not speeding during the three months of the intervention. All variance inflation

factors had a value lower than four, which in those specific cases reflected the strong

correlations. As such, there was no need to remove those variables from the analysis.

Lastly, the assumptions were met in the one-way ANCOVA tests. In some of the

two-way ANCOVA tests, the assumption of equality of variance was not met. No α

adjustments were necessary when there was no significant interaction effect. In one


case with a significant interaction effect, a lower, more conservative α (.025) was

adopted when exploring the data (Wickens & Keppel, 2004).


Consistent with Patton and Stanford (1995), the internal reliability analysis of

BIS-11 data, collected at Time 1, revealed high internal consistency with a Cronbach's

α of 0.84. The generally acceptable limit is 0.7 (DeVellis, 2016). The participants

(n=480) had a mean score of 58.62 (SD=10.33).

The internal reliability check of SPSRQ data, collected at Time 2, revealed high

internal consistency in both components, sensitivity to punishment (Cronbach's α of

0.84) and sensitivity to reward (Cronbach's α of 0.81). The participants (n=210) had a

mean score of 13.51 (SD=5.60) on the sensitivity to punishment scale and 11.50

(SD=4.60) on the sensitivity to reward scale.

7.5 RESULTS

7.5.1 Participants' intention not to speed before the intervention (RQ2.1, H.1 - H.3)

The following analysis of data, collected at baseline, was guided by RQ2.1 (What

did we know about the participants before the intervention, and to what extent could

the extended TPB framework predict their intention not to speed?), and by H.1, H.2

and H.3 respectively (see Section 7.3).

7.5.1.1 Means, standard deviations and bivariate correlations.

Table 7.1 below presents the means, standard deviations and Pearson's r

correlations for the TPB variables. Although on average participants considered that

they were in control of their not speeding (mean perceived controllability of 7.41 on a

9-point scale), they were not very certain they will be able to perform not speeding

(mean self-efficacy of 5.70 on a 9-point scale). Such uncertainty may partially explain

their comparatively low (close to the scale mid-point) reported past behaviour of not

speeding score (5.54 on a 9-point scale). The lowest mean score (5.14 on a 9-point

scale) was given for descriptive norm, which can be interpreted that everyone thought

their important others are speeding. On all measures, a reasonably-wide variability of

the scores was observed with the full range of possible responses being used.


Table 7.1. Means, standard deviations and bivariate correlations for the TPB variables at Time 1 (n=480).

Mean SD 1 2 3 4 5 6 7 8

1. Past behaviour of not speeding 5.54 2.38 - .83** .58** .56** .32** .39** .56** .16**

2. Intention not to speed 6.27 2.31 - .62** .58** .39** .37** .55** .18**

3. Instrumental attitude 6.26 2.21 - .66** .50** .35** .43** .15**

4. Affective attitude 5.44 2.47 - .40** .26** .41** .14**

5. Subjective norm 7.03 2.26 - .34** .29** .14**

6. Descriptive norm 5.14 2.15 - .33* .06

7. Self-efficacy 5.70 2.69 - .37**

8. Perceived controllability 7.41 2.11 -** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed).

Table 7.1 shows consistency with TPB in that past behaviour of not speeding is

highly correlated with intention not to speed (r=0.83, p < .001). Although it meant that

whoever speeded in the past would do that again in the future, it confirmed a strong

link to be explored. The table also shows that both past behaviour of not speeding and

intention not to speed were significantly correlated with all of their underlying

constructs. However, subjective norm and descriptive norm were moderately

correlated with them, while the correlations with instrumental attitude and affective

attitude were strong. Self-efficacy was strongly correlated with both past behaviour of

not speeding and intention not to speed, while perceived controllability exhibited a

weak correlation with the two constructs.

7.5.1.2 Predictors of intention not to speed

Following the order of entry described in Section 4.4.4, a 3-step hierarchical

multiple regression was conducted to assess which measures (demographics, TPB and

additional predictors), and to what extent, account for the variance in the participants’

self-reported intention not to speed for the whole sample (n=480) at Time 1.

As shown in Table 7.2, at Time 1, the demographic variables explained a

significant 6% (adj. R2 = .06, p < .001) of the variance in intention not to speed.

Nevertheless, the explained variance was very small, and age did not emerge as a

significant predictor. However, gender (β=.20, p < .001) and driving license (β=-.17,

p = .006) were statistically significant independent predictors.

These results were consistent with H.1, which predicted that demographic

variables would account for a significant variation in intention not to speed.


Table 7.2. 3-step hierarchical multiple regression analysis, predicting intention not to speed for all participants at Time 1, with demographic factors, TPB variables and additional variables as predictors

(n=480).

Step Variables R2 R2change Fchange Step 1 β Step 2 β Step 3 β Step 3

sr2

Bivariate

R2

1 Gender 0.06** 0.06 10.28** 0.204** 0.005 0.047 0.002 0.04**

Age 0.063 0.067 -0.006 <.001 <.01

Driving license -0.166* -0.111* -0.008 <.001 0.02*

2 Instrumental

attitude

0.54** 0.48 80.88** 0.284** 0.143** 0.008 0.39**

Affective

attitude

0.212** 0.054 0.001 0.33**

Subjective norm 0.019 0.051 0.002 0.15**

Descriptive

norm

0.141** 0.026 <.001 0.14**

Self-efficacy 0.309** 0.103* 0.006 0.31**

Perceived

controllability

-0.020 0.003 <.001 0.03**

3 Past behaviour

of not speeding

0.73** 0.19 66.82** 0.638** 0.183 0.69**

Impulsivity -0.008 <.001 0.05**

Perceived risk -0.061* 0.002 0.11**

Moral norm -0.004 <.001 0.19**

Peers' norm 0.015 <.001 0.17**

All beta weights are standardised. * p < .05 ** p < .001

Adding the TPB variables, at Step 2, significantly increased the explained

variance (ΔR2 = .48, p < .001). Thus, the explained variance exceeded 50%. Four TPB

variables emerged as significant predictors: instrumental attitude (β=.28, p < .001),

affective attitude (β=.21, p < .001), descriptive norm (β=.14, p < .001) and self-efficacy

(β=.31, p < .001), as well as driving license (β=-.11, p = .009).

The results were consistent with H.2, which predicted that TPB constructs would

account for a significant variation in intention not to speed, over and above the



Adding the additional predictors, at Step 3, significantly increased the explained

variance, over and above TPB (ΔR2 = .19, p < .001). The statistically significant

independent predictors in the final regression equation were past behaviour of not

speeding (β=.64, p < .001), instrumental attitude (β=.14, p < .001), self-efficacy

(β=.10, p = .001) and perceived risk (β=-.06, p = .046).

Exploring the individual bivariate relations between the DV and the IVs (final

column in Table 7.2) showed that, if considered separately, all IVs, except age, were

statistically significant predictors of intention not to speed. The three strongest

individual predictors were past behaviour of not speeding, instrumental attitude and

affective attitude, which explained 69%, 39% or 33% of the variance, respectively.

However, when all IVs were considered in an overall model, past behaviour of not

speeding uniquely explained the highest percentage of the variance, 18% in this case.

Assessing the contribution of sensitivity to punishment and sensitivity to reward

as additional predictors required the regression test to be run only for the 210

participants who completed the second survey (see Table 7.3), as SPSRQ was

administered only at Time 2. With sensitivity to punishment and sensitivity to reward

added as additional predictors at the multiple hierarchical regression Step 3, the

explained variance over and above TPB was a significant 75% (adj. R2 = .73, p < .001),

2% more than in the equation assessing the full sample. The statistically significant


speeding (β=.58, p < .001), instrumental attitude (β=.13, p = .038) and self-efficacy

(β=.18, p < .001).

Exploring the individual bivariate relations showed that, while sensitivity to

reward was a statistically significant predictor of intention not to speed, sensitivity to

punishment was not. Past behaviour of not speeding remained the strongest individual

predictor of intention not to speed, explaining 67% of the variance. Once again, when

all IVs were considered in an overall model, past behaviour of not speeding uniquely

explained the most variance, 16%. Self-efficacy also emerged as a noticeable unique

predictor, explaining 1.7% of the variance.


Table 7.3. Linear multiple regression analysis predicting Intention not to speed at Time 1 with demographic factors, TPB variables and additional variables, including sensitivity, as predictors

(n=210).


sr2

Bivariate

R2

1 Gender 0.07* 0.07 5.08* 0.188* 0.017 0.028 0.001 0.04*

Age 0.101 0.082 0.007 <.001 <.01

Driving license -0.236* -0.107 -0.019 <.001 0.03*

2 Instrumental

attitude

0.58** 0.51 39.90** 0.307** 0.128* 0.006 0.39**

Affective

attitude

0.191* 0.076 0.003 0.34**

Subjective norm 0.009 0.081 0.004 0.16**

Descriptive

norm

0.111* 0.009 <.001 0.10**

Self-efficacy 0.385** 0.178** 0.017 0.35**

Perceived

controllability

-0.067 -0.030 0.001 0.03*

3 Past behaviour

of not speeding

0.75** 0.17 19.43** 0.577** 0.159 0.67**

Impulsivity 0.006 <.001 0.04*

Perceived risk -0.004 <.001 0.12**

Moral norm -0.036 0.001 0.20**

Peers' norm 0.026 <.001 0.18**

Sensitivity to

punishment

-0.018 <.001 <.01

Sensitivity to

reward

-0.053 0.002 0.11**


The results provide support for H.3, which predicted that additional predictors

(past speeding behaviour, perceived risk, moral norm, peers' norm, impulsivity,

sensitivity to reward and sensitivity to punishment) would account for a significant

variation in intention not to speed, over and above the TPB variables.


7.5.2 Changes in salient beliefs (RQ2.2, H.4 - H.7)

As discussed in Section 7.2.5, a total of 210 completed questionnaires were

collected at Time 2, 126 from Control group participants and 84 from Intervention

group participants. 31 of the 84 Intervention group participants could be linked to the

data collected in the app leaderboard. Therefore, the analysis in the current section is

based on those 31 entries. It is guided by RQ2.2. Did the intervention change the

participants' salient beliefs, as depicted by the TPB constructs?, as well as by H.4,

H.5, H.6 and H.7.

7.5.2.1 Means, standard deviations and bivariate correlations.

At Time 2, approximately three months after the intervention was deployed, the

relations between the TPB measures remained statistically significant (see Table 7.4).

The most notable differences in comparison to Time 1 were that participants reported

lower scores on average.

Table 7.4. Means, standard deviations and bivariate correlations for the standard TPB variables at Time 2 (n=157).

Mean SD 1 2 3 4 5 6 7 8

1. Behaviour of not speeding during

the three months of the intervention5.24 2.21

-.82**.55**.63**.29**.35**.61**.21**

2. Intention not to speed 5.85 2.20 - .52**.67**.27**.34**.60**.23**

3. Instrumental attitude 6.22 1.83 - .67**.40**.29**.42**.30**

4. Affective attitude 5.34 2.21 - .33**.28**.43**.21**

5. Subjective norm 6.69 2.27 - .42**.23**.25**

6. Descriptive norm 5.05 1.97 - .22** .13

7. Self-efficacy 5.34 2.54 - .29**

8. Perceived controllability 7.33 1.95 -** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed).

7.5.2.2 Impact of the intervention

A series of one-way ANCOVA tests was performed to evaluate the effect of the

intervention on the DVs, so a) intention not to speed (measured at Time 2) and b)

behaviour of not speeding during the three months of the intervention. The condition,

so Control group and Intervention group, was the fixed factor IV. The covariates to

control for pre-existing conditions within the two groups were a) intention not to speed

(measured at Time 1) and b) past behaviour of not speeding (measured at Time 1).

A series of two-way ANCOVAs assessed whether a number of individual

differences and personal characteristics, previously discussed as potential influencers


on the way young people behave on the road (see Section 3.6), moderated the effect of

the participants' condition. Those were the mentioned-earlier demographic variables

(gender and driving experience, denoted by driving license) and additional predictors

(impulsivity, sensitivity to reward and sensitivity to punishment). Those were chosen

as they are stable in time and, thus, are not expected to be influenced by an

intervention.

In the case of intention not to speed, an inspection of the mean scores revealed

that the Control group reported a greater decrease of their intention not to speed than

the Intervention group (see Table 7.5).

Table 7.5. Means and standard deviations for the Control and the Intervention groups' intention not to speed at Time 1 and Time 2 (n=157).

Condition Mean Std. Deviation NTime 1 Control 6.38 2.243 126

Intervention 6.36 2.165 31Total 6.37 2.221 157

Time 2 Control 5.78 2.256 126Intervention 6.13 1.958 31Total 5.85 2.199 157

After adjusting for the participants' self-reported intention not to speed before

the intervention, no significant difference between the Control group and the

Intervention group was found in intention not to speed at Time 2, F (1, 154) = 1.14, p

= .28, ηp2 = .007. There was a statistically significant (p < .001) strong relationship

between intention not to speed at Time 1 and intention not to speed at Time 2, as

indicated by a ηp2 = .417. After finding the non-significant effect of the intervention

between the two groups in respect to their intention not to speed, two-way ANCOVAs

found no significant effects on the result with personality characteristics as moderators

either (see Table 7.6).

Table 7.6. Interaction effects between Condition and personality characteristics, intention not to speed, adjusted for Time 1 values (n=157).

Moderator F (1, 152) p ηp2

Gender .161 .689 .001

Driving experience 2.140 .146 .014

Impulsivity .042 .959 .001

Sensitivity to punishment 2.280 .133 .015

Sensitivity to reward .369 .544 .002


These results did not support H.4, which predicted that participants in the

Intervention group would report significantly greater intention not to speed in the

future than the Control group participants.

Similar to intention not to speed, for behaviour of not speeding (self-reported

frequency of not speeding before and during the intervention), an inspection of the

mean scores revealed that the Control group reported a greater reduction of its score

than the Intervention group (see Table 7.7).

Table 7.7. Means and standard deviations for the Control and the Intervention groups' behaviour of not speeding at Time 1 and Time 2 (n=157).

Condition Mean Std. Deviation N Time 1 Control 5.54 2.26 126

Intervention 5.84 2.44 31 Total 5.60 2.29 157

Time 2 Control 5.15 2.23 126 Intervention 5.61 2.14 31 Total 5.24 2.21 157

After adjusting for the participants' self-reported past behaviour of not speeding

before the intervention, no significant difference between the Control group and the

Intervention group was found in past behaviour of not speeding during the three

months of the intervention, F (1, 154) = .67, p = .41, ηp2 = .004. There was a statistically

significant (p < .001) strong relationship between past behaviour of not speeding

before the intervention and past behaviour of not speeding during the three months of

the intervention, as indicated by a ηp2 = .510. After finding the non-significant effect

of the intervention between the two groups in respect to their past behaviour of not

speeding during the three months of the intervention, two-way ANCOVAs found no

significant effects on the result, with personality characteristics as moderators, either

(see Table 7.8). The assumption of equality of variance was not met when investigating

the interaction effect between the group condition and driving experience, and

impulsivity. Despite that this created a bias in the obtained result, given that there was

no significant interaction effect, no α adjustments were necessary.


Intervention group would report significantly greater behaviour of not speeding during

the three months of the intervention than the Control group participants.


Table 7.8. Interaction effects between Condition and personality characteristics, past behaviour of not speeding during the three months of the intervention, adjusted for Time 1 values (n=157).


Gender .586 .445 .004



Sensitivity to punishment .264 .608 .002


Thus, the intervention did not have any significant effect on either of the DVs,

intention not to speed and past behaviour of not speeding during the three months of

the intervention. No significant effect was found on any of the other potentially-

modifiable Time 2 extended TPB variables, after adjusting for Time 1 values, either

(see Table 7.9).

Table 7.9. Effect of the intervention on TPB variables, adjusted for Time 1 values (n=157).

Variable F (1, 154) p ηp2

Instrumental attitude .094 .759 .001

Affective attitude .007 .935 < .001

Subjective norm .616 .434 .004

Descriptive norm .080 .778 .001

Self-efficacy 1.093 .297 .007

Perceived controllability 1.392 .240 .009

Moral norm .718 .398 .005

Peers' norm .348 .556 .002

Perceived risk 1.488 .224 .010

These results did not support H.6, which predicted that the safe-driving app

intervention would have positively influenced the Intervention group participants'

instrumental attitude, affective attitude, self-efficacy and perceived controllability,

moral norm and peers' norm directly as well as subjective norm, descriptive norm and

perceived risk indirectly.

Three more variations of the Intervention group were investigated, to build a

more comprehensive picture of the potential effects of the intervention: 1) a subgroup


of 18 highly engaged Intervention participants, ones that had a score generated in more

than half of the intervention period, were compared to all 126 entries in the Control

group; 2) all 126 entries in the Control group were compared with all 84 entries in the

Intervention group; and 3) a sub-sample of 31 participants from the Control group was

selected to match, as close as possible, the sample of 31 active Intervention group

participants. The selection in the third case was made by looking at the participants'

demographics in the following order: state of origin, gender, age, and driving licence.

When a complete match was not possible, a close one was selected, e.g. the same

gender, age and driving licence but in a different state, or the same state of origin,

gender and driving licence but age, plus or minus 1. Those results are provided as

Appendix D.

7.5.2.3 Leaderboard driving data

Out of 243 invited participants, 62 successfully joined the GoOz group in the

safe-driving app. Thus, the overall installation success rate was 26%. After three

months of intervention, only 18 out of the 62 participants in the in-app leaderboard

maintained an active status, i.e. had a generated score in their profile. Fifty of the

participants had a score at one or another time during the intervention, which suggested

that 12 could not get the app to collect their driving data at all. Only 15 of the

participants had scores generated in each of the 13 intervention weeks.

Given that score generation depends on route, driving style, car, etc., comparing

scores between participants would not provide a lot of insight; neither would

comparing the absolute change in scores for the respective participant. Thus the

relative change in scores was computed for each of the participants, that had more than

one generated score, by subtracting their first score value from their last one and

dividing the result by their first score. The average result was an improvement of 2%,

with values ranging from -92% to 380%. Fifteen out of 50 participants with scores

registered improvement in their driving score, 22 showed deterioration, and for 13

there was no change. Potential improvement seemed not to play a role in whether a

participant remained active until the end of the study. Out of the 18 participants that

were active at week 13, 10 registered an overall increase in their score, while 8 had a

decrease.

These results provided limited support for H.7, which predicted that the

intervention would improve the participants' driving, as represented by the observed


driving scores in the safe-driving app leaderboard. The hypothesis was valid for 30%

of the observed cases and not valid for the other 70%, i.e. the majority of the sample.

7.5.3 Predictors of behaviour of not speeding during the intervention (RQ 2.3, H.8 - H.10)

The analysis of the data collected after the intervention was guided by RQ2.3

(Using the extended TPB framework, to what extent could the data available before

the intervention predict the participants' behaviour of not speeding during the

intervention?), and by H.8, H.9 and H.10.

In this analysis, data from all 210 participants who completed the second

questionnaire was assessed, irrespective of their condition, Control or Intervention,

whether they managed to join the app leaderboard, or how active they were, as no

evidence for any statistically significant effect from the intervention was found in the

previous analysis. A 3-step hierarchical multiple regression analysis was performed to

identify which measures (demographics, TPB and additional predictors), and to what

extent, account for the variance in the participants’ self-reported behaviour of not

speeding during the three months of the intervention. The order of entering IVs was

guided by the model, described in Section 4.4.4.

As shown in Table 7.10, the demographic variables from Time 1 explained a

significant 10% (adj. R2 = .08, p < .001) of the variance in behaviour of not speeding

during the three months of the intervention. Gender (β=.17, p = .011) and driving

license (β=-.34, p < .000) were statistically significant independent predictors. Age did

not emerge as a significant predictor.

The results were consistent with H.8, which predicted that demographic

variables would account for a significant variation in behaviour of not speeding during

the three months of the intervention.

Adding the TPB variables, at Step 2, significantly increased the explained

variance, over and above demographics (ΔR2 = .40, p < .001). Thus, the explained

variance reached 50%. Two TPB variables emerged as significant predictors. The TPB

variables, intention not to speed (β=.51, p < .001) and self-efficacy (β=.16, p = .014),

were statistically significant independent predictors, as well as driving license (β=-.20,

p = .007).


The results were consistent with H.9, which predicted that TPB constructs would

account for a significant variation in behaviour of not speeding during the three months

of the intervention, over and above the demographic variables.

Table 7.10. 3-step hierarchical multiple regression analysis, predicting behaviour of not speeding during the three months of the intervention for all participants at Time 2, with demographic factors,

TPB variables and additional variables as predictors (n=210).


sr2

Bivariate

R2

1 Gender 0.10** 0.10 7.25** 0.169* 0.046 0.025 <.001 0.03*

Age 0.149 0.086 0.031 <.001 <.01

Driving license -0.338** -0.196* -0.156* 0.011 0.05*

2 Intention not to

speed

0.50** 0.40 53.90** 0.513** -0.001 <.001 0.44**

Self-efficacy 0.164* 0.086 0.004 0.27**

Perceived

controllability

0.098 0.092 0.007 0.06**

3 Past behaviour

of not speeding

0.64** 0.14 11.33** 0.578** 0.103 0.55**

Impulsivity 0.010 <.001 0.05*

Perceived risk 0.082 0.004 0.12**

Moral norm -0.024 <.001 0.16**

Peers' norm 0.043 0.001 0.13**

Sensitivity to

punishment

-0.038 0.001 <.01

Sensitivity to

reward

-0.188** 0.026 0.15**


Adding the additional predictors, at Step 3, significantly increased the explained

variance, over and above TPB (ΔR2 = .14, p < .001). The statistically significant


speeding (β=.58, p < .001), sensitivity to reward (β=-.19, p < .001) and driving license

(β=-.16, p = .013). Investigating the individual bivariate relations between the DV and

the IVs showed that, if considered separately, all IVs, except for age and sensitivity to

punishment, were statistically significant predictors of behaviour of not speeding


during the three months of the intervention. The three strongest individual predictors

were past behaviour of not speeding, intention not to speed and self-efficacy, which

explained 55%, 44% or 27% of the variance, respectively. However, when all IVs were

considered in an overall model, past behaviour of not speeding, sensitivity to reward

and driving licence uniquely explained the most variance in behaviour of not speeding

during the three months of the intervention, 10.3%, 2.6% and 1.1%, respectively.

The results provided support for H.10, which predicted that the additional

predictors (past speeding behaviour, perceived risk, moral norm, peers' norm,

impulsivity, sensitivity to reward and sensitivity to punishment) would account for a

significant variation in behaviour of not speeding during the three months of the

intervention, over and above the TPB variables.

7.5.4 Potential negative effects: Self-reported smartphone engagement (RQ2.4, H.11)

This analysis investigated whether the intervention did not introduce additional

risks, i.e. increased distraction, despite the good intentions behind its implementation.

It was guided by RQ2.4 (How did the intervention influence the participants’

engagement with their smartphones?) and by H.11.

7.5.4.1 Internal reliability, means, standard deviations, bivariate correlations and frequencies

Three repeated measures were used to assess the participants' interaction with

their smartphones before and after the intervention, focusing on if they initiated (less)

communication, monitored/read (less) communication, or responded (less) to

communication. The scale had a high internal consistency with a Cronbach's α of .91.

Table 7.11. Mean, standard deviation and bivariate correlations for the phone interaction variables at Time 1 (n=157).

Mean SD 1 2 3 Initiate (less) communication 5.27 2.05 - .72** .80** Monitor/read (less) communication 4.84 2.14 - .78** Respond (less) to communication 5.00 2.12 -

** Correlation is significant at the 0.01 level (2-tailed).

The three items were highly correlated between each other, with averages close

to the value of 5, meaning that people on average reported one to two interactions for

the past three months (see Table 7.11). This, however, was shaped by the large

proportion of people reporting no interaction at all (see Table 7.12). Despite that Q-Q

plots examination suggested normal distribution, the items were negatively skewed. If


"no interaction" answers were removed, the skewness would become close to zero,

and the histograms would suggest normal distribution, too.

Table 7.12. Self-reported phone interaction at Time 1 and Time 2 (n=157).

How often do you do the following on your smartphone while driving:

Initiate communication

on social interactive

technology?

Monitor/read social

interactive

technology?

Respond to

communication on social

interactive technology?

Time 1 Time 2 Time 1 Time 2 Time 1 Time 2

More than once per

day

6.4% 7.6% 8.3% 12.7% 7.0% 8.9%

Daily 6.4% 7.0% 10.2% 8.9% 8.9% 12.1%

1–2 times per week 12.1% 15.9% 14.6% 20.4% 14.6% 14.6%

1–2 times per month 10.8% 9.6% 7.0% 8.9% 10.2% 13.4%

1–2 times per 3

months

8.3% 10.2% 16.6% 9.6% 10.2% 14.0%

Once a year 5.7% 6.4% 2.5% 5.7% 3.8% 7.0%

Never 50.3% 43.3% 40.8% 33.8% 45.2% 29.9%

Table 7.12 shows how often (in %) participants (n=157) reported engaging in

initiating, monitoring/reading, and responding to social interactive technology on their

smartphone while driving at Time 1. For example, 40.8% of the participants, in the

current study, reported never monitoring/reading, 45.2% reported never responding,

and 50.3% reported never initiating communication while driving. On the other hand,

12.8% of the participants initiated communication at least once per day, 33.1%

monitored/read communication at least once per week, and 40.7% responded to

communication at least once per month.

At Time 2, there was a noticeable difference in the reported frequencies. While

initiating communication remained the least common behaviour, the most commonly

observed one was responding. The proportion of people who reported never engaging

reduced to 33.8% in monitoring/reading communication, to 29.9% in responding to

communication, and to 43.3% in initiating communication while driving. The

reduction in scores was confirmed by the means at Time 2 (see Table 7.13).


Nevertheless, the mean scores still indicated that people reported, on average, one to

two interactions for the past three months. The correlation coefficients, at Time 2, did

not visually differ much from the observed coefficients at Time 1.

Table 7.13. Mean, standard deviation and bivariate correlations for the phone interaction variables at Time 2 (n=157).

Mean SD 1 2 3 Initiate (less) communication 5.00 2.10 - .71** .82** Monitor/read (less) communication 4.46 2.21 - .76** Respond (less) to communication 4.52 2.07 -

** Correlation is significant at the 0.01 level (2-tailed).

A deeper inspection of the mean scores revealed that both groups reported a

reduction in their scores, meaning a greater smartphone use while driving during the

intervention period of three months (see Table 7.14).

Table 7.14. Phone interactions' means and standard deviations for the Control (n=126) and the Intervention group (n=31) at Time 1 and Time 2.

Condition

Time 1 Time 2 Change in means (T1-T2) Mean SD Mean SD

Initiate (less) communication

Control 5.17 2.05 4.91 2.09 -0.26 Intervention 5.65 2.04 5.36 2.13 -0.29

Monitor/read (less) communication


Respond (less) to communication



To further assess if the intervention did not, indeed, have a negative influence

on the participants' smartphone use, three one-way ANCOVA tests were performed on

the DVs (initiating (less) communication, monitoring/reading (less) communication,

and responding (less) to communication at Time 2) with condition (Control and

Intervention) as a fixed factor and IVs (initiating (less) communication,

monitoring/reading (less) communication, and responding (less) to communication at

Time 1). No significant differences between the Control group and the Intervention

group were found on any of the DVs (See Table 7.15).

Table 7.15. Effect of the intervention on phone interaction variables, adjusted for Time 1 values, with Condition as a fixed factor (n=157).


Initiate (less) communication .343 .559 .002

Monitor/read (less) communication .107 .744 .001

Respond (less) to communication .868 .353 .006


After finding the non-significant effect of the intervention on the two groups in

respect to any of the DVs, two-way ANCOVAs were performed to investigate whether

any personality characteristics did not moderate that results. The assumption of

equality of variance was not met when investigating the interaction effect between the

group condition and gender in the case of monitor/read (less) communication. Despite

that it created a bias in the obtained result, given that there was no significant

interaction effect, no adjustments were necessary. Significant interaction effects, with

a low to medium effect sizes, were found between the group condition and gender and

sensitivity to punishment for the participants' initiating (less) communication and

between the group condition and gender for the participants' responding (less) to

communication (see Table 7.16).

Table 7.16. Interaction effects between Condition and personality characteristics, phone interaction variables, adjusted for Time 1 values (n=157).

Variable Moderator F (1, 152) p ηp2

Initiate (less) communication

Gender 4.935 .028* .031

Driving experience .243 .623 .002


Sensitivity to punishment 4.306 .040* .028


Monitor/read (less) communication

Gender 1.994 .160 .013





Respond (less) to communication

Gender 5.816 .017* 0.37

Driving experience .015 .901 < .001




* p < .05


Further analysis was undertaken to follow-up on the observed significant

interaction effects between the group condition and gender. No statistically significant

main effects were found in neither the case of initiating (less) communication

(condition (F (1, 152) = .28, p =. 60, ηp2 = .002); gender (F (1, 152) = 2.15, p = .14,

ηp2 = .014)) nor in the case of responding (less) to communication (condition (F (1,

152) = .78, p =. 38, ηp2 = .005); gender (F (1, 152) = 1.98, p = .16, ηp2 = .013)). The

lack of main effects suggested that males and females behaved differently, depending

on their condition. Two-way gender-split ANCOVAs did not show a statistically

significant effect for the female participants' initiating (less) communication (F (1, 73)

= 1.22, p = .27, ηp2 = .016) or responding (less) to communication (F (1, 73) = .98, p

= .33, ηp2 = .013). However, the effect for the male participants was statistically

significant both when initiating (less) communication (F (1, 78) = 4.34, p = .04, ηp2 =

.053) and when responding (less) to communication (F (1, 78) = 6.50, p = .01, ηp2 =

.077). Investigating the mean scores revealed that the males in the Intervention group

reported higher initiating (less) communication mean score (5.84), i.e. lower rate of

initiating communication, than the males in the Control group (4.81), as well as higher

responding (less) to communication mean score (5.45), i.e. lower rate of responding to

communication, than the males in the Control group (4.27).

In a follow-up analysis on the significant interaction effect between the group

condition and the participants' sensitivity to punishment in the case of initiating (less)

communication was performed, no statistically significant main effect of condition (F

(1, 152) = .42, p =. 52, ηp2 = .003) was found. However, there was a main effect of

sensitivity to punishment (F (1, 152) = 4.27, p =.040, ηp2 = .027). In a two-way

ANCOVA, split by sensitivity to punishment, the assumption of equality of variance

was not met in the case of low-sensitive participants. Thus a lower, more conservative,

α (.025) was adopted when assessing the result (Wickens & Keppel, 2004). The

obtained result did not show a statistically significant effect for the participants neither

with low (F (1, 70) = 4.05, p = .048, ηp2 = .055) nor with high sensitivity to punishment

(F (1, 81) = .76, p = .39, ηp2 = .009).

The results were consistent with H.11, which predicted that the intervention

would not significantly increase the Intervention group drivers' distraction in terms of

initiating, monitoring/reading or responding to communication, in comparison to the



7.6 DISCUSSION

The study was split and analysed in four parts: First, to establish a baseline for

the implemented smartphone safe-driving app intervention, it sought to understand

more about the young drivers that took part in the intervention (RQ2.1). Second, to

investigate if driving for three months with a safe-driving app, installed on a

participant's smartphone, produced statistically significant effects (RQ2.2). Third, to

help understand, irrespective of condition, how much of the participants' self-reported

behaviour during the three months of the intervention could have been predicted before

the intervention happened, with the data collected at Time 1 (RQ2.3). And forth, to

investigate whether, along with the good intentions, the intervention did not, indeed,

have a negative influence on the participants' smartphone use (RQ2.4).

7.6.1 Findings

Demographic variables play an important role in planning a road safety

intervention. Thus, their predictive contributions were investigated first, in relation to

the participants’ intention not to speed before the intervention (RQ2.1) and their

behaviour of not speeding during the three months of the intervention (RQ2.3). In line

with previous research (Horvath et al., 2012; Scott-Parker, 2012), gender was found

to be a significant contributor in explaining both DVs of interest, intention not to speed

and behaviour of not speeding during the three months of the intervention. In the

obtained results, females were generally associated with higher scores, i.e. they were

drivers who reported less risky behaviour. Driving experience, as depicted by driving

licence, was also found to be a significant contributor. Greater risky behaviour was

reported by drivers with more driving experience (with an open drivers' licence). Age

did not emerge as a significant contributor.

After the demographic variables were explored, RQ2.1 and RQ2.3 guided the

investigation of the contribution of the extended TPB framework in the regression

equations. Consistent with the literature (Elliott & Thomson, 2010), the extended TPB

framework was a good fit for the study. It was predicted that the TPB constructs would

account for a significant variation in intention not to speed, over and above the

demographics, before the intervention took place. Consistent with Horvath et al.

(2012), TPB explained large amounts of significant variance, over and above the

demographic variables. A regression test was run on data from all 480 participants to

identify which of the constructs played a role within the sample. Instrumental attitude,


affective attitude, descriptive norm and self-efficacy, from the TPB variables, emerged

as significant contributors. Thus the study provided support for the predictive validity

of TPB, with 48% additional variance in intention not to speed being explained. It was

also predicted that the standard TPB constructs would account for a significant

variation in behaviour of not speeding during the three months of the intervention, over

and above the demographics. Intention not to speed and self-efficacy emerged as

significant contributors in the performed linear regression test. TPB added a significant

40% of additional explained variance, over and above the demographics.

As a third and final step in the regression equations, additional predictors were

assessed in relation to RQ2.1 and RQ2.3. Several additional predictors (past behaviour

of not speeding, perceived risk, moral norm, peers' norm, impulsivity, sensitivity to

punishment and sensitivity to reward) were examined to investigate whether they

contributed to explaining additional variance in either intention not to speed or

behaviour of not speeding during the three months of the intervention as DVs, over

and above TPB. Consistent with Elliott and Thomson (2010), past behaviour of not

speeding was a strong predictor in both regression models. Past behaviour of not

speeding was also the strongest individual predictor, as well as the predictor,

explaining the most unique variance, in both DVs.

The observed predictive power of past behaviour of not speeding suggested that

offenders might be a good fit for a smartphone safe-driving app intervention.

Nevertheless, the value of awareness-raising might be higher in prevention efforts,

which suggests that prevention shall target all drivers, irrespective of their past

behaviour. Still, designing different interventions for drivers with different behaviour

might be worth considering.

Contrary to Gannon et al. (2014), perceived risk was a significant independent

predictor of intention not to speed. However, that was true only until sensitivity to

reward and sensitivity to punishment were added, which, this time, confirmed Gannon

et al. (2014) findings. The two sensitivity measures were added when analysing the

smaller sample of 210 participants, as those two constructs were measured only at

Time 2. The observed difference might be due to the two sensitivity measures

accounting for the predictive power of perceived risk and diluting its predictive

validity. Perceived risk was not a significant independent predictor of behaviour of not


speeding during the three months of the intervention, which agreed with Gannon et al.

(2014) findings.

Consistent with Constantinou et al. (2011) and Pearson et al. (2013), impulsivity

did not explain additional variance over and above TPB in intention not to speed or in

behaviour of not speeding during the three months of the intervention. In the case of

intention not to speed, neither did sensitivity to reward and sensitivity to punishment,

nor the additional normative influences of moral norm and peers' norm, leaving past

behaviour of not speeding as the only additional predictor with significance, over and

above the TPB constructs. The picture was almost the same for behaviour of not

speeding during the three months of the intervention as a DV. The difference was that,

consistent with Castellà and Pérez (2004) and Scott-Parker and Weston (2017),

sensitivity to reward emerged as a second significant predictor in the final model.

Participants, who reported greater sensitivity to reward, reported more speeding.

Given that personality characteristics are considered stable and hard to change,

the findings suggested that focusing on them is of little use when TPB is measured.

Consistent with Ajzen's (2006) recommendations, targeting salient beliefs, generally

accepted as modifiable, is what interventions might better focus on. Nevertheless, if

TPB measures are not taken, and there is information on personality, it might be

considered useful, as the respective bivariate models were statistically significant. For

example, in the case of the current sample, designing an intervention, focusing on

participants with high sensitivity to reward, might be beneficial.

Despite the predictive power of the model, the intervention itself was not found

to be able to influence any of the TPB constructs over time (RQ2.2). Although mean

scores for measures, including both intention not to speed and behaviour of not

speeding, decreased for both the Control group and the Intervention group in the three

months of the intervention, that change was not explained by the intervention. A more

in-depth exploration, through two-way ANCOVAs, investigating whether any stable

characteristic, i.e. gender, driving experience, impulsivity, sensitivity to reward or

sensitivity to punishment, moderated that result, did not find anything significant,

either. Separate ANCOVA tests did not find significant differences over time as a

result of the intervention between the Control group and the Intervention group on any

of the participants' salient beliefs.


As a result of applying the COTs' selection criteria (see Section 4.2), there was

an initial expectation of the participants' salient belief to change. Flo was expected to

be able to influence the extended TPB framework dichotomised attitude (instrumental

attitude and affective attitude), dichotomised PBC (self-efficacy and perceived

controllability), moral norm and peers' norm (see Section 6.4). The objective of the

current program of research was to find evidence for that influence in a safe-driving

app intervention implemented as close to the real world as possible. The collected data,

however, did not provide evidence to support this initial expectation.

The collected data on driving scores from the app’s leaderboard supported the

notion of the intervention having no effect. For the participants for which driving

scores were calculated, the results were mixed, with marginal average improvement.

Many participants had no improvement at all. Some registered a score deterioration.

An additional objective of the study was to investigate whether the implemented

intervention, which intended to reduce speeding, did not produce any side effects, and

more particularly did not encourage smartphone distraction (RQ2.4). The study

specifically focused on three previously investigated associated behaviours (initiating

(less) communication, monitoring/reading (less) communication, and responding

(less) to communication) (Gauld et al., 2016). The findings showed that a substantial

number of young drivers did not engage at all with their smartphones while driving.

The number of those, that never interacted was higher than the number, reported by

Gauld et al. (2016), which might be attributed to the more diverse sample of the current

study. Gauld et al. (2016) sample comprised 79% of university students.

Consistent with previous research (Gauld et al., 2016), initiating (less)

communication was the least common behaviour, while monitoring/reading (less)

communication was the most common one. No evidence was found that the

intervention significantly changed the self-reported phone use of the participants when

condition was the only IV in the implemented ANCOVAs. However, when separate

two-way ANCOVAs tests were performed, with gender, driving experience,

impulsivity, sensitivity to reward or sensitivity to punishment as a second IV, evidence

was found that the intervention might have impacted distraction. Male participants in

the Intervention group reported significantly higher mean scores in both initiating

(less) communication and responding (less) to communication. This suggested a

potential positive side effect of the intervention in relation to the generally-perceived-


as-riskier male group. Despite the intervention having no effect on the males' intention

and self-reported behaviour, it might have made them more focused on the driving

task, and less prone to be distracted by their smartphones.

7.6.2 Strengths

Study 2 deployed a novel low-cost intervention approach, which can be

potentially easily replicated without substantial funds and technological upgrades,

which is the case in Creaser et al. (2015). Nevertheless, an effort was made to involve

a comparatively large sample to improve generalisability of the findings in comparison

with other lower-cost research interventions, focused on young drivers (Fitz-Walter et

al., 2017; Zhang et al., 2014). The intervention was designed to be as far from the

laboratory conditions as possible, trying to mimic a casual release of a safe-driving

app to the general public, where no expectations are imposed on adoption and usage.

The driving was intentionally unsupervised, and potential rewards to the participants

were not tied to anything related to the app.

Study 2 addressed several limitations reported in previous studies. Musicant and

Lotan (2015) considered the app they used as forgiving, i.e. the app was not providing

too much critical feedback on the participants’ behaviour. Since improving driving

behaviour was the focus in the current study, rather than adoption and usage, Flo was

perceived as providing critical feedback. There was no interest in it being forgiving.

The app was providing real-time alerts or post-trip analysis, depending on the

preference of the participant. It also had a self-starting capability, thus, reducing the

effort required from the participant, another limitation, reported by Musicant and

Lotan (2015). Different from Musicant and Lotan (2015), but similar to Creaser et al.

(2015), a baseline for evaluating behavioural change was established. Creaser et al.

(2015) used parental control to motivate behavioural change. Parental control, and

external influence, in general, is less likely to exist than not in reality, when the

decision to use the app is taken by the driver and is not coordinated with other parties.

In the current study, such external influence was not deployed.

Another strength of Study 2 was the diverse information, collected for the

participants, on top of the sample’s demographic representativeness. Diversity was

achieved in terms of gender, geographical coverage (in Australia), age (as long as it is

within the predefined frame of 18 to 25) and driving experience (despite the low

number of learner drivers, which may be due to 18 being the minimum required age to


participate). Recruitment reached out beyond the university campuses, which are a

common source for study participants' recruitment, and, thus, usually reported as a

limitation of findings. The study also allowed a Control group to be established to

control for any general influence, which might have been experienced by the study

participants. Personality characteristics (impulsivity, sensitivity to punishment and

sensitivity to reward) were explored together with other additional predictors (past

behaviour, perceived risk, moral norm and peers' norm), over and above the TPB

constructs. The diversity of the collected variables allowed a more complete effects’

picture to be established.

Another particular strength of the current study was its span across three months.

Few studies focus on smartphone use changes over time within a particular sample. A

more traditional approach is to focus on a time snapshot when distraction is analysed

and to explore the underlying causes. The current study adopted a more

unconventional approach, and did not look at distraction as separate risky driving

behaviour, but rather observed it as part of more complex research design, focused on

speeding. Although not in the primary focus, from a systematic perspective, the

smartphone use results provided insights into the effects of the intervention as a

potential disruptor. An increased smartphone interaction, as denoted by lower mean

scores for all participants, was observed between Time 1 and Time 2. The change in

the self-reported scores was not found to be significantly influenced by the

intervention. However, a further exploration found a significant effect of the

intervention on the Intervention group male participants' smartphone interaction. They

increased their self-reported mean score, i.e. interacted less with their smartphones, as

a result of the intervention.

7.6.3 Limitations

The surveys’ design provided no room for committing unintentional errors.

Nevertheless, the data was collected through online questionnaires and might have

been susceptible to bias, which the technique inherits. The anonymous nature of the

data collection, the impossibility of consequences for reporting speeding, as well as

the fact that rewards were offered irrespective of provided data, should have minimised

bias.

To increase the diversity of information collected for the Intervention group

participants, an attempt was made to incorporate real driving data by observing their


scores in the Flo leaderboard. Unfortunately, this effort had a limited added value due

to the scarce collected data, which emerged as a limitation. In addition to the scarcity,

there was also uncertainty about the correctness of the collected data. That uncertainty

stemmed from the inability of Flo to distinguish when a participant is a driver and

when they are not (see Subsection 7.2.3). The leaderboard did not provide additional

information for each participant other than the driving score, the ranking and the

number of trips. Through the leaderboard, it could not be inferred whether the

participants knew each other, either. All participants observed the same information in

the leaderboard. However, it cannot be inferred whether they saw it as representative

of their driving behaviour, or whether it was well correlated with their driving. The

research team had no information on the algorithms used to generate the leaderboard

information.

Another limitation was that no specific information was systematically collected

on the potential difficulties the participants might have experienced in working with

the app. Individual participants reported difficulties in installing the app, getting the

app to collect data, or in joining the app leaderboard group. The interest in using the

app was observed as fading away for participants that did not experience major

problems. Very few remained an active status by the end of the study. Ultimately, most

of them dropped from using the app. This might have been partially because using the

app was not inherently required by the study, and was not a prerequisite to enter the

prize draw. The reason for not listing such a requirement was that the situation should

be as close to reality as possible, where rewards, beyond the promise to improve one's

driving skills, hardly exist.

It also has to be acknowledged that, because the intervention was intended to

resemble real-world conditions, no checks were conducted to determine the extent of

participants' prior familiarity with COTs, as such are not likely to be performed in the

real world. The participants’ COTs familiarity might be another explanation for the

large drop-out rate. It is possible that participants, more familiar with COTs and, more

specifically, with apps and safe-driving apps, were more likely to download and use

Flo, than those without such an experience.

Data on how many of the Intervention group participants were interested in using

the safe-driving app, in general, was not collected, which emerged as a limitation

during analysis. For many participants, who completed the second survey, there was


no evidence that the app was actually used. Thus, the only reference number of true

Intervention group participants was provided by the number of users that joined the

safe-driving app group, which was much less than the total number of participants.

Problems with installation and data collection in the app could be inferred from

the observed leaderboard scores, too. Not all participants joined the leaderboard. For

the ones that joined, score generation did not happen all the time during the

intervention. However, information was not gained as to why some participants were

not able, and others were, to download and install the app, as well as to subsequently

join the leaderboard. This information would have been useful, as using the app was a

condition for the Intervention group. Such problems might help explain the large drop-

out rate.

Overall, Study 2 had a significant drop-out rate, greater than initially expected.

56.25% of the participants did not complete the second survey. The Intervention group

was additionally reduced, as a result of the encountered difficulty in confirming actual

participation in the intervention. Not all leaderboard scores could be linked with

specific self-reports due to data coding. Although the collected 157 cases were more

than the initially planned 120, this fact reduced the predictive potential of the study

and limited the validity of the findings.

An additional limitation of the study was that the underlying salient beliefs of

distraction were not explored. As the main focus was on speeding, attempts to collect

additional data would have increased the amount of input required from the

participants, thus, further reducing the potential to retain them. However, the collected

data revealed changes in the reported smartphone use behaviours over time, which

pointed in a positive direction, i.e. reduced distraction. Such results might be due to

participants, becoming more vigilant to their behaviour, but robust conclusions cannot

be made with the currently available data.

7.7 CONCLUSION

Study 2 was implemented as close to a free-living environment as possible and

was the first to examine potential real-world safety benefits from an intervention in

which the general public adopts a free non-obligation off-the-shelf safe-driving app.

With safe-driving apps coming out of both academia and businesses (see Chapter 5

and Chapter 6) and the natural propensity of their developers to stress their perceived


advantages, there was a need to investigate whether a safe-driving app makes a

significant contribution in terms of real-world safety benefits. An additional question,

with no less importance, was whether it introduced additional risks while being used.

This was a valid concern, given that mobile phones, including smartphones, are

identified as a major source of distraction (WHO, 2011), and using them, while

driving, increases four times the chances of a crash occurring (White et al., 2011).

Study 2 was implemented to consider a multitude of psychological influences

(Scott-Parker, 2012) as well as demographic parameters (Horvath et al., 2012). Such

design was expected to provide more in-depth insights on the subsequent intervention

results.

Study 2 exhibited several strengths such as real-world replicable nature, a

comparatively large and diverse sample, a control for any general influence, diverse

collected information, and a focus on long-term effects. Nevertheless, the study had

its limitations, e.g. potential for self-report bias, poor real driving data, lack of

information about the participants’ familiarity with COT and what problems they

experienced with Flo, and large drop-out rate.

Overall, the findings of the current study offered support for the use of an

extended TPB framework to explain young drivers' speeding intention and behaviour.

Personality characteristics did not play the expected role, as suggested by the literature,

indicating that a clear picture of the salient beliefs might be sufficient to design an

intervention. Nevertheless, while investigating RQ2 (How do young drivers’ self-

reported behaviour of not speeding and intention not to speed alter in their free-living

environment, as a result of exposure to a smartphone safe-driving app intervention?),

it was found that the implemented intervention did not succeed in positively

influencing neither the participants’ self-reported behaviour of not speeding nor their

intention not to speed or underlying constructs. However, there was evidence that it

significantly decreased smartphone engagement while driving, i.e. distraction,

amongst the male participants. Thus, future research should be focused on

investigating additional possible routes for influence, preferably with larger samples,

and with the possibility to analyse both self-reports and naturalistic driving data.

140 Chapter 8: Study 3 - Systematic review of VR simulations of risky driving

Chapter 8: Study 3 - Systematic review of VR simulations of risky driving

Chapter 8 describes the third study of the present program of research. Study 3

investigated the extent to which VR was previously explored in road safety research

and whether any safety benefits were reported. It addressed RQ3:

How is VR applied in road safety research to motivate behavioural change in

young drivers?

Following a PRISMA design, Study 3 reviewed the available literature in terms

of VR deployment and effects on young drivers' behaviour and safety. First, the search

results from the two used databases are presented, before outlining key findings,

followed by a discussion.

8.1 RATIONALE FOR CONDUCTING A SYSTEMATIC REVIEW

The young drivers' propensity to adopt and explore technology (Lee, 2007)

offers an opportunity that shall not be limited to using for road safety already

ubiquitous technologies, such as smartphones (see Chapter 5 and Chapter 6). A

potentially higher added value to reduce road trauma amongst young drivers may be

hidden in emerging technologies, such as VR. Furthermore, VR, unlike smartphones,

is less likely to generate unwanted additional risks, such as an increase in distraction.

VR has the capability to simulate life-threatening situations in safety. Thus, it may

potentially improve behaviour (van Loon et al., 2018). Whether VR can facilitate the

adoption of safer behaviour on the road is yet to be seen.

Little research has been done in the domain of VR. Although there is already

evidence of VR's potential to increase empathy (Garner, 2017; Ingram et al., 2019; van

Loon et al., 2018), the evidence for successful behavioural change is mixed (Ahn et

al., 2014; Morina et al., 2015; Schwebel et al., 2017; Theng et al., 2015; van Loon et

al., 2018). This called for a systematic investigation of the available literature. To the

author's best knowledge, to date, there is no systematic review exploring the available

evidence for VR effects in road safety.

Chapter 8: Study 3 - Systematic review of VR simulations of risky driving 141

8.2 METHOD

Study 3 mirrored the Study 1 methodology and used the PRISMA guidelines as

a systematic review framework. The review protocol followed the same steps:

1. Development of the research question.

2. Identification of search databases.

3. Definition of scope, inclusion, and exclusion criteria.

4. Definition of a search term.

5. A systematic search for information.

6. Screening and selection of studies (see PRISMA flowchart as Figure 8.1).

7. Review of selected articles.

8. Summarising of findings.

8.2.1 Search databases

Similar to Study 1, relevant papers were identified through searches in TRID

(https://trid.trb.org) and Scopus (https://www.scopus.com).

8.2.2 Literature search criteria

Papers published from 2010 onwards, and written in English, were considered

for inclusion in the review. The investigation focused on the actual and potential

application and utility of VR in (a) road safety research more generally, (b) road safety

practice more generally, (c) young drivers road safety research specifically, and (d)

young drivers road safety practice specifically. Papers were excluded in the cases when

there was no connection to drivers, when they were not relevant to driving or when

they referred to augmented reality, literature reviews, evaluation of traffic data,

medical, technical solutions, traffic modelling, theoretical discussions, and

pedestrians.

8.2.3 Search term

The search term deployed in TRID was (road OR driver) safety "virtual reality"

( headset OR "HTC Vive" OR "Oculus Rift" OR "PlayStation VR" OR "Google

Daydream View" OR "Samsung Gear"). The search term deployed in Scopus was (

ALL ( road OR driver ) AND ALL ( safety ) AND ALL ( "virtual reality" ) AND

ALL ( headset OR "HTC Vive" OR "Oculus Rift" OR "PlayStation VR" OR

"Google Daydream View" OR "Samsung Gear" ) ). Initially, the brands of the most

common headsets were not part of the search term. However, without specifying those,


the returned search result contained a large number of regular driving simulator

studies, referring to themselves as VR ones, which is technically correct but does not

reflect the notion of VR as a consumer-oriented technology (COT). Thus a decision

was made to specify in the search term itself what hardware was considered to be

relevant to VR as a COT in the context of this thesis.

8.3 SEARCH AND SCREENING RESULTS

The search in TRID, executed on October 08, 2018, returned one record. The

article was in the domain of pedestrian safety and, thus, not relevant to the current

study.

The search in Scopus covered the years 2010 to 2017. The output was limited to

documents in English. The search returned 50 records. The titles of all records were

screened for relevance. Ten articles were selected for abstract review. After the

abstract screening, seven were retained for full-text review and were downloaded for

the purpose. One of the studies was subsequently excluded as it deployed a dynamic

driving simulator while the headset was for recording electroencephalographic

changes, not VR experience. Thus, 6 papers were retained to be included in the

qualitative synthesis (see Figure 8.1).

Figure 8.1. Data extraction flowchart based on the PRISMA statement.


8.4 FINDINGS

Table 8.1 shows the final selection of 6 papers to be analysed in detail. It includes

the elements of interest to the current study, such as the type and number of

participants, the measures taken, and summaries of the studies’ findings. Although the

level of relevance of the shortlisted studies to the current thesis is debatable, a decision

was taken to include them in the synthesis. Some of the studies did not evaluate VR as

a real-world intervention, similar to the one that is implemented in the current program

of research. For example, Ropelato, Zünd, Magnenat, Menozzi, and Sumner (2017)

focused on simulator sickness only, Gaibler, Faber, Edenhofer, and von Mammen

(2015) used a keyboard to control the software. The design that is closest to the Study

4 (Chapter 9) is the design in Orfila et al. (2015). The authors focused on eco-driving

in an event for the general public. Despite the challenges of finding similarities

between the selected studies, all six were included in the analysis because, given their

low number, they can still provide useful information on how VR is used in road safety.

In contrast with Study 1, in Study 3, fewer studies’ characteristics were included

in the qualitative analysis. The studies’ design was not considered separately as all VR

studies happen in simulated settings, i.e. they cannot be implemented in naturalistic

driving settings. VR also does not offer the diversity of sensors, present in

smartphones, which made the topic irrelevant for Study 3.


Table 8.1. Virtual reality in the road safety literature

Authors VR use Type of participants Number of participants Measures taken Summary of findings

Gonzalez et al.

(2017)

The VR simulated the risk of

tractor overturning while being

driven on a farm road.

Aged 16 to 56, 93

male and 34 female

127 Self-reported perception

of the risk and safety

Participants reported positive

experience from using the tractor

driving simulator that can

potentially help them drive safer,

but they would need more training.

Ropelato et al.

(2017)

The VR simulated driving

through a virtual city with

streets, 519 buildings, and 40

other cars.

5 female and 12 male,

mean age 29.5 years

17 Self-reported simulator

sickness

Driving the simulator slightly

increased discomfort. The test ran

for under 15 minutes.

Nevertheless, 4 participants

aborted the simulation.

Agrawal, Knodler,

Fisher, and Samuel

(2017)

The VR simulated latent

hazards. An example is given

about a truck blocking the view

immediately before a pedestrian

crossing.

Young novice drivers,

aged 18 to 25 years

24 Eye movements, hazards

anticipation

The young drivers improved their

ability to detect threats. Compared

to the control groups, V-RAPT

users anticipated significantly

more latent hazards.

Gaibler et al. (2015) The VR simulated DUI by

introducing lags in the responses

of the software to steering and

acceleration commands.

Students 40 User-friendliness Players were engaged in the drink-

driving game and reported

enjoying it although it was not set

to achieve great realism.


Authors VR use Type of participants Number of participants Measures taken Summary of findings

Chen, Xu, Lin, and

Radwin (2015)

The VR simulated driving in

which the driver had to check

their blind spots before changing

lanes. The drivers had to step on

a pedal if they detected a white

truck in the blind spot.

14 between 18 and 35

years and 12 between

65 to 75 years, 15

female and 11 male

26 Target detection Younger drivers were twice more

successful in target detection in

less time than older ones. They

also rotated their trunks on

average in two-times greater

radius than the older drivers.

Orfila et al. (2015) The VR simulated driving of an

automatic car with a petrol

engine.

Random visitors 1900 Fuel consumption Moderate acceleration and

constant speed improve fuel

consumption. No immersion

realism of the simulation.


8.4.1 Studies’ samples

The common impression from all six studies in regards to characterising their

samples is that the provided information seems scarce. Three studies provided

information about both their participants’ age and gender (Chen et al., 2015; Gonzalez

et al., 2017; Ropelato et al., 2017). One study specified the age of the participants but

not the gender while stressing on their experience, i.e. young novice drivers (Agrawal

et al., 2017). Gaibler et al. (2015) also implied the importance of experience by

recruiting students. However, they do not provide any further information about their

participants. Orfila et al. (2015) were also vague, only reporting that their participants

were random visitors to their intervention stand. The reported age range is between 16

(Gonzalez et al., 2017) and 75 (Chen et al., 2015). The reported numbers of involved

participants also vary greatly between the studies, ranging from 17 (Ropelato et al.,

2017) to 1,900 (Orfila et al., 2015).

8.4.2 Measures

The variability of the studies in respect to their samples continues in what was

measured within them and, thus, their focus. Three of the studies focused on self-

reports, collecting data on risk and safety perception (Gonzalez et al., 2017), simulator

sickness (Ropelato et al., 2017) and VR simulation user-friendliness (Gaibler et al.,

2015). The other three studies focused on the participants driving, looking into their

hazards anticipation (Agrawal et al., 2017), target detection (Chen et al., 2015) and

fuel consumption (Orfila et al., 2015).

8.4.3 Benefits

The full texts revealed that research on VR was at its very early stages, which

makes it hard to draw an overall conclusion or to find support for findings across

studies. For example, Ropelato et al. (2017) focused on VR-triggered simulator

sickness, rather than on driving-related measures. Gaibler et al. (2015) focused on the

VR user-friendliness rather than on the risky behaviour the VR simulated. If the VR

experience is not set to achieve great realism, assigning a positive result to it might be

questionable, as in the reportedly-enjoyable Gaibler et al. (2015) drunk-driving game.

Orfila et al. (2015) could not achieve great VR realism either. The authors reported no

immersion realism of the simulation. However, they found that moderate acceleration

and constant speed improved fuel consumption.


Focused on driving behaviour, Agrawal et al. (2017) measured eye movements

and hazards anticipation. The authors found that the VR improved the VR-trained

participants’ ability to detect threats in comparison with the control participants. Chen

et al. (2015) also focused on detection. However, in their study, the VR was used as a

measurement tool to compare younger and older participants, rather than to produce

any effect on them. Gonzalez et al. (2017) concluded that VR driving simulations

could potentially help participants drive safer. The authors exposed their participants

to virtual tractor driving simulations. However, they also concluded that the

participants would need more training to achieve such a positive result.

8.5 SUMMARY

Although all analysed studies reported, or implied, the involvement of young

drivers, the target group of the current program of research, only Agrawal et al. (2017)

specifically targeted them. Two of the studies targeted specific driving behaviour,

drink-driving (Gaibler et al., 2015) and eco-driving (Orfila et al., 2015). However, both

studies reported that the deployed VR technology as a game (Gaibler et al., 2015) or

in an intervention with free public access (Orfila et al., 2015) was far from realistic

and immersive. Only Agrawal et al. (2017) reported safety benefits as a result of using

VR to train participants in latent hazards anticipation. However, the authors did not

investigate whether these beneficial effects are sustained in the long-term. Thus, while

addressing RQ3 (How is VR applied in road safety research to motivate behavioural

change in young drivers?), this systematic review further confirmed the conclusion,

made in Chapter 2, i.e. that little is known about the use of VR in road safety. Very

little is known about its potential for safety impact in real-world interventions, too,

especially when young drivers are involved and specific driving behaviours, such as

DUI, are targeted.

8.6 DISCUSSION

As discussed in Chapter 2, the literature provides evidence on the negative

impact of DUI on young drivers. For example, the young drivers' crash risk increases

five times as a result of engaging in DUI (Peck et al., 2008), making DUI a target for

much-needed prevention efforts. As a response, current prevention campaigns utilise

VR technology to raise awareness on the risks of DUI. Such use of VR takes place in

the light of limited knowledge around its potential to deliver road safety benefits.


Given that limited evidence, the current PhD project had an opportunity to expand the

available knowledge in the identified underexplored research space.

VR, not only in road safety research but in any other field, currently cannot be

delivered on anything else than through a simulation. As such, in road safety, it is

delivered through a simulator. Simulators are known for causing simulator sickness.

Ropelato et al. (2017) found that driving the VR simulator increased discomfort. Their

test run for under 15 minutes, and despite it being comparatively short, 4 participants

aborted the simulation. The work of Ropelato et al. (2017) in investigating self-

reported simulator sickness provided support strict exclusion criteria to be used when

recruiting participants. Thus, the requirement "Have no history of seizures or epilepsy"

was used when recruiting the Intervention group in Study 4.

Chapter 9: Study 4 - Intervention with VR simulations of risky driving 149

Chapter 9: Study 4 - Intervention with VR simulations of risky driving

Chapter 9 presents the fourth, and final, study of the current program of research.

Study 4 assessed if a VR intervention influenced young drivers' self-reported DUI

behaviour and intention.

This chapter 9 first provides a brief introduction of Study 4, which is followed

by an outline of the study key aims, method and hypotheses. The chapter then explores

the study results and, finally, is concluded with a discussion.

9.1 INTRODUCTION

Drink-driving is identified as a main contributor to 30% of the fatal and 9% of

the non-fatal injuries (ATC, 2011). Still, research shows that 7.8% of the young driver

respondents had driven after drinking alcohol, while 20% rode with a driver who had

been drinking (CDCP, 2016). Drug-driving is a main behavioural factor in 7% of the

fatal and 2% of the non-fatal crashes (ATC, 2011). DUI impairs driving performance,

increasing the risk of crashes (Hingson et al., 2002) five times for young drivers (Peck

et al., 2008) who, in turn, report high engagement in DUI (Ward et al., 2018).

Study 4 examined the effects of a VR intervention to influence self-reported DUI

behaviour and intention among young drivers. It assessed the impact of a novel COT-

based practice-oriented approach, implemented as part of real-world awareness-raising

intervention, to persuade young drivers to adopt safer and more responsible driving

behaviour. The intervention enabled participants to choose their "high" and step behind

the wheel of a virtual car. The intervention's effect was measured through two

questionnaires, one before the intervention, and another, three months after it, to

answer RQ4 (How do young drivers’ self-reported behaviour of not DUI and intention

not to DUI alter in their free-living environment as a result of a VR intervention?).

Similar to Study 2, the variables identified of most interest were intention not to

DUI and behaviour of not DUI during the three months after the intervention. Intention

not to DUI was assessed before the intervention. The construct reflected what the

drivers planned to do during the following three months without being influenced.

150 Chapter 9: Study 4 - Intervention with VR simulations of risky driving

Behaviour of not DUI during the three months after the intervention was assessed after

the intervention. The construct reflected what the drivers had actually been doing after

the intervention took place. It was expected that, as a result of the intervention,

participants in the Intervention group would report significantly greater past behaviour

of not DUI during the three months after the intervention, DUI being defined as driving

above the legal BAC limit for alcohol or under the influence of illegal drugs, as well

as significantly greater intention not to DUI, in comparison with the Control group.

The evaluation of Study 4 was grounded in an extended TPB (see Section 3.7),

consistent with the evaluation undertaken as part of Study 2. It was analysed in three

parts to address the overarching RQ4. The first part looked at the sample of participants

before some of them were exposed to the intervention (RQ4.1). The analysis focused

on the participants’ intention not to DUI and its predictors. The second part focused

on the changes the intervention might or might not have triggered in regards to the

TPB constructs (RQ4.2). Finally, the third part answered the question of how much of

the self-reported behaviour of not DUI during the three months after the intervention

could have been predicted with data, available before the intervention (RQ4.3).

9.2 METHOD

Study 4 was designed as a controlled experiment with an Intervention group and

a Control group. This section provides details on the VR tool and how it was used in

the implemented intervention. Subsequently, it discusses the involved participants,

how they were recruited for the intervention, and the subsequent data collection

procedure. Details are presented, in turn, in separate subsections.

9.2.1 VR tool and intervention

In contrast to smartphone safe-driving apps (see Section 6.4), there was less

choice of VR apps available to select from. Due to the novelty of the VR technology,

the VR apps stores were not as readily accessible and as richly stocked. As a result,

road safety VR software for the ordinary consumer was not available. For that reason,

the current program of research did not incorporate a separate process for choosing

appropriate VR software as the case was with the safe-driving apps.

At the time, the VR software "3D Tripping" was deployed in road safety

interventions in Europe, Asia and South America – interventions which, arguably, did

not undergo theory-based evaluation. For the purpose of evaluation within the current


program of research, "3D Tripping" was sourced free of charge from its developers.

Thus, in Study 4, "3D Tripping" became the VR-based intervention tool.

Initial criteria supporting using "3D Tripping" as a tool in a VR intervention

were derived from the literature (see Chapter 8). Lack of realism (Gaibler et al., 2015)

and immersion (Orfila et al., 2015) were reported limitations in VR studies. VR realism

would include realistic sound and intuitive controls for manipulating the vehicle in the

virtual world. VR immersion would enable driving interaction in the virtual world of

the simulation. "3D Tripping" addressed those two limitations convincingly enough

for the purpose of this program of research.

"3D Tripping" immerses users into VR-simulated DUI. At first, they enter the

car to drive on a straight stretch of a road, without any other traffic participants, and

without any DUI impairment. This experience allows users time to get used to

managing the VR driving simulator. It also serves as a beginning to an overarching

story of the VR experience, in which it is suggested that users drive completely sober

to a night club (see Figure 9.1).

Figure 9.1. A user is getting used to managing the VR driving simulator.

After this initial familiarisation drive, the story continues at the night club, where

the vehicle is parked, and the premise is entered (see Figure 9.2).


Figure 9.2. The VR software visualises parking of the vehicle before entering the night club.

Once inside the night club, users are given a choice between alcohol and drugs,

and they need to make a selection. If drugs are selected, then an additional choice,

between ecstasy, cannabis and magic mushrooms, is given (see Figure 9.3).

Figure 9.3. A choice to experience impaired driving as a result of ecstasy, cannabis or magic mushrooms influence is given to users.

Once a selection is made, a picture with people on a dance floor continues the

narrative before users would find themselves back in the car and ready to drive back

home (see Figure 9.4) under the influence of the selected alcohol or drug. Until that


moment, all users have the same experience using this tool. However, from that

moment on the different choices lead to different road trips, events and DUI VR

experiences.

Figure 9.4. A user driving under the VR-simulated influence of magic mushrooms.

The VR alters the DUI experience in the following ways:

- Alcohol, the vision area is reduced. There is a delay between the vehicle's

response to a given command.

- Ecstasy, everything moves at an increased pace. Sensors are sharpened.

Everything is very colourful and flashy, also getting blurry at intervals.

- Cannabis, everything is very slow. Colours are calm. Vision does not stretch

very far, very much the opposite of ecstasy.

- Magic mushrooms, the world is unreal, with imaginary sceneries and

characters. The vehicle behaves opposite to the commands it receives.

The chosen DUI experience is additionally reinforced by different types of

music. Each experience is composed of several parts, happening on different types of

roads, e.g. motorway or rural. If the user crashes, they are taken back to the beginning

of the respective part, thus, being given a chance to correct their behaviour. The

simulation is over when a point is reached where a Police car appears and pulls the

user over.


In summary, using "3D Tripping" as part of the intervention allowed the young

participants of this study (see Subsection 9.2.4 Participants) to "drive stoned" in a safe

virtual environment, and to learn about the dangers of mixing substance abuse and

driving. The VR simulated experiences revealed to them precisely how their

perception of reality and, therefore, driving competence was impaired.

Overall, “3D Tripping” offered features that seemed worth investigating in terms

of using VR as a COT road safety intervention. The lack of choice when selecting the

VR intervention tool did not allow its expected influence on the drivers to be compared

to the influence of other VR software packages. Nevertheless, the theory-based COTs

selection criteria (see Section 4.2) were applied to "3D Tripping" to generate a better

understanding of its potential to influence the participants' salient beliefs.

During an intervention, the software negatively affects the participants' driving

abilities while they are sober. Thus, the participants can make a conclusion on whether

such behaviour is favourable or unfavourable (SC1, see Table 4.1). The software

significantly alters their experience when DUI. Thus, they can see that they are able to

execute full control over the behaviour only by choosing not to DUI (SC3). "3D

Tripping" is designed to be used during interventions with free access. As a

consequence, the VR simulation is typically performed in front of public (SC2),

including participant's friends and peers (SC5), which allows the participants' norms

to be influenced. Those spectators are welcome to comment and discuss the behaviour

which can serve as a moral benchmark (SC4). The VR simulation ends by either a car

crash or by the participant being caught by the Police. Both outcomes can potentially

increase the perceived risk associated with DUI (SC6). As a result, "3D Tripping",

when used as part of a public installation, scores highly on the theory-derived selection

criteria (see Section 4.2). Scoring high raised an expectation that the VR would be able

to influence the participants' extended TPB salient beliefs successfully within the

framework of the current study.

For the purpose of the study, "3D Tripping" was operated on a driving simulator

console, consisting of Oculus Rift goggles, driving seat, driving wheel Logitech G29

and a computer (see Figure 9.5). The simulator was installed at venues with free public

access to encourage the drivers' interactions with the public and their peers. The

scenarios, happening inside the virtual environment, were visualised for the spectators

on a large TV.


Figure 9.5. A participant, operating the VR driving simulator in front of their peers.

To achieve the maximum possible effect, each VR simulation was preceded with

making sure the respective participant was as comfortable as possible. First, the

participant was invited to take a seat. Second, the distance between the seat and the

simulator pedals was adjusted to the participant’s preference. Third, the VR headset

was adjusted to fit the participant (see Figure 9.6). After the VR driving simulator was

adjusted for a participant’s comfort, the VR software was started.

Figure 9.6. A participant is operating the VR software on a fully adjusted VR driving simulator.

9.2.2 Recruitment

The intervention group participants were recruited face-to-face at events where

the VR driving simulator with the "3D Tripping" was installed. The simulator was


installed at venues with free public access, similar to the way this type of simulator

would be used in the real world as a road safety intervention. A pilot test was carried

out at the Brisbane Queen Street Mall on the 28th of May as part of a larger Rotary

Club of Brisbane event. This allowed testing the equipment and the recruitment

procedure.

As a result of the pilot test, an option for age "more than 25" was added as a

possible selection under "age", to avoid both disappointments in the public, and

misunderstandings on behalf of some potential recruits. Data from entries with "more

than 25" selected were not included in the analysis. Furthermore, to aid recruitment,

only spots with high student traffic at the QUT campuses were considered for the VR

driving simulator setup, such as the QUT library lobbies or the QUT Cube.

The QUT Garden Point Campus HiQ reception space, level 3 of P block, was

eventually chosen as the most appropriate location that was also available at the time.

It is an open access area with a constant flow of people. It also has security cameras

and is locked outside business hours. The main recruitment took place from 16th to the

27th of July 2018, from 10:00 a.m. till 4 p.m., Monday to Friday.

A social media campaign was implemented on Facebook to recruit Control group

participants. The objective was to collect data from the general public in parallel with

collect data for the Intervention group participants. The Control group recruitment ran

from 19th to 31st of July 2018. The Facebook campaign was set to target people who

were aged 18 to 25, resided in Australia, spoke English and possessed a driving license.

A critical criterion that was considered when designing the Facebook campaign

was to produce as little impact on the people seeing it as possible, while still informing

them adequately of the purpose of the study to be able to obtain their informed consent

for participation. The purpose of the Facebook campaign was not to influence the

potential participants, rather to preserve the quality of the recruited group as a Control

condition. The sole purpose was to invite Facebook users to complete the online survey

after they provide their informed consent to participate in the study.

The Facebook campaign showed an ad with text and a static image of a white

car, taking a turn while driving on a night city road. The image background had blurry

city lights and reflections. The researcher regarded those lights and reflections as

potentially suggestive of a DUI theme. The text was encouraging the viewer to help


investigate whether VR influences young people's behaviour on the road. Lastly, it

clarified that the viewer was not expected to take part in the VR intervention itself and

that the survey was going to take 7 minutes of their time.

Overall, the Facebook campaign reached slightly less than 20,000 people. The

click-through rate was approximately 2%, more than two times lower than the

campaign in Study 2. Extrinsic incentives (vouchers) were offered to all participants.

All participants (both Control and Intervention) were offered to enter into a pool for a

random draw. The pool comprised 10 Amazon vouchers of 50 AUD in Survey 1, and

10 Amazon vouchers of 100 AUD in Survey 2. An additional $10 voucher was offered

only to the Intervention group participants. These vouchers were offered for the

additional effort of driving the VR driving simulator with "3D Tripping" software. The

$10 vouchers were offered until the minimum required number of 200 participants was

reached. Due to financial restrictions, the $10 vouchers were not offered to participants

above that minimum number. Nevertheless, young people were welcome to take part

in the intervention without that extrinsic incentive. Not offering the extrinsic incentive

did not reduce the number of interested potential participants (see Subsection 4.4.5 for

details on the obtained ethical clearance).

9.2.3 Data collection procedure

"3D Tripping" did not allow for any data to be collected with respect to the

participants' driving performance. Data were collected only through self-completion

questionnaires.

At Time 1 (July 2018), self-completion questionnaires were used to collect data

on demographics, TPB constructs and additional predictors. A participants'

information sheet was provided online to the participants, as a cover sheet, before they

started completing the survey. The Intervention group was required to complete a

driving scenario in the "3D Tripping" VR environment after completing the survey.

The same survey was completed by the Control group. The Control group had no task

other than filling the surveys at Time 1 and Time 2.

At Time 2 (approximately three months after Time 1, November 2018), an

invitation to complete the second survey was sent to all participants. Following the

invitation, two reminders were sent to the participants. The self-generated participants'


anonymous identifiers were used to link datasets from Survey 1 and Survey 2,

originating from the same participant.

9.2.4 Participants

Initially, 282 participants took part in the VR intervention. Twenty-two entries

were subsequently removed, as the age option "more than 25" was selected. One

duplicate case was removed. Seventy people completed the Control group survey.

Partially completed surveys were not considered. No people requested in writing that

they wanted to opt-out of the study after they had completed the first survey. In the

end, 329 cases (237 male; Mage = 20.92 years, SD = 2.16) were retained for analysis.

The average driving experience of the sample was 3.25 years (SD = 2.07).

At Time 2, 138 young drivers (91 male; Mage = 20.93 years, SD = 2.22)

completed the second survey, 99 Intervention group participants and 39 Control group

participants. Their driving experience ranged from 0 to 9 years (M = 3.38 years, SD =

2.07). The dropout rate of 58.10% exceeded initial expectations. Both groups had less

than the initially required number of participants (see Section 4.4.1). From the

Intervention group, 52 participants reported that they had chosen to experience alcohol

in the VR simulator; 20 reported choosing magic mushrooms; 15 – cannabis; and 12 –

ecstasy.

9.3 HYPOTHESES

To investigate the predictors of DUI at baseline (RQ4.1), it was hypothesised

that:

H.12. Demographic variables (gender, age and driving experience) would

account for a significant variation in intention not to DUI. DUI is a major

contributor to crashes (ATC, 2011), and increased crash risk is shown to

associate with both inexperience and age (McCartt et al., 2009). Gender is

often identified as playing a role in the young drivers' risky behaviours, too

(Scott-Parker, 2012).

H.13. TPB constructs (instrumental attitude, affective attitude, subjective

norm, descriptive norm, self-efficacy and perceived controllability) would

account for a significant variation in intention not to DUI, over and above the



According to Sniehotta et al. (2014), TPB may lack sufficient predictive power

in longitudinal studies, or in ones with samples coming outside the university

campuses. Other researchers, such as Conner (2015), do not share that view. Away

from the theoretical discussion, findings from longitudinal studies, such as Elliott and

Thomson (2010), a study the current one leverages on (see Subsection 4.4.2), show the

TPB as having sufficient predictive power. As suggested by Conner (2015), an

extended theoretical framework may answer a great deal of the criticism, and address

the TPB limitations, as well as provide additional insights on the effect of the deployed

intervention. Additional predictors may also explain additional variation in the DVs

after controlling for TPB variables. Thus, it was hypothesised that:

H.14. Additional predictors (past behaviour of not DUI, perceived risk, moral

norm, peers' norm, impulsivity, sensitivity to reward and sensitivity to

punishment) would account for a significant variation in intention not to DUI,

over and above the TPB variables.

To explore the effects of the intervention (RQ4.2), it was hypothesised that, after the

intervention:


intention not to DUI than the Control group participants.


behaviour of not DUI than the Control group participants.

H.17. Due to the expectation that the VR software can influence all salient

beliefs (see Subsection 9.2.1), the VR intervention would have positively

influenced the Intervention group participants' instrumental attitude, affective

attitude, subjective norm, descriptive norm, self-efficacy, perceived

controllability, moral norm, peers' norm and perceived risk.

Finally, looking at the predictors of behaviour of not DUI during the three

months after the intervention, it was assessed how much of the behaviour after the

intervention, reported at Time 2, could have been predicted with the information

available before the intervention, at Time 1 (RQ4.3). In that respect, it was

hypothesised that:


H.18. Demographic variables (gender, age and driving experience) would

account for a significant variation in behaviour of not DUI during the three

months after the intervention.

H.19. TPB constructs (intention not to DUI, self-efficacy and perceived

controllability) would account for a significant variation in behaviour of not

DUI during the three months after the intervention, over and above the


H.20. Additional predictors (past behaviour of not DUI, perceived risk, moral

norm, peers' norm, impulsivity, sensitivity to reward and sensitivity to

punishment) would account for a significant variation in behaviour of not DUI

during the three months after the intervention, over and above the TPB

variables.

9.4 PRELIMINARY ANALYSIS

Preliminary data analysis was performed before studying the results from the

different studies in detail. It dealt with missing data, transforming data, deciding how

to deal with dropouts, and establishing the participants' profile in regards to their

personality characteristics.

9.4.1 Missing data

There was no missing data. The setup of the data collection required all questions

to be compulsorily answered with a limited number of answer options to choose from,

except for the driving experience measure, which required an integer to be entered.

Thus successful submission of a questionnaire could happen only if it contained all

questions completed.

9.4.2 Dropouts

A preliminary one-way between-groups MANOVA was performed in regards to

the provided answers on the extended TPB variables (11 DVs) at Time 1. The IV was

whether Time 2 questionnaire was completed or not. Homogeneity assumption was

not met with a Box's M p < .001. The equality of variance assumption was violated for

a number of constructs (intention, subjective norm, self-efficacy, perceived

controllability, moral norm, peers' norm and past DUI behaviour). The obtained

Wilks' Lambda value was .93 with a p = .018 and a ηp2 = .069, F (11, 317) = 2.13.


Thus, due to the assumptions' violations, it could not be concluded whether there was

a significant difference amongst the people who completed both surveys, and those

that completed only Survey 1. Nevertheless, the descriptive statistics revealed that the

participants who completed only Survey 1 scored on average lower on all constructs.

This meant that the data from comparatively riskier participants were not available at

Time 2. Thus, a decision was made to retain the data from participants who did not

complete the Survey 2, for the purpose of the subsequent analysis where possible, e.g.

at Time 1.

9.4.3 Assumption checks and data transformation

The design of the survey provided little room for committing unintentional

errors. Errors were possible only when answering the question around driving

experience, and as a result, errors were observed. In three cases, a discrepancy between

the reported age and driving experience was identified. The reported driving

experience was considered to be a too big number when compared to expected, also

possible and legal, values of age. All three cases did not complete the second survey.

Given that it could not be concluded whether the mismatch was due to malicious intent

or to a data entry error, a decision was made to replace those scores with the median

score for driving experience.

Following data collection, measures for intention, attitudes, norms (with the

exclusion of moral norm) and past DUI behaviour were recoded (transformed) so that

higher scores indicated greater agreement with the construct (perceived negatively-

geared answer to the left of the scale, smaller value, and perceived positively-geared

answer to the right of the scale, higher value).

The two separate questions "To what extent do you intend to drive under the

influence of alcohol or drugs over the next 3 months?" (A great extent to no extent at

all after recoding) and "How often do you think you will drive under the influence of

alcohol or drugs in the next 3 months?" (All the time to never after recoding) for the

construct intention yielded very strong significantly correlated results (Pearson's r =

0.73, p < .001), thus, they were combined (through finding an average) into a single

measure intention not to DUI.

The questions "If you were to drive under the influence of alcohol or drugs over

the next 3 months, how much would you worry about being involved in a road crash?"


(Not at all worried to worried very much) and "If you were to drive under the influence

of alcohol or drugs over the next 3 months, how much would you worry about being

caught by the Police?" (Not at all worried to worried very much) for the construct

perceived risk yielded very strong significantly correlated results (Pearson's r = 0.72,

p < .001), thus, they were combined (through finding an average) into a single measure

perceived risk.

Gender was recoded, from a string, into a numeric variable, with "0" denoting

males and "1" denoting females.

Normality of the DVs, intention not to DUI at Time 1 and behaviour of not DUI

during the three months after the intervention (past behaviour of not DUI, measured

at Time 2), was assessed statistically, via skewness and kurtosis, and visually, via

histograms and Q-Q plots. Intention not to DUI was negatively skewed at Time 1 (-

4.94, std. error = .13). The respective values for kurtosis was 28.88 (std. error = .27).

Behaviour of not DUI during the three months after the intervention was also

negatively skewed (-5.52, std. error = .21) with a positive kurtosis (38.13, std. error =

.41). The values suggested a departure from normality, which was confirmed by a

visual examination of the histograms and the Q-Q plots. Boxplots also revealed

outliers and supported the suggestion for lack of normal distribution. However, such

distribution was hardly surprising given the behaviour that was being examined, and

the questions asked. The recommended choice, in such situations, is that

nonparametric tests are used when analysing the data. Those tests rely on medians

rather than on means.

The medians of intention not to DUI at Time 1 and behaviour of not DUI during

the three months after the intervention were examined. The results showed that the

medians were at the maximum of the scale, 9. If nonparametric tests were applied to

the data with medians at the maximum of the scale, the result would be that the

intervention did not produce any effect. An investigation of the frequencies of the two

variables showed that never (the maximum of the scale, 9) was selected in 85.1% of

the intention not to DUI cases, and in 84.1% of the behaviour of not DUI during the

three months after the intervention cases. Efforts to normalise the data, through

transformations (Lg10, Sqrt), did not satisfactorily improve the parameters. Thus, a

decision was made to recode the two variables into categorical scales with two values,


"0", denoting answers other than never, and "1", denoting a selection of the value 9

(never) as an answer, and to use non-parametric tests (see Subsection 4.4.4).

In some cases, the minimum expected cell frequency assumption for a chi-square

test for independence was violated. In those cases, the results of Fisher's exact tests

were reported.


Consistent with Patton and Stanford (1995), the internal reliability check of the

BIS-11, impulsivity, revealed a high internal consistency with a Cronbach's α of 0.81.

The value was above the generally accepted limit of 0.7 (DeVellis, 2016). The

participants (n=329) had a mean score of 60.95 (SD=9.22).

The internal reliability check of the SPSRQ revealed a high internal consistency

in both components, sensitivity to punishment (Cronbach's α of 0.87) and sensitivity to

reward (Cronbach's α of 0.75). The participants (n=138) had a mean score of 12.97

(SD=5.62) on the sensitivity to punishment scale, and of 12.04 (SD=4.31) on the

sensitivity to reward scale.

9.5 RESULTS

9.5.1 Participants' intention not to DUI before the intervention (RQ4.1, H.12 - H.14)

The analysis of collected baseline data was guided by RQ4.1 (What did we know

about the participants before the intervention, and to what extent the extended TPB

framework could predict their intention not to DUI?), and by H.12, H.13 and H.14

(see Section 9.3).

9.5.1.1 Frequencies, means, standard deviations and bivariate correlations.

Table 9.1 below presents the means, standard deviations and Spearman's r

correlations for the TPB constructs. On average participants scored very high on all

measures, above 7 on a 9-point scale. As discussed earlier, the scores on all measures

were negatively skewed.

Table 9.1 shows consistency with TPB. Past behaviour of not DUI is highly

correlated with intention not to DUI (r=0.66, p < .001). Although it means that whoever

DUI in the past will DUI in the future, it provided support for both choosing TPB as a

framework, and extending the TPB framework with the construct of past behaviour

(see Chapter 3). Another strong link was the correlation within the PBC construct, i.e.


between self-efficacy and perceived controllability (r=0.65, p < .001). The low and

non-significant correlation coefficients of self-efficacy and perceived controllability

with descriptive norm came as a surprise, as well as their low and comparatively less

significant correlations with instrumental attitude and affective attitude.

Table 9.1. Frequencies, means, standard deviations and bivariate correlations for the TPB variables at Time 1 (n=329).

Frequency of

max score (9)Mean SD 1 2 3 4 5 6 7 8

1. Past behaviour of not DUI 81.8% 8.60 1.14 - .66**.28**.44**.44**.27**.37**.34**

2. Intention not to DUI 85.1% 8.71 .98 - .34**.37**.40**.25**.41**.35**

3. Instrumental attitude 68.7% 8.25 1.45 - .59**.30**.28**.15**.19**

4. Affective attitude 68.1% 8.20 1.53 - .37**.25**.19**.18**

5. Subjective norm 85.4% 8.55 1.41 - .35**.25**.20**

6. Descriptive norm 41.6% 7.54 1.78 - -.05 .07

7. Self-efficacy 63.8% 7.43 2.69 - .65**

8. Perceived controllability 67.5% 7.60 2.57 - ** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed).

9.5.1.2 Predictors of intention not to DUI

Following the methodology described in Section 4.4.4, a 3-step multiple logistic

regression was conducted to assess the strength of the intention not to DUI predictors

at Time 1 for the whole sample (n=329).

Step 1 of the model, which contained the demographic factors (gender, age and

driving experience), was statistically significant, χ2 (3, N = 329) = 21.40, p < .001,

indicating that the model was able to distinguish between participants who never

intended to DUI and participants who intended to do so. The model explained between

6.3% (Cox and Snell R squared) and 11.1% (Nagelkerke R squared) of the variance in

intention not to DUI, and correctly classified 84.5% of the cases. Nevertheless, the

explained variance is very small. As shown in Table 9.2, only Gender made a unique,

statistically significant contribution to the model (p = .020), with an odds ratio of .34,

i.e. female participants were more likely to indicate an intention not to DUI. Driving

experience and age did not emerge as significant predictors.


variables would account for a significant variation in intention not to DUI.


Table 9.2. Logistic regression analysis, predicting Intention not to DUI for all participants at Time 1, with demographic factors as predictors (n=329).

B S.E. Wald df p Exp(B)

95% C.I.for EXP(B)

Lower Upper

Gender -1.070 .462 5.372 1 .020 .343 .139 .848

Age -.144 .089 2.590 1 .108 .866 .727 1.032

Driving experience -.156 .087 3.225 1 .073 .855 .721 1.014

Constant 6.229 1.825 11.648 1 .001 507.231

Step 2 of the model contained the TPB variables (instrumental attitude, affective

attitude, subjective norm, descriptive norm, self-efficacy and perceived

controllability), together with the demographics (gender, age and driving experience).

It was statistically significant, χ2 (9, N = 329) = 80.91, p < .001, indicating that the

model was able to distinguish between participants, who never intended to DUI and

such that did intend to DUI, too. The model explained between 21.8% (Cox and Snell

R squared) and 38.3% (Nagelkerke R squared) of the variance in intention not to DUI,

and correctly classified 89.4% of the cases. Instrumental attitude, descriptive norm

and self-efficacy from the TPB variables emerged as significant predictors. As shown

in Table 9.3, the strongest predictor of intention not to DUI was gender (p = .042),

with an odds ratio of 2.94, i.e. once again, female participants were more likely to

indicate an intention not to DUI. The other statistically significant unique contributors

were instrumental attitude (p = .030, odds ratio = 1.31), descriptive norm (p = .003,

odds ratio = 1.35), self-efficacy (p = .013, odds ratio = 1.25) and driving experience (p

= .028, odds ratio = .80).

Table 9.3. Logistic regression analysis predicting Intention not to DUI for all participants at Time 1 (n=329) with demographic factors and TPB variables as predictors.

B S.E. Wald df p Exp (B)

95% C.I.for EXP(B)

Lower Upper

Gender 1.080 .531 4.142 1 .042 2.944 1.041 8.331

Age -.095 .107 .788 1 .375 .909 .737 1.122

Driving experience -.224 .102 4.843 1 .028 .799 .654 .976

Instrumental attitude .271 .125 4.722 1 .030 1.311 1.027 1.674

Affective attitude .161 .113 2.013 1 .156 1.175 .940 1.467

Subjective norm .214 .110 3.814 1 .051 1.239 .999 1.536

Descriptive norm .299 .101 8.795 1 .003 1.348 1.107 1.643

Self-efficacy .223 .090 6.106 1 .013 1.250 1.047 1.491

Perceived controllability .010 .092 .011 1 .918 1.010 .843 1.210

Constant -4.556 2.546 3.203 1 .074 .011


The results were consistent with H.13, which predicted that TPB constructs

would account for a significant variation in intention not to DUI, over and above the


Adding the additional predictors (past behaviour of not DUI, perceived risk,

moral norm, peers' norm, impulsivity) to the model with demographic factors (gender,

age and driving experience) and the TPB variables (instrumental attitude, affective

attitude, subjective norm, descriptive norm, self-efficacy and perceived

controllability), at Step 3, produced a statistically significant result, χ2 (14, N = 329) =

117.81, p < .001. The model explained between 30.1% (Cox and Snell R squared) and

52.9% (Nagelkerke R squared) of the variance in intention not to DUI, and correctly

classified 90.3% of the cases. As shown in Table 9.4, the strongest predictor of

intention not to DUI was past behaviour of not DUI (p < .001), with an odds ratio of

3.90, i.e. participants who DUI in the past were reporting a significantly greater

intention to DUI in the future. The other statistically significant unique contributors

were instrumental attitude (p = .045, odds ratio = 1.33) and descriptive norm (p = .017,

odds ratio = 1.35).

Table 9.4. Logistic regression analysis predicting Intention not to DUI for all participants at Time 1 (n=329) with demographic factors, TPB variables and additional variables as predictors.


95% C.I.for EXP(B)

Lower Upper

Gender 1.139 .660 2.984 1 .084 3.125 .858 11.386

Age -.034 .127 .070 1 .791 .967 .753 1.241



Affective attitude -.012 .146 .007 1 .932 .988 .742 1.314

Subjective norm .007 .153 .002 1 .962 1.007 .747 1.359

Descriptive norm .300 .125 5.729 1 .017 1.349 1.056 1.724

Self-efficacy .145 .117 1.532 1 .216 1.156 .919 1.453


Past behaviour of not DUI 1.362 .317 18.469 1 .000 3.903 2.097 7.263

Perceived risk -.122 .137 .802 1 .370 .885 .677 1.157

Moral norm .098 .137 .514 1 .473 1.103 .843 1.443

Peers' norm .082 .126 .421 1 .516 1.085 .848 1.388

Impulsivity .035 .025 1.959 1 .162 1.036 .986 1.088

Constant -16.612 4.623 12.915 1 .000 .000


Assessing the contribution of sensitivity to punishment and sensitivity to reward

as additional predictors required the regression test to be run only for the 138

participants who completed the second survey, as SPSRQ was administered only at

Time 2. Adding sensitivity to punishment and sensitivity to reward as additional

predictors to the model produced a statistically significant result, χ2 (16, N = 138) =

62.60, p < .001. The model explained between 36.5% (Cox and Snell R squared) and

73.3% (Nagelkerke R squared) of the variance in intention not to DUI, and correctly

classified 93.5% of the cases. As shown in Table 9.5, the strongest predictor of

intention not to DUI was still past behaviour of not DUI (p = .006), with an odds ratio

of 43.08. The other statistically significant unique contributors were instrumental

attitude (p = .044, odds ratio = 2.66) and impulsivity (p = .047, odds ratio = 1.18).

Thus, from the additional predictors, past behaviour not to DUI was the strongest

unique predictor. Impulsivity was the other statistically significant unique additional

contributor in the full equation.

Table 9.5. Logistic regression analysis, predicting Intention not to DUI for all participants at Time 1, with demographic factors, TPB variables and all additional variables as predictors (n=138).


95% C.I.for EXP(B)

Lower Upper

Gender -.154 1.553 .010 1 .921 .857 .041 17.974

Age -.291 .384 .573 1 .449 .748 .352 1.588



Affective attitude -.478 .425 1.264 1 .261 .620 .269 1.427

Subjective norm .277 .742 .140 1 .709 1.320 .308 5.653

Descriptive norm .342 .446 .590 1 .442 1.408 .588 3.372

Self-efficacy .031 .437 .005 1 .943 1.032 .438 2.428

Perceived controllability .520 .502 1.070 1 .301 1.681 .628 4.501

Past behaviour of not DUI 3.763 1.373 7.509 1 .006 43.084 2.920 635.669

Perceived risk -.187 .652 .082 1 .774 .829 .231 2.978

Moral norm -.265 .302 .770 1 .380 .767 .425 1.386

Peers' norm .011 .345 .001 1 .975 1.011 .514 1.988

Impulsivity .164 .083 3.950 1 .047 1.178 1.002 1.385

Sensitivity to punishment .225 .121 3.492 1 .062 1.253 .989 1.587

Sensitivity to reward -.308 .226 1.870 1 .171 .735 .472 1.143

Constant -37.731 16.730 5.086 1 .024 .000

The results provided support for H.14, which predicted that additional predictors

(past behaviour of not DUI, perceived risk, moral norm, peers' norm, impulsivity,


sensitivity to reward and sensitivity to punishment) would account for a significant

variation in intention not to DUI, over and above the TPB variables.

9.5.2 Changes in salient beliefs (RQ4.2, H.15 - H.17)

The analysis in this section was based on 138 survey entries collected at Time 2.

It was guided by RQ4.2 (Did the intervention change the participants' salient beliefs,

as depicted by the TPB constructs?), and by H.15, H.16 and H.17.

9.5.2.1 Frequencies, means, standard deviations and bivariate correlations

At Time 2, approximately three months after the intervention, the relations

between the TPB variables remained statistically significant (see Table 9.6). The most

notable differences in comparison to Time 1 were: 1) the mean score for participants'

perceived controllability increased above 8, on the 9-point scale; 2) the previously

weak and non-significant correlation of self-efficacy and perceived controllability with

descriptive norm became moderate and significant; and 3) both subjective norm and

descriptive norm decreased and weakened their correlations with most variables, other

than with self-efficacy and perceived controllability.

Table 9.6. Frequencies, means, standard deviations and bivariate Spearman correlations for the TPB variables at Time 2 (n=138).

Frequency of

max score (9)Mean SD 1 2 3 4 5 6 7 8

1. Past behaviour of not DUI 84.1% 8.72 .93 - .62**.20**.30** .11 .18*.36**.29**

2. Intention not to DUI 79.0% 8.76 .60 - .34**.38** .21*.36**.51**.45**

3. Instrumental attitude 71.0% 8.41 1.19 - .52**.32** .21*.29** .12

4. Affective attitude 64.5% 8.10 1.62 - .26** .13 .25** .21*

5. Subjective norm 86.2% 8.80 .58 - .16 .19* .17*

6. Descriptive norm 32.6% 7.64 1.42 - .25**.27**

7. Self-efficacy 62.3% 7.80 2.31 - .55**

8. Perceived controllability 75.4% 8.30 1.81 - ** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed).


From the Intervention group, 47 participants reported that they had chosen to

experience one of the drug simulations. Fifty-two participants reported that they had

chosen to experience alcohol in the VR simulator. As the effects of using drugs or

alcohol might differ, the following analysis regarded them as two separate Intervention

groups. The frequencies for each variable of interest (intention not to DUI at Time 1,


intention not to DUI at Time 2, past behaviour of not DUI at Time 1, and behaviour of

not DUI during the three months after the intervention) are presented in Table 9.7.

Table 9.7. Dichotomised DVs' frequencies per group condition (n=138)

Variable Dichotomised

value

Alcohol

intervention group

Drugs intervention

group Control group

Intention not to

DUI at Time 1

Never 47 40 36

Other than never 5 7 3

Intention not to

DUI at Time 2

Never 46 33 30


Past behaviour of

not DUI at Time 1

Never 44 40 35


Behaviour of not

DUI during the

three months after

the intervention

Never 47 39 30


A series of chi-square tests for independence were run for each group (Alcohol

or Drugs Intervention group) on each variable (intention not to DUI at Time 1,

intention not to DUI at Time 2, past behaviour of not DUI at Time 1, and behaviour of

not DUI during the three months after intervention) to investigate whether the

frequencies of the answers never and other than never for each of those four variables

were significantly different than the ones given by the Control group (see Table 9.7).

A series of McNemar's tests were performed to evaluate the effect of the intervention

on the Intervention groups' intention not to DUI and behaviour of not DUI (past

behaviour of not DUI at Time 1 and behaviour of not DUI during the three months

after the intervention) to understand whether there was a change in the proportion of

the answers never and other than never in the samples. A series of Wilcoxon Signed

Ranks Test were performed to evaluate for any changes in other TPB constructs

(instrumental attitude, affective attitude, subjective norm, descriptive norm, self-

efficacy, and perceived controllability) within the Intervention groups.

9.5.2.2.1 Alcohol Intervention group

Between-group analysis

When investigating intention not to DUI at Time 1, the minimum expected cell

frequency assumption for a Chi-square test for independence was violated. Fisher's


Exact Test returned p = 1.00, indicating no significant association between group

condition and intention not to DUI at Time 1. A Chi-square test for independence (with

Yates Continuity Correction) indicated no significant association between group

condition and intention not to DUI at Time 2 for the Alcohol Intervention group, either,

χ2 (1, N = 91) = 1.40, p = .24, phi =.15.


Intervention group would report significantly greater intention not to DUI than the


A Chi-square test for independence (with Yates Continuity Correction) indicated

no significant association between group condition, neither for past behaviour of not

DUI at Time 1 (χ2 (1, N = 91) = .16, p = .69, phi = -.08) nor for behaviour of not DUI

during the three months after the intervention (χ2 (1, N = 91) = 2.15, p = .14, phi =

.19).

Overall, these results did not support H.16, which predicted that participants in

the Intervention group report significantly greater behaviour of not DUI during the

three months after the intervention than the Control group participants.

Within-group analysis

For the Alcohol Intervention group, the McNemar's test did not show a

statistically significant difference in intention not to DUI (N = 52, Exact Sig. = 1.00).

At Time 2, one more person selected other than never as their intention not to DUI, in

comparison to Time 1 (see Table 9.7). The difference in behaviour of not DUI (N =

52, Exact Sig. = .45), before and three months after the intervention, was not

statistically significant, either. Three people less reported other than never as a recall

of their behaviour of not DUI, in comparison to Time 1 (see Table 9.7).

The results of a Wilcoxon Signed Ranks Test indicated that there were no

significant differences for the Alcohol Intervention group (n=52) before the

intervention and three months after the intervention, in none of the potentially-

modifiable extended TPB variables: instrumental attitude (z=-.72, p = .47), affective

attitude (z=-.19, p = .85), subjective norm (z=-.06, p = .95), descriptive norm (z=-1.32,

p = .19), self-efficacy (z=-.79, p = .43), perceived controllability (z=-.72, p = .47),

moral norm (z=-.96, p = .34), peers' norm (z=-1.94, p = .05) and perceived risk (z=-

.23, p = .82).


These results did not support H.17, which predicted that the VR intervention

would have positively influenced the Intervention group participants' instrumental

attitude, affective attitude, subjective norm, descriptive norm, self-efficacy, perceived


9.6.2.2.2 Drugs Intervention group

Between-group analysis

When investigating intention not to DUI at Time 1, the minimum expected cell

frequency assumption for a Chi-square test for independence was violated. Fisher's

Exact Test returned p = .34, indicating no significant association between group

condition and intention not to DUI at Time 1. A Chi-square test for independence (with

Yates Continuity Correction) indicated no significant association between group

condition and intention not to DUI at Time 2 for the Drugs Intervention group, χ2 (1,

N = 86) = .21, p = .65, phi =-.08.


Intervention group would report significantly greater intention not to DUI in the future

than the Control group participants.

When investigating past behaviour of not DUI at Time 1 for the Drugs

Intervention group, the minimum expected cell frequency assumption for a Chi-square

test for independence was violated. Fisher's Exact Test returned p = .75, indicating no

significant association between group condition and past behaviour of not DUI at Time

1. A Chi-square test for independence (with Yates Continuity Correction) indicated no

significant association between group condition and behaviour of not DUI during the

three months after the intervention for the Drugs Intervention group, χ2 (1, N = 86) =

.19, p = .67, phi = .08.


Intervention group would report significantly greater behaviour of not DUI during the

three months after the intervention than the Control group participants.

Within-group analysis

For the Drugs Intervention group, the McNemar's test did not show a statistically

significant difference in intention not to DUI (N = 47, Exact Sig. = .09). At Time 2,

seven more people selected other than never as their intention not to DUI, in

comparison to Time 1 (see Table 9.7). The difference in behaviour of not DUI (N =


47, Exact Sig. = 1.00), before and three months after the intervention was not

statistically significant either. One person less reported other than never as a recall of

their behaviour of not DUI, in comparison to Time 1 (see Table 9.7).

The results of a Wilcoxon Signed Ranks Test indicated that there were no

significant differences for the Drugs Intervention group (n=47), before the intervention

and three months after the intervention, in none of the potentially-modifiable extended

TPB variables: instrumental attitude (z=-.81, p = .42), affective attitude (z=-1.05, p =

.29), subjective norm (z=-1.03, p = .31), descriptive norm (z=-.04, p = .97), self-

efficacy (z=-.87, p = .40), perceived controllability (z=-.19, p = .85), moral norm (z=-

.68, p = .50), peers' norm (z=-.27, p = .79) and perceived risk (z=-1.71, p = .09).

These results did not support H.17, which predicted that the VR intervention

would have positively influenced the Intervention group participants' instrumental

attitude, affective attitude, subjective norm, descriptive norm, self-efficacy, perceived


9.5.3 Predictors of behaviour of not driving under the influence of drugs or alcohol after the intervention (RQ4.3, H.18 - H.20)

The analysis of data, collected after the intervention, was guided by RQ4.3

(Using the extended TPB framework, to what extent could the data available before

the intervention predict the participants' behaviour of not DUI after the intervention?),

and by H.18, H.19 and H.20.

In this analysis, data from 138 participants who completed the second

questionnaire was assessed, irrespective of their condition, Control or Intervention, as

no evidence for any statistically significant effect from the intervention was found in

the previous analysis. A 3-step direct logistic regression was conducted to assess the

strength of the predictors of behaviour of not DUI during the three months after the

intervention, following the order of entry described in Section 4.4.4 and already

applied in Study 2.

Step 1 of the model contained demographic factors (gender, age and driving

experience) and was statistically significant, χ2 (3, N = 138) = 11.02 p = .012,

indicating that the model was able to distinguish between participants who never

performed DUI during the three months after the intervention and such that did. The

model explained between 7.7% (Cox and Snell R squared) and 13.1% (Nagelkerke R


squared) of the variance in behaviour of not DUI during the three months after the

intervention, and correctly classified 84.8% of the cases. Nevertheless, the explained

variance is small. As shown in Table 9.8, only driving experience made a unique,

statistically significant contribution to the model (p = .004), with an odds ratio of .61.

Gender and age did not emerge as significant predictors.


variables would account for a significant variation in behaviour of not DUI during the

three months after the intervention.

Table 9.8. Logistic regression analysis predicting behaviour of not DUI during the three months after the intervention for participants at Time 2 (n=138) with demographic factors as predictors.

B S.E. Wald df p Exp(B)

95% C.I.for EXP(B)

Lower Upper

Gender -.320 .539 .352 1 .553 .726 .253 2.088

Age .264 .166 2.538 1 .111 1.302 .941 1.801


Constant -1.760 2.992 .346 1 .556 .172

At Step 2, the model contained the TPB variables (intention, self-efficacy and

perceived controllability), together with the demographics. It was statistically

significant, χ2 (6, N = 138) = 16.54, p = .011. The model explained between 11.3%

(Cox and Snell R squared) and 19.3% (Nagelkerke R squared) of the variance in

behaviour of not DUI during the three months after the intervention, and correctly

classified 85.5% of the cases. However, only intention not to DUI emerged as a

significant predictor from the TPB constructs. As shown in Table 9.9, the two

statistically significant unique contributors were intention not to DUI (p = .032, odds

ratio = 4.54), and driving experience (p = .007, odds ratio = .61).

Table 9.9. Logistic regression analysis, predicting Behaviour of not DUI during the three months after intervention for participants at Time 2, with demographic factors and TPB from Time 1 as predictors

(n=138).


95% C.I.for EXP(B)

Lower Upper

Gender .244 .549 .197 1 .657 1.276 .435 3.739

Age .300 .178 2.846 1 .092 1.350 .953 1.914


Intention not to DUI 1.513 .707 4.586 1 .032 4.541 1.137 18.136

Self-efficacy -.211 .257 .671 1 .413 .810 .490 1.341


Constant -14.471 7.201 4.038 1 .044 .000


The results were consistent with H.19, which predicted that TPB constructs

would account for a significant variation in behaviour of not DUI during the three

months after the intervention, over and above the demographic variables.

At Step 3 of the model, the contribution over and above the TPB variables of

additional predictors (past behaviour of not DUI, perceived risk, moral norm, peers'

norm, impulsivity, sensitivity to reward and sensitivity to punishment) was

investigated. The model was statistically significant, χ2 (13, N = 138) = 38.36,

p < .001. It explained between 24.3% (Cox and Snell R squared) and 41.5%

(Nagelkerke R squared) of the variance in behaviour of not DUI during the three

months after the intervention, and correctly classified 88.4% of the cases. Peers' norm

and sensitivity to punishment emerged as statistically significant unique additional

predictors. As shown in Table 9.10, the three statistically significant unique

contributors were peers' norm (p = .005, odds ratio = 1.63), sensitivity to punishment

(p = .021, odds ratio = 1.16) and driving experience (p = .020, odds ratio = .59).

Table 9.10. Logistic regression analysis, predicting Behaviour of not DUI during the three months after intervention for participants at Time 2, with demographic factors, TPB and additional predictors

from Time 1 as predictors (n=138).


95% C.I.for EXP(B)

Lower Upper

Gender -.787 .757 1.079 1 .299 .455 .103 2.009

Age .187 .198 .896 1 .344 1.206 .819 1.776


Intention not to DUI .776 1.137 .466 1 .495 2.173 .234 20.187

Self-efficacy -.287 .322 .792 1 .374 .751 .399 1.411


Past behaviour of not DUI -.101 .911 .012 1 .911 .903 .152 5.386

Perceived risk .257 .247 1.079 1 .299 1.293 .796 2.099

Moral norm -.352 .432 .663 1 .416 .703 .302 1.641

Peers' norm .491 .174 7.936 1 .005 1.634 1.161 2.300

Impulsivity -.044 .034 1.665 1 .197 .957 .896 1.023

Sensitivity to punishment .149 .065 5.296 1 .021 1.161 1.022 1.318

Sensitivity to reward -.122 .080 2.327 1 .127 .885 .756 1.035

Constant -4.076 9.047 .203 1 .652 .017

The results provided support for H.20, which predicted that the additional

predictors (past behaviour of not DUI, perceived risk, moral norm, peers' norm,

impulsivity, sensitivity to reward and sensitivity to punishment) would account for a


significant variation in behaviour of not DUI during the three months after

intervention, over and above the TPB variables.

9.6 DISCUSSION

Study 4 was analysed in three parts. The purpose of the first part was to establish

a baseline, understanding more about the young drivers as well as how much of their

intention not to DUI before they took part in the VR intervention (RQ4.1). The purpose

of the second part was to investigate if the intervention with the VR software "3D

Tripping" produced any statistically significant changes in regards to the assessed TPB

constructs (RQ4.2). The purpose of the third part was to help understand how much of

the participants' future DUI could have been predicted before the intervention

happened, with the data collected at Time 1 (RQ4.3).

9.6.1 Findings

The predictive contributions of demographic factors (gender, age and driving

experience) were investigated first, before the TPB predictive validity in relation to

the participants’ intention not to DUI before the intervention (RQ4.1) and behaviour

of not DUI during the three months after the intervention (RQ4.3). Consistent with

previous research (Scott-Parker, 2012), gender was statistically significant in

predicting intention not to DUI. However, gender did not have significant predictive

power when predicting behaviour of not DUI during the three months after the

intervention, which disagreed with Scott-Parker (2012). In the case of behaviour of not

DUI during the three months after the intervention, driving experience was the

strongest demographic predictor. The observed difference with existing literature

might be due to the nature of the investigated behaviour, DUI, while Scott-Parker

(2012) looked into young drivers' risky driving behaviour in more general terms.

Another reason might be the systematic differences between the two samples, before

and after the intervention, used in the regression models for intention not to DUI and

for behaviour of not DUI during the three months after the intervention. Many

participants who reported riskier behaviour at Time 1, did not complete Survey 2,

which collected the data used to assess behaviour of not DUI during the three months

after the intervention.

After the demographic factors were explored, RQ4.1 and RQ4.3 guided the

investigation of the contribution of the extended TPB framework in the regression


equations. Consistent with Study 2, and more broadly with the literature (Elliott &

Thomson, 2010; Horvath et al., 2012), the extended TPB framework predicted

additional variation, over and above the demographic factors. Instrumental attitude,

descriptive norm and self-efficacy were significant predictors of intention not to DUI.

Intention not to DUI was a significant predictor of behaviour of not DUI during the

three months after the intervention.

As a third and final step in the regression equations, additional predictors were

assessed in relation to RQ4.1 and RQ4.3. Consistent with Elliott and Thomson (2010),

past behaviour of not DUI before the intervention was a strong, unique predictor of

intention not to DUI, over and above TPB. The other additional predictor that

contributed significantly to the final regression equation was impulsivity, a finding

inconsistent with Constantinou et al. (2011) and Pearson et al. (2013). The

inconsistency might be due to the very specific behaviour, targeted by Study 4, DUI,

while Constantinou et al. (2011) and Pearson et al. (2013) looked into risky driving

behaviour, in general.

The contributions of additional predictors were different when predicting

behaviour of not DUI during the three months after the intervention. Past behaviour

of not DUI did not emerge as a statistically significant contributor, which was

inconsistent with Elliott and Thomson (2010). This might be due to, once again, the

different examined behaviours. Elliott and Thomson (2010) focused on speeding rather

than on DUI. However, it might also be due to the participants' characteristics. Elliott

and Thomson (2010) studied offenders, while in Study 4, many participants that

reported DUI at Time 1, i.e. likely offenders, did not complete Survey 2, and thus, their

data could not be assessed.

Consistent with Sela‐Shayovitz (2008), peers' norm emerged as a statistically

significant contributor in the final regression equation with behaviour of not DUI

during the three months after the intervention as a DV. Also, consistent with the

literature (Elliott & Thomson, 2010), moral norm did not emerge as a statistically

significant contributor in that regression equation. Inconsistency with the reviewed

literature (Constantinou et al., 2011) was observed in the predictive validity of

sensitivity to punishment and sensitivity to reward. Sensitivity to punishment emerged

as a statistically significant contributor in the final regression equation with behaviour

of not DUI during the three months after the intervention as a DV, while sensitivity to


reward did not, which does not offer support for the findings in Constantinou et al.

(2011). Given that Constantinou et al. (2011) studied young drivers, too, the difference

in the results might be coming from the different behaviours being investigated.

Constantinou et al. (2011) explored risky and aggressive driving through DBQ, while

Study 4 focused only on DUI.

Despite the found evidence for significant correlations between the variables, the

VR intervention was not found to be able to influence any of the extended TPB

constructs (RQ4.2). Separate tests did not find significant differences between the

Control group and the Intervention group neither on self-reported behaviour nor on

intention, over time. The results were the same on any of the potentially-modifiable

constructs within the Intervention groups, too. Similar to the safe-driving app findings,

such results came at a surprise because "3D Tripping" was initially seen as potentially

capable of influencing all salient beliefs in the adopted theoretical model (see

Subsection 9.2.1).

Such results might be rooted in the inherent social unacceptability of DUI

behaviour. In line with general expectations, participants reported that they neither did

DUI in the past nor intended to do so in the future. Positively changing behaviour that

is already positive is inherently challenging. In the future, it is suggested to look into

the effect of the intervention for offenders. Alternatively, other constructs might be

explored about a more general target group. Detailed recommendations about how

future research can build on the Study 4 findings are presented in Section 10.3.

9.6.2 Strengths

Study 4 was designed as a real-world intervention, focused on evaluating the

long-term impact of example VR simulations of risky driving software. It was

implemented similar to interventions undertaken by road safety advocacy groups in

their regular activities. Such interventions are easily replicable once the initial

investment of purchasing the necessary hardware and software is made. However,

many such interventions are not are robustly evaluated. Thus, the collected self-

reported data provided information that was not readily available from other sources.

Initially, an effort was made to involve a comparatively large sample (n=282) as

an Intervention group, to improve generalisability of the findings in comparison with

other lower-cost studies, focused on young drivers (Fitz-Walter et al., 2017; Zhang et


al., 2014). A particular strength of the study was that a Control group was established

to control for any general influence, which might have been experienced by the

Intervention group participants. Another strength was the sample’s demographic

diversity in terms of gender, age (as long as it is within the predefined frame of 18 to

25) and driving experience. Personality characteristics (impulsivity, sensitivity to

punishment and sensitivity to reward) were explored together with other additional

predictors (perceived risk, moral norm, peers' norm and past behaviour of not DUI)

on top of the TPB constructs. This allowed a more complete picture to be established

for each participant, both at a given moment and over time.

9.6.3 Limitations

Data was collected through online questionnaires, and self-reports are known to

be inherently susceptible to bias. It has to be acknowledged that the investigated

behaviour of DUI is much less socially acceptable than the investigated speeding

behaviour in Study 2, for example. Thus, higher pressure to report socially acceptable

answers may be expected. The anonymous nature of the data collection, the

impossibility of consequences for reporting DUI, as well as the fact that rewards were

offered, irrespective of provided answers, should have minimised bias.

Study 4 data were collected twice, before and three months after the intervention.

Thus, the evaluation of the long-term intervention effect did not incorporate the

assessment of changes in the constructs of interest immediately after the intervention.

It is acknowledged that with the participants being still present at the intervention

venue, additional data could have been collected. However, at the time of evaluation

design, this additional data collection was considered as imposing on the participants

a time-consuming effort, misaligned with the overall research focus. In such a

situation, and to limit the potential dropout rates, a decision was taken not to

overburden the Intervention group participants. Rather, the study focused on the

possible long-term effects only, which are much less often explored in the literature.

Another limitation of the study was the initial sample gender distribution (91

females, 236 males). Such distribution might not be a fair representation of the

Australian young drivers' population. An additional limitation of the sample was that

the Intervention group was recruited predominantly at the QUT campus. A negligible

number of 5 participants was recruited during the pilot intervention. University

campuses are a common source for study participants' recruitment, as in Zhang et al.


(2014) or Scott-Parker (2012), but it limits the generalisability of findings. Students in

psychology represent a significant proportion of the subjects in psychological research,

which introduces a known bias in findings (Smart, 1966). Although Study 4 involved

QUT students primarily, due to the choice of the intervention venue, it can be assumed

that the participants had diverse backgrounds and were pursuing different degrees in

the university.

The study was limited by the number of participants who completed both

surveys. Although the collected data from 138 participants at Time 2 was a

considerable sample, it was still insufficient for robust conclusions. Furthermore, the

participants who completed both surveys were, on average, less risky than the

participant who dropped out. The challenges associated with evaluating the effect of

the intervention on a riskier population represented a significant limitation. The

systematic bias in the retained data might be a partial explanation for the lack of

significant effects of the intervention.

Another explanation for the lack of significant effects of the intervention could

be the participants were allowed to choose from four different scenarios. The

participants were given the opportunity to make their own choice in regards to the DUI

they wanted to experience to maintain the real-world nature of the intervention. Using

different scenarios in combination with the large drop-out might have reduced the

experimental control over the mechanisms causing shifts in the participants' salient

beliefs before and after the intervention. As a result, the within-group variability might

have been potentially increased, which, in turn, is likely to result in a decreased

likability to detect differences. To increase the future studies potential to detect

differences, Section 10.3 discusses a proposal for analysing data collected through

interventions with different scenarios as a condition.

Violations of assumptions were another observed data-related issue. The

violations forced the application of non-parametric tests, limiting the validity of the

findings. Similar to Study 2, case-targeted measures to reduce dropout rate may prove

useful in future research, although this may reduce the real-world resemblance of the

study.

Another limitation was that the perceived ease or difficulty of driving under the

different conditions was not assessed with respect to the individual participants. It is

acknowledged that a participant might have found driving in the impaired VR


simulation easy, or potentially more challenging and fun. In such a case, the

intervention could have promoted the DUI, despite the good intention behind its

implementation. However, the intervention was delivered "as-is" in the real world and

was evaluated as such. Furthermore, the data did not provide evidence for increased

DUI three months after the intervention.

9.7 CONCLUSION

Raising awareness of driving-related risks is not new, and education programs

have been studied in the past (Lewis, Fleiter, & Smith, 2015). In such cases,

researchers were able to provide essential insights into understanding the full

intervention implications, and subsequently, suggest strategies for addressing gaps in

their implementation. For example, researchers suggest that COTs are potentially

persuasive instruments that might help young drivers adopt safer driving behaviour

when used as intervention tools (Schroeter et al., 2012; Steinberger et al., 2015). VR

is one such COT. However, VR is a new tool. With the VR technology becoming

increasingly available and finding its way into prevention efforts, the question of how

VR can potentially improve road safety becomes necessary to investigate. At the same

time, there is a very limited number of VR studies in road safety (see Chapter 8).

Study 4 assessed the impact of a real-world VR intervention with "3D Tripping"

as an intervention tool. By doing so, it exhibited several strengths besides its real-world

nature, such as a comparatively large and diverse sample, a control for any general

influence, and a focus on long-term effects. Nevertheless, the study had its limitations,

e.g. potential for self-report bias, large drop-out rate, and problematic data. Despite

those limitations, Study 4 not only expanded the knowledge around using VR

simulations of risky driving in road safety but extended the knowledge into the context

of DUI. To achieve that, Study 4 utilised self-report items, drawn directly from the

literature and subsequently adapted to fit the study (see Subsection 4.4.3). Their

number was intentionally kept to a minimum to maximise response rate, as suggested

by Hart et al. (2005).

While answering RQ4 (How do young drivers’ self-reported behaviour of not

DUI and intention not to DUI alter in their free-living environment as a result of a VR

intervention?), it was found that the implemented intervention did not produce a

statistically significant effect on the participants’ self-reported behaviour of not DUI

and intention not to DUI. The observed results might be due to the overly positive self-


reported behaviour of the participants in general. Nevertheless, the findings of the

current study supported the predictive validity of the extended TPB framework to

explore young drivers' DUI when a VR intervention is deployed. Study 4 also assessed

additional predictors, over and above TPB. The assessment of those additional

predictors revealed mixed results. Such findings provided support salient beliefs to be

considered as potentially better targets of VR interventions than personality

characteristics.

Overall, Study 4 provided insights into the complexity of DUI as a behaviour. It

showed that positively influencing DUI might be neither easy nor straightforward. The

obtained results from the "3D Tripping" intervention showed that targeting DUI

intention and DUI behaviour is not one and the same thing. Different constructs shall

be considered leading, depending on whether the objective is to influence intention or

behaviour. Specific insights on understanding the difference, stemming from the Study

4 VR simulations of risky driving, are presented in Subsection 9.6.1 above.

Suggestions on how to build on the obtained results are discussed in Section 10.3.

182 Chapter 10: General discussion

Chapter 10: General discussion

Chapter 10 begins with a summary of the rationale for this PhD research

program's focus on young drivers' safety and its overall contribution. The chapter

continues with integrating research findings, strengths and limitations, followed by a

discussion of directions for further research.

10.1 OVERALL CONTRIBUTION

Despite the consistent efforts of the international community to reduce road

trauma, fatalities continue to rise (WHO, 2018). Young people are reported across

jurisdictions as overrepresented in crashes (BITRE, 2018; NHTSA, 2018a; EC, 2018;

WHO, 2018). COTs are regarded as one of the contributors to these statistics, while,

at the same time, they do not contribute to the driving tasks (Parliament of Victoria

Road Safety Committee, 2006). This is particularly valid for young drivers, who are

recognised as early technology adopters (Lee, 2007). Nevertheless, researchers

suggest that COTs could become part of the efforts to reduce road trauma (Schroeter

et al., 2012; Steinberger et al., 2015).

Both academia and businesses promise and subsequently offer technological

solutions to increase safety. To potentially access those solutions, COTs are

increasingly available to the general consumer. Thus, COTs carry a potential for

ubiquitous low-cost outreach, but there is limited knowledge about the safety benefits

they deliver.

The current PhD program of research examined the effects of two examples of

COT-based interventions to reduce risky driving behaviour among young drivers. The

first intervention used a smartphone safe-driving app that aimed to transform the

smartphone, from an existing risk source, into a tool for reducing speeding. The second

intervention deployed VR simulations of risky driving under the influence to reduce

DUI. This was one of the first programs of research to comprehensively explore the

safety benefits on young drivers, of these two examples of off-the-shelf readily-

available COTs, in the participants' free-living environment. This was achieved across

four studies (see Figure 10.1).

Chapter 10: General discussion 183

SAFE-DRIVING APPS

Study 1 (Chapter 5)

Systematic Review of Safe-driving Apps

RQ: What is the state of the art evidence of the safety benefits of smartphone safe-driving apps for young

drivers?

METHOD: Systematic Review (adhering to the PRISMA guidelines)

ANALYSIS: Narrative

FINDINGS: Of 80 papers, selected for full-text review, 22 papers were found to be relevant for the

current program of research, with only 4 of them being explicitly focused on young drivers in naturalistic

settings. Only three articles of the four reported safety benefits for the participating young drivers,

stemming from using a smartphone safe-driving app. The interventions, reported in the three studies,

were considered to be not representative of real-world adoption and usage of smartphones safe-driving

apps.

Chapter 6

Selecting a safe-driving app

METHOD: Focus group (n=10), Systematic Review

ANALYSIS: Narrative, user testing

FINDINGS: To be considered useful as an intervention tool with persuasive potential, a safe-driving

app should be free for users, not requiring additional hardware, safe (not distracting), self-starting, not

geographically restricted and informative, i.e. provide detailed feedback. At the time, Flo was identified

as the most suitable candidate to be integrated into a real-world intervention.

Study 2 (Chapter 7)

Intervention with an off-the-shelf smartphone safe-driving app

RQ: How do young drivers’ self-reported behaviour of not speeding and intention not to speed alter in

their free-living environment, as a result of exposure to a smartphone safe-driving app intervention?

METHOD: A randomised controlled intervention with cross-sectional pre- (n=480) and post-surveys

(n=157, 126 controls), complemented by participants' (n=31) driving scores from safe-driving app

leaderboard.

ANALYSIS: Hierarchical multiple regression, Analysis of covariance.

FINDINGS: As a result of the smartphone safe-driving app intervention with Flo as an intervention tool,

no statistically significant effect was found on any of the TPB measures, including the participants' self-

reported behaviour of not speeding during the three months of the intervention and their intention not to

speed in the future. No statistically significant increased distraction was found, either. The only

statistically significant effect found was a decrease in distraction for the Intervention group male

participants.


VIRTUAL REALITY

Study 3 (Chapter 8)

Systematic Review of VR

RQ: How is VR applied in road safety research to motivate behavioural change in young drivers?

METHOD: Systematic Review (adhering to the PRISMA guidelines)

ANALYSIS: Narrative

FINDINGS: VR is still underused in road safety. Limited evidence on positive effects on young drivers

was found from using VR in laboratory conditions in small-scale studies. Of 7 papers selected for full-

text review, 6 were found to be partially relevant for the current program of research, with only 1

reporting safety benefits. VR simulations of risky driving were not evaluated as part of real-world

interventions.

Study 4 (Chapter 9)

Intervention with VR simulations of risky driving

RQ: How do young drivers’ self-reported behaviour of not DUI and intention not to DUI alter in their

free-living environment as a result of a VR intervention?

METHOD: A controlled intervention, with convenience-based assignment to groups, with cross-

sectional pre- (n=329) and post-surveys (n=138, 39 controls).

ANALYSIS: Logistic regression, McNemar's test, Chi-square test for independence, Wilcoxon Signed

Ranks Test.

FINDINGS: No statistically significant effect as a result of a VR intervention, with "3D Tripping" as an

intervention tool, was found on any of the TPB measures, including the participants' self-reported

behaviour of not DUI during the three months after the intervention and their intention not to DUI in the

future.

Figure 10.1. Outline of the thesis studies and findings

The four implemented studies were expected to deliver two main outcomes

within the framework of the current program of research. The first foreseen outcome

was a contribution to a better understanding of the safety benefits from using two

examples of COT applications, a smartphone safe-driving app and VR simulations of

risky driving, for risk prevention purposes in a road safety context. This first outcome

is synthesised in the following Section 10.2. The second outcome was informing the

future evaluations of COTs-based interventions, leveraging smartphone safe-driving

apps and VR simulations of risky driving, that aim to reduce risky driving behaviours

in young drivers. This second outcome is discussed in the subsequent Section 10.3.


10.2 INTEGRATION OF FINDINGS, STRENGTHS AND LIMITATIONS

10.2.1 Theoretical considerations

This PhD program of research investigated the effects of two COTs, a

smartphone safe-driving app and VR simulations of risky driving under the influence.

The evaluations of the two interventions were grounded in TPB (Ajzen, 1988). While

criticism of TPB and calls for its retirement have emerged (Sniehotta et al., 2014), the

theory is still widely applied across health domains with researchers suggesting paths

for its improvement (Conner, 2015). The findings from within the current program of

research added to the conversation around the TPB relevance.

Contributing to the theoretical discussion, the current program of research

followed Conner's (2015) suggestion to extend the TPB. Additional constructs,

previously being utilised to extend the theory, were drawn directly from the literature.

The current program of research utilised past behaviour (Conner & Sparks, 2005;

Elliott & Thomson, 2010; Gauld et al., 2016; Haque et al., 2012), moral norm (Conner

& Sparks, 2005; Elliott & Thomson, 2010; Manstead, 2000), peers' norm (Conner &

Sparks, 2005; Fleiter et al., 2006) and perceived risk (Gannon et al., 2014; Haque et

al., 2012; Rhodes & Pivik, 2011). As mentioned earlier, personality characteristics

(impulsivity, sensitivity to punishment and sensitivity to reward), previously found to

be relevant to young drivers (Scott-Parker, 2012), were also included in the extended

TPB framework. Including those constructs explained additional variance in the

regression models, investigating each of the DVs of primary interest (intention not to

speed, behaviour of not speeding during the three months of the intervention, intention

not to DUI and behaviour of not DUI during the three months after the intervention).

However, none of the additional predictors emerged consistently as a significant

predictor for all four DVs. An exception was past behaviour of not speeding, which

was found to uniquely explain the most variance in both intention not to speed and

behaviour of not speeding during the three months of the intervention. The past

behaviour of not DUI was the strongest predictor of intention not to DUI. Nevertheless,

the additional predictors' contribution cannot be generalised and should be analysed

carefully, in relation to each behaviour of interest, as done in Chapter 7 and in Chapter

9.

The overall research design (see Chapter 4) focused on addressing some

observed TPB limitations (see Section 3.5). For example, both interventions


implemented as part of the current program of research took a form with an

experimental and practice-oriented nature, thus, making the criticism that the theory

has a static explanatory nature (Sniehotta et al., 2014) less relevant. Sheeran et al.

(2013) argued that TPB focuses on people's rational reasoning and does not account

for unconscious influences. This argument was the reason the constructs of impulsivity,

sensitivity to punishment and sensitivity to reward, were added to the model.

Mackenzie (2016) criticised the TPB for the assumption that it sees the person

as having the resources and skills to enact the behaviour of interest. Safer behaviour

on the road is believed to be a result of driver training (Watson, 1997). However, the

conventional driver training experience does not necessarily lead to safer driving

behaviour (Vernick et al., 1999; Watson, 1997). In the two implemented COTs-based

interventions, the two example COTs were perceived as providing unconventional

driving experience. Thus, the participants were given opportunities to acquire new

skills or experiences and, in turn, to perform the behaviour of interest. For example,

Flo was providing critical feedback to the young drivers, which was based on

measuring their driving performance. In turn, that critical feedback was expected to

enable the study participants to manage their speed better. "3D Tripping" was putting

the participants in driving simulations, revealing the challenges of DUI. Thus, the

participants were equipped with new personalised knowledge of how their driving

performance was influenced by simulated drugs or alcohol. In turn, this new

knowledge should help the participants make a safe decision should they have to

choose between DUI and other alternatives. As a result, the criticised TPB assumption

of people having the resources and skills to enact the behaviour of interest was less

relevant.

Sniehotta et al. (2014) shared concerns that TPB might not always have

sufficient predictive power on its own. While findings from both interventions

confirmed TPB as a good fit, those findings revealed mixed evidence about the level

of the predictive power of the theory. Study 2, which focused on speeding, potentially

a more common and socially acceptable behaviour, found TPB to predict more than

50% of the variance in both intention not to speed and behaviour of not speeding

during the three months of the intervention. Thus, it did not offer support for Sniehotta

et al. (2014). However, in Study 4, which focused on DUI, a potentially more extreme

behaviour with few people committing it, and much fewer feel comfortable to report


their engagement in it, the TPB's predictive power did not pass the 50% threshold,

which supported Sniehotta et al. (2014).

It has to be noted that the data analysis in Study 4 had limitations. For example,

the collected data was not normally distributed. Also, a substantial number of people

who reported some level of DUI before the intervention did not complete the second

survey. This missing information created a gap in the collected data. These limitations

need to be addressed in future research to make a more informed conclusion on

whether TPB is suitable as an overarching framework in interventions targeting DUI.

For example, future research may focus on recruiting a larger sample, or on DUI

offenders as a target group, which may reduce the impact of the encountered

challenges.

10.2.2 Practical considerations

Going beyond the discussion around the TPB relevance for road safety

researchers, the current findings may have implications for persuasion literature,

especially literature that explores technology. While previous research (see Chapter 5

and Chapter 8) reported some positive effects, the findings from the current research

did not provide such evidence for the two examples of COTs when used in the

participants' free-living environment. For example, findings from the smartphone safe-

driving app intervention did not find speed-related safety benefits such as those

reported by Zhang et al. (2014) for facilitating speed limits compliance. This may be

due to the small sample size, the length of the intervention period or the COT being

used. Zhang et al. (2014) examined only five people on five predefined routes. It

cannot be inferred from Zhang et al. (2014) study whether the behaviour could be

sustained for a long time, on different routes and without supervision, which was the

case in Study 2. The current findings were also different than Creaser et al. (2015)

findings. The authors reported reduced risky driving behaviours of their participants,

in general. Similar to Creaser et al. (2015), Flo was alerting the drivers of their risky

behaviour. However, Study 2 did not employ in-vehicle monitoring and parental

notifications, which might have served their deterrent purpose well in Creaser et al.

(2015) study. Thus, the results from both Study 2 and Study 4 provided insights into

how effective such COTs-based interventions are if there is no external influence on

the young drivers, i.e. when the young drivers take their own decisions in regards to

their behaviour on the road.


In the case of Study 4, the VR intervention with "3D Tripping" as an intervention

tool did not result in a statistically significant change in DUI. This result is different

than the result reported by Agrawal et al. (2017). This might be due to the very

different investigated behaviours, which was hazards anticipation in Agrawal et al.

(2017), while Study 4 focused on DUI. Agrawal et al. (2017) also did not examine

long-term effects. Nevertheless, Study 4 addressed limitations reported in the

systematic review (see Chapter 8), such as lack of VR realism and immersion (Gaibler

et al., 2015; Orfila et al., 2015). Furthermore, the Chapter 8 systematic review did not

identify another real-world VR intervention study. Study 4, therefore, investigated the

effects of a real-world VR intervention, making a unique contribution to the literature.

In the real world of road safety practitioners, such as social entrepreneurs, the

findings from this program of research may provide evidence that COTs need to be

investigated well in advance. Such evidence may provide the most added value before

COTs integration into road safety interventions or their release to market. The

suggestion may be increasingly valid in light of the evidence for lack of statistically

significant safety benefits, delivered through the two evaluated examples of COTs.

The observed evidence suggests that, although examples of technology may be

enjoyable and easily adopted by young drivers, COTs may have no significant positive

effect on their behaviour when used for road safety. Nevertheless, positive effects were

found by other authors (see Chapter 5 and Chapter 8). For example, Agrawal et al.

(2017) found VR simulations of risky driving to improve hazards anticipation.

Smartphone safe-driving apps intervention contributed to lowering speed (Musicant &

Botzer, 2016), decreasing fatigue and reducing negative mood (He et al., 2017;),

preventing collisions (Botzer et al., 2017), improving distance perception

(Schartmüller & Riener, 2015), and reducing drivers’ risk-taking (Creaser et al., 2015).

However, due to the differences in the focus of those studies as well as in the used

COTs’ features, it is challenging to define consistent reasons as to why similar

beneficial effects were not observed in Study 2 and Study 4. A common observation

in both systematic reviews (see Chapter 5 and Chapter 8), though, was that the

implemented COTs-based interventions were not resembling real-world interventions.

Thus, more research is required to investigate whether the claims for potential safety

benefits from particular examples of COTs are sustained in free living environment

contexts and over longer periods of time.


From a practical perspective, the present program of research showed the

potential synergies that could be leveraged when researchers and road safety

practitioners, such as social entrepreneurs (see Section 1.4), work together. While

theoretically-grounded evaluations of interventions are the norm amongst researchers,

they do not seem to be an equally common approach amongst road safety social

entrepreneurs. The current program of research went outside the laboratory, and into

spaces where 1) road safety social entrepreneurs usually operate, and 2) researchers do

not execute control. It generated new knowledge about the effects of the two

implemented interventions. At the same time, this less common approach to research

methodology design did not impact the quality of the research findings.

10.2.3 Methodological considerations

The two implemented interventions with COTs as intervention tools were

assessed for safety benefits for young drivers, i.e. reducing their speeding and DUI,

respectively. They were implemented in the way they are unconditionally offered to

the general public, simulating their real-world adoption. No expectations were

imposed on any facet of participants’ use of these interventions. As a result, the current

PhD program of research added to the literature evidence on two novel complementary

approaches to persuade young drivers to reduce their risk-taking in respect of speeding

and DUI. It is believed to be the first comprehensive program of research to do so.

Thus arguably, a substantial strength of the current research was that it went into the

real world of road safety interventions as far from simulated research settings and into

the young drivers' free-living environments as possible.

By going outside the laboratory, the current program of research generated new

knowledge around real-world adoption and retention rates of the particular COTs, i.e.

the safe-driving app Flo. Flo was selected, taking into consideration recommendations,

coming from road safety practitioners and researchers (see Chapter 6). The final choice

was made after comparing Flo with many other safe-driving apps. Nevertheless, the

success in getting young people to use it was lower than expected. At the same time,

it has to be acknowledged that study incentives were not tied to using the smartphone

safe-driving app in Study 2, which may be one of the reasons for the low success rate.

Yet, the lack of such incentives is a realistic aspect in free-living environments.

Although there is a body of evidence suggesting that incentives, such as rewards,

may influence driving behaviour (Musicant & Lotan, 2015; Schroeter et al., 2014),


exploring their effect fell outside the scope of the current project. Using performance-

based incentives would have reduced the real-world nature of the interventions, where

such are not always available. In fact, the Study 2 retention rate without incentives was

higher than the retention rate reported in Musicant and Lotan (2015) with incentives.

Musicant and Lotan's (2015) participants stopped using the studied smartphone safe-

driving app before the study was completed once they obtained all incentives. The

generated information on real-world adoption and retention rates in Study 2 may be

found useful as a comparison when reviewing existing technological solutions. Such

information can also be used to question developers’ claims when preliminarily

assessing the potential of specific safe-driving apps.

As part of the two implemented interventions' investigations, the current

program of research provided an understanding of how other COTs, namely social

media, can be used to recruit participants. The structure and the nature of research

samples are a common concern in road safety research. Social media could be explored

as an additional route to reach out to more people that can benefit from an intervention.

However, despite the initial success in recruiting larger sample sizes with the help of

social media, unexpected high dropout rates resulted in the current program of research

final data collection time point sample size requirements not being met. Therefore, the

presented findings require careful interpretation, which, nevertheless, may provide a

starting point for future research to build on.

It is important to acknowledge that the findings from the present program of

research are preliminary in nature. The intervention studies would require replication

to confirm or to disagree with the findings, preferably with other COTs as intervention

tools and with larger samples.

10.3 FUTURE RESEARCH DIRECTIONS

The TPB (Ajzen, 1988) underpinned the evaluation within the current program

of research. However, several other theories were also considered before the final

choice to use TPB was made (see Chapter 3). Given that the current TPB-based

evaluation could not find intervention effects within the Study 2 and Study 4 collected

data, future research may consider using other theories to underpin evaluation. For

example, the TTM (Prochaska & Velicer, 1997) assigns a single standard scale score

to each participant which determines their behavioural change stage. Moving through


those stages would signify significant changes in behaviour. Exploring the participants'

individual scores might be a simple way to identify behavioural changes. Nevertheless,

those changes will still be self-reported.

The reviewed literature showed that a large proportion of road safety research

relied on self-reports to collect data. Self-reports are likely to be easier to collect and

in larger numbers. However, it may come with the disadvantage of potential bias.

Thus, it could be argued that objective on-road behavioural measures are a better fit to

assess the impact of an intervention. Such objective data may be very difficult to

collect for DUI, and, in that case, self-reports might still be the best available tool.

However, in the case of speeding, currently available technology claims to have the

capacity to offer data-collection capabilities.

In naturalistic driving studies focused on speeding, future research that aims to

evaluate potential safety benefits delivered through smartphone safe-driving apps

could build on the knowledge presented in the current thesis through integrating the

analysis of anonymous raw data from the participants' smartphones. Available sensors

(clock, GPS, accelerometer, gyroscope and magnetometer) can provide driving

information, as well as information on interactions (e.g. using an app, answering a

phone call, writing a message). Researchers can estimate potential critical events, such

as speeding, hard acceleration, hard braking and fast cornering, if there was an

interaction with the smartphone, and the nature of the interaction. However,

researchers should consider that it is not easy to find a common definition of a critical

event threshold in the literature. Different studies used different fixed G-force levels

to identify critical events. For example, Paefgen, Kehr, Zhai, and Michahelles (2012)

used as thresholds an accelerometer output of 0.1g for acceleration and braking events,

and 0.2g for steering. Fazeen, Gozick, Dantu, Bhukhiya, and González (2012) used a

g-force of more than ±0.3g on the y-axis to determine critical events. The threshold

used by Freidlin et al. (2018) was 0.45g.

The approach of using thresholds might be considered too general to allow for

reliable comparison of driving in different conditions, on various roads and with

different cars. Adaptive algorithms that generate driver profiles for each participant

may provide a more targeted solution (Castignani et al., 2017; Saiprasert,

Thajchayapong, Pholprasit, & Tanprasert, 2014). For example, Castignani et al. (2017)

would classify an event as critical only if it deviates from the driver's normal driving


style, i.e. an outlier event is observed. This would allow scores, and the related changes

in the driving style, to be calculated independently for each driver, taking into

consideration their car and their route. Thus, drivers can be objectively compared,

based on the achieved change in reducing the number of critical events (number of

outliers), rather than on the triggered G-force events, which may be normal and

necessary in the specific driver's situation. Such a solution, however, may increase the

workload on the smartphone hardware and may consume additional power. Constantly

running two power-consuming apps on one smartphone may negatively affect the

device's battery life. This may result in higher participants' drop-out rates, triggered by

the need to recharge smartphones often. The challenge, of additional power

consumption, may be overcome by:

1. Collaborating with the developer of the off-the-shelf app to be deployed

as an intervention tool. However, in that case, there might be a conflict of

interest, and the researcher should consider the possibility for the data to be

filtered or manipulated before being made available.

2. Offloading some of the computation tasks by acquiring direct access to

the status of the various vehicle systems through an OBD2 reader. The OBD2

reader would help collect potentially more accurate data than the one generated

by smartphone sensors. However, in that case, the OBD2 reader’s quality, its

price and the ability of the participants to install, and properly use it, become

valid considerations. The availability of a trustworthy open-source interface to

transfer the collected data to a research server becomes another issue. If there

are means for the researcher to purchase the devices, then assisting with

installation and with getting the devices to work on the participants' vehicles will

impose constraints, both geographically and in numbers.

3. Offloading some of the computation tasks to a second smartphone with

the necessary software, installed by the researcher in advance. In that case,

delivery and installation on the participants' vehicles are still valid constraints.

They could be limited if the researcher assesses in advance constructs in relation

to participants' propensity and ease to adopt new technology. This will allow

efforts to be channelled towards people that need more help.

Once such infrastructure is established, a researcher will be able to link objective

naturalistic driving data with subjective self-report measures. In turn, this will allow


for a comprehensive evaluation not only of safe-driving app interventions but also of

any interventions that involve independent driving on behalf of the participants.

Apart from the described opportunities for future research from a technological

perspective, such can build on the current program of research by continuing to

examine interventions with the current technology limitations but with a different

design and larger samples. For example, further research can consider splitting the

participants into more than two groups, to control for potential influences (see Figure

10.2).

Figure 10.2. Example of future safe-driving app intervention design

Such a design, in the case of a future safe-driving apps research, can be

operationalised after recruitment, with participants randomly assigned into one of the

four groups (three intervention and one no intervention/control). Depending on the

group a participant is assigned to, they may: a) have no additional tasks; b) periodically

receive peers' feedback, e.g. a leaderboard by email, with their relative achievements

in comparison to their peers in the subgroup; c) install and use an off-the-shelf safe-

driving app while driving; or d) same as “c”, but with the additional requirement to

participate in a leaderboard group in the chosen safe-driving app.

A similar design can be implemented, to build on the results of the current Study

4, by separately exploring the effect of each of the scenarios, in which a participant

drives under the influence of drugs (Figure 10.3). This, however, would require a

larger sample to be recruited or the researcher may have to choose the "3D Tripping"

VR experience, instead of the participant.


Figure 10.3. Example of a future DUI VR intervention design

Both interventions could be further stratified to control for participants’ gender

or experience. Both gender and experience were shown to be significant contributors

towards explaining the variance of intentions and behaviour in the two interventions

(see Chapter 7 and Chapter 9). Males were shown to be less distracted as a result of

the smartphone safe-driving app intervention. Future research may involve measures

to leverage such influence. Those design variations will allow further, and more in-

depth research to focus on the examination of COTs-backed interventions, focusing

on more than one condition.

10.3.1 A researcher's wish list to COT developers

This section reports on reflections by the author to challenge COT engineers and

developers to create apps and technology needed to foster further behavioural research

and to propose specific design recommendations. It is noted that those

recommendations are not based on the collected data within the current program of

research. Basing them on only one example of a safe-driving app and one example of

VR simulations of risky driving would be insufficient. Furthermore, the selected

methodology was not aimed at drawing out such recommendations. It is acknowledged

that it would require a research framework similar to Vaezipour (2018), including 1)

integrating a theory that looks at user experience and technology acceptance into the

research framework, 2) implementing a qualitative inquiry into the users' experience

and preferences, and 3) comparing experience and preferences as a result of more than

one example of the same COT type used as intervention tools.

While the current research was not guided by such design-centred methodology,

there is value in reflecting on the author’s experiences conducting the behavioural


research using the COTs as interventions. The following paragraphs discuss a number

of challenges with respect to both VR simulations of risky driving and smartphone

safe-driving apps.

The VR simulations of risky driving did not collect any data during the current

program of research. However, driving data could potentially be collected by the VR

software while the participants are experiencing DUI driving scenarios. Comparing

experience both in "normal" and in DUI mode can be highly beneficial. To enable such

comparisons, the VR software developers can embed the same situations in both

modes. For example, a situation that would trigger a participant's braking reaction can

collect data about the time needed to react in DUI and compare it to the reaction time

in normal mode. Such driving tasks can be more complicated, e.g. lane-keeping or

overtaking, but can provide the participant with a realistic understanding of their

driving abilities (SC3, see Section 4.2). Data can also be collected with respect to

obeying traffic rules or to the number of crashes and potential victims when DUI. Such

a variety of quantitative data can serve as a basis of a much more in-depth evaluation

of the drivers' experience. By presenting the data to the respective driver, the overall

experience may become more meaningful and informative. Increasing the informative

value of the experience can potentially influence the participants DUI attitudes (SC1)

and norms (SC 2). This influence on attitudes and norms may have a higher potential

to trigger the desired behaviour change.

While the VR environment is comparatively resilient to collecting unrelated data

because all events happen as part of predetermined scenarios, the case with the

smartphone safe-driving apps is different. Thus, future safe-driving apps to be used for

research purposes will provide much more utility if they are able to:

- Always run when a participant drives. Currently, self-starting safe-driving

apps are limited in detecting then a participant actually drives (as in operates)

a vehicle (and when not). They require reaching a certain speed threshold to

start recording data. Sometimes, the smartphones' sensors do not

communicate properly delaying self-start. In other cases, the vehicle might

move slower in traffic than the threshold which would not trigger self-start

at all. A similar problem arises with the apps' self-stop. Currently, they

require a certain amount of time to pass during which the speed of the

smartphone complies with certain criteria before the app stops recording.


Issues arise when the vehicle is in a traffic jam for a prolonged period of time

or when the person stops the vehicle and immediately starts walking. A

related issue is that safe-driving apps do not distinguish between the

participant being a driver and not being a driver. Currently, they use their

algorithms to detect speed triggers regardless of how the speed is generated.

As a result, an app would start recording when the participant is a passenger,

rides on public transport or on a bicycle. Data in those cases should not be

collected as it does not reflect the participant's driving. In conclusion, the

ability to collect all the relevant data and only the relevant data is critical to

improve subsequent data analysis and foster future behavioural research.

- Similar trips comparison. Currently, some safe-driving apps offer detailed

trip feedback. A user is able to look at similar trips and compare them

manually. However, it would be useful if the comparison of such trips would

be supported by more sophisticated algorithms. Such algorithms could

generate nuanced reports, listing significant improvements and significant

deteriorations in different driving situations of identical, and therefore,

comparable driving trips. Such reports should be made available to both

users and researchers. Users can use the information and reflect on their

behaviour. Such reflections can influence study participants' salient beliefs

and potentially lead to improvements in their behaviour. In respect to

researchers, such reports would allow an assessment of how much, in reality,

the specific app meets the COTs selection criteria (see Section 4.2). In the

end, the researchers can make conclusions about what behavioural changes

were triggered by the provided app feedback.

- Work on all devices. The variety of available smartphones is considerable,

which can cause issues with apps' deployment and use. Currently, safe-

driving apps do not seem to work on all available smartphone devices. In the

framework of the current program of research, problems were observed with

getting Flo to work on an iPhone version. In respect to another tested app,

missing sensors in a Huawei model caused problems with collecting

comparable data. To improve usability across smartphones, app developers

might consider focusing on simpler algorithms that are built to utilise only

components, common amongst all smartphones. A challenge in such an


endeavour would be the competition amongst hardware developers which

generally stands in the way of hardware standards harmonisation. The

variability of those standards would require the app developers to account

for many possible platforms to host their apps. Despite those challenges and

as discussed in Section 7.6.3, reported and unreported technical problems

might be one of the reasons for the Study 2 large drop-out rate. If those

problems are overcome, two benefits might occur. The first one is that more

users will be attracted to the technology. For examples, it is not uncommon

peers to have similar devices. Thus, if one group member shows interest,

potentially others will follow. Such a scenario leads to the second potential

benefit, which is recruiting and retaining larger research samples. With larger

research samples and less technical challenges, the behavioural research

would enjoy more robust results and conclusions.

10.4 CHAPTER SUMMARY

Road safety interventions are intended to reduce road trauma. They are

considered an effective countermeasure to encourage individuals to adopt less risky

behaviours while driving. While previous research provided knowledge with respect

to road safety interventions, which may potentially influence young drivers, one area

seemed underexplored – COTs. The reason for the limited knowledge may not be that

people do not know about COTs, or are not interested in the opportunities they create.

The reason may be that COTs are novel, while well-designed research takes time.

To establish a comprehensive baseline of expanding the knowledge around using

COTs in road safety, this research used the PRISMA guidelines to systematically

review the available evidence in respect to the two deployed COTs, a smartphone safe-

driving app and VR simulations of risky driving. The findings confirmed the gap,

revealed in the traditional literature review. Previous research at the intersection of

road safety, psychology and human-computer interaction provided little evidence of

safety benefits, delivered through smartphone safe-driving apps. Such, in the domain

of VR, was practically not existent, where probably the most original contribution of

the current thesis sits.

The program of research incorporated theoretical constructs, an extended TPB

framework, into real-world road safety interventions, which used a smartphone safe-


driving app and VR simulations of risky driving as intervention tools. The framework

fitted well into the two interventions, helping to explain a substantial amount of

variance in regards to the DVs of interest: 75% in intention not to speed, 64% in

behaviour of not speeding during the three months of the intervention, between 36.5%

and 73.3% in intention not to DUI, and between 24.3% and 41.5% in behaviour of not

DUI during the three months after intervention.

As argued throughout this dissertation, novel technologies, such as the two

deployed COTs, are here to stay. However, using them for the right purpose requires

both an understanding of the technology applications themselves as well as an

understanding of the interventions that leverage them. This research program provided

an original contribution to knowledge by examining the effects of two COT-based

interventions to reduce risky driving behaviour among young drivers. By separately

answering each of the four research questions (see Section 10.1), it addressed the

following fundamental research question:

How do COTs-based interventions influence young drivers?

The current PhD program of research did not find evidence for statistically

significant influence on young drivers as a result of the two COTs-based interventions

with Flo and "3D Tripping" as intervention tools. Neither intention not to perform the

respective behaviour of interest nor the participants’ self-reported behaviour were

significantly different between the Intervention and Control groups of the respective

intervention.

Both Control groups were not exposed to the interventions themselves.

However, the control participants were well aware of the studies and their aims, due

to the provided participants' information before their consent to participate was

obtained, followed by responding to the baseline survey about their driving behaviour.

This fact alone might have made them more aware not only of their respective

behaviour but also of the behaviour of the people around them. Such increased

awareness might have influenced the Control groups' participants’ answers in the post-

intervention surveys. The possibility that data collected from them might have been

influenced has to be acknowledged. Such influence might have contributed to the lack

of statistically significant effects between the groups in Study 2 and Study 4.


None of the other TPB constructs was significantly influenced as a result of the

Study 2 and Study 4 interventions. So, within the limitations of the current program of

research, no evidence was found that either of the two implemented COT-based

interventions managed to influence the involved young drivers’ safer driving

positively. However, when Flo was used as a tool, evidence was found that the

intervention resulted in significantly decreased self-reported smartphone interactions

for male participants.

Irrespective of the lack of significant effects on speeding and DUI, significant

positive effects on the Intervention group male participants were observed in Study 2.

The positive effects were observed in regards to their smartphone interactions, i.e. self-

reported initiating (less) communication and responding (less) to communication.

Such effects were not observed for the Intervention group female participants, though.

Thus, the findings are partially consistent with Creaser et al. (2015), who reported that

both their Intervention groups interacted significantly less with their smartphones than

the Control group. It has to be noted that no pressure, e.g. parental notifications, was

used in Study 2, which might have facilitated the consistent results amongst the groups

in Creaser et al. (2015). These findings highlighted the need that, when technology is

deployed in road safety, a more complex systematic approach should be considered.

This need is equally valid when COTs were not designed for research purposes but are

readily available, which was the case with Flo and "3D Tripping".

Notwithstanding the design of an intervention or its target group, the only certain

thing is that improving countermeasures, encouraging safer driving behaviours, must

remain in focus. Given that road crashes continue to be the leading cause of death in

the 5-29 age group (WHO, 2018), there is a constant need for improvement and

innovation. The current program of research provided insights and discussed

considerations on which road safety intervention designers, both researchers and

practitioners, may draw upon in their future endeavours. Such interventions may or

may not persuade individuals to reduce their engagement in risky driving behaviours.

However, not trying is the only certain way not to contribute to road trauma reduction.

References 201

References

Adminaite, D., Calinescu, T., Jost, G., Stipdonk, H., & Ward, H. (2018). 12th Annual Road Safety Performance Index (PIN) Report.

Agrawal, R., Knodler, M., Fisher, D. L., & Samuel, S. (2017). Advanced Virtual Reality Based Training to Improve Young Drivers’ Latent Hazard Anticipation Ability. Paper presented at the Proceedings of the Human Factors and Ergonomics Society Annual Meeting.

Ahn, S. J. G., Bailenson, J. N., & Park, D. (2014). Short-and long-term effects of embodied experiences in immersive virtual environments on environmental locus of control and behavior. Computers in Human Behavior, 39, 235-245.

Ajzen, I. (1988). Attitudes, personality, and behavior. Chicago: Dorsey Press. Ajzen, I. (1991). The theory of planned behavior. Organizational behavior and human

decision processes, 50(2), 179-211. Ajzen, I. (2002). Perceived behavioral control, self‐efficacy, locus of control, and the

theory of planned behavior 1. Journal of Applied Social Psychology, 32(4), 665-683.

Ajzen, I. (2006). Behavioral interventions based on the theory of planned behavior. In. Retrieved from https://people.umass.edu/aizen/pdf/tpb.intervention.pdf.

Ajzen, I., & Fishbein, M. (1980). A theory of reasoned action: Some applications and implications.

Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behaviour.

Armitage, C., & Conner, M. (2001). Efficacy of the theory of planned behaviour: A meta‐analytic review. British Journal of Social Psychology, 40(4), 471-499.

Audit Office of New South Wales. (2011). Improving road safety: Young drivers – Roads and traffic authority of NSW.

Australian Institute of Health and Welfare. (2008). Injury among young Australians. Bulletin 60.

Australian Transport Council. (2011). National Road Safety Strategy 2011–2020. 2011. Australian Transport Council: Canberra.

Austroads. (2008). The crash and offence experience of newly licensed young drivers in South Australia.

Aveyard, P., Cheng, K., Almond, J., Sherratt, E., Lancashire, R., Lawrence, T., . . . Evans, O. (1999). Cluster randomised controlled trial of expert system based on the transtheoretical (“stages of change”) model for smoking prevention and cessation in schools. Bmj, 319(7215), 948-953.

Bandura, A. (1977). Self-efficacy: toward a unifying theory of behavioral change. Psychological review, 84(2), 191.

Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory: Prentice-Hall, Inc.

Banks, J. (1998). Handbook of simulation: principles, methodology, advances, applications, and practice: John Wiley & Sons.

Banks, T. D. (2008). An investigation into how work-related road safety can be enhanced. (PhD), Retrieved from http://eprints.qut.edu.au/29683/

Bazargan-Hejazi, S., Teruya, S., Pan, D., Lin, J., Gordon, D., Krochalk, P., & Bazargan, M. (2017). The theory of planned behavior (TPB) and texting while driving behavior in college students. Traffic Injury Prevention, 18(1), 56-62.

202 References

Beanland, V., Goode, N., Salmon, P. M., & Lenné, M. G. (2013). Is there a case for driver training? A review of the efficacy of pre- and post-licence driver training. Safety Science, 51(1), 127-137. doi:https://doi.org/10.1016/j.ssci.2012.06.021

Bellotti, F., Berta, R., & De Gloria, A. (2014). A Social Serious Game Concept for Green, Fluid and Collaborative Driving. In A. De Gloria (Ed.), Applications in Electronics Pervading Industry, Environment and Society (Vol. 289, pp. 163-170): Springer International Publishing.

Bhagavathula, R., Williams, B., Owens, J., & Gibbons, R. (2018). The Reality of Virtual Reality: A Comparison of Pedestrian Behavior in Real and Virtual Environments. Paper presented at the Proceedings of the Human Factors and Ergonomics Society Annual Meeting.

Bingham, C. R., Barretto, A. I., Walton, M. A., Bryant, C. M., Shope, J. T., & Raghunathan, T. E. (2011). Efficacy of a web-based, tailored, alcohol prevention/intervention program for college students: 3-month follow-up. Journal of drug education, 41(4), 405-430.

Birrell, S., Fowkes, M., & Jennings, P. (2013). A smart driving smartphone application: Real-world effects on driving performance and glance behaviours. Paper presented at the 3rd International Conference on Driver Distraction and Inattention.

Birrell, S., Young, M., Stanton, N., & Jennings, P. (2017). Using Adaptive Interfaces to Encourage Smart Driving and Their Effect on Driver Workload. In Advances in Human Aspects of Transportation (pp. 31-43): Springer.

Birrell, S. A., & Fowkes, M. (2014). Glance behaviours when using an in-vehicle smart driving aid: A real-world, on-road driving study. Transportation research part F: traffic psychology and behaviour, 22, 113-125.

Birrell, S. A., Fowkes, M., & Jennings, P. A. (2014). Effect of using an in-vehicle smart driving aid on real-world driver performance. IEEE Transactions on Intelligent Transportation Systems, 15(4), 1801-1810.

Blanco, M., Biever, W. J., Gallagher, J. P., & Dingus, T. A. (2006). The impact of secondary task cognitive processing demand on driving performance. Accident Analysis & Prevention, 38(5), 895-906. doi:http://dx.doi.org/10.1016/j.aap.2006.02.015

Blascovich, J., Loomis, J., Beall, A. C., Swinth, K. R., Hoyt, C. L., & Bailenson, J. N. (2002). Immersive virtual environment technology as a methodological tool for social psychology. Psychological Inquiry, 13(2), 103-124.

Botzer, A., Musicant, O., & Perry, A. (2017). Driver behavior with a smartphone collision warning application–a field study. Safety Science, 91, 361-372.

Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative research in psychology, 3(2), 77-101.

Briggs, G. F., Hole, G. J., & Land, M. F. (2016). Imagery-inducing distraction leads to cognitive tunnelling and deteriorated driving performance. Transportation research part F: traffic psychology and behaviour, 38, 106-117. doi:http://dx.doi.org/10.1016/j.trf.2016.01.007

Bureau of Infrastructure Transport and Regional Economics. (2018). Road trauma Australia 2017 statistical summary.

Castellà, J., & Pérez, J. (2004). Sensitivity to punishment and sensitivity to reward and traffic violations. Accident Analysis & Prevention, 36(6), 947-952. doi:https://doi.org/10.1016/j.aap.2003.10.003

References 203

Castignani, G., Derrmann, T., Frank, R., & Engel, T. (2017). Smartphone-Based Adaptive Driving Maneuver Detection: A Large-Scale Evaluation Study. IEEE Transactions on Intelligent Transportation Systems.

Catford, J. (1998). Social entrepreneurs are vital for health promotion—but they need supportive environments too. Health Promotion International, 13(2), 95-97.

Centers for Disease Control and Prevention. (2016). Youth Risk Behavior Surveillance—United States, 2015. MMWR. Surveillance Summaries, 65.

Chan, D. C. N., Wu, A. M. S., & Hung, E. P. W. (2010). Invulnerability and the intention to drink and drive: An application of the theory of planned behavior. Accident Analysis & Prevention, 42(6), 1549-1555. doi:https://doi.org/10.1016/j.aap.2010.03.011

Chen, H.-Y. W., Donmez, B., Hoekstra-Atwood, L., & Marulanda, S. (2016). Self-reported engagement in driver distraction: An application of the Theory of Planned Behaviour. Transportation research part F: traffic psychology and behaviour, 38, 151-163. doi:http://dx.doi.org/10.1016/j.trf.2016.02.003

Chen, K. B., Xu, X., Lin, J.-H., & Radwin, R. G. (2015). Evaluation of older driver head functional range of motion using portable immersive virtual reality. Experimental gerontology, 70, 150-156.

Compeau, D. R., & Higgins, C. A. (1995). Application of social cognitive theory to training for computer skills. Information systems research, 6(2), 118-143.

Conner, M. (2015). Extending not retiring the theory of planned behaviour: a commentary on Sniehotta, Presseau and Araújo-Soares. Health psychology review, 9(2), 141-145. doi:10.1080/17437199.2014.899060

Conner, M., & Sparks, P. (2005). Theory of planned behaviour and health behaviour. Predicting health behaviour, 2, 170-222.

Constantinou, E., Panayiotou, G., Konstantinou, N., Loutsiou-Ladd, A., & Kapardis, A. (2011). Risky and aggressive driving in young adults: Personality matters. Accident Analysis & Prevention, 43(4), 1323-1331. doi:https://doi.org/10.1016/j.aap.2011.02.002

Creaser, J., Morris, N., Edwards, C., Manser, M., Cooper, J., Swanson, B., & Donath, M. (2015). Teen Driver Support System (TDSS) Field Operational Test.

Creaser, J. I., Edwards, C. J., Morris, N. L., & Donath, M. (2015). Are cellular phone blocking applications effective for novice teen drivers? Journal of Safety Research, 54, 75. e29-78.

Davis, R., Campbell, R., Hildon, Z., Hobbs, L., & Michie, S. (2015). Theories of behaviour and behaviour change across the social and behavioural sciences: a scoping review. Health psychology review, 9(3), 323-344.

De Nooijer, J., Van Assema, P., De Vet, E., & Brug, J. (2005). How stable are stages of change for nutrition behaviors in the Netherlands? Health Promotion International, 20(1), 27-32.

Deery, H. A. (2000). Hazard and risk perception among young novice drivers. Journal of Safety Research, 30(4), 225-236.

DeVellis, R. F. (2016). Scale development: Theory and applications (Vol. 26): Sage publications.

Di Noia, J., Contento, I. R., & Prochaska, J. O. (2008). Computer-mediated intervention tailored on transtheoretical model stages and processes of change increases fruit and vegetable consumption among urban African-American adolescents. American journal of health promotion, 22(5), 336-341.

Diewald, S., Möller, A., Roalter, L., Stockinger, T., & Kranz, M. (2013). Gameful design in the automotive domain: review, outlook and challenges. Paper

204 References

presented at the Proceedings of the 5th International Conference on Automotive User Interfaces and Interactive Vehicular Applications.

Dinh-Zarr, T. B., Sleet, D. A., Shults, R. A., Zaza, S., Elder, R. W., Nichols, J. L., . . . Services, T. F. o. C. P. (2001). Reviews of evidence regarding interventions to increase the use of safety belts. American Journal of Preventive Medicine, 21(4), 48-65.

Donovan, R. (2011). Theoretical models of behaviour change. The SAGE handbook of social marketing, 15-31.

Duggan, K., & Shoup, K. (2013). Business gamification for dummies: John Wiley & Sons.

Ecker, R., Holzer, P., Broy, V., & Butz, A. (2011). EcoChallenge: a race for efficiency. Paper presented at the Proceedings of the 13th International Conference on Human Computer Interaction with Mobile Devices and Services, Stockholm, Sweden.

Elliott, A., & Woodward, W. (2007). Statistical analysis quick reference guidebook: With SPSS examples: Sage.

Elliott, M., & Thomson, J. (2010). The social cognitive determinants of offending drivers’ speeding behaviour. Accident Analysis & Prevention, 42(6), 1595-1605. doi:http://dx.doi.org/10.1016/j.aap.2010.03.018

Engström, I., Gregersen, N. P., Hernetkoski, K., Keskinen, E., & Nyberg, A. (2003). Young novice drivers, driver education and training. Literature Review. VTI Report A, 491.

European Commission. (2018). Road safety in European Union: trends, statistics and main challenges. Report.

Falk, E. B., Cascio, C. N., O'Donnell, M. B., Carp, J., Tinney, F. J., Bingham, C. R., . . . Simons-Morton, B. G. (2014). Neural responses to exclusion predict susceptibility to social influence. JOURNAL OF ADOLESCENT HEALTH, 54(5), S22-S31.

Fazeen, M., Gozick, B., Dantu, R., Bhukhiya, M., & González, M. C. J. I. T. o. I. T. S. (2012). Safe driving using mobile phones. 13(3), 1462-1468.

Ferguson, S. A. (2003). Other high-risk factors for young drivers—how graduated licensing does, doesn't, or could address them. Journal of Safety Research, 34(1), 71-77. doi:http://dx.doi.org/10.1016/S0022-4375(02)00082-8

Fernandes, R., Hatfield, J., & Soames Job, R. F. (2010). A systematic investigation of the differential predictors for speeding, drink-driving, driving while fatigued, and not wearing a seat belt, among young drivers. Transportation research part F: traffic psychology and behaviour, 13(3), 179-196. doi:https://doi.org/10.1016/j.trf.2010.04.007

Fife‐Schaw, C., Sheeran, P., & Norman, P. (2007). Simulating behaviour change interventions based on the theory of planned behaviour: Impacts on intention and action. British Journal of Social Psychology, 46(1), 43-68.

Fineout-Overholt, E., Melnyk, B. M., Stillwell, S. B., & Williamson, K. M. (2010). Evidence-based practice step by step: critical appraisal of the evidence: part I. The American Journal of Nursing, 110(7), 47-52.

Fisher, D., Pradhan, A., Pollatsek, A., & Knodler Jr, M. (2007). Empirical evaluation of hazard anticipation behaviors in the field and on driving simulator using eye tracker. Transportation Research Record: Journal of the Transportation Research Board(2018), 80-86.

Fitz-Walter, Z., Johnson, D., Wyeth, P., Tjondronegoro, D., & Scott-Parker, B. (2017). Driven to drive? Investigating the effect of gamification on learner driver

References 205

behavior, perceived motivation and user experience. Computers in Human Behavior, 71, 586-595.

Fleiter, J. J., Watson, B. C., Lennon, A. J., & Lewis, I. M. (2006). Significant others, who are they?-Examining normative influences on speeding.

Fogg, B. (2009). Creating Persuasive Technologies: An Eight-Step Design Process. Freeman, J., Liossis, P., Schonfeld, C., Sheehan, M., Siskind, V., & Watson, B. (2005).

Self-reported motivations to change and self-efficacy levels for a group of recidivist drink drivers. Addictive Behaviors, 30(6), 1230-1235. doi:http://dx.doi.org/10.1016/j.addbeh.2004.10.007

Frei, A., Svarin, A., Steurer-Stey, C., & Puhan, M. A. (2009). Self-efficacy instruments for patients with chronic diseases suffer from methodological limitations-a systematic review. Health and quality of life outcomes, 7(1), 86.

Freidlin, R. Z., Dave, A. D., Espey, B. G., Stanley, S. T., Garmendia, M. A., Pursley, R., . . . uHealth. (2018). Measuring risky driving behavior using an mhealth smartphone app: development and evaluation of gforce. 6(4), e69.

Gaibler, F., Faber, S., Edenhofer, S., & von Mammen, S. (2015). Drink & drive: A serious but fun game on alcohol-induced impairments in road traffic. Paper presented at the Games and Virtual Worlds for Serious Applications (VS-Games), 2015, 7th International Conference.

Gannon, B., Rosta, L., Reeve, M., Hyde, M. K., & Lewis, I. (2014). Does it matter whether friends, parents, or peers drink walk? Identifying which normative influences predict young pedestrian’s decisions to walk while intoxicated. Transportation research part F: traffic psychology and behaviour, 22, 12-24.

Garner, T. A. (2017). Echoes of Other Worlds: Sound in Virtual Reality: Past, Present and Future: Springer.

Gauld, C. S., Lewis, I. M., White, K. M., & Watson, B. (2016). Young drivers’ engagement with social interactive technology on their smartphone: Critical beliefs to target in public education messages. Accident Analysis & Prevention, 96, 208-218.

Ghavami, M., Harandy, T. F., & Kabir, K. (2016). The effect of educational intervention in promoting safe behaviors in a sample of Iranian primary school students: an application of the health belief model. Global journal of health science, 8(11), 242.

Gillibrand, R., & Stevenson, J. (2006). The extended health belief model applied to the experience of diabetes in young people. British journal of health psychology, 11(1), 155-169.

Gold, M. A., Tzilos, G. K., Stein, L., Anderson, B. J., Stein, M. D., Ryan, C. M., . . . DiClemente, C. (2016). A Randomized Controlled Trial to Compare Computer-assisted Motivational Intervention with Didactic Educational Counseling to Reduce Unprotected Sex in Female Adolescents. Journal of pediatric and adolescent gynecology, 29(1), 26-32.

Gonzalez, D. O., Martin-Gorriz, B., Berrocal, I. I., Morales, A. M., Salcedo, G. A., & Hernandez, B. M. (2017). Development and assessment of a tractor driving simulator with immersive virtual reality for training to avoid occupational hazards. Computers and Electronics in Agriculture, 143, 111-118.

Gosselin, D., Gagnon, S., Stinchcombe, A., & Joanisse, M. (2010). Comparative optimism among drivers: An intergenerational portrait. Accident Analysis & Prevention, 42(2), 734-740.

206 References

Hagger, M., & Chatzisarantis, N. (2005). First‐and higher‐order models of attitudes, normative influence, and perceived behavioural control in the theory of planned behaviour. British Journal of Social Psychology, 44(4), 513-535.

Haque, M. M., & Washington, S. (2013). Effects of mobile phone distraction on drivers’ reaction times. Journal of the Australasian College of Road Safety, 24(3), 20-29.

Haque, R., Clapoudis, N., King, M., Lewis, I., Hyde, M. K., & Obst, P. (2012). Walking when intoxicated: An investigation of the factors which influence individuals’ drink walking intentions. Safety Science, 50(3), 378-384. doi:https://doi.org/10.1016/j.ssci.2011.09.017

Harré, N., Foster, S., & O'Neill, M. (2005). Self‐enhancement, crash‐risk optimism and the impact of safety advertisements on young drivers. British Journal of Psychology, 96(2), 215-230.

Hart, T. C., Rennison, C. M., & Gibson, C. (2005). Revisiting respondent “fatigue bias” in the National Crime Victimization Survey. Journal of Quantitative Criminology, 21(3), 345-363.

Hatfield, J., Fernandes, R., & Job, R. S. (2014). Thrill and adventure seeking as a modifier of the relationship of perceived risk with risky driving among young drivers. Accident Analysis & Prevention, 62, 223-229.

He, Y., Yan, X., Chu, D., Wu, C., Liu, J., & Wang, X. (2017). Development and Evaluation of a Smartphone-based Curve-speed Warning System for Heavy Duty Vehicles. Retrieved from

Hingson, R., Heeren, T., Levenson, S., Jamanka, A., & Voas, R. (2002). Age of drinking onset, driving after drinking, and involvement in alcohol related motor-vehicle crashes. Accident Analysis & Prevention, 34(1), 85-92. doi:http://dx.doi.org/10.1016/S0001-4575(01)00002-1

Hochbaum, G., Rosenstock, I., & Kegels, S. (1952). Health belief model. United States Public Health Service.

Horswill, M. S., Waylen, A. E., & Tofield, M. I. (2004). Drivers' Ratings of Different Components of Their Own Driving Skill: A Greater Illusion of Superiority for Skills That Relate to Accident Involvement1. Journal of Applied Social Psychology, 34(1), 177-195.

Horvath, C., Lewis, I., & Watson, B. (2012). Peer passenger identity and passenger pressure on young drivers’ speeding intentions. Transportation research part F: traffic psychology and behaviour, 15(1), 52-64.

Hu, X., Deng, J., Zhao, J., Hu, W., Ngai, E. C.-H., Wang, R., . . . Leung, V. (2015). SAfeDJ: a crowd-cloud codesign approach to situation-aware music delivery for drivers. ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 12(1s), 21.

Hughes, J. R., Keely, J. P., Fagerstrom, K. O., & Callas, P. W. (2005). Intentions to quit smoking change over short periods of time. Addictive Behaviors, 30(4), 653-662.

Ifinedo, P. (2016). Examining students' intention to continue using blogs for learning: Perspectives from technology acceptance, motivational, and social-cognitive frameworks. Computers in Human Behavior. doi:http://dx.doi.org/10.1016/j.chb.2016.12.049

Ingram, K. M., Espelage, D. L., Merrin, G. J., Valido, A., Heinhorst, J., & Joyce, M. (2019). Evaluation of a virtual reality enhanced bullying prevention curriculum pilot trial. Journal of adolescence, 71, 72-83.

References 207

International Transport Forum. (2015). Why Does Road Safety Improve When Economic Times Are Hard?

Janz, N. K., & Becker, M. H. (1984). The health belief model: A decade later. Health Education & Behavior, 11(1), 1-47.

Jiang, Y., Zhang, J., Chikaraishi, M., Seya, H., & Fujiwara, A. (2017). Effects of a GPS-enabled smart phone App with functions of driving safety diagnosis and warning information provision on over-speeding violation behavior on expressways. Transportation research procedia, 25, 1815-1823.

Kashevnik, A., & Lashkov, I. (2018). Decision Support System for Drivers and Passengers: Smartphone-Based Reference Model and Evaluation. Paper presented at the Proceedings of the 23rd Conference of Open Innovations Association FRUCT.

Kaye, S.-A. (2014). Individual differences in the processing of punishment and reward cues : an application to road safety messages. (PhD), Retrieved from http://eprints.qut.edu.au/79616/

Kim, H.-Y. (2013). Statistical notes for clinical researchers: assessing normal distribution (2) using skewness and kurtosis. Restorative dentistry & endodontics, 38(1), 52-54.

Kircher, K., Ahlström, C., & Patten, C. (2011). Mobile telephones and other communication devices and their impact on traffic safety. VTI Report A, 729.

Klauer, S. G., Dingus, T. A., Neale, V. L., Sudweeks, J. D., & Ramsey, D. J. (2006). The impact of driver inattention on near-crash/crash risk: An analysis using the 100-car naturalistic driving study data.

Kohler, C. L., Grimley, D., & Reynolds, K. (1999). Theoretical approaches guiding the development and implementation of health promotion programs. In Handbook of health promotion and disease prevention (pp. 23-49): Springer.

Kotler, P., Hessekiel, D., & Lee, N. (2012). Good Works!: Marketing and Corporate Initiatives that Build a Better World... and the Bottom Line: John Wiley & Sons.

Kowalski, K., Jeznach, A., & Tuokko, H. A. (2014). Stages of driving behavior change within the Transtheoretical Model (TM). Journal of Safety Research, 50, 17-25. doi:http://dx.doi.org/10.1016/j.jsr.2014.01.002

Laapotti, S., Keskinen, E., Hatakka, M., & Katila, A. (2001). Novice drivers' accidents and violations—a failure on higher or lower hierarchical levels of driving behaviour. Accident Analysis & Prevention, 33(6), 759-769.

Lamble, D., Laakso, M., & Summala, H. (1999). Detection thresholds in car following situations and peripheral vision: Implications for positioning of visually demanding in-car displays. Ergonomics, 42(6), 807-815.

Lang, Y., Liang, W., Xu, F., Zhao, Y., & Yu, L.-F. (2018). Synthesizing Personalized Training Programs for Improving Driving Habits via Virtual Reality.

Lawton, R., Parker, D., Manstead, A. S. R., & Stradling, S. G. (1997). The Role of Affect in Predicting Social Behaviors: The Case of Road Traffic Violations. Journal of Applied Social Psychology, 27(14), 1258-1276. doi:10.1111/j.1559-1816.1997.tb01805.x

Lee, C. (1989). Theoretical weaknesses lead to practical problems: The example of self-efficacy theory. Journal of Behavior Therapy and Experimental Psychiatry, 20(2), 115-123. doi:http://dx.doi.org/10.1016/0005-7916(89)90044-X

Lee, J. D. (2007). Technology and teen drivers. Journal of Safety Research, 38(2), 203-213. doi:http://dx.doi.org/10.1016/j.jsr.2007.02.008

208 References

Legislative Assembly of Queensland: Parliamentary Travelsafe Committee. (2005). Driving on empty: Fatigue driving in Queensland: Government Press.

Lennon, A., Oviedo-Trespalacios, O., & Matthews, S. (2017). Pedestrian self-reported use of smart phones: Positive attitudes and high exposure influence intentions to cross the road while distracted. Accident Analysis & Prevention, 98, 338-347. doi:http://dx.doi.org/10.1016/j.aap.2016.10.028

Lettice, F., & Parekh, M. (2010). The social innovation process: themes, challenges and implications for practice. International Journal of Technology Management, 51(1), 139-158.

Lewis, I., Fleiter, J., & Smith, J. (2015). Students' responses to the RACQ docudrama program.

Lewis, I., Watson, B., White, K. M., & Elliott, B. (2013). The beliefs which influence young males to speed and strategies to slow them down: informing the content of antispeeding messages. Psychology & Marketing, 30(9), 826-841.

Lewis, I. M., Watson, B., & White, K. M. (2010). Response efficacy: The key to minimizing rejection and maximizing acceptance of emotion-based anti-speeding messages. Accident Analysis & Prevention, 42(2), 459-467. doi:http://dx.doi.org/10.1016/j.aap.2009.09.008

Lewis, I. M., Watson, B. C., & Tay, R. S. (2007). Examining the effectiveness of physical threats in road safety advertising: The role of the third-person effect, gender, and age. Transportation research part F: traffic psychology and behaviour, 10(1), 48-60. doi:10.1016/j.trf.2006.05.001

Lewis, I. M., Watson, B. C., Tay, R. S., & White, K. M. (2007). The Role of Fear Appeals in Improving Driver Safety: A Review of the Effectiveness of Fear-arousing (threat) Appeals in Road Safety Advertising. International Journal of Behavioral and Consultation Therapy, 3(2), 203-222.

Li, Q., Qiao, F., Qiao, Y., & Yu, L. (2016). Implications of smartphone messages on driving performance along local streets. Bridging the East and West, 282.

Liberati, A., Altman, D. G., Tetzlaff, J., Mulrow, C., Gøtzsche, P. C., Ioannidis, J. P., . . . Moher, D. J. P. m. (2009). The PRISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate health care interventions: explanation and elaboration. 6(7), e1000100.

Littell, J. H., & Girvin, H. (2002). Stages of change A critique. Behavior Modification, 26(2), 223-273.

Louveton, N., Mccall, R., Koenig, V., Avanesov, T., & Engel, T. (2016). Driving while using a smartphone-based mobility application: evaluating the impact of three multi-choice user interfaces on visual-manual distraction. Applied Ergonomics, 54, 196-204.

Luk, J. W., Trim, R. S., Karyadi, K. A., Curry, I., Hopfer, C. J., Hewitt, J. K., . . . Wall, T. L. (2017). Unique and interactive effects of impulsivity facets on reckless driving and driving under the influence in a high-risk young adult sample. Personality and Individual Differences, 114, 42-47.

Mackenzie, J. E. (2016). Mothers' sleepiness and driving in the postpartum period. (PhD), Retrieved from http://eprints.qut.edu.au/95190/

Mair, J., & Noboa, E. (2006). Social entrepreneurship and social transformation: An exploratory study. University of Navarra-IESE Business School Working Paper Series, 955.

Manstead, A. (2000). The role of moral norm in the attitude–behavior relation. In Attitudes, behavior, and social context: The role of norms and group

References 209

membership. (pp. 11-30). Mahwah, NJ, US: Lawrence Erlbaum Associates Publishers.

Markey, A. R. (2014). Three essays on boredom. Mauriello, L. M., Ciavatta, M. M. H., Paiva, A. L., Sherman, K. J., Castle, P. H.,

Johnson, J. L., & Prochaska, J. M. (2010). Results of a multi-media multiple behavior obesity prevention program for adolescents. Preventive medicine, 51(6), 451-456.

Mayhew, D. R., Simpson, H. M., & Pak, A. (2003). Changes in collision rates among novice drivers during the first months of driving. Accident Analysis & Prevention, 35(5), 683-691.

McCartt, A. T., Mayhew, D. R., Braitman, K. A., Ferguson, S. A., & Simpson, H. M. (2009). Effects of age and experience on young driver crashes: review of recent literature. Traffic Injury Prevention, 10(3), 209-219.

McEvoy, S., & Stevenson, M. (2007). An exploration of the role of driver distraction in serious road crashes. Distracted driving. Sydney, Australasian College of Road Safety, 189-211.

McEvoy, S. P., Stevenson, M. R., & Woodward, M. (2006). The impact of driver distraction on road safety: results from a representative survey in two Australian states. Injury prevention : journal of the International Society for Child and Adolescent Injury Prevention, 12(4), 242-247. doi:10.1136/ip.2006.012336

McGwin Jr, G., & Brown, D. B. (1999). Characteristics of traffic crashes among young, middle-aged, and older drivers. Accident Analysis & Prevention, 31(3), 181-198.

McKinsey&Company. (2014). Connected car, automotive value chain unbound. McKnight, A. S., & McKnight, A. J. (2003). Young novice drivers: careless or

clueless? Accident Analysis and Prevention, 35(6), 921-925. doi:10.1016/S0001-4575(02)00100-8

Miller, S. M., Shoda, Y., & Hurley, K. (1996). Applying cognitive-social theory to health-protective behavior: breast self-examination in cancer screening. Psychological bulletin, 119(1), 70.

Moan, I., & Rise, J. (2011). Predicting intentions not to “drink and drive” using an extended version of the theory of planned behaviour. Accident Analysis Prevention, 43(4), 1378-1384.

Morina, N., Ijntema, H., Meyerbröker, K., & Emmelkamp, P. M. (2015). Can virtual reality exposure therapy gains be generalized to real-life? A meta-analysis of studies applying behavioral assessments. Behaviour research and therapy, 74, 18-24.

Morrongiello, B. A., Corbett, M., Foster, A., & Koutsoulianos, S. (2018). Using Virtual Reality to Examine the Effects of Peer Modeling on Child Pedestrian Behavior. Retrieved from

Mulrow, C. (1994). Systematic reviews: rationale for systematic reviews. Bmj, 309(6954), 597-599.

Murray, W., White, J., & Ison, S. (2012). Work-related road safety: A case study of Roche Australia. Safety Science, 50(1), 129-137. doi:http://dx.doi.org/10.1016/j.ssci.2011.07.012

Musicant, O., & Botzer, A. (2016). The Safety Benefits Of Collision Warning Applications–Evidence From On-road Data. International Journal of Safety and Security Engineering, 6(2), 362-371.

210 References

Musicant, O., & Lotan, T. (2015). Can novice drivers be motivated to use a smartphone based app that monitors their behavior? Transportation research part F: traffic psychology and behaviour.

National Highway Traffic Safety Administration. (2010). Overview of the National Highway Traffic Safety Administration’s Driver Distraction Program. Report No. DOT HS, 811, 299.

National Highway Traffic Safety Administration. (2015). Traffic safety facts: Research note. Distracted driving 2013 (Report No. DOT HS 812 132). In.

National Highway Traffic Safety Administration. (2017). 2016 Fatal Motor Vehicle Crashes: Overview.

National Highway Traffic Safety Administration. (2018a). 2016 Young Drivers Traffic Safety Facts.

National Highway Traffic Safety Administration. (2018b). 2017 Fatal Motor Vehicle Crashes: Overview.

Ng, B.-Y., Kankanhalli, A., & Xu, Y. C. (2009). Studying users' computer security behavior: A health belief perspective. Decision Support Systems, 46(4), 815-825.

Obst, P., Armstrong, K., Smith, S., & Banks, T. (2011). Age and gender comparisons of driving while sleepy: Behaviours and risk perceptions. Transportation research part F: traffic psychology and behaviour, 14(6), 539-542. doi:http://dx.doi.org/10.1016/j.trf.2011.06.005

Orfila, O., Gruyer, D., Judalet, V., & Revilloud, M. (2015). Ecodriving performances of human drivers in a virtual and realistic world. Paper presented at the Intelligent Vehicles Symposium (IV), 2015 IEEE.

Organisation for Economic Co-operation and Development. (2006). Young Drivers: The Road to Safety.

Otto, J., Ward, N., Swinford, S., & Linkenbach, J. (2014). Engaging worksite bystanders to reduce risky driving. Transportation research part F: traffic psychology and behaviour, 26, 370-378. doi:https://doi.org/10.1016/j.trf.2014.02.006

Oviedo-Trespalacios, O., Haque, M. M., King, M., & Washington, S. (2016). Understanding the impacts of mobile phone distraction on driving performance: A systematic review. Transportation Research Part C: Emerging Technologies, 72, 360-380.

Oviedo-Trespalacios, O., Truelove, V., Watson, B., & Hinton, J. (2019). The impact of road advertising signs on driver behaviour and implications for road safety: A critical systematic review. Transportation Research Part A: Policy, 122, 85-98.

Paefgen, J., Kehr, F., Zhai, Y., & Michahelles, F. (2012). Driving behavior analysis with smartphones: insights from a controlled field study. Paper presented at the Proceedings of the 11th International Conference on mobile and ubiquitous multimedia.

Parliament of Victoria Road Safety Committee. (2006). Inquiry into driver distraction: final report. In: Victoria Government.

Parr, M. N., Ross, L. A., McManus, B., Bishop, H. J., Wittig, S. M. O., & Stavrinos, D. (2016). Differential impact of personality traits on distracted driving behaviors in teens and older adults. Accident Analysis & Prevention, 92, 107-112. doi:https://doi.org/10.1016/j.aap.2016.03.011

Patten, C. J. D., Kircher, A., Östlund, J., Nilsson, L., Svenson, O., Psykologiska, i., . . . Samhällsvetenskapliga, f. (2006). Driver experience and cognitive workload

References 211

in different traffic environments. Accident Analysis and Prevention, 38(5), 887-894. doi:10.1016/j.aap.2006.02.014

Patton, J. H., & Stanford, M. S. (1995). Factor structure of the Barratt impulsiveness scale. Journal of clinical psychology, 51(6), 768-774.

Pearson, M. R., Murphy, E. M., & Doane, A. N. (2013). Impulsivity-like traits and risky driving behaviors among college students. Accident Analysis & Prevention, 53, 142-148.

Peck, R. C., Gebers, M. A., Voas, R. B., & Romano, E. (2008). The relationship between blood alcohol concentration (BAC), age, and crash risk. Journal of Safety Research, 39(3), 311-319. doi:http://dx.doi.org/10.1016/j.jsr.2008.02.030

Pickrell, T. M., & Liu, C. (2015). Occupant Restraint Use in 2013: Results From the NOPUS Controlled Intersection Study. Retrieved from

Polacsek, M., Rogers, E. M., Woodall, W. G., Delaney, H., Wheeler, D., & Rao, N. (2001). MADD victim impact panels and stages-of-change in drunk-driving prevention. Journal of Studies on Alcohol, 62(3), 344-350.

Pollatsek, A., Fisher, D. L., & Pradhan, A. (2006). Identifying and Remedying Failures of Selective Attention in Younger Drivers. Current Directions in Psychological Science, 15(5), 255-259. doi:10.1111/j.1467-8721.2006.00447.x

Porter, M. E., & Kramer, M. R. (2011). The big idea: Creating shared value. Harvard Business Review, 89(1), 2.

Potard, C., Kubiszewski, V., Camus, G., Courtois, R., & Gaymard, S. (2018). Driving under the influence of alcohol and perceived invulnerability among young adults: An extension of the theory of planned behavior. Transportation research part F: traffic psychology and behaviour, 55, 38-46. doi:https://doi.org/10.1016/j.trf.2018.02.033

Prato, C. G., Toledo, T., Lotan, T., & Taubman - Ben-Ari, O. (2010). Modeling the behavior of novice young drivers during the first year after licensure. Accident Analysis & Prevention, 42(2), 480-486. doi:https://doi.org/10.1016/j.aap.2009.09.011

Prochaska, J. O., DiClemente, C. C., Velicer, W. F., & Rossi, J. S. (1993). Standardized, individualized, interactive, and personalized self-help programs for smoking cessation. Health Psychology, 12(5), 399.

Prochaska, J. O., Evers, K. E., Castle, P. H., Johnson, J. L., Prochaska, J. M., Rula, E. Y., . . . Pope, J. E. (2012). Enhancing multiple domains of well-being by decreasing multiple health risk behaviors: a randomized clinical trial. Population health management, 15(5), 276-286.

Prochaska, J. O., & Velicer, W. F. (1997). The transtheoretical model of health behavior change. American journal of health promotion, 12(1), 38-48.

Quine, L., Rutter, D. R., & Arnold, L. (1998). Predicting and understanding safety helmet use among schoolboy cyclists: a comparison of the theory of planned behaviour and the health belief model. Psychology and Health, 13(2), 251-269.

Quine, L., Rutter, D. R., & Arnold, L. (2000). Comparing the theory of planned behaviour and the health belief model: the example of safety helmet use among schoolboy cyclists. Understanding and changing health behaviour: From health beliefs to self-regulation, 73-98.

Quine, L., Rutter, D. R., & Arnold, L. (2001). Persuading school‐age cyclists to use safety helmets: Effectiveness of an intervention based on the Theory of Planned Behaviour. British journal of health psychology, 6(4), 327-345.

212 References

Rahman, R., Qiao, F., Li, Q., & Yu, L. (2016). Developing a smartphone based warning system application to enhance the safety at work zones. Retrieved from

Rana, N. P., & Dwivedi, Y. K. (2015). Citizen's adoption of an e-government system: Validating extended social cognitive theory (SCT). Government Information Quarterly, 32(2), 172-181. doi:http://dx.doi.org/10.1016/j.giq.2015.02.002

Recarte, M. A., & Nunes, L. M. (2000). Effects of verbal and spatial-imagery tasks on eye fixations while driving. Journal of Experimental Psychology: Applied, 6(1), 31.

Redshaw, S. (2006). Peer Reviewed Paper Dangerous Gender Performances: 'Hydraulic Masculinity' as a Norm for Young Male Drivers.

Regan, M., Williamson, A., Friswell, R., Hatfield, J., & Grzebieta, R. (2012). Submission to Staysafe Parliamentary Inquiry into Distracted Driving.

Regan, M. A., Lee, J. D., & Young, K. L. (2009). Driver distraction: theory, effects, and mitigation. Boca Raton: CRC Press.

Rhodes, N., & Pivik, K. (2011). Age and gender differences in risky driving: The roles of positive affect and risk perception. Accident Analysis & Prevention, 43(3), 923-931.

Riener, A., & Reder, J. (2014). Collective data sharing to improve on driving efficiency and safety. Paper presented at the Adjunct Proceedings of the 6th International Conference on Automotive User Interfaces and Interactive Vehicular Applications.

Rivis, A., & Sheeran, P. (2003). Descriptive norms as an additional predictor in the theory of planned behaviour: A meta-analysis. 22(3), 218-233.

Rodrigues, M. A. F., Macedo, D. V., Serpa, Y. R., & Serpa, Y. R. (2015). Beyond fun: an interactive and educational 3D traffic rules game controlled by non-traditional devices. Paper presented at the Proceedings of the 30th Annual ACM Symposium on Applied Computing.

Rollnick, S., Heather, N., Gold, R., & Hall, W. (1992). Development of a short ‘readiness to change’ questionnaire for use in brief, opportunistic interventions among excessive drinkers. British Journal of Addiction, 87(5), 743-754. doi:10.1111/j.1360-0443.1992.tb02720.x

Ropelato, S., Zünd, F., Magnenat, S., Menozzi, M., & Sumner, R. (2017). Adaptive Tutoring on a Virtual Reality Driving Simulator. Paper presented at the 1st Workshop on Artificial Intelligence Meets Virtual and Augmented Worlds (AIVRAR) in conjunction with SIGGRAPH Asia 2017.

Rowden, P. J., & Watson, B. C. (2013). Mobile phone use and driving : the message is just not getting through. Paper presented at the Proceedings of the Australasian College of Road Safety Conference 2013, National Wine Centre of Australia, Adelaide, SA. http://eprints.qut.edu.au/64520/

Rutter, J., & Quine, L. (2002). Changing health behaviour (Vol. 17): Citeseer. Ryder, B., Gahr, B., Egolf, P., Dahlinger, A., & Wortmann, F. (2017). Preventing

traffic accidents with in-vehicle decision support systems-The impact of accident hotspot warnings on driver behaviour. Decision Support Systems, 99, 64-74.

SafetyNet. (2009). Novice Drivers, retrieved 18.10.2016. Saiprasert, C., Thajchayapong, S., Pholprasit, T., & Tanprasert, C. (2014, 3-7 Nov.

2014). Driver behaviour profiling using smartphone sensory data in a V2I environment. Paper presented at the 2014 International Conference on Connected Vehicles and Expo (ICCVE).

References 213

Schartmüller, C., & Riener, A. (2015). Field studies to investigate safety distance violation with a low-cost observation system. Paper presented at the Adjunct Proceedings of the 7th International Conference on Automotive User Interfaces and Interactive Vehicular Applications.

Schroeder, P., Meyers, M., & Kostyniuk, L. (2013). National survey on distracted driving attitudes and behaviors--2012. Retrieved from

Schroeter, R., Oxtoby, J., & Johnson, D. (2014). AR and gamification concepts to reduce driver boredom and risk taking behaviours. Paper presented at the Proceedings of the 6th International Conference on Automotive User Interfaces and Interactive Vehicular Applications, Seattle, WA, USA. http://eprints.qut.edu.au/76134/

Schroeter, R., Rakotonirainy, A., & Foth, M. (2012). The social car : new interactive vehicular applications derived from social media and urban informatics. Paper presented at the AutomotiveUI'12, Portsmouth, New Hampshire. http://eprints.qut.edu.au/53208/

Schwarzer, R., & Renner, B. (2000). Social-cognitive predictors of health behavior:

action self-efficacy and coping self-efficacy. Health Psychology, 19(5), 487. Schwebel, D. C., Combs, T., Rodriguez, D., Severson, J., & Sisiopiku, V. (2016).

Community-based pedestrian safety training in virtual reality: A pragmatic trial. Accident Analysis & Prevention, 86, 9-15.

Schwebel, D. C., McClure, L. A., & Porter, B. E. (2017). Experiential exposure to texting and walking in virtual reality: a randomized trial to reduce distracted pedestrian behavior. Accident Analysis & Prevention, 102, 116-122.

Scialfa, C. T., Deschênes, M. C., Ference, J., Boone, J., Horswill, M. S., & Wetton, M. (2011). A hazard perception test for novice drivers. Accident Analysis & Prevention, 43(1), 204-208.

Scott-Parker, B. (2012). A compehensive investigation of the risky driving behaviour of young novice drivers. Thesis.

Scott-Parker, B., Goode, N., & Salmon, P. (2015). The driver, the road, the rules… and the rest? A systems-based approach to young driver road safety. Accident Analysis & Prevention, 74, 297-305.

Scott-Parker, B., King, M., & Watson, B. (2015). The psychosocial purpose of driving and its relationship with the risky driving behaviour of young novice drivers. Transportation research part F: traffic psychology and behaviour, 33, 16-26.

Scott-Parker, B., & Oviedo-Trespalacios, O. (2017). Young driver risky behaviour and predictors of crash risk in Australia, New Zealand and Colombia: Same but different? Accident Analysis & Prevention, 99, 30-38. doi:https://doi.org/10.1016/j.aap.2016.11.001

Scott-Parker, B., Watson, B., King, M. J., & Hyde, M. K. (2013). A further exploration of sensation seeking propensity, reward sensitivity, depression, anxiety, and the risky behaviour of young novice drivers in a structural equation model. Accident Analysis & Prevention, 50, 465-471. doi:http://dx.doi.org/10.1016/j.aap.2012.05.027

Scott-Parker, B., & Weston, L. (2017). Sensitivity to reward and risky driving, risky decision making, and risky health behaviour: A literature review. Transportation research part F: traffic psychology and behaviour, 49, 93-109. doi:https://doi.org/10.1016/j.trf.2017.05.008

Seelos, C., & Mair, J. (2004). Social entrepreneurship-The contribution of individual entrepreneurs to sustainable development.

214 References

Sela‐Shayovitz, R. (2008). Young drivers’ perceptions of peer pressure, driving under the influence of alcohol and drugs, and involvement in road accidents. Criminal Justice Studies, 21(1), 3-14.

Senserrick, T. M. (2007). Recent developments in young driver education, training and licensing in Australia. Journal of Safety Research, 38(2), 237-244. doi:http://dx.doi.org/10.1016/j.jsr.2007.03.002

Sheeran, P., Gollwitzer, P. M., & Bargh, J. A. (2013). Nonconscious processes and health. Health Psychology, 32(5), 460.

Shekari Soleimanloo, S. (2016). Effects of light and caffeine on human sleepiness and alertness: A simulated driving experiment. (PhD), Retrieved from https://eprints.qut.edu.au/95888/

Sherman, W. R., & Craig, A. B. (2018). Understanding virtual reality: Interface, application, and design: Morgan Kaufmann.

Shope, J. T., & Bingham, C. R. (2008). Teen Driving: Motor-Vehicle Crashes and Factors That Contribute. American Journal of Preventive Medicine, 35(3, Supplement), S261-S271. doi:http://dx.doi.org/10.1016/j.amepre.2008.06.022

Silverans, P., Alvarez, J., Assum, T., Evers, C., & Mathijssen, R. (2007). Alcolock programmes for professional and non-professional drivers in a European field trial. Paper presented at the The 8th International Annual Ignition Interlock Symposium, Seattle, WA.

Şimşekoğlu, Ö., & Lajunen, T. (2008). Social psychology of seat belt use: A comparison of theory of planned behavior and health belief model. Transportation research part F: traffic psychology and behaviour, 11(3), 181-191.

Smart, D., Vassallo, S., Sanson, A., Cockfield, S., Harris, A., & Harrison, W. (2005). In the driver's seat: Understanding young adults' driving behaviour.

Smart, R. G. (1966). Subject selection bias in psychological research. Canadian Psychologist/Psychologie canadienne, 7a(2), 115-121. doi:10.1037/h0083096

Smedslund, J. (1978). Bandura's theory of self‐efficacy: A set of common sense theorems. Scandinavian Journal of Psychology, 19(1), 1-14.

Sniehotta, F. F., Presseau, J., & Araújo-Soares, V. (2014). Time to retire the theory of planned behaviour. In: Taylor & Francis.

Stanford, M. S., Mathias, C. W., Dougherty, D. M., Lake, S. L., Anderson, N. E., Patton, J. H. J. P., & differences, i. (2009). Fifty years of the Barratt Impulsiveness Scale: An update and review. 47(5), 385-395.

Staton, C., Vissoci, J., Gong, E., Toomey, N., Wafula, R., Abdelgadir, J., . . . Hocker, M. (2016). Road Traffic Injury Prevention Initiatives: A Systematic Review and Metasummary of Effectiveness in Low and Middle Income Countries. PloS one, 11(1), e0144971. doi:10.1371/journal.pone.0144971

Stead, M., Tagg, S., MacKintosh, A. M., & Eadie, D. (2005). Development and evaluation of a mass media Theory of Planned Behaviour intervention to reduce speeding. Health education research, 20(1), 36-50.

Steinberger, F., Proppe, P., Schroeter, R., & Alt, F. (2016). CoastMaster: An ambient speedometer to gamify safe driving. Paper presented at the Proceedings of the 8th International Conference on Automotive User Interfaces and Interactive Vehicular Applications.

Steinberger, F., Schroeter, R., Foth, M., & Johnson, D. (2017). Designing gamified applications that make safe driving more engaging. Paper presented at the Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems.

References 215

Steinberger, F., Schroeter, R., Lindner, V., Fitz-Walter, Z., Hall, J., & Johnson, D. M. (2015). Zombies on the road: A holistic design approach to balancing gamification and safe driving. Paper presented at the Proceedings of the 7th International Conference on Automotive User Interfaces and Interactive Vehicular Applications, Nottingham, UK. http://eprints.qut.edu.au/84799/

Tabachnick, & Fidell. (2007). Using multivariate statistics: Allyn & Bacon/Pearson Education.

Tao, D., Zhang, R., & Qu, X. (2017). The role of personality traits and driving experience in self-reported risky driving behaviors and accident risk among Chinese drivers. Accident Analysis & Prevention, 99, 228-235. doi:https://doi.org/10.1016/j.aap.2016.12.009

Tay, R. (2005). The effectiveness of enforcement and publicity campaigns on serious crashes involving young male drivers: Are drink driving and speeding similar? Accident Analysis & Prevention, 37(5), 922-929. doi:http://dx.doi.org/10.1016/j.aap.2005.04.010

Theng, Y.-L., Lee, J. W., Patinadan, P. V., & Foo, S. S. (2015). The use of videogames, gamification, and virtual environments in the self-management of diabetes: a systematic review of evidence. Games for health journal, 4(5), 352-361.

Torrubia, R., Ávila, C., Moltó, J., & Caseras, X. (2001). The Sensitivity to Punishment and Sensitivity to Reward Questionnaire (SPSRQ) as a measure of Gray's anxiety and impulsivity dimensions. Personality and Individual Differences, 31(6), 837-862. doi:http://dx.doi.org/10.1016/S0191-8869(00)00183-5

Tranter, P., & Warn, J. (2008). Relationships between interest in motor racing and driver attitudes and behaviour amongst mature drivers: An Australian case study. Accident Analysis & Prevention, 40(5), 1683-1689.

Vaezipour, A. (2018). Design and development of an in-vehicle human machine interface for eco-safe driving. (PhD), Retrieved from https://eprints.qut.edu.au/118058/

Van Duyn, M. A. S., Heimendinger, J., Russek-Cohen, E., DlClemente, C. C., Sims, L. S., Subar, A. F., . . . Kahle, L. L. (1998). Use of the transtheoretical model of change to successfully predict fruit and vegetable consumption. Journal of Nutrition Education, 30(6), 371-380.

van Loon, A., Bailenson, J., Zaki, J., Bostick, J., & Willer, R. (2018). Virtual reality perspective-taking increases cognitive empathy for specific others. PloS one, 13(8), e0202442.

Velicer, W. F., Redding, C. A., Paiva, A. L., Mauriello, L. M., Blissmer, B., Oatley, K., . . . Prochaska, J. O. (2013). Multiple behavior interventions to prevent substance abuse and increase energy balance behaviors in middle school students. Translational behavioral medicine, 3(1), 82-93.

Vernick, J. S., Li, G., Ogaitis, S., MacKenzie, E. J., Baker, S. P., & Gielen, A. C. (1999). Effects of high school driver education on motor vehicle crashes, violations, and licensure. American Journal of Preventive Medicine, 16(1), 40-46. doi:10.1016/S0749-3797(98)00115-9

Walker, E. (2014). Australian graduated licensing scheme - policy framework. Wallace, L. S., Buckworth, J., Kirby, T. E., & Sherman, W. M. (2000). Characteristics

of Exercise Behavior among College Students: Application of Social Cognitive Theory to Predicting Stage of Change. Preventive medicine, 31(5), 494-505. doi:http://dx.doi.org/10.1006/pmed.2000.0736

Ward, N. J., Otto, J., Schell, W., Finley, K., Kelley-Baker, T., & Lacey, J. H. (2017). Cultural predictors of future intention to drive under the influence of cannabis

216 References

(DUIC). Transportation research part F: traffic psychology and behaviour, 49, 215-225. doi:https://doi.org/10.1016/j.trf.2017.06.013

Ward, N. J., Schell, W., Kelley-Baker, T., Otto, J., & Finley, K. (2018). Developing a theoretical foundation to change road user behavior and improve traffic safety: Driving under the influence of cannabis (DUIC). Traffic Injury Prevention, 19(4), 358-363. doi:10.1080/15389588.2018.1425548

Warner, H. W., & Åberg, L. (2008). Drivers’ beliefs about exceeding the speed limits. Transportation research part F: traffic psychology and behaviour, 11(5), 376-389.

Watson, B. C. (1997). When common sense just won't do: Misconceptions about changing the behaviour of road users. Paper presented at the The 2nd International Conference on Accident Investigation, Reconstruction, Interpretation and the Law, Brisbane, Queensland. http://eprints.qut.edu.au/7295/

Webb, T. L., & Sheeran, P. (2006). Does changing behavioral intentions engender behavior change? A meta-analysis of the experimental evidence. Psychological bulletin, 132(2), 249.

Wells-Parker, E., Burnett, C., Dill, P., & Williams, M. (1997). Initial development of self-efficacy scales for controlling drinking and driving. Paper presented at the Proceedings of the 14th International Conference on Alcohol, Drugs and Traffic Safety.

Wells-Parker, E., Williams, M., Dill, P., & Kenne, D. (1998). Stages of change and self-efficacy for controlling drinking and driving: A psychometric analysis. Addictive Behaviors, 23(3), 351-363.

West, R. (2005). Time for a change: putting the Transtheoretical (Stages of Change) Model to rest. Addiction, 100(8), 1036-1039.

Weston, L., & Hellier, E. (2018). Designing road safety interventions for young drivers – The power of peer influence. Transportation research part F: traffic psychology and behaviour, 55, 262-271. doi:https://doi.org/10.1016/j.trf.2018.03.003

White, K. M., Hyde, M. K., Walsh, S. P., & Watson, B. (2010). Mobile phone use while driving: An investigation of the beliefs influencing drivers’ hands-free and hand-held mobile phone use. Transportation research part F: traffic psychology and behaviour, 13(1), 9-20.

White, M. J., Cunningham, L. C., & Titchener, K. (2011). Young drivers’ optimism bias for accident risk and driving skill: Accountability and insight experience manipulations. Accident Analysis & Prevention, 43(4), 1309-1315.

Wickens, T. D., & Keppel, G. (2004). Design and analysis: A researcher's handbook: Pearson Prentice-Hall.

Williams, K. J., Peters, J. C., & Breazeal, C. L. (2013). Towards leveraging the driver's mobile device for an intelligent, sociable in-car robotic assistant. Paper presented at the Intelligent Vehicles Symposium (IV), 2013 IEEE.

World Health Organization. (2011). Mobile Phone Use: A Growing Problem of Driver Distraction.

World Health Organization. (2018). Global status report on road safety 2018. Wundersitz, L. (2012). An analysis of young drivers involved in crashes using in-depth

crash investigation data. Decision making (Total), 62(61.0), 115. Yıldırım-Yenier, Z., Vingilis, E., Wiesenthal, D. L., Mann, R. E., & Seeley, J. (2016).

Relationships between thrill seeking, speeding attitudes, and driving violations

References 217

among a sample of motorsports spectators and drivers. Accident Analysis & Prevention, 86, 16-22. doi:http://dx.doi.org/10.1016/j.aap.2015.09.014

You, C.-W., Lane, N. D., Chen, F., Wang, R., Chen, Z., Bao, T. J., . . . Torresani, L. (2013). CarSafe app: alerting drowsy and distracted drivers using dual cameras on smartphones. Paper presented at the Proceeding of the 11th annual international conference on Mobile systems, applications, and services.

Zhang, J., Jiang, Y., Sasaki, K., Tsubouchi, M., Matsushita, T., Kawai, T., & Fujiwara, A. (2014). A GPS-enabled smart phone app with simplified diagnosis functions of driving safety and warning information provision. Paper presented at the Proceedings of the 21st world congress of intelligent transport systems. Detroit, USA.

Appendices 219

Appendices

221

Appendix A

Smartphone safe-driving apps on Google Play and iTunes

Store: Google Play

Keyword Name Relevance road app Road Drivers: Legacy Road Mode DailyRoads Voyager Colorado Roads Real-time feedback Road Sidekick Lite On The Road Road Radar Speed limit alerts Easy Roads - Road Trip

Planner

Cross Road Traffic Voyage: Usa Roads CoPilot GPS - Navigationsmart driving RTA Smart Drive Speed Limit alerts while driving Smart Drive Records trips which can be shared

with friends and family. Dash - Drive Smart Provides driver score and insights to

help improve your performance. Checks engine light notifications and provides an explanation. Shows leaderboard to compare the best drivers. Provides multi-vehicle support with automatic VIN de-coding. In driving mode shows real-time data on MPG, as well as audio alerts. There are parental / alert features active for extreme deceleration, curfew, and geo-fencing.

Smart Drivers (SG) SMART DRIVING Real-time diagnostics about the car

with the following key features: Speed, RPM, coolant temperature, airflow temperature, on-board voltage, throttle position, engine load and oxygen sensors. Provides an opportunity to plan trips, calculate costs, search for attractions along the road, and others. Provides social functions to invite friends and share experience.

SmartDriver Collects driving information through the accelerometer and the GPS,

222

including distance, location, braking and acceleration, to determine driving behaviour and provide feedback.

Smart Control Free (OBD / ELM)

Shows messages about trip start and end, performance, and speed limits alert.

Driving G Monitor Uses GPS and acceleration sensors to monitor acceleration and determine driving comfort.

Auto4iBlack-DriveRecorder

Provides event alarm (shock, motion) and emergent recording. Sends 6 SOS SMS.

DriveSafe.Smart Bars incoming and outgoing calls and text messages.

Safe driving Safe driving SafeDrive Blocks calls and texts. It is free and

not geographically restricted. Tries to connect gamification (earning points) with the real world (receiving rewards). Works on auto-start.

Drive Safe Safe Driver Safety Driving Safe Driving Assistant Safe Driving + Auto SMS DashDroid - Safe Driving

App

Safe Driving Ltd. Drive Care - Safe driving Visualises odometer and distance

travelled, speeding alerts and trip details. Sends SMS alerts with details.

SEAT Safe Drive Works in the background even with the device blocked. Uses the device proximity sensor to activate the app without distractions. Visualises trips statistics only once the car has stopped.

Drive Mode Drive Safe Safe Driving Text

Machine

Drive Alive Lite iOnRoad Augmented

Driving Lite Real-time feedback

Safe Driving

223

Safe Driving Flo - Driving Insights Provides live feedback to the driver

or can run invisibly in the background providing detailed after-trip feedback on a Google map. It is free for users and has a web interface. There is no geographic restriction. Does not need a dongle.

Safe Driving App Records harsh acceleration and deceleration as well as speed limit exceeded by more than 10% with location and time.

Driver Safety - Automatic SMS

Drive Now Text Later Safe Drive Enforcer Anti Texting Safe Driving

App.

Safe Driving App App4Drivers safe teen

driver Feedback

Ride Safe Way - Safe Driving

Safe Driving + Auto SMS + TTS

Drive Control. AAMI Safe Driver Embeds a good mix of gamified

elements (scores, badges). Monitors for exact speed limits, not the general ones. Runs in the background. Provides a detailed after-trip feedback on a Google map. Does not need a dongle and is free for users.

iOnRoad Augmented Driving Pro

Safe Driver Text Response DraVA Driving Coach Driving Coach GPS Safe Driving Tracker Drive Safe Text Safe Drive Safe DrivingBuddy Generates a leaderboard with

achievements and levelling. There is an opportunity for bracket competition and socialization on Facebook.

DriveSafe Mode Drive SafeOBD game Olivia Drive | OBD2 -

ELM327

224

OBD2 scanner bluetooth Elm327

RealDash Android driver U-Scan Gears Pro (OBD 2 & Car) Drivemode: Driving

interface

Store: iTunes

Keyword Name Relevance road app Waze - GPS navigation,

Maps and Real-time traffic

smart driving Metromile - smart drive Provides daily driving insights. DriveSmart - Drive &

Save money Records events such as breaking, acceleration, and cornering.

ŠKODA Drive Calculates route efficiency, average speed, route distance, and saved money.

Greenlight - Safer Driving Starts Here

Generates a unique driver score based on driving data.

Driveway - Smart Driving Records events such as breaking, acceleration, and cornering.

Try and Drive Real-time data collection. T-Connect TH Smart System Monitor Start Smart: California

Teen Driver License Guide

Drive ULU Business Collects driver performance statistics (speeding, braking, driving style, eco-driving) and generates a weekly score.

Smart Drive by ÖkoTaxi Travel

Driver360 Powered by Agero Travel

Works in the background while driving to avoid driver distraction. Provides a personal driver score. Let's routes, times and trips be reviewed.

DriveProfiler Smart Lifestyle

Shows trips which can be submitted into a logbook. Keeps track of distance travelled on a graphical dashboard. Shows driver scores and provides driver behaviour feedback.

Drive Protected Travel Monitors driving speed using Apple watch.

YouDrive Travel

225

Mojio Shift Lifestyle Monitors driving behaviour to increase safety and save money on gas.

Terra Drive Navigation Safe driving EverDrive™ - Safe

Driving Lifestyle Keeps past drives with detailed feedback on maps. Helps improve driving with personalized tips. Allows for competition with friends, family, and other drivers. Reduces distracted driving and encourages safe driving.

OBD game RealDash Utilities Monitors vehicle speed and current location on the map. Times laps. Gathers performance measurements (with limited accuracy).

226

Appendix B

Study 2 Questionnaire

Time 1 before intervention (Sections 1, 2 & 3) Time 2 after 3 months driving with a smartphone app (Sections 1, 2 & 4) Conditions to participate:

1. Young driver, aged 18 to 25; 2. Have a valid open or provisional Australian driver's license. 3. Drive a car as the only means of transport; 4. Drive a minimum of 100 kilometres per month. N Question Possible answers

1. Demographic data 1.1 Age 18-19-20-21-22-23-

24-251.2 Gender Male / Female / Other 1.3 Type of license Learner / Provisional

(Year 1) / Provisional (Year 2) / Open

1.4 State Queensland / South Australia / New South Wales / Victoria / Australian Capital Territory / Northern Territory / Western Australia / Tasmania

2. Measuring social cognitive determinants (TPB-based) of drivers’ speeding behaviour (Elliott & Thomson, 2010). All items are measured using 9-point scales (scored 1–9).

2.1 To what extent do you intend to drive faster than the speed limit over the next 3 months? (measuring Intention to speed)

(No extent at all to A great extent)

2.2 How often do you think you will drive faster than the speed limit in the next 3 months? (measuring Intention to speed)

(Never to All the time)

2.3 How bad or good would it be for you personally if you drove faster than the speed limit over the next 3 months? (measuring Instrumental attitude)

(Extremely bad to Extremely good)

2.4 How unenjoyable or enjoyable would it be for you personally if you drove faster than the speed limit over the next 3 months? (measuring Affective attitude)

(Extremely unenjoyable to Extremely enjoyable)

2.5 Would the people who are important to you disapprove or approve of you driving faster than the speed limit over the next 3 months? (measuring Subjective norm)

(Definitely disapprove to Definitely approve)

227

2.6 How often do you think the people who are important to you will drive faster than the speed limit over the next 3 months? (measuring Descriptive norm)


2.7 How confident are you that you will be able to avoid driving faster than the speed limit over the next 3 months? (measuring Self-efficacy)

(Not at all confident to Extremely confident)

2.8 Over the next 3 months, how much do you feel that avoiding driving faster than the speed limit is under your control? (measuring Perceived controllability)

(Not at all to Very much so)

2.9 How wrong do you think it would be for you to drive faster than the speed limit over the next 3 months? (measuring Moral norm)

(Not at all wrong to Extremely wrong)

2.10 How often did you drive faster than the speed limit over the last 3 months? (measuring Past speeding behaviour)


2.11 Would your friends disapprove or approve of you driving over the speed limit over the next 3 months? (measuring Peers' norm)


Adapted from Gannon et al. (2014) to measure perceived risk.2.12 If you were to drive over the speed limit over the

next 3 months, how much would you worry about being involved in a road crash?

(Not at all worried to Worried very much)

2.13 If you were to drive over the speed limit over the next 3 months, how much would you worry about being caught by the Police?


Measuring frequencies (%) of Initiating, Monitoring/reading, and Responding to Social Interactive Technology on Smartphones while Driving (Gauld et al., 2016). How often do you do the following on your smartphone while driving: 2.11 Initiate communication on social interactive

technology? (Starting a communication) More than once per day; Daily; 1–2 times per week; 1–2 times per month; 1–2 times per 3 months; Once a year; Never

2.12 Monitor/read social interactive technology? (Checking for communication)

More than once per day; Daily; 1–2 times per week; 1–2 times per month; 1–2 times per 3 months; Once a year; Never

2.13 Respond to a communication on social interactive technology? (Replying to communication)

More than once per day; Daily; 1–2 times per week; 1–2 times per month; 1–2 times

228

per 3 months; Once a year; Never

3. Measuring impulsiveness (Barratt Impulsiveness Scale Version 11) (Patton & Stanford, 1995). All items are measured using 4-point scales.

3.1 I plan tasks carefully. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.2 I do things without thinking. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.3 I make-up my mind quickly. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.4 I am happy-go-lucky. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.5 I don’t “pay attention.” 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.6 I have “racing” thoughts. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.7 I plan trips well ahead of time. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.8 I am self controlled. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.9 I concentrate easily. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.10 I save regularly. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.11 I “squirm” at plays or lectures. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.12 I am a careful thinker. 1 (Rarely/Never); 2 (Occasionally); 3

229

(Often); 4 (Almost Always/Always)

3.13 I plan for job security. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.14 I say things without thinking. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.15 I like to think about complex problems. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.16 I change jobs. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.17 I act “on impulse.” 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.18 I get easily bored when solving thought problems. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.19 I act on the spur of the moment. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.20 I am a steady thinker. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.21 I change residences. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.22 I buy things on impulse. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.23 I can only think about one thing at a time. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.24 I change hobbies. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

230

3.25 I spend or charge more than I earn. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.26 I often have extraneous thoughts when thinking. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.27 I am more interested in the present than the future. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.28 I am restless at the theater or lectures. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.29 I like puzzles. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.30 I am future oriented. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

4. Measuring reward and punishment sensitivity through Sensitivity to Punishment and Sensitivity to Reward Questionnaire (SPSRQ) (Torrubia, Ávila, Moltó, & Caseras, 2001).

4.1 Do you often refrain from doing something because you are afraid of it being illegal?

Yes / No

4.2 Does the good prospect of obtaining money motivate you strongly to do some things?

Yes / No

4.3 Do you prefer not to ask for something when you are not sure you will obtain it?

Yes / No

4.4 Are you frequently encouraged to act by the possibility of being valued in your work, in your studies, with your friends or with your family?

Yes / No

4.5 Are you often afraid of new or unexpected situations?

Yes / No

4.6 Do you often meet people that you find physically attractive?

Yes / No

4.7 Is it difficult for you to telephone someone you do not know?

Yes / No

4.8 Do you like to take some drugs because of the pleasure you get from them?

Yes / No

4.9 Do you often renounce your rights when you know you can avoid a quarrel with a person or an organisation?

Yes / No

4.10 Do you often do things to be praised? Yes / No

231

4.11 As a child, were you troubled by punishments at home or in school?

Yes / No

4.12 Do you like being the centre of attention at a party or a social meeting?

Yes / No

4.13 In tasks that you are not prepared for, do you attach great importance to the possibility of failure?

Yes / No

4.14 Do you spend a lot of your time on obtaining a good image?

Yes / No

4.15 Are you easily discouraged in difficult situations? Yes / No 4.16 Do you need people to show their affection for you

all the time? Yes / No

4.17 Are you a shy person? Yes / No 4.18 When you are in a group, do you try to make your

opinions the most intelligent or the funniest?Yes / No

4.19 Whenever possible, do you avoid demonstrating your skills for fear of being embarrassed?

Yes / No

4.20 Do you often take the opportunity to pick up people you find attractive?

Yes / No

4.21 When you are with a group, do you have difficulties selecting a good topic to talk about?

Yes / No

4.22 As a child, did you do a lot of things to get people's approval?

Yes / No

4.23 Is it often difficult for you to fall asleep when you think about things you have done or must do?

Yes / No

4.24 Does the possibility of social advancement, move you to action, even if this involves not playing fair?

Yes / No

4.25 Do you think a lot before complaining in a restaurant if your meal is not well prepared?

Yes / No

4.26 Do you generally give preference to those activities that imply an immediate gain?

Yes / No

4.27 Would you be bothered if you had to return to a store when you noticed you were given the wrong change?

Yes / No

4.28 Do you often have trouble resisting the temptation of doing forbidden things?

Yes / No

4.29 Whenever you can, do you avoid going to unknown places?

Yes / No

4.30 Do you like to compete and do everything you can to win?

Yes / No

4.31 Are you often worried by things that you said or did?

Yes / No

4.32 Is it easy for you to associate tastes and smells to very pleasant events?

Yes / No

4.33 Would it be difficult for you to ask your boss for a raise (salary increase)?

Yes / No

4.34 Are there a large number of objects or sensations that remind you of pleasant events?

Yes / No

4.35 Do you generally try to avoid speaking in public? Yes / No

232

4.36 When you start to play with a slot machine, is it often difficult for you to stop?

Yes / No

4.37 Do you, on a regular basis, think that you could do more things if it was not for your insecurity or fear?

Yes / No

4.38 Do you sometimes do things for quick gains? Yes / No4.39 Comparing yourself to people you know, are you

afraid of many things? Yes / No

4.40 Does your attention easily stray from your work in the presence of an attractive stranger?

Yes / No

4.41 Do you often find yourself worrying about things to the extent that performance in intellectual abilities is impaired?

Yes / No

4.42 Are you interested in money to the point of being able to do risky jobs?

Yes / No

4.43 Do you often refrain from doing something you like in order not to be rejected or disapproved of by others?

Yes / No

4.44 Do you like to put competitive ingredients in all of your activities?

Yes / No

4.45 Generally, do you pay more attention to threats than to pleasant events?

Yes / No

4.46 Would you like to be a socially powerful person? Yes / No4.47 Do you often refrain from doing something because

of your fear of being embarrassed?Yes / No

4.48 Do you like displaying your physical abilities even though this may involve danger?

Yes / No

233

Appendix C

Study 4 Questionnaire

Eligibility to participate: 1. Young driver, aged 18 to 25; 2. Have a valid driver's license. 3. Have no history of seizures or epilepsy. (Only for the intervention group!)

Survey 1

Anonymou

s identifier

Survey 1: You are invited to participate in the research project "Driving under the influence virtual experience" information about which can be found in the Participants Information Sheet at www (link to the file). If you would like to participate in this project, please, generate your anonymous identifier below in order to proceed to Survey 1. It shall include your: day of birth, first letter of name, first letter of family name and last two digits of mobile number (example 24DL08).

(Short text, 6 characters limit)

N Question Possible answers 1. Demographic data

1.1 How old are you (in years)? (whole number) 1.2 What is your gender? Male / Female / Other1.3 How much is your driving experience (in years)? (whole number)

2. Measuring standard and extended TPB constructs on the influence of alcohol or drugs. All items are measured using 9-point scales (scored 1–9).

Adapted from Elliott & Thomson (2010) to measuring social cognitive determinants.2.1 To what extent do you intend to drive under the

influence of alcohol or drugs over the next 3 months? (measuring Intention to drive under the influence of alcohol or drugs)


2.2 How often do you think you will drive under the influence of alcohol or drugs in the next 3 months? (measuring Intention to drive under the influence of alcohol or drugs)


2.3 How bad or good would it be for you personally if you drove under the influence of alcohol or drugs over the next 3 months? (measuring Instrumental attitude)


2.4 How unenjoyable or enjoyable would it be for you personally if you drove under the influence of alcohol or drugs over the next 3 months? (measuring Affective attitude)


234

2.5 Would the people who are important to you disapprove or approve of you driving under the influence of alcohol or drugs over the next 3 months? (measuring Subjective norm)


2.6 How often do you think the people who are important to you will drive under the influence of alcohol or drugs over the next 3 months? (measuring Descriptive norm)


2.7 How confident are you that you will be able to avoid driving under the influence of alcohol or drugs over the next 3 months? (measuring Self-efficacy)


2.8 Over the next 3 months, how much do you feel that avoiding driving under the influence of alcohol or drugs is under your control? (measuring Perceived controllability)


2.9 How wrong do you think it would be for you to drive under the influence of alcohol or drugs over the next 3 months? (measuring Moral norm)


2.10 How often did you drive under the influence of alcohol or drugs over the last 3 months? (measuring Past behaviour of not DUI)


2.11 Would your friends disapprove or approve of you driving under the influence of alcohol or drugs over the next 3 months? (measuring Peers' norm)


Adapted from Gannon et al. (2014) to measure perceived risk.2.12 If you were to drive over the next 3 months under

the influence of alcohol or drugs, how much would you worry about being involved in a road crash?


2.13 If you were to drive over the next 3 months under the influence of alcohol or drugs, how much would you worry about being caught by the Police?


3. Measuring impulsiveness (Barratt Impulsiveness Scale Version 11) (Patton & Stanford, 1995). All items are measured using 4-point scales.

3.1 I plan tasks carefully. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.2 I do things without thinking. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.3 I make-up my mind quickly. 1 (Rarely/Never); 2 (Occasionally); 3

235


3.4 I am happy-go-lucky. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.5 I don’t “pay attention.” 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.6 I have “racing” thoughts. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.7 I plan trips well ahead of time. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.8 I am self controlled. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.9 I concentrate easily. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.10 I save regularly. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.11 I “squirm” at plays or lectures. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.12 I am a careful thinker. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.13 I plan for job security. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.14 I say things without thinking. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.15 I like to think about complex problems. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

236

3.16 I change jobs. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.17 I act “on impulse.” 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.18 I get easily bored when solving thought problems. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.19 I act on the spur of the moment. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.20 I am a steady thinker. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.21 I change residences. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.22 I buy things on impulse. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.23 I can only think about one thing at a time. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.24 I change hobbies. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.25 I spend or charge more than I earn. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.26 I often have extraneous thoughts when thinking. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.27 I am more interested in the present than the future.

1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.28 I am restless at the theater or lectures. 1 (Rarely/Never); 2 (Occasionally); 3

237


3.29 I like puzzles. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

3.30 I am future oriented. 1 (Rarely/Never); 2 (Occasionally); 3 (Often); 4 (Almost Always/Always)

Contact

Please, enter your e-mail address: /If you would like to enter into a random draw of 10 Amazon vouchers of 50 AUD each, please, enter your e-mail. We will use your e-mail for the sole purpose of communication in case you are drawn to win one of the vouchers and in three months to supply you with a link to Survey 2./

(Short text)

Survey 2 – Intervention group Eligibility to participate: Completed Survey 1 as participant in the Intervention group.

Anonymou

s identifier

You are invited to complete Survey 2 of the research project "Driving under the influence virtual experience" information about which can be found in the Participants Information Sheet at www (link to the file). If you would like to continue your participation in this project, please, enter the anonymous identifier you used in Survey 1 below in order to proceed to Survey 2. It shall include your: day of birth, first letter of name, first letter of family name and last two digits of mobile number (example 24DL08).


N Question Possible answers 1. Chosen experience on the driving simulator

1.1 What experience did you choose when driving the driving simulator with 3D Tripping virtual reality software?

Alcohol/Ecstasy/Magic mushrooms/Cannabis

2. Measuring standard and extended TPB constructs on the influence of alcohol or drugs. All items are measured using 9-point scales (scored 1–9).




238


























239

3. Measuring reward and punishment sensitivity through Sensitivity to Punishment and Sensitivity to Reward Questionnaire (SPSRQ) (Torrubia et al., 2001).


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No

3.10 Do you often do things to be praised?

Yes / No


Yes / No


Yes / No


Yes / No


Yes / No

3.15 Are you easily discouraged in difficult situations? Yes / No 3.16 Do you need people to show their affection for

you all the time?Yes / No




Yes / No


Yes / No


Yes / No


Yes / No

240


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No

3.35 Do you generally try to avoid speaking in public? Yes / No3.36 When you start to play with a slot machine, is it

often difficult for you to stop?Yes / No


Yes / No




Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No

241

3.46 Would you like to be a socially powerful person? Yes / No 3.47 Do you often refrain from doing something

because of your fear of being embarrassed?Yes / No


Yes / No

Comments

Is there anything else you would like to share with the research team such as impressions, comments and/or suggestions?

(Short text)

Contact

Please, enter your e-mail address: /If you would like to enter into a random draw of 10 Amazon vouchers of 100 AUD each, please, enter your e-mail. We will use your e-mail for the sole purpose of communication in case you are drawn to win one of the vouchers./

(Short text)

Survey 2 – Control group Eligibility to participate: Completed Survey 1 as participant in the Control group.

Anonymou

s identifier

You are invited to complete Survey 2 of the research project "Driving under the influence virtual experience" information about which can be found in the Participants Information Sheet at www (link to the file). If you would like to continue your participation in this project, please, enter the anonymous identifier you used in Survey 1 below in order to proceed to Survey 2. It shall include your: day of birth, first letter of name, first letter of family name and last two digits of mobile number (example 24DL08).


N Question Possible answers 1. Measuring standard and extended TPB constructs on the influence of

alcohol or drugs. All items are measured using 9-point scales (scored 1–9).








242






















2. Measuring reward and punishment sensitivity through Sensitivity to Punishment and Sensitivity to Reward Questionnaire (SPSRQ) (Torrubia et al., 2001).


Yes / No


Yes / No

243


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No

2.10 Do you often do things to be praised?

Yes / No


Yes / No


Yes / No


Yes / No


Yes / No

2.15 Are you easily discouraged in difficult situations? Yes / No 2.16 Do you need people to show their affection for

you all the time?Yes / No




Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No

244


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No

2.35 Do you generally try to avoid speaking in public? Yes / No2.36 When you start to play with a slot machine, is it

often difficult for you to stop?Yes / No


Yes / No




Yes / No


Yes / No


Yes / No


Yes / No


Yes / No


Yes / No

2.46 Would you like to be a socially powerful person? Yes / No2.47 Do you often refrain from doing something

because of your fear of being embarrassed?Yes / No


Yes / No

245

Comments

Is there anything else you would like to share with the research team such as impressions, comments and/or suggestions?

(Short text)

Contact

Please, enter your e-mail address: /If you would like to enter into a random draw of 10 Amazon vouchers of 100 AUD each, please, enter your e-mail. We will use your e-mail for the sole purpose of communication in case you are drawn to win one of the vouchers./

(Short text)

246

Appendix D

Additional Study 2 Effects of the intervention models

1. A subgroup of 18 highly engaged Intervention participants, ones that had a score

generated in more than half of the intervention period, were compared to all 126 entries

in the Control group.

A one-way ANCOVA test was performed to evaluate the effect of the

intervention of the DVs intention not to speed and past behaviour of not speeding

during the three months of the intervention, as described in Section 4.4.4. After

adjusting for the participants' self-reported intention not to speed before the

intervention, no significant difference between the Control group and the Intervention

group was found in intention not to speed, F (1, 141) = .90, p = .34, ηp2 = .006. There

was a statistically significant (p < .001) strong relationship between intention not to

speed at Time 1 and at Time 2, as indicated by a ηp2 = .416. After finding the non-

significant effect of the intervention between the two groups in respect to their

intention not to speed, two-way ANCOVAs found no significant effects with

personality characteristics as moderators of the result, either (see Table 10.1).

Table 10.1. Interaction effects between Condition and personality characteristics, intention not to speed, adjusted for Time 1 values (n=144).


Gender .185 .667 .001







future than the Control group participants.

After adjusting for the participants self-reported past behaviour of not speeding



247

months of the intervention, F (1, 141) = .55, p = .46, ηp2 = .004. There was a statistically






significant effects with personality characteristics as moderators of the result, either

(see Table 10.2). The assumption of equality of variance was not met when

investigating the interaction effect between the group condition and impulsivity.

Despite the created bias in the obtained result, given that there is no significant

interaction effect, no adjustments were necessary.



Gender .361 .549 .003




Sensitivity to reward < .001 .986 < .001






the intervention. No such effect was found on any of the other potentially-modifiable

Time 2 extended TPB variables, after adjusting for Time 1 values, either (see Table

10.3).






248



Instrumental attitude .001 .975 < .001


Subjective norm .001 .972 < .001




Moral norm .468 .495 .003

Peers' norm .153 .696 .001

Perceived risk 1.537 .217 .011

2. All 126 entries in the Control group were compared with all 84 entries in the

Intervention group.



during the three months of the intervention, as described in Section 4.4.4. After

adjusting for the participants' self-reported intention not to speed before the

intervention, no significant difference between the Control group and the Intervention

group was found in intention not to speed, F (1, 207) = 1.28, p = .26, ηp2 = .006. There

was a statistically significant (p < .001) strong relationship between intention not to

speed at Time 1 and at Time 2, as indicated by a ηp2 = .459. After finding the non-

significant effect of the intervention between the two groups in respect to their

intention not to speed, two-way ANCOVAs provided information about whether

personality characteristics moderated the result (see Table 10.4). The assumption of

equality of variance was not met when investigating the interaction effect between the

group condition and gender. Despite the created bias in the obtained result, given that

there is no significant interaction effect, no adjustments were necessary.

249

10.4. Interaction effects between Condition and personality characteristics, intention not to speed, adjusted for Time 1 values (n=210).


Gender .803 .371 .004




Sensitivity to reward 1.247 .265 .006

As shown in Table 10.4, a significant interaction effect was found between the

group condition and driving experience, after adjusting for the participants' intention

not to speed before the intervention. Investigating further, neither of the main effects

was statistically significant: condition (F (1, 205) = 1.09, p =. 30, ηp2 = .005) or driving

experience (F (1, 205) = .22, p = .64, ηp2 = .001). The lack of main effects suggested

that provisionally and openly licenced drivers behaved differently, depending on their

condition. A two-way ANCOVA, split by driving licence, showed a statistically

significant effect for the provisionally-licenced drivers (F (1, 107) = 4.32, p = .040,

ηp2 = .039) and a non-significant effect for the open-licenced drivers (F (1, 97) = .53,

p = .47, ηp2 = .005). Investigating the mean scores revealed that the provisionally-

licenced drivers in the Intervention group reported greater intention not to speed mean

score (6.42), i.e. lower intention to speed, than the provisionally-licenced drivers in

the Control group (5.68).

These results provided partial support for H.4, which predicted that participants

in the Intervention group would report significantly greater intention not to speed in

the future than the ones in the Control group, only in the case of provisionally-licenced

drivers.




months of the intervention, F (1, 207) = .05, p = .83, ηp2 < .001. There was a statistically




250



significant effects with personality characteristics as moderators of the result, either

(see Table 10.5).



Gender .245 .621 .001




Sensitivity to reward 1.858 .174 .009








10.6).





Subjective norm .043 .835 < .001

Descriptive norm .031 .860 < .001

Self-efficacy .481 .489 .002

Perceived controllability .085 .771 < .001

Moral norm .097 .756 < .001

Peers' norm .137 .712 .001

Perceived risk .499 .481 .002

251






3. A sub-sample of 31 participants from the Control group was selected to match as

close as possible the sample of 31 active Intervention participants.



during the three months of the intervention, as described in Section 4.4.4. Due to the

low number of participants considered in this sample, the moderating effects of gender,

driving experience, impulsivity, sensitivity to reward and sensitivity to punishment,

were not investigated as IVs.

After adjusting for the participants' self-reported intention not to speed before

the intervention, no significant difference between the Control group and the

Intervention group was found in intention not to speed, F (1, 59) = .44, p = .51, ηp2 =

.007. There was a statistically significant (p < .001) strong relationship between

intention not to speed at Time 1 and at Time 2, as indicated by a ηp2 = .301.



future than the ones in the Control group.




months of the intervention, F (1, 59) < .01, p = .96, ηp2 < .001. There was a statistically



the intervention, as indicated by a ηp2 = .464.



the three months of the intervention than the participants in the Control group.

252





10.7).




Affective attitude .236 .610 .004

Subjective norm .301 .585 .005




Moral norm .986 .325 .016

Peers' norm .048 .828 .001

Perceived risk .014 .906 <.001






Documents

SMARTPHONE APPS AND VIRTUAL REALITY AS ROAD SAFETY I … · smartphone safe-driving app as an intervention tool. The off-the-shelf safe-driving app "Flo - driving insights" (Flo)