IMPROVING COST ESTIMATION
PERFORMANCE: AN INVESTIGATION OF
PREDICTION TECHNIQUE AND PERSON-
ENVIRONMENT INTERACTION
Bo Xiong
BEng, MEng
Submitted in fulfilment of the requirements for the degree of
Doctor of Philosophy
School of Civil Engineering and Built Environment
Faculty of Science and Engineering
Queensland University of Technology
2016
Improving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment
Interaction i
Abstract
Job performance of construction cost engineers is critical to the successful
operation of projects and organisations. From a review of previous studies, two
approaches — exploring efficient estimation techniques and examining person-
environment interactions — are found to be valuable for improving cost estimation.
Following this logic, this thesis by publications makes contributions to both
objectives.
A hybrid approach based on Akaike information criterion (AIC) and principal
component regression (PCR) is firstly proposed to solve overfitting and collinearity
problems which are common in cost estimation. Although there have been many
studies of estimation and model fitting in construction, few have focused on
addressing the overfitting and collinearity problems that frequently occur in
developing predictive models. In an application of estimating the cost of construction
project preliminaries, the AIC-PCR approach is demonstrated in comparison with
alternative regression models and three data mining techniques of artificial neural
networks, case based reasoning and support vector machines. In addition to reducing
the risks of overfitting and collinearity, experimental results show that the AIC-PCR
approach presents a good predictive accuracy. In addition to the new approach,
effects of early cost drivers on determination of contingencies are also examined.
Besides of skills on technical tools, Job performance is influenced by
interactions between people and their environment. To identify such factors, a
literature review is firstly conducted. Building on the person-environment (P-E) fit
theory and the stimulus-organism-response (S-O-R) paradigm, a conceptual model to
understand the job performance of construction professionals is developed.
Psychological reactions such as work stress and job satisfaction need to be specially
emphasised for their mediating role in linking environmental factors and individual
differences on job performance. Structural equation modelling (SEM) is used as the
principal statistical method to explore interactions between people and their
environment. However, previous applications of SEM in construction management
area are not very satisfactory. Therefore, this research critically reviews extent SEM
iiImproving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment Interaction
applications for solving problems related to construction management. Some
common drawbacks (such as comparatively small sample size, questionable
construct validity, and low GOF) of these applications are pointed out, with
suggestions for improvement.
Work stress is explicitly examined, since cost engineers face a high level of
uncertainty and much responsibility. The perceived stress questionnaire (PSQ) is
used to measure cost professionals’ work stress. Principal component analysis and
confirmatory factor analysis are utilised to test the dimensions of occupational stress;
this area has mostly been overlooked in previous research on stress in the
construction context. Analysis of the results identifies one stressor — demand — and
three secondary emotional reactions to the work situation — worry, tension and lack
of joy.
Job satisfaction, another indicator of person-environment fit, is also important
to employee performance. Recent decades have seen an increasing number of
theoretical explorations and empirical demonstrations of the nexus between job
satisfaction and job performance. Some argue that “happier workers produce more”,
while others insist that workers with better performance achieve satisfaction through
greater rewards. This study conducts a fine-grained analysis to propose a new
conceptual model on satisfaction-performance (S-P) nexus. Job satisfaction is
subdivided into economic satisfaction and noneconomic satisfaction. This
assumption is validated in this study by a principal component analysis of empirical
evidence from a questionnaire survey of construction cost engineers. Additionally,
this model is tested with construction project participants, using survey data from
construction companies about their most recent project experiences.
Improving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment
Interaction iii
Keywords
Akaike Information Criterion
Collinearity diagnosis
Construction cost estimation
Cost drivers
Cost engineers
Job performance,
Job satisfaction
Multiple linear regression
Overfitting diagnosis
Person-environment fit
Principal component analysis
Principal component regression
Stimulus-organism-response
Stimulus-reaction-performance
Psychological reaction
Structural equation modelling
Work stress
ivImproving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment Interaction
List of Publications by Candidate
Journal papers published
Bo Xiong*, Martin Skitmore, Bo Xia. A critical review of structural equation
modelling applications in construction research. Automation in Construction,
49, 59-70, 2015. (ERA:A) (Chapter 3, Section 3.2)
Bo Xiong*, Martin Skitmore, Bo Xia. Exploring and validating the internal
dimensions of occupational stress: Evidence from construction cost
estimators in China. Construction Management and Economics, 33(5-6), 495-
507, 2015. (ERA:A) (Chapter 4)
Bo Xiong*, Martin Skitmore, Bo Xia, Md Asrul Masrom, Kunhui Ye, Adrian
Bridge. Examining the influence of participant performance factors on
contractor satisfaction: A structural equation model. International Journal of
Project Management. 32(3), 482-491, 2014. (ERA: A) (Chapter 5, Section
5.2)
Bo Xiong*, Weisheng Lu, Martin Skitmore, K.W. Chau, & Meng Ye,
Virtuous nexus between corporate social performance and financial
performance: A study of construction enterprises in China, Journal of Cleaner
Production, 129, 223-233, 2016. (ERA: A)
Bo Xia, Bo Xiong, Martin Skitmore, Peng Wu, Fang Hu, Investigating the
impact of project definition clarity on project performance: a Structural
Equation Modelling (SEM) Study, Journal of Management in Engineering,
32(1), 04015022,2016. (ERA: A*)
Xiaolong Gan, Jian Zuo, Kunhui Ye, Martin Skitmore, Bo Xiong, Why
sustainable construction? Why not? An owner's perspective, Habitat
International, 47, 61-68, 2015. (ERA:A)
Brendon Lim, Madhav Nepal, Martin Skitmore, Bo Xiong, Drivers of the
accuracy of developers’ early stage cost estimates in residential construction,
Journal of Financial Management of Property and Construction, 21(1), 4-20.
(ERA:C)
Improving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment
Interaction v
Journal manuscripts
Bo Xiong*, Martin Skitmore, Bo Xia, Sidney Newton, A hybrid approach for
reducing overfitting and collinearity: an application in construction cost
estimation. Submitted to Journal of Civil Engineering and Management
(Chapter 2, Section 2.1)
Bo Xiong*, Martin Skitmore, Md Asrul Masrom, Bo Xia, A fine-grained
analysis of contractor satisfaction in promoting project management
performance. Submitted to Project Management Journal (Chapter 5, Section
5.3)
International conference papers
Bo Xiong (Oral presenter), Exploring dimensions of job satisfaction and
relationships with performance: evidences from construction professionals,
CIB World Building Congress 2016, Tampere, Finland, May 30 – June 3,
2016. (Chapter 5, Section 5.1)
Bo Xiong (Oral presenter), Bo Xia, Examining the impacts of early cost
drivers on contingencies with path analyses, 2014 ASCE Construction
Research Congress, Atlanta, USA, May 19-21, 2014, pp. 1518-1527.
(Chapter 2, Section 2.2)
Bo Xiong (Oral presenter), Martin Skitmore, Bo Xia, Exploring the internal
dimensions of work stress: Evidence from construction cost estimators, 2014
30th ARCOM, UK, Sep 1-3, 2014, pp 321-329.
Bo Xiong (Oral presenter), Will lean construction be paid off: lessons learnt
from BREEAM buildings, 2015 IGLC Summer School, Perth, Australia,
August 1-2 2015, pp 51-55.
Bo Xiong (Oral presenter), The role of person-environment fit in promoting
job performance: towards a conceptual model and a research agenda, EPPM
2015, Gold Coast, Australia, September 1-3, 2015, pp 476-484.
viImproving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment Interaction
Table of Contents
Abstract .................................................................................................................................................... i
Keywords .............................................................................................................................................. iii
List of Publications by Candidate .......................................................................................................... iv
Table of Contents ................................................................................................................................... vi
List of Figures ..................................................................................................................................... viii
List of Tables ......................................................................................................................................... ix
List of Abbreviations ............................................................................................................................. xi
Acknowledgements ............................................................................................................................... xv
Statement of Original Authorship ....................................................................................................... xvii
CHAPTER 1: INTRODUCTION ....................................................................................................... 1
1.1 Background .................................................................................................................................. 1 1.1.1 Improving construction estimation by technique innovation ............................................ 1 1.1.2 Improving performance by considering person-environment interactions ....................... 2
1.2 LITERATURE REVIEW ............................................................................................................ 3 1.2.1 Literature review on driving factors and techniques of cost estimation ........................... 3 1.2.2 Literature review on job performance .............................................................................. 9
1.3 RESEARCH QUESTIONS AND OBJECTIVES ..................................................................... 13
1.4 Thesis Outline ............................................................................................................................ 14 1.4.1 Chapter 2: Construction cost estimation techniques ....................................................... 14 1.4.2 Chapter 3: Conceptual framework and structural equation modelling ........................... 16 1.4.3 Chapter 4: Work stress ................................................................................................... 17 1.4.4 Chapter 5: Job satisfaction .............................................................................................. 18 1.4.5 Chapter 6: Conclusions ................................................................................................... 20
CHAPTER 2: CONSTRUCTION COST ESTIMATION TECHNIQUES ................................... 21
2.1 A new cost estimation approach ................................................................................................ 21 Statement of contribution ........................................................................................................... 21 2.1.1 Introduction .................................................................................................................... 23 2.1.2 Literature review ............................................................................................................ 25 2.1.3 A hybrid approach .......................................................................................................... 27 2.1.4 Application in construction cost estimation .................................................................... 31 2.1.5 Conclusions .................................................................................................................... 36
2.2 Impacts of early cost drivers ...................................................................................................... 38 Statement of contribution ........................................................................................................... 38 2.2.1 Introduction .................................................................................................................... 40 2.2.2 Early cost drivers ............................................................................................................ 41 2.2.3 Research method ............................................................................................................ 42 2.2.4 Path analysis modelling .................................................................................................. 45 2.2.5 Findings and discussions ................................................................................................ 47 2.2.6 Conclusion ...................................................................................................................... 48
CHAPTER 3: CONCEPTUAL FRAMEWORK AND STRUCTURAL EQUATION
MODELLING 51
3.1 Towards a conceptual framework of job performance .............................................................. 51 3.1.1 Introduction .................................................................................................................... 51 3.1.2 Development of the conceptual framework .................................................................... 52 3.1.3 Discussion ...................................................................................................................... 57 3.1.4 Conclusions .................................................................................................................... 59
Improving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment
Interaction vii
3.2 Structural equation modelling .................................................................................................... 60 Statement of contribution ........................................................................................................... 60 3.2.1 Introduction .................................................................................................................... 62 3.2.2 Methodology ................................................................................................................... 63 3.2.3 Critical issues in the application of SEM ........................................................................ 69 3.2.4 Discussion and recommendations ................................................................................... 83 3.2.5 Conclusions .................................................................................................................... 85
CHAPTER 4: WORK STRESS ........................................................................................................ 87
Statement of contribution ...................................................................................................................... 87
4.1 Introduction ................................................................................................................................ 89
4.2 Literature review ........................................................................................................................ 90 4.2.1 Occupational stress and its effects .................................................................................. 90 4.2.2 Stressors and coping strategies ....................................................................................... 93 4.2.3 Measures of occupation stress and divisibility ............................................................... 94
4.3 Research method ........................................................................................................................ 95 4.3.1 Perceived stress questionnaire ........................................................................................ 96 4.3.2 Translation and back translation ..................................................................................... 97 4.3.3 Data collection and demographics .................................................................................. 98 4.3.4 Data reliability ................................................................................................................ 99
4.4 Data analysis and discussion .................................................................................................... 100 4.4.1 Principal component analysis ....................................................................................... 100 4.4.2 Discussion-PCA results ................................................................................................ 101 4.4.3 Validation with SEM .................................................................................................... 103 4.4.4 Validation with SEM Discussion of the CFA and SEM results .................................... 106
4.5 Conclusion ............................................................................................................................... 107
CHAPTER 5: JOB SATISFACTION ............................................................................................. 109
5.1 The nexus between job satisfaction and job performance of construction cost engineers ....... 109 5.1.1 Introduction .................................................................................................................. 109 5.1.2 Literature review ........................................................................................................... 110 5.1.3 Research method ........................................................................................................... 113 5.1.4 Results .......................................................................................................................... 114 5.1.5 Discussion and conclusions .......................................................................................... 116
5.2 Examining the influence of participant performance factors on contractor satisfaction: A
structural equation model .................................................................................................................... 118 Statement of contribution ......................................................................................................... 118 5.2.1 Introduction .................................................................................................................. 120 5.2.2 Introduction .................................................................................................................. 121 5.2.3 Research method ........................................................................................................... 122 5.2.4 Results .......................................................................................................................... 130 5.2.5 Findings and discussion ................................................................................................ 134 5.2.6 Conclusions .................................................................................................................. 137
5.3 The nexus between contractor satisfaction and project Management performance ................. 140 Statement of contribution ......................................................................................................... 140 5.3.1 Introduction .................................................................................................................. 142 5.3.2 Theoretical background and hypotheses development ................................................. 143 5.3.3 Methodology ................................................................................................................. 145 5.3.4 Results .......................................................................................................................... 152 5.3.5 Discussion ..................................................................................................................... 159 5.3.6 Conclusions .................................................................................................................. 160
CHAPTER 6: CONCLUSIONS ...................................................................................................... 163
6.1 Summary and discussion .......................................................................................................... 163
6.2 Limitations and recommendations ........................................................................................... 165
BIBLIOGRAPHY ............................................................................................................................. 169
viiiImproving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment Interaction
List of Figures
Figure 1.1 Cost estimate accuracy range change curves with project progress ....................................... 4
Figure 1.2 Number of related articles by journal .................................................................................. 11
Figure 2.1 Collinearity diagnostics for Model 3 ................................................................................... 35
Figure 2.2 The initial model .................................................................................................................. 42
Figure 2.3 The next to last path analysis model .................................................................................... 45
Figure 2.4 Final path analysis model .................................................................................................... 46
Figure 3.1 Main P-E fits and psychological reactions ........................................................................... 54
Figure 3.2 Proposed conceptual framework .......................................................................................... 55
Figure 3.3 Schematic diagram of a structural equation model .............................................................. 64
Figure 3.4 Article selection ................................................................................................................... 67
Figure 3.5 Number of SEM-based articles by journals and year ........................................................... 69
Figure 4.1 Effects of dimensions of stress on organizational commitment ......................................... 106
Figure 5.1 Main conceptual models of the S-P nexus ......................................................................... 112
Figure 5.2 Proposed conceptual model for this study ......................................................................... 113
Figure 5.3 Model evaluations by regression analysis .......................................................................... 115
Figure 5.4 Structural component ......................................................................................................... 126
Figure 5.5 Measurement component ................................................................................................... 131
Figure 5.6 Final SEM model results ................................................................................................... 133
Figure 5.7 Conceptual model 1 ........................................................................................................... 147
Figure 5.8 Conceptual model 2 ........................................................................................................... 148
Figure 5.9 Model 1A testing H2A: COS causes CPMP ...................................................................... 154
Figure 5.10 Model 1B testing H2B: CPMP causes COS .................................................................... 154
Figure 5.11 Model 2 ............................................................................................................................ 157
Improving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment
Interaction ix
List of Tables
Table 2.1 Sample descriptions - part 1 .................................................................................................. 32
Table 2.2 Sample descriptions - part 2 .................................................................................................. 33
Table 2.3 Elemental cost items framework ........................................................................................... 33
Table 2.4 Developed regression models ................................................................................................ 34
Table 2.5 Comparison of results for Application 1, price estimating .................................................... 36
Table 2.6 Description of variables ........................................................................................................ 44
Table 2.7 Model fit indices ................................................................................................................... 46
Table 2.8 Standardized direct, indirect and total effects of variables .................................................... 47
Table 3.1 Issues related to research design ........................................................................................... 75
Table 3.2 Issues related to model development ..................................................................................... 78
Table 3.3 GOF evaluation criteria and practical results ........................................................................ 81
Table 3.4 Description of reported GOF indices .................................................................................... 82
Table 3.5 Recommendations for selected issues in SEM ...................................................................... 84
Table 4.1 Perceived stress questionnaire ............................................................................................... 96
Table 4.2 Translation and back translations .......................................................................................... 97
Table 4.3 PCA with varimax rotation ................................................................................................. 100
Table 4.4 Standardized regression weights ......................................................................................... 104
Table 4.5 Goodness of fit .................................................................................................................... 105
Table 5.1 Measures of Job satisfaction ............................................................................................... 113
Table 5.2 Principal component analysis with varimax rotation .......................................................... 114
Table 5.3 Correlations between factors ............................................................................................... 115
Table 5.4 forms of effects ................................................................................................................... 116
Table 5.5 Constructs and measurement of SEM ................................................................................. 128
Table 5.6 Details of respondents ......................................................................................................... 129
Table 5.7 Reliability test of the questionnaire responses .................................................................... 130
Table 5.8 Standardized regression weights and SMCs ....................................................................... 131
Table 5.9 Results of goodness of fit (Adapted from Ong and Musa (2012)) ...................................... 132
Table 5.10 P values and indirect effects (Sobel test) ........................................................................... 134
Table 5.11 Description of projects ...................................................................................................... 148
Table 5.12 Measurement constructs and items .................................................................................... 149
Table 5.13 Reliability test ................................................................................................................... 150
Table 5.14 Validity test results ............................................................................................................ 153
Table 5.15 Results of hypothesis tests ................................................................................................. 155
Table 5.16 Standardized regression weights and SMCs ...................................................................... 155
Table 5.17 Goodness of fit .................................................................................................................. 157
Table 5.18 Hypothesis direct effects ................................................................................................... 157
xImproving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment Interaction
Table 5.19 Standardized direct/indirect/total effects ........................................................................... 158
Improving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment
Interaction xi
List of Abbreviations
AACE: American Association of Cost Engineering
ADF: Asymptotically distribution-free
AGFI: Adjusted goodness-of-fit index
AHP: Analytic hierarchy process
ANN: Artificial neural networks
AVE: Average variance extracted
AUTCON: Automation in Construction
AIC: Akaike information criterion
ASCE: American Society of Civil Engineers
B&E: Building and Environment
BCIS: Building Cost Information Service
BREEAM: Building Research Establishment Environmental Assessment
Methodology
CBR: Case based reasoning
CFA: Confirmatory factor analysis
CFI: Comparative fit index
CI: Condition index
CIOB: Chartered Institute of Building
CME: Construction Management and Economics
CNY: Chinese yuan
CR: Composite reliability
CV-SEM: Covariance-based structural equation modelling
CVP: Coefficient variance proportion
CWB: counterproductive work behaviour
xiiImproving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment Interaction
D-A: Demands-abilities
ECAM: Engineering, Construction and Architectural Management
ES: Economic satisfaction
EW: Equal weights
GA: Genetic algorithm
GDM: Gradient descent method
GFI: Goodness-of-fit index
GLS: Generalized least square
GFA: Gross floor area
GOF: Goodness of fit
IFI: Incremental fit index
IJPM: International Journal of Project Management
JME: Journal of Management in Engineering
JCEM: Journal of Construction Engineering and Management
KNN: K-nearest neighbour
LEED: Leadership in Energy and Environmental Design
LOOCV: Leave one out cross validation
LV: Latent variable
MAE: Mean absolute error
MAER: Mean absolute error rate
MAPE: Mean absolute percentage error
ML: Maximum likelihood
MLR: Multiple linear regression
MSE: Mean squared error
MV: Manifest variable
N-S: Needs-supplies
Improving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment
Interaction xiii
OCB: Organizational citizen behaviour
PA: Path analysis
PCA: Principal component analysis
PCFI: parsimony comparative fit index
PCR: Principal component regression
PSQ: perceived stress questionnaire
P-E: Person-environment
PLS: Partial least squares
PLS-SEM: partial least squares path modelling
PNFI: Parsimony normed-fit index
PS: Production-related/ noneconomic satisfaction
RICS: Royal Institution of Chartered Surveyors
RMR: Root mean square residual
RMSE: Root mean squared error
RMSEA: Root mean square error of approximation
RR: Ridge regression
RSS: Residual sum of squares
SEM: Structural equation modelling
SMC: Squared multiple correlation
S-O-R Stimulus-organism-response
S-P: Satisfaction-performance
SRMR: Standardized root mean square residual
S-R-P: Stimulus-reaction-performance
SSE: Sum of squares error
SVM: Supportive vector machine
SVR: Supportive vector regression
xivImproving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment Interaction
TLI: Tucker-Lewis Index
TP: Task performance
TSS: Total sum of squares
ULS: unweighted least squares
UK: United Kingdom
US: United States of America
VIF: Variance inflation factor
WHI: Work–home Interference
Improving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment
Interaction xv
Acknowledgements
This thesis would not be possible without the help of many people. Firstly, I
would like to thank my supervisor Professor Martin Skitmore, who has been a
fabulous mentor and supporter of my research and career development. I would like
to thank Martin for planning reasonable goals and encouraging me to complete tasks
in a timely manner. Thanks to Martin’s example, I understand that hard work, time
management and integrity are important for scholarly success.
I would like to thank my associate supervisor Dr Bo Xia, research project
leader Associate Professor Sidney Newton and external supervisor Dr Pablo
Ballesteros-Pérez for providing invaluable advice on research and career
development whenever I was in need of their help. In particular, I would like to thank
Bo for helping me to refine my research topic, guiding me in the transition to post-
doctoral studies and giving me the opportunity to assist with several courses.
I am thankful to Professor Kunhui Ye, the supervisor of my master by research
program, for providing advice on many issues such as research planning,
questionnaire distribution, grant applications and my academic career in general.
My gratitude extends to thesis committee members for encouraging my study
and giving constructive feedback in key completion stages of this thesis: Professor
Stephen Kajewski, Professor Hannes Zacher, Professor Simon Washington and
Professor Laurie Buys. I would also like to thank Dr Le Chen for always sharing her
opinions on academic development.
I wish to acknowledge valuable support from academics in the discipline of
Construction and Project Management. Special thanks go to Professor Jay Yang,
Associate Professor Adrian Bridge, Associate Professor Karen Manley, Dr Madhav
Nepal, Dr Fiona Lamari, Dr Melissa Teo and Dr Carol Hon. I would also like to
thank my senior Dr Md Asrul Masrom for letting me access his data, and Dr Mei Li,
Dr Yulin Liu, Miss Hao Zhang and Mr Xin Hu for their help with questionnaire
development and discussions.
I would like to thank Professor Abdol R. Chini for hosting and directing my
short-term visiting research in the M.E. Rinker, Sr. School of Construction
xviImproving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment Interaction
Management at the University of Florida. I thank Professor Charles J. Kibert for
sharing the inspiring conference proceeding and providing me a workplace in the
Powell Centre for Construction and Environment.
I would like to thank Professor Jack Goulding for inviting me to engage in
short-term visiting research in the Centre for Sustainable Development at the
University of Central Lancashire, UK, for supervising my visiting research project
and providing advice on academic life. There, suggestions from Professor Akintola
Akintoye and Dr Farzad Pour Rahimian were inspiring and sincerely appreciated.
I would like to thank Professor Andrew Baldwin for inviting me to conduct
short-term visiting research in the School of Civil and Building Engineering at
Loughborough University, UK; I also thank Dr Mohammed Osmani for admirably
constructed supervision and many suggestions for career development.
I would like to thank Professor K W Chau for inviting me to conduct short-
term visiting research in the Department of Real Estate and Construction at the
University of Hong Kong; I also thank Associate Professor Wilson Lu for his kind
directions, inspiring revisions and timely encouragements.
Besides the academic advisors mentioned, I made many friends in visited
universities and when I attended conferences. I would like to express my special
thanks to them as well as all my friends in Brisbane for their fellowship and support.
I am grateful to the School of Civil Engineering and Built Environment for
providing wonderful facilities and services. I would also like to thank Professor Paul
Burnett and Ms Linda Clay for inviting me to serve as a member of QUT Research
Student Center User Advisory Group in 2015, which turned out to be a precious
experience. Supports from QUT SEF HDR team are also sincerely appreciated. I
thank Dr Christina Houen for editing substantial parts of my thesis to the standards
and guidelines of the Institute of Professional Editors (IPEd).
Lastly, and most importantly, I wish to thank my family for their support and
encouragement. My parents have always given me selfless love, generous support
and timely encouragement throughout my life. They are always happy to hear me tell
stories of my experiences. And I need to thank them in Chinese now:
感谢父母、亲友们一直的关爱与支持,让我心无旁骛的完成了博士阶段的学习。我一定
继续努力,不辜负你们的期望!
Improving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment
Interaction xvii
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: _______1/07/2016_________
QUT Verified Signature
xviiiImproving Cost Estimation Performance: An Investigation of Prediction Technique and Person-Environment Interaction
Chapter 1: Introduction 1
Chapter 1: Introduction
1.1 BACKGROUND
It is common to find that the final costs of projects greatly exceed estimates
(Williams, Lakshminarayanan, & Sackrowitz, 2005). An analysis of 258
transportation infrastructure projects worth US$90 billion reveal that nine out of ten
cost overruns are caused by inaccurate estimation in the early stages (Flyvbjerg,
Bruzelius, & Rothengatter, 2003). Similarly, 74% of cost growth in projects
undertaken by chemical, oil, and mineral industries in North America are caused by
underestimation in the early stages (Merrow, Chapel, & Worthing, 1979). Therefore,
accurate cost estimation is critical to project success (Lowe, Emsley, & Harding,
2006; Skitmore, Stradling, Tuohy, & Mkwezalamba, 1990).
For organizations like government authorities or real estate developers,
inaccurate early estimates will cause low efficiency in the use of money, missed
development opportunities and unsuccessful project management (Oberlender &
Trost, 2001). The difficult task of cost estimation is always assumed by construction
cost engineers (quantity surveyors in the UK). Quantity surveyors advise clients on
the likely tender price of proposed projects, and assist in setting a budget accordingly
(Lowe, et al., 2006). In addition to estimation techniques, the estimation performance
of these professionals is affected by psychological reactions like work stress (Leung,
Zhang, & Skitmore, 2008). Therefore, this thesis by thesis explores both aspects of
this challenging task.
1.1.1 Improving construction estimation by technique innovation
Cost estimation in the early stages of construction, with limited information
and unclear scope definition, is a complicated and stressful work. Estimators are
involved in many subjective decisions because of the complexity and uncertainty of
construction work. Skitmore (1985) attributes the ability to make good subjective
judgements to estimating expertise. Through a pioneering series of experiments to
measure the early stage estimating abilities of quantity surveyors, Skitmore (1985)
confirms the role of expertise in achieving accuracy, and finds that experts are more
2 Chapter 1: Introduction
relaxed and confident. This expertise is further linked with experience, that is, an
experienced quantity surveyor can give a more accurate estimate (Skitmore et al.,
1990). However, becoming an expert by increasing one’s experience consumes a lot
of resources and time. Millions of dollars may be at risk through poor and inefficient
estimation practices, and bigger mistakes may occur during an estimator’s early
years. Additionally, distrust and job ambiguity can exacerbate estimators’ stress
(Leung, Ng, Skitmore, & Cheung, 2005). This dilemma calls for understanding how
to build cost drivers in the early stages, and developing some inexpensive, quick and
reasonably accurate estimating techniques; these objectives motivate this research.
Analysing historical data with statistical methods can help to improve
prediction accuracy. This is consistent with Skitmore’s (1985) finding that experts
are able to recall the estimating details of previous projects and are good at adjusting
to new requirements. Some techniques like activity based cost estimates and
estimates of elements such as floor area are criticised for unsuitability or inaccuracy
in the early stages. To date, three methods usually recommended for forecasting
early estimates are multiple linear regression (MLR), artificial neural networks
(ANN) and case based reasoning (CBR) (Kim, An, & Kang, 2004). Although MLR
is a comparative way that is widely used, practitioners face overfitting and
collinearity problems in modelling. ANN needs much training time for each use.
Given the comparative inaccuracy of MLR and stiffness of ANN to add new case,
CBR is more flexible in adding new cases for continuous improvement. However,
current applications of CBR in early estimates (An, Kim, & Kang, 2007; Kim, et al.,
2004; Kim & Kim, 2010; Kim, Choi, Kim, & Kang, 2005) are defective in
determining similarity weights and adapting the model to new cases.
A hybrid approach based on Akaike information criterion and principal
component regression is proposed to address these problems and validated in this
thesis. In addition, this approach can help to improve the general MLR method (see
Chapter 2). Effects of cost drivers are discussed in Chapter 1 and Chapter 2.
1.1.2 Improving performance by considering person-environment interactions
Cost engineers are important professionals in construction who work mainly in
offices. Therefore, their job performance, including task performance and
Chapter 1: Introduction 3
organisational citizen behaviour (OCB), are influenced by environmental factors
such as organisational support and organisational politics, and psychological
reactions to job satisfaction, work stress and so on.
However, the interaction of these factors in the work performance of cost
construction professionals has been little explored. Leung, Zhang et al. (2008)
examined the effects of organisational support on stress via mediation of some
stressors (such as unfair rewards) among cost engineers in Hong Kong. Cost
estimation is an experience based task (Skitmore, 1985); experiential learning is very
important in improving cost estimating abilities (Lowe & Skitmore, 1994).
Therefore, the organisational learning climate is critical to the job performance of
cost engineers. This link was not found to be significant in the study of Lowe and
Skitmore (2007), which may be attributed to the researchers’ focus on specific
measurement of task performance and not taking mediating variables into account.
Many studies (such as Egan, Yang, and Bartlett, 2004) acknowledge that a better
learning climate would enhance the transfer of knowledge in organisations. On the
other hand, Bergeron, Shipp, Rosen, and Furst (2011) point out that the task
performance of the individual may not necessarily improve significantly by spending
extra time on OCB. However, organisational performance should benefit from
organisational citizen behaviours (Mikkelsen & Grønhaug, 1999). Bearing these
challenges in mind, it is worth investigating whether job performance can be
improved through further exploration of person-environment interactions.
Therefore, a thorough literature review of studies of driving factors that are key
to job performance published in managerial and psychological journals is presented,
and a comprehensive framework is proposed in Chapter 3, based on the theory of
Person-Environment (P-E) fit. Psychological reactions including work stress and job
satisfaction are explored in Chapters 4 and Chapter 5 respectively.
1.2 LITERATURE REVIEW
1.2.1 Literature review on driving factors and techniques of cost estimation
Influencing cost drivers
Estimation in the early stage is mostly inaccurate, yielding limited and vague
information. In this study, early stage refers to the pre-design stage, including a
4 Chapter 1: Introduction
feasibility study. Barnes (1974) proposes the accuracy of estimation at this stage is
around -40% to +20%, as shown in Figure 1.1 (Barnes, 1974; Skitmore, 1987b). It is
reported that inaccuracy of these estimates is around 30% in Germany, and this
inaccuracy is mainly caused by simply multiplying the floor area with an indicator,
which is inaccurately measured in certain projects because of uncertain cost drivers
(Stoy & Schalcher, 2007). These findings are consistent with AACE International’s
general cost estimate classification system across stages including concept screening,
feasibility, budget authorization, control and bid/tender (Christensen & Dysert,
2005). For example, the expected accuracy ranges for feasibility are -15% to -30%
(low) and +20% to +50% (high). Skitmore (1987b) states that during the pre-tender
stage, the accuracy of forecasting can improve as the design progresses for more
gradual release of information. Despite the risks of inaccuracy, an inexpensive,
quick, and comparatively accurate pre-design estimation is important for its effects
on decision making and feasibility studies (Li, Shen, & Love, 2005). To achieve this,
a review of previous studies on building cost relevant drivers at the early stage
should be conducted first, and the impacts of these drivers on building cost formation
should also be examined, but this is rarely considered in previous research.
Figure 1.1 Cost estimate accuracy range change curves with project progress
Skitmore (1987a) proposes that building prices can be seen as a result of a
series of interdependent causal mechanisms, and emphasises the market effect on the
formation of building prices. Primary cost drivers include building type, size,
complexity and quality, type of client, contractor selection, contractual arrangements,
location, and the economic and legal environment of the project location. Skitmore
Chapter 1: Introduction 5
and Ng (2003) pinpointed the effect of the contractor period on construction costs by
establishing a time-cost simultaneous model with details of 93 Australian projects.
They found that the errors in predicted actual construction costs become smaller as
the contract period increases. Similarly, Gunner and Skitmore (1999) found that three
variables, floor area, number of storeys above ground and contract period, have
comparatively high correlations with contract sums after conducting correlation
analysis for Singapore data.
Li et al. (2005) constructed two early estimate regression models for reinforced
concrete(RC) office buildings and steel office buildings in Hong Kong, China. They
selected seven variables: average floor area, total floor area, average storey height,
number of above-ground storeys, total building height, number of basements and
completion year, and found that total floor area, total building height, and average
floor area are important ones for modelling. While total building height is hard to
know from the beginning, the height of a floor is associated with design
specifications and some situations such as with or without air conditioning systems
(Stoy & Schalcher, 2007). An inherent limitation with this study is that the total
sample size and the two sub-sets used to develop the two models are small (37, 12, 7
respectively) and their representativeness of office buildings in Hong Kong is not
clear.
Elhag, Boussabaine, and Ballal (2005) conducted a questionnaire survey to
separate 67 factors into six categories. Although six ranking lists by importance are
proposed, impacts of these variables are still unknown and most factors are not
available in the pre-design stage. But it is interesting to find that complexity of
building services ranks as the top two in the project characteristics category. This is
consistent with the fact that building projects have become more complex nowadays,
and complexity is a critical characteristic to project success (Xia & Chan, 2012).
From the contractors’ perspective, a tender price is the sum of total costs
(including direct and indirect costs) and mark up (expected profit) (Runeson &
Skitmore, 1999; Yuan, 2011). Factors that influence a contractor’s mark-up
percentage should be mentioned here. Li, Shen, and Love (1999) conducted a study
to predict contractors’ mark-up percentage based on ten independent variables,
including project size, location, market conditions, number of competitors, project
type, project complexity and four other contractor-specific characteristics. Similarly,
6 Chapter 1: Introduction
Liu and Ling (2003) used three variables, market condition, project complexity and
project size, to estimate mark-up percentage. Tendering theory has been developed
for more than 50 years since the first proposal by Lawrence Friedman in 1956 and
many versions have evolved from that (Runeson & Skitmore, 1999).
Building costs can also represented by contract price/winner’s bid price, since
it is the cost for a client of completing a project. The actual contractor cost is always
ambiguous in the market place. Although variation in cost from a good detailed
estimate should be small (within 5%), Park and Chapin (1992) found that the actual
cost can vary almost ±20% from the estimated costs (Park & Chapin, 1992).
Therefore, using contractors’ expected cost is possibly unreliable because of its
inaccuracy.
In summary, understanding these building cost drivers will help clients and
quantity surveyors to avoid vagueness in scope, which is a major cause of cost
overrun (Akinci & Fischer, 1998). Soetanto and Proverbs’ (2002) study in the UK
indicates that contractor satisfaction increases with the perception that clients do not
know what they want. On the other hand, Xiong et al. (2014) found that this
conclusion is not applicable to Malaysian cases, where contractor satisfaction
increases with the client’s clarity of objectives. Despite the differences between these
findings, the importance of clear scope should be emphasised for achieving project
success (Xia, Xiong, Skitmore, Wu, & Hu, 2015). The above literature review seeks
to support this objective.
Building cost modelling techniques
The term “building cost modelling” was formally mentioned in the Building
Cost Research Conference held in 1982 (Newton, 1991). Cost models are the basis
for cost forecasting, and understanding their properties is vital to effective control
and development of future techniques (Skitmore & Marston, 1999a). Newton (1991)
proposed an agenda for this area of research, and classified 56 relevant published
works from 1960-1988 by nine dimensions. In terms of technique, the usage of
expert systems and networking was less than 5% (Newton, 1991). However, case-
based reasoning (CBR) as an expert system and artificial neural networks (ANN)
have improved greatly in efficiency with the aid of developing computer techniques
in the last two decades (Chou & Tseng, 2011).
Chapter 1: Introduction 7
Traditional estimation methods at early stage include element based floor area
models, probabilistic models and regression models (Raftery, 1987). The problem of
the first method is that the floor area is not the only factor affecting cost. According
to Stoy and Schalcher (2007), inaccuracy of these estimates by simply multiplying
the floor area by a certain indicator is around 30% in Germany. Acceptability of the
probabilistic model for cost estimating, usually in the form of a Monte Carlo
simulation, is questionable (Chau, 1997; Fellows, 1996; Li, et al., 2005).
The multiple linear regression (MLR) method has been regarded as a powerful
tool in early estimates for many years (Li et al., 2005; Skitmore & Patchell, 1990). Li
et al (2005) established regression cost equations by seven basic variables in early
stages for two types of office buildings in Hong Kong. But this research is limited by
the use of small samples. Aiming at optimal predictive ability with the current
sample, regression has a principle of parsimony (step-wise regression). This
character reduces the possibility of inputting a large number of predictors, and
reduces regression’s ability to explain changes (Kim, et al., 2004).
The artificial neural networks model (ANN) simulates the learning process of
the human brain by forming thousands of simulated neurons, and is widely used for
its predicting ability in many fields (Kim et al., 2004; Kim et al., 2005). Kim et al.
(2004) point out that many previous researchers have proved that the accuracy of
applying ANN is higher than that of the regression method in forecasting cost, and
they confirm that opinion by analysing 530 historical costs in Korea. Besides having
a mysterious process, another problem of ANN is that adding a new case means
retesting models with all the data again. This is time-consuming and does not support
sustainable improvement. In regard to forecasting ability, Kim et al. (2004) compare
Mean Absolute Error Rate (MAER) between ANN and CBR, and find that CBR’s
MAER (=4.81%) is less than the average of 75 ANN models (=5.65%), although
higher than the best ANN model (2.97%). Kim et al. (2005) find that ANN’s mean
error (6.66%) is almost twice CBR’s mean error (3.68%), in a study of 540 apartment
buildings in South Korea. Thus the accuracy of ANN is not greater than that of CBR,
especially considering the randomness in determining weights in Kim et al.’s study
(2004).
Case based reasoning (CBR) is a method of solving a new case by using
previous experience (Aamodt & Plaza, 1994; Xu, 1994). The inherent logic of CBR
8 Chapter 1: Introduction
is consistent with Skitmore’s (1985) finding that experts are good at recalling
estimating details of previous projects and then adjusting to new requirements
(Skitmore, 1985). After a new solution is achieved, the new case can be used to
enrich the current database, which means that CBR is a method that suports
sustainable improvement. Xu (1994) compares CBR with rule based expert systems
and finds that CBR is superior in (1) getting a solution with partial understanding; (2)
providing a closer match to actual human reasoning; (3) providing more explanation
capability (Xu, 1994).
CBR is a widely used tool, and there have been many studies on solving
construction related problems during the last decade (Kim & Kim, 2010). Despite
CBR’s suitability, there are only a few studies on using CBR to conduct an early
estimate (An et al., 2007; Kim et al., 2004; Kim & Kim, 2010; Kim et al., 2005).
Kim et al. (2005) used 540 apartment buildings’ cost data to compare the quality of
estimates of ANN and CBR, and found that ANN’s error rate was twice that of CBR.
Kim et al. (2004) examined the estimating capability of different methods by using
MLR, ANN and CBR separately with 530 historical cases, and found that CBR was
better than MLR and the average accuracy of 75 ANN models. Although CBR’s
error rate is bigger than the best ANN model, Kim et al. (2004) pointed out the
limitation of ANN models for updating new cases. The difference between these two
studies also indicates the importance of methods for determining variable importance
weights when calculating similarity indexes.
Kim et al. (2004) use the gradient descent method (GDM) available in
ESTEEM, software developed from CBR. An et al. (2007) compare estimate quality
of three weight deciding methods: equal weights (EW), GDM, and the analytic
hierarchy process (AHP) for 580 residential buildings’ cost data in Korea, and find
that AHP-CBR is more accurate. AHP is time consuming, and carries the risk of
subjectivity. In a study of cost estimation for 216 pre-stressed concrete beam bridges
completed in Korea, Kim and Kim (2010) propose connecting CBR with a genetic
algorithm (GA) to avoid the subjectivity that can occur with AHP. The validity of
this method is unknown without comparing it to other methods (EW, GDM), and
subjectivity in the adaptation process is another problem. Only length and width are
chosen as the basis for a ratio to adjust cost, and the importance of the width’s weight
is 0, as calculated by using GA. Kim et al. (2005) compare the performance of the
Chapter 1: Introduction 9
GDM and regression methods and find that using regression is better. It is reasonable
to assume that regression performs better because the weights are not decided by
intuition or randomly, but by sampling the data statistically. Regression’s parsimony
principle is also reflected in Kim et al.’s (2005) study in three factors: storey, unit per
storey and finishing grades do not have weights.
1.2.2 Literature review on job performance
Conceptualisation of job performance
Job performance is the central construct in occupational psychology
(Viswesvaran & Ones, 2000) and even the ultimate goal in organisational
management practices (Judge, Thoresen, Bono, & Patton, 2001). Theories of job
performance can be traced back to Taylor’s Scientific Management, providing
techniques such as synthesis and standardisation to improve efficiency of the
production process and productivity of workers. “Fordism”, a further application of
“Taylorism”, is well known for high productivity generated by machines, but higher
wages provided to attract workers to do “boring” works on assembly lines. In the era
of “post-Fordism”, non-technical factors such as organisational culture are believed
to be critical to achieving success (Bonanno & Constance, 2001). Review of previous
research into job performance indicates the three main kinds of job performance to
be task performance, organisational citizen behaviour and counterproductive work
behaviour (Viswesvaran & Ones, 2000).
Early studies measuring job performance focused on task performance,
indicating the extent to which employees complete the professional duties specified
in their work descriptions. Task performance is defined as
the proficiency with which incumbents perform activities that are formally
recognized as part of their jobs; activities that contribute to the organization’s
technical core either directly by implementing a part of its technological
process, or indirectly by providing it with needed materials or services.
(Borman & Motowidlo, 1993b, p73)
For example, task performance was used in the Hawthorne studies exploring
the linkages between job satisfaction and the task performance of workers.
Organisational citizen behaviour (OCB) — assuming job responsibilities and
innovation for the benefit of an organisation without reward expectations
10 Chapter 1: Introduction
(Eisenberger, Fasolo, & Davis-LaMastro, 1990) — has been increasingly emphasised
in many organisational studies. It has been proposed that OCB should comprise five
dimensions of altruism, conscientiousness, sportsmanship, courtesy and civic virtue
(LePine, Erez, & Johnson, 2002; Organ, 1988a). Following this typology, Podsakoff,
Ahearne, and MacKenzie (1997) conducted a study measuring performance in terms
of the quantity and quality of 218 people working in a paper factory, and found that
altruism and sportsmanship led to better performance. However, such OCB
dimensions are not significantly discriminating (LePine et al., 2002). Smith, Organ,
and Near (1983) identify two main kinds of OCB behaviours, including generalised
compliance (indicating conscientious self-disciplined behaviours) and altruism
(indicating a willingness to help others). A positive relationship between
organisational support and OCB was found by Eisenberger et al. (1990).
Counterproductive work behaviour (CWB) is behaviour conducted to
intentionally harm corporate legitimate interests (Dalal, 2005). Such behaviours
include property/equipment sabotage, substance abuse (Sackett & Wanek, 1996), and
behaviours reducing the effectiveness of employees (Fox, Spector, & Miles, 2001).
CWB is assumed to share similar antecedents with OCB and task performance
(Dalal, 2005). For example, Fox et al. (2001) found that job stressors including
organisational constraints, interpersonal conflict and perceived injustice result in
CWB via the mediation of negative emotion. Additionally, the relationships between
job stressors and CWB are stronger for individuals with higher level negative
affectivity (Penney & Spector, 2005) or low in conscientiousness (Bowling &
Eschleman, 2010).
Studies of job performance of construction professionals
Maloney and McFillen in an early (1983) study argued that no validated model
of worker performance existed for the construction industry, although the importance
of organisational constraints, job satisfaction, and motivation had been increasingly
acknowledged outside the industry. To understand research progress into the job
performance of construction professionals, a literature review was carried out. The
keyword “job performance” for article selection was applied to a group of high
impact construction journals, comprising: the Journal of Construction Engineering
and Management (JCEM) and Journal of Management in Engineering (JME) from
the ASCE library; the International Journal of Project Management (IJPM),
Chapter 1: Introduction 11
Automation in Construction (AUTCON) and Building and Environment (B&E) from
Elsevier; Construction Management and Economics from Taylor and Francis; and
Engineering, Construction and Architectural Management (ECAM) from Emerald.
227 search records were initially found. These records were browsed to identify
articles where the job performance of construction professionals is the key theme,
and 13 articles were selected for detailed review. The studies cover construction
professionals, including architects and engineers, quantity surveyors and project
managers. The number of articles by journal is presented in Figure 2.1.
Figure 1.2 Number of related articles by journal
Environmental and individual factors are significant determinants of job
performance. Aiming to improve the work performance of construction project
managers, Pheng and Chuan (2006) point out the importance of the working
environment, and conducted a study to explore job related, project related and
organisational related factors. The differences in these factors between contractor and
consultant project managers are explored by Pheng and Chuan (2006). In addition to
environmental factors, Carr, De La Garza, and Vorster (2002) point to the necessity
of linking individual personality traits to job performance for engineering and
architectural professionals providing project design services. For example, a person
with a personality preference for “judging” performs better in preparing contract
documentation than others with a preference for “perception” (Carr et al., 2002).
Building on motivation theory, Tuuli and Rowlinson (2009) explore the relationships
between the psychological empowerment and job performance of project
0
1
2
3
4
5
AUTCON B&E ECAM IJPM C&E JME JCEM
12 Chapter 1: Introduction
management staff, finding that empowered employees have better a job performance.
In addition to the significant contribution of knowledge of job techniques, time
management abilities, problem solving and relationship management are also critical
predictors of the performance of project managers in mass housing building projects
(Ahadzie et al., 2008a).
Work stress and job satisfaction are popular topics. Leung et al. (2005), for
example, examine the impact of stress on the estimation performance of professional
cost engineers in Hong Kong, finding that stress negatively affects overall
performance in both linear and inverted U-shaped forms. In addition to the effects of
stress, Leung et al. (2006) also explore the effects of stress-coping behaviours on
estimation performance to show that, for instance, both preparatory action and
support seeking actions can improve estimation performance. Although not testing
the significance of the mediation effect of career commitment, Leung, Yu, and
Chong (2015) further demonstrated the negative effect of stress on career
commitment, and the positive effect of career commitment on cost estimation
accuracy. These findings indicate a proactive personality is also important for
improving job performance (Leung, Shan Isabelle Chan, & Dongyu, 2011). For
construction project managers, Leung et al. (2011) explore the nexus between stress
and performance through structural equation modelling (SEM) to demonstrate the
negative effect of job stress on task performance. Job satisfaction is another
psychological factor that affects the performance of construction professionals. As
pointed out by Ling and Loo (2013), job characteristics (such as work autonomy) and
personality characteristics (such as work knowledge and skills) affect the satisfaction
of construction project managers and their work performance.
Other studies focus on conceptualising job performance. In an attempt to
develop competency-based performance measures for construction project managers,
for example, Ahadzie, Proverbs, and Olomolaiye (2008b) draw on empirical
evidence from Ghana to point out the necessity of distinguishing between task
performance behaviours and contextual performance behaviours. Liu and Fellows
(2008), on the other hand, investigated the OCB of quantity surveyors in Hong Kong,
and found that an individualistic orientation was negatively correlated with OCB,
whereas collectivism is positively correlated with OCB. In an another study, Dainty
et al. (2005), aiming to assist human resource management decisions by a better
Chapter 1: Introduction 13
understanding of behavioural competencies, identified the core competencies of
construction project managers and developed a model to predict performance. They
found that self-control and team leadership are critical predictors of project
management performance.
In reviewing previous studies, it is apparent that there are some limitations.
Firstly, there is no consideration of mediators or moderators. Regression analysis is
usually applied in a one-shot approach, with predictors on one side of the equation
and a dependent variable on the other, without considering the interactions between
the predictors. Recently developed statistical methods such as SEM can be helpful in
this situation (Xiong, Skitmore, & Xia, 2015a). Another problem is the definition of
concepts. For example, Ling (2002) identified that both hard attributes (such as job
knowledge) and soft attributes (such as commitment) affect the performance of
architects and engineers in design-build projects. However, these attributes seem to
be a mixed combination of predictors of performance and measures of performance.
For example, Ling (2002) found that performance can be predicted by the attribute of
the speed of producing design drawings, which is really a measure of performance
rather than a predictor. Another problem is “scope-matching”, in that items should be
measured at the same level or cross level analyses are needed. In the study by Ling
and Loo (2013), for instance, performance was measured at the project level, while
satisfaction was measured at the individual level.
1.3 RESEARCH QUESTIONS AND OBJECTIVES
Based on the research background and literature review, this thesis therefore
seeks to examine two main research questions:
• What can be done to improve prediction technique by dealing with overfitting
and multicollinearity problems frequently occurred in construction research?
• What is role of psychological reactions on job performance of cost engineers?
By addressing these questions, this study contributes greatly to improving
construction cost estimation by technique innovation and understanding the
organisation-individual interactions of construction cost engineers. Primary research
objectives are thus developed as:
14 Chapter 1: Introduction
• To develop a hybrid approach based on Akaike information criterion (AIC)
and principal component regression (PCR) to deal with overfitting and
multicollinearity problems
• To evaluate the efficiency of the AIC-PCR approach with an application of
construction cost estimation.
• To examine the role of psychological reactions in promoting job performance
by developing a comprehensive framework.
• To examine job satisfaction and work stress, and their relationships with job
performance with empirical evidences from construction cost engineers.
1.4 THESIS OUTLINE
This thesis is presented by publication. Besides of the introduction and
conclusion chapters, several papers/manuscripts comprising the main content of the
dissertation are presented.
1.4.1 Chapter 2: Construction cost estimation techniques
In this chapter, a new estimation approach is firstly proposed to deal with
problems of overfitting and collinearity; the improved predictability of this approach
is compared with three widely used methods including artificial neural network
(ANN), case-based reasoning (CBR), and supportive vector regression (SVR). An
early version of this study was presented in PhD Student Poster Session of the 2014
Construction Research Congress in Atlanta, US. A conference paper exploring the
cost drivers of contingencies is also presented in Section 2.2.
Bo Xiong*, Martin Skitmore, Bo Xia, Sidney Newton. A hybrid approach for
reducing overfitting and collinearity: an application in construction cost
estimation. Submitted to Journal of Civil Engineering and Management.
Paper Abstract
Chapter 1: Introduction 15
Although many estimation and modelling studies have been conducted, little
research has focused on addressing the overfitting and collinearity problems that
frequently occur in developed predicative models in construction. This study
concerns itself with providing a hybrid approach based on Akaike information
criterion (AIC) and principal component regression (PCR) for those problems. An
application of estimating the preliminaries of construction projects demonstrates the
method and to test its effectiveness in comparison with competing models including
alternative regression models and three data mining techniques of artificial neural
networks, case based reasoning and support vector machines. The experimental
results show that the AIC-PCR approach presents a good predictive accuracy.
Therefore, the hybrid model is a promising alternative for avoidance of overfitting
and collinearity. An abstract should be a brief summary of significant items of the
main paper. An abstract should give concise information about the content of the
core idea of the paper and clearly describe methods and the major findings reported
in the manuscript.
Bo Xiong*, Bo Xia. Examining the impacts of early cost drivers on
contingencies with path analyses. 2014 ASCE Construction Research
Congress, Atlanta, USA, May 19-21, 2014, pp. 1518-1527.
Paper Abstract
The accuracy of early cost estimates is critical to the success of construction
projects. In previous research, the selected tender price (clients' building cost) is seen
as a holistic dependent variable when examining early stage estimates. Unlike other
components of construction cost, the amount of contingencies is decided by
clients/consultants with consideration of early project information. Cost drivers of
contingencies estimates are associated with uncertainty and complexity, and include
project size, schedule, ground condition, construction site access, market conditions,
and so on.
A path analysis of 133 UK school building contracts was conducted to identify
the impacts of nine major cost drivers on the determination of contingencies by
different clients/cost engineers. This research finds that gross floor area (GFA),
schedule, and requirements for air conditioning have statistically significant impacts
16 Chapter 1: Introduction
on contingency determination. The mediating role of schedule between gross floor
area and contingencies (GFAScheduleContingencies) was confirmed with the
Soble test. The total effects of the three variables on contingencies estimates were
obtained with the consideration of this indirect effect. The squared multiple
correlation (SMC) of contingencies (=0.624) indicates that the identified three
variables can explain 62.4% variance of contingencies, which is comparatively
satisfactory considering the heterogeneity of different estimators, unknown
estimating techniques and different projects.
1.4.2 Chapter 3: Conceptual framework and structural equation modelling
This chapter covers two sections before introductions of detailed studies on job
satisfaction and work stress. The first section reviews previous studies on the job
performance of construction professionals and develops a conceptual framework
based on the person-environment fit theory to reveal the role of psychological
reactions in promoting job performance. An early version of this section was
partially presented at the 6th International Conference on Engineering, Project, and
Production Management (EPPM, 2015) held in Gold Coast, Australia.
Section 3.2 presents a critical review of structural equation modelling (SEM),
since SEM is the main data analysis approach in followed studies in examining
psychological reactions on job performance of construction cost engineers.
Bo Xiong,* Martin Skitmore, Bo Xia. A critical review of structural equation
modelling applications in construction research. Automation in Construction,
(ERA: A, IF=1.822). Published January 2015, 49, 59-70.
Paper Abstract
Structural equation modelling (SEM) is a versatile multivariate statistical
technique, and applications have been increasing since its introduction in the 1980s.
This paper provides a critical review of 84 articles involving the use of SEM to
address construction related problems over the period 1998-2012 including, but not
limited to, seven top construction research journals. After conducting a yearly
publication trend analysis, it was found that SEM applications have been accelerating
Chapter 1: Introduction 17
over time. However, there are inconsistencies in the various recorded applications
and several recurring problems exist. The important issues that need to be considered
are examined in research design, model development and model evaluation and are
discussed in detail with reference to current applications. A particularly important
issue concerns the construct validity. Relevant topics for efficient research design
also include longitudinal or cross-sectional studies, mediation and moderation effects,
sample size issues and software selection. A guideline framework is provided to help
future researchers in construction SEM applications.
1.4.3 Chapter 4: Work stress
In this chapter, sub-dimensions of work stress are revealed by utilising the
perceived stress questionnaire (PSQ) with cost professionals. These findings benefit
model development and future research.
Bo Xiong*, Martin Skitmore, Bo Xia. Exploring and validating the internal
dimensions of occupational stress: Evidence from construction cost
estimators in China, Construction Management and Economics (ERA: A).
33(5-6), pp. 495-507.
Paper Abstract
A recurring feature of modern practice is occupational stress among project
professionals, which has debilitating effects on the people concerned and indirectly
affects project success. Previous research outside the construction industry has
involved the use of a psychology perceived stress questionnaire (PSQ) to measure
occupational stress, resulting in the identification of one stressor – demand – and
three sub-dimensional emotional reactions in terms of worry, tension and joy. The
PSQ is translated into Chinese with a back translation technique and used in a survey
of young construction cost professionals in China. Principal component analysis and
confirmatory factor analysis are used to test the divisibility of occupational stress,
which is little mentioned in previous research on stress in the construction context. In
addition, structural equation modelling is used to assess nomological validity by
testing the effects of the three dimensions on organisational commitment; the main
finding is that lack of joy is the sole significant effect. The three-dimensional
18 Chapter 1: Introduction
measurement framework facilitates the standardising measurement of occupational
stress. Further research will establish if the findings are also applicable in other
settings and explore the relations between stress dimensions and other managerial
concepts.
1.4.4 Chapter 5: Job satisfaction
In this chapter, relationships between satisfaction and performance are
explored. The first section explores the nexus at an individual level as represented by
construction cost engineers. The second section examines the performance of other
project participants on two dimensions of contractor satisfaction. The third section
explores the nexus of construction contractors. Three studies demonstrated the fine-
grained model proposed for explaining S-P nexus.
Bo Xiong. Exploring dimensions of job satisfaction and relationships with
performance: Evidence from construction professionals. CIB World Building
Congress 2016 (ERA: A), Tampere, Finland, May 30–June 3, 2016.
Paper Abstract
Theoretical explorations and empirical demonstrations of the nexus between
job satisfaction and job performance have never ceased. Some argue “happier
workers produce more”, while some insist that workers with better performance
achieve satisfaction through bigger chances of rewards. In a review of previous
studies, weak empirical evidence may be attributed to changing definitions of
concepts. This study conducts a fine-grained analysis to propose a new conceptual
model based on the S-P nexus. Firstly, job satisfaction is divided into economic
satisfaction (ES) and production-related/noneconomic satisfaction (PS). This
assumption is validated in this study by principal component analysis of empirical
evidence from a questionnaire survey of construction professionals in China. It is
found that the effects of ES and PS on job performance are different and warrant
further study. The proposed model will be helpful to both academics and
practitioners when investigating the nature of the satisfaction-performance nexus and
making strategic decisions on personnel management.
Chapter 1: Introduction 19
Bo Xiong,* Martin Skitmore, Bo Xia, Md Asrul Masrom, Kunhui Ye, Adrian
Bridge. Examining the influence of participant performance factors on
contractor satisfaction: A structural equation model. International Journal of
Project Management, 32(3), 482-491, 2014.
Paper Abstract
Participant performance is critical to the success of projects. At the same time,
enhancing the satisfaction of participants not only helps with problem solving but
also improves their motivation and cooperation. However, previous research related
to participant satisfaction is primarily concerned with clients and customers and
relatively little attention has been paid to contractors.
This paper investigates how the performance of project participants affects
contractor project satisfaction in terms of the client's clarity of objectives (OC) and
promptness of payments (PP), designer carefulness (DC), construction risk
management (RM), the effectiveness of their contribution (EW) and mutual respect
and trust (RT). With 125 valid responses from contractors in Malaysia, a contractor
satisfaction model is developed based on structural equation modelling.
The results demonstrate the necessity for dividing abstract satisfaction into
two dimensions, comprising economic-related satisfaction (ES) and production-
related satisfaction (PS), with DC, OC, PP and RM having significant effects on ES,
while DC, OC, EW and RM influence PS. In addition, the model tests the indirect
effects of these performance variables on ES and PS. In particular, OC indirectly
affects ES and PS through mediation of RM and DC respectively. The results also
provide opportunities for improving contractor satisfaction and supplementing the
contractor selection criteria for clients.
Bo Xiong,* Martin Skitmore, Md Asrul Masrom, Bo Xia. A fine-grained
analysis of contractor satisfaction in promoting project management
performance. Submitted to Project Management Journal.
Paper Abstract
20 Chapter 1: Introduction
Despite the fast growth of project-based companies and industries, studies on
the satisfaction-performance (S-P) nexus of project participants are lacking. This
study aims to explore the role of contractor satisfaction in affecting contractor project
management performance along with considering external effects from other key
participants. Fine-grained hypothesized models are developed by using two broad
dimensions of satisfaction toward noneconomic factors and economic factors.
Structural equation modelling techniques are applied with data collected from 117
projects. Modelling results show that it is insufficient to simply conclude that
contractor satisfaction influences project managerial performance and the vice versa,
and the satisfaction disaggregation is necessity. Additionally, it is found that
noneconomic satisfaction contributes to performance, which in turn contributes to
economic satisfaction. The theoretical and practical implications are further
discussed.
1.4.5 Chapter 6: Conclusions
In this chapter, the objectives of the thesis are reviewed through describing the
contributions of the studies conducted in the process of this research. Implications for
future research are discussed in detail.
Chapter 2: Construction cost estimation techniques 21
Chapter 2: Construction cost estimation
techniques
2.1 A NEW COST ESTIMATION APPROACH
Statement of contribution
The authors listed below have certified that:
1. They meet the criteria for authorship in that they have participated in the
conception, execution, or interpretation, of at least that part of the publication in their
field of expertise;
2. They take public responsibility for their part of the publication, except for
the responsible author who accepts overall responsibility for the publication;
3. There are no other authors of the publication according to these criteria;
4. Potential conflicts of interest have been disclosed to (a) granting bodies, (b)
the editor or publisher of journals or other publications, and (c) the head of the
responsible academic unit, and
5. They agree to the use of the publication in the student’s thesis and its
publication on the Australasian Research Online database consistent with any
limitations set by publisher requirements.
In the case of this chapter:
A new cost estimation approach
Bo Xiong*, Martin Skitmore, Bo Xia, Sidney Newton, A hybrid approach for
reducing overfitting and collinearity: an application in construction cost estimation,
Submitted to Journal of Civil Engineering and Management.
Contributor Statement of contribution
Bo Xiong
Conducted a literature review, designed the research, wrote the
manuscript and acted as the corresponding author.
27/06/2016
Martin Skitmore Directed and guided this study, and proofread the manuscript.
Bo Xia Directed and guided this study.
Sidney Newton Directed and guided this study, and proofread the manuscript.
22 Chapter 2: Construction cost estimation techniques
Principal Supervisor Confirmation
I have sighted email or other correspondence from all Co-authors confirming their
certifying authorship.
Martin Skitmore
___________________ _____________________ _________________
Name Signature Date
Chapter 2: Construction cost estimation techniques 23
2.1.1 Introduction
Estimates such as of scope, cost and schedule are needed for most projects and
many papers have demonstrated the use of estimation methods such as multiple
linear regression (MLR) for this purpose (Cheung & Skitmore, 2006; Li, et al., 2005;
Skitmore, et al., 1990). Two main problems are collinearity and overfitting, as the
existence of either can produce significantly biased results. Overfitting occurs when
too many independent variables are incorporated into the developed (training) model.
An extreme example is where there are as many variables as cases so that, although a
perfect fit is obtained with the sample data, the model has little chance of
representing the population and predicting accurately. Collinearity occurs when the
independence assumption is violated. That is, when the independent variables are
highly correlated and can be largely represented by other variables. These problems
make some researchers seek other methods such as artificial neural networks (ANN),
case-based reasoning (CBR) and support vector machines (SVM) for solutions.
Unlike these black box or indirect approaches, MLR produces the desired parameter
estimates directly and accurately if collinearity and overfitting are dealt with
properly.
The collinearity problem in ordinary least squares (OLS) regression was
recognised several decades ago (Skitmore & Marston, 1999a) and many treatments
have been developed. Although it is possible to improve a model by simply deleting
one or more predictors with a high 𝑅𝑖2 (see Eq. (8)) (O’brien, 2007), keeping or
removing a variable should depend on the theoretical underpinning involved
(Andersen & Bro, 2010). Ridge regression (RR), partial least squares regression
(PLS) and principal component regression (PCR) (see Liu, Kuang, Gong, and Hou
(2003)) are three popular methods developed to deal with collinearity and avoid the
loss of information when deleting variables (Næs & Martens, 1988; Vigneau,
Devaux, Qannari, & Robert, 1997). Although these methods are comparable in
predictive ability, RR and PLS still produce biased estimates of the regression
coefficients of the predictor variables (O’brien, 2007). PCR, on the other hand, is
more consistent with stepwise MLR and collinearity diagnostics. In addition,
although PCR does not necessarily lead to improved predictions relative to OLS,
such improvements do nevertheless occur quite often in practice (Næs & Martens,
1988). The currently recommended method of overcoming collinearity problems is
24 Chapter 2: Construction cost estimation techniques
therefore to use the PCR procedure to correct parameter estimates (Liu, et al., 2003).
This involves the application of the default stepwise regression approach in selecting
predictor variables by simultaneously minimizing the sum of squares error (SSE) and
maximizing adjusted R2 in principal component selection.
To deal with overfitting problems, the principal of parsimony needs to be
considered in variable selection (Andersen & Bro, 2010). Including too many
variables leads to a high variance in parameter estimates and an overfit model with
weak generalizability, while too few variables leads to a lack of necessary
information and decreased model fit (Johnson & Omland, 2004). For overfitting
problems, the Akaike Information Criterion (AIC), an asymptotically unbiased
estimator of the expected relative Kullback-Leibler information quantity (Kullback &
Leibler, 1951), has been recommended for choosing suitable predictor variables
(Akaike, 1974). This statistic represents the amount of information lost in the model
fit when adding predictor variables to help avoid overfitting with a comparatively
small sample size (Posada & Buckley, 2004) and is given by:
AIC = -2ι + 2K (2.1)
with maximized log-likelihood (𝜄) and 𝐾 estimable parameters. Despite the
“superficial” form of the AIC formula, it is well founded in information theory and
with a non-arbitrary “penalty term” 2𝐾 (Burnham & Anderson, 2002).
This paper aims to provide a solution collinearity and overfitting in estimation
exist simultaneously. As pointed out by Xu (1994), using traditional PCR to counter
multicollinearity problems increases the risk of overfitting. Using the SSE and
adjusted R2 criteria can result in some irrelevant variables being input to the OLS
regression model (Xu, 1994). The Akaike information criterion (AIC) is the most
commonly used information theoretic approach to measuring how much information
is lost between a selected model and the true model. It has been used widely as an
effective model selection method in many scientific fields, including ecology
(Johnson & Omland, 2004) and phylogenetics (Sullivan & Joyce, 2005). Compared
with the use of adjusted R2 to evaluate the model solely on fit, AIC also takes model
complexity into account (Johnson & Omland, 2004). In addition, AIC has several
important advantages over the likelihood ratio test (Posada & Buckley, 2004).
Chapter 2: Construction cost estimation techniques 25
Acknowledging the effectiveness of AIC in model selection, this paper
presents a hybrid Akaike information criterion-Principal component regression (AIC-
PCR) approach to deal with the problems of overfitting and collinearity frequently
occurring in OLS regression. Additionally, comparisons with other estimation
methods using an artificial neural network (ANN), case based reasoning (CBR) and
support vector machines (SVM) are conducted within an objectively quantitative
source dataset and a Likert-scaled dataset for validation.
2.1.2 Literature review
Construction cost estimation
The term “building cost modelling” was formally introduced in the Building
Cost Research Conference held in 1982 (Newton, 1991). The accuracy of early stage
construction cost estimates is very important (Lowe et al., 2006; Skitmore et al.,
1990). Understanding the properties of a cost model is therefore vital for the
effective control and development of future techniques (Skitmore and Marston,
1999). Although the accuracy of cost estimating is expected to improve as more
information is released as design evolves (Skitmore, 1987), clients still require
accurate cost advice before design work to assist in assessing the feasibility of
different development proposals. For organizations such as government authorities
and real estate developers, inaccurate early estimates can result in the inefficient use
of money, missed development opportunities and unsuccessful project management
(Oberlender and Trost, 2001). Furthermore, it has become increasingly common for
the final cost of projects to exceed the estimated costs and by an increasing margin
(Williams et al., 2005). For example, Flyvbjerg et al. (2003) analysed 258
transportation infrastructure projects worth US$90 billion and found that 9 out of 10
cost overrun projects are a direct result of inaccurate estimation in the early project
stages. Similarly, Merrow et al. (1979) found that 74% of the cost growth of projects
undertaken by the chemical, oil, and minerals industries in North America is also
caused by underestimation in the early project stages.
Applications of MLR, CBR, ANN and SVM
The regression method has been used as an effective tool in estimation of
project performance for decades. For example, Williams (2003) uses regression
26 Chapter 2: Construction cost estimation techniques
models developed using data from five transportation agencies in the US to predict
the final cost of highway projects. Li, et al. (2005) construct regression estimate
models for office buildings in Hong Kong. To optimize predicative ability within the
sample, the stepwise regression approach can be applied to meet the principle of
parsimony. For example, Masrom et al. (2013) apply forward and backward stepwise
regression to identify key items from 95 possible factors of contraction satisfaction
and Guerrero, Villacampa, and Montoyo (2014) use stepwise regression modelling to
predict the construction time of 168 Spanish building projects. Despite its
applicability in many situations, the regression method are taken for granted as all
variable input approach other than the stepwise one, which makes the method
unfairly weak in comparisons. For example, Son, Kim, and Kim (2012) use the full
variable input MLR rather than stepwise regression when comparing with the SVM
method in the dataset with severe collinearity.
Applications of Artificial neural networks (ANN) and case-based reasoning
(CBR) methods accounted for less than 5% of 56 publications related to construction
cost estimation during 1960-1988 (Newton, 1991), but have developed rapidly since
then with the aid of improved computer techniques (Chou & Tseng, 2011). ANN
simulates the learning process of the human brain by representing variables as input-
output nodes in a weighted network trained on data, and has been used to make
predictions in a variety of fields (Kim, et al., 2004; Kim, et al., 2005). Kim, et al.
(2004) develop an ANN for cost estimation using data from 530 projects in Korea
and show the accuracy to be slightly higher than that provided by regression.
Cheung, Wong, Fung, and Coffey (2006) use ANN to predict project performance
based on information available at the bidding stage from the Hong Kong Housing
Authority. However, ANN is a “black box” method and suffers the potential
drawback of having to retrain the model completely with all data whenever a new
case is added. Additionally, ANN studies have difficulties of generalization because
of overfitting nature (Min and Lee, 2005).
CBR is a method that uses previous experience to solve new cases (Aamodt &
Plaza, 1994; Xu, 1994). It is particularly suited to: (1) obtaining a solution with
partial understanding; (2) providing a reasonably close match to actual human
reasoning; and (3) providing more explanation of its working (Xu, 1994). The
inherent logic of CBR is consistent with Skitmore (1985)’s finding that construction
Chapter 2: Construction cost estimation techniques 27
experts predict by recalling the estimating details of previous projects and then
adjusting these to suit new requirements. The number of publications applying CBR
to construction related problems during the last decade is also increasing (Kim &
Kim, 2010). For example, Kim, et al. (2005) use both ANN and CBR to model the
construction cost of 540 Korean apartment buildings, finding that CBR performs
particularly well. Kim, et al. (2004) examine the estimating capabilities of MLR,
ANN and CBR using data from 530 projects to find that CBR outperforms both
MLR and an average of 75 alternative ANN models.
Support vector machines (SVM) are developed mainly by Vapnik (2000) based
on structural risk minimization, and have been shown to ensure good generalization
(Movahedian Attar et al., 2013). An et al. (2007b) apply SVM to classify the
accuracy of cost estimations for 62 Korean building projects and for regression
purposes. Attar, Khanzadi, Dabirian, and Kalhor (2013) use support vector
regression (SVR) to forecast how far construction costs deviate from client
expectations during contractor prequalification, and find that SVR performs better
than ANN. Son, Kim, et al. (2012) use PCA-SVR, a SVM approach aided by
principal component analysis to reduce dimensions, to predict the construction costs
of 84 building projects. However, they ignore the severe collinearity of the 64
predicting variables used in their dataset and the overfitting that results from using so
many variables/principal components given such a relatively small sample size.
However, there has been little study of how effectively the methods handle data
where the risk of overfitting and collinearity is significant in construction research.
2.1.3 A hybrid approach
The treatment of modelling problems is usually considered in terms of
detection and correction (Farrar & Glauber, 1967). Various diagnoses are possible
and several need to be considered before an appropriate approach can be proposed to
solve collinearity and overfitting.
Collinearity diagnosis
When several variables/predictors in a multivariate regression model are highly
correlated, one variable can be linearly and largely explained by the other variables.
The coefficient estimates of a multiple regression with this problem may change
28 Chapter 2: Construction cost estimation techniques
erratically in response to small changes in the model or the data, rendering the
coefficient estimates unreliable. The variance inflation factor (VIF) is widely used as
a standard way to detect collinearity, with larger VIF values indicating more severe
correlation. In the ideal situation, when the predictors are not correlated, all 𝑅𝑖2 = 0
and the VIF values of all variables have a minimum value of 1. A larger 𝑅𝑖2
(dependency on other predictors) leads to a larger VIF. A VIF larger than 10 is
usually used to indicate significant collinearity (Neter et al., 1989). However, high
VIF values do not necessarily worsen the regression analysis and the influence of
other factors on the variance of regression coefficients should also be considered
(O’brien, 2007).
Two questions are particularly important in collinearity diagnostics: (1) how
many dimensions in the predictor space are nearly collinear; and (2) which predictors
are most strongly implicated in each of those dimensions (Friendly & Kwan, 2009).
To address these questions, Belsley, Kuh, and Welsch (2005) propose a strategy
involving principal component analysis, known as Belsley collinearity diagnostics.
The strategy seeks to identify collinearity by introducing two statistics: a condition
index (CI) and coefficient variance proportion (CVP). CI is defined as CIk =
√λ1/λk where is the Eigenvalue in collinearity diagnostics. Belsley, et al.
(2005) recommend to be cautious with 𝐶𝐼>10. Friendly and Kwan (2009) regard
𝐶𝐼k<5 as “ok”, 5<𝐶𝐼k<10 as “warning” and 𝐶𝐼k> 10 as “danger”. CVP indicates the
proportion of variance of each variable associated with each principal component as
a decomposition of the coefficient variance for each dimension (Belsley, et al., 2005;
Friendly & Kwan, 2009) Caution is needed with two or more 𝐶𝑉𝑃k>0.5.
Overfitting diagnosis
Leave one out cross validation (LOOCV) is a commonly used method in model
selection to detect overfitting and compare predictive ability (Xu, 1994). Compared
with the insufficient data utilization and unreliability of the traditional separate
holdout-set, in-out sample performance, LOOCV does not waste data and has better
reliability in model predictive ability comparisons (Moore, 2001). LOOCV is easy to
understand in that the 𝑖 th model is developed by training the remaining dataset
without the 𝑖th case and then using this model to predict the 𝑖th case, and calculating
the mean error after repeating this exercise N (i.e. sample size) times with
replacement. Although widely applied in model development and selection in many
Chapter 2: Construction cost estimation techniques 29
other scientific areas (such as chemistry) this technique is comparatively new to the
construction management and economics field. For example, Cheung and Skitmore
(2006) used the technique to compare the efficiency of the storey enclosure area
method and four other traditional methods in very early design stage cost forecasting.
Despite having wide application and well known predictive properties,
however, LOOCV is often criticized for being time-consuming and performs
comparatively poorly in selecting linear models when compared with more classical
statistical methods (Rivals & Personnaz, 1999). For this reason LOOCV is only used
here to compare regression models developed by other statistical linear model
development criteria (i.e. SSE, adjusted R2, and AIC).
AIC-PCR procedure and formulas
If a MLR has overfitting problems and the model with lowest AIC still suffers
from collinearity problems, then the AIC-PCR procedure may be useful. AIC-PCR is
described in eight steps, as follows:
Step 1: Proceed with the AIC criterion stepwise regression, with a column
comprising the actual values of the dependent variable Y and a matrix X comprising
all independent variables to obtain a model with the lowest AIC. The logic is to add
the one variable that most helps to reduce the AIC of the model, and then repeat
adding another variable or removing an existing variable. Whichever most helps
reduce AIC is actioned until the lowest AIC with k predictors is achieved. Software
such as MATLAB has an automatic command for this task. Note that, whilst SPSS
can provide the AIC values of a stepwise regression model using Syntax
programming, these models are still selected according to the SSE criterion.
Step 2: Proceed to obtain collinearity diagnostics including VIF, CI and 𝐶𝑉𝑃.
The variance inflation factor (VIF) of the 𝑖 th predictor variable, indicating its
collinearity with other predictors is given by:
Xi = α + β1X1 + ⋯ + βnXn + error (Xi is excluded in the right side) (2.2)
Total sum of squares (TSS) = ∑ (Xi-Xi)2n
1 (2.3)
Explained sum of squares (ESS) = ∑ (Xi-Xi)2n
1 (2.4)
Residual sum of squares (RSS) = ∑ (Xi-Xi)2n
1 (2.5)
30 Chapter 2: Construction cost estimation techniques
𝑇𝑆𝑆𝑖 = 𝐸𝑆𝑆𝑖 + 𝑅𝑆𝑆𝑖 (2.6)
𝑅𝑖2 = 𝐸𝑆𝑆𝑖/𝑇𝑆𝑆𝑖 (2.7)
𝑉𝐼𝐹 =1
1−𝑅𝑖2 = 𝑇𝑆𝑆𝑖/𝑅𝑆𝑆𝑖 (2.8)
where, 𝑋i is the mean value of 𝑋i, and ��i is the estimated value of 𝑋i. TSS is
the sum of the squared differences between each observation and the overall mean;
ESS is the sum of the squared deviations between the estimated values and mean
values of each variable; and RSS is the sum of the squared residuals. CI and CVP can
be determined by applying the collinearity diagnostics command in software such as
MATLAB and SPSS, or calculated using equations 2.2-2.8 as provided.
Step 3: Proceed with the principal component analysis (PCA) with software
such as MATLAB and SPSS to transform the k correlated variables to a set of
uncorrelated principal components, 𝐶i. All components should be extracted at this
stage, and they should account for 100% of the variance.
Step 4: Compute the standardized dependent variable, the 𝑝 standardized
independent variables and the values of the 𝑝 principal components respectively in
preparation for establishing 𝑝 standardized principal component regression
equations:
𝑌′ = (𝑌 − ��)/𝑆𝑌 (2.9)
Xi' =
Xi-Xi
SXi
(i = 1, … . , k) (2.10)
Cj = a1jX1' + a2jX2
' + ⋯ + akjXk' (i = 1, … . , k; j = 1, … . , p) (2.11)
where, 𝑌′ denotes the standardized dependent variable; 𝑌 the dependent
variable; 𝑆Y the standard deviation of the dependent variable; �� the mean of the
dependent variable; 𝑋i′ the 𝑖 th standardized independent variable; 𝑋i the 𝑖 th
independent variable; 𝑋i the mean of the 𝑖th independent variable; 𝑆Xithe standard
deviation of the 𝑖th independent variable, 𝐶j the 𝑗th principal component and 𝑎ij the
coefficient of the principal component matrix (the matrix consists of 𝐶j and 𝑋i).
Chapter 2: Construction cost estimation techniques 31
Step 5: Proceed with the AIC criterion stepwise regression of principal
components to estimate 𝑌′ in MATLAB and if not all principal components are
significant, select the lowest AIC regression equation, as in:
y' = ∑ Bj' Cj (j = 1, … , q ≤ p) (2.12)
Step 6: Transform Eq. (12) with Eq. (11) to obtain:
y' = ∑ bi' Xi
' (i = 1, … k) (2.13)
where ��′ is the standardized estimate of the linear regression equation and 𝑏i′
the 𝑖 th standardized regression coefficient of the standardized linear regression
equation (Liu, et al., 2003).
Step 7: Calculate the regression coefficients and constant, and transform the
standardized linear regression equation into a general linear regression equation
𝑏𝑖 = 𝑏𝑖′(
𝐿𝑦𝑦
𝐿𝑥𝑖𝑥𝑖
)1/2 (2.14)
b0 = Y- ∑ bi Xi (i = 1, … , k) (2.15)
y = b0 + ∑ bi Xi (i = 1, … , k) (2.16)
where, 𝑏i is the regression coefficient of 𝑖th variable; 𝐿yy the sum of squares of
dependent variable Y; 𝐿xixi the sum of squares of the ith independent variable 𝑋i;and
𝑏0 is the constant of the new linear model.
Step 8: Calculate the mean squared error (MSE) of the final model from
MSE =1
n∑ (Yi-Yi)
2n1 (2.17)
2.1.4 Application in construction cost estimation
Preliminaries
According to the standard elemental cost categorization used by the Building
Cost Information Service (BCIS), construction costs comprise eight components:
substructure, superstructure, internal finishes, fittings, services, external works,
preliminaries, and contingencies. The quantities of the items involved in most of
these cost components are relatively straight-forward to determine given the level of
32 Chapter 2: Construction cost estimation techniques
information typically available at an early project stage, and the value of
contingencies is decided by the clients/consultants prior to tendering. However, the
preliminaries component contains an allocation of budget to general overheads, site
overheads, risk contingency and profit, and this is often a key competitive
component not determined until the tendering stage. The actual allocation is
influenced by a complex combination of past and recent experience on the part of the
contractor, the current workload of the contractor, market conditions, and project
characteristics often determined by the contractor, such as contract duration
(Akintoye, 2000; Tah et al., 1994). The cost of subcontractor rework is another
expenditure component generally included as a component of the preliminaries
(Love and Li, 2000). The mark up strategy for preliminaries is also different to that
of other cost components and can be used to achieve an unbalanced tender that
significantly improves the cash flow of a contractor (Kaka, 1996).
Sample projects
The sample cases comprise 204 UK school building projects completed during
2000-2012, and selected at random from a large commercial cost database. Project
differences due to geographical location, construction year and rate of inflation are
addressed by rebasing all prices to the same date (Fourth quarter, 2012) and location
(Greater London district), using the BCIS Construction Price Index. Some important
characteristics of the sample cases are presented in Tables 2.1 and 2.2. The left hand
column provides the categories, or cost drivers. These are blank in places/for some
projects due to a lack of complete information on such features as building height
and contract type.
Table 2.1 Sample descriptions - part 1
Category Type Frequency Percentage
Building
function
Primary schools 86 42.16%
Secondary schools 64 31.37%
Nursery schools 29 14.22%
Special schools 13 6.37%
Sixth form/tertiary colleges 12 5.88%
Structure
Brick construction 70 34.31%
Steel framed 115 56.37%
Timber framed 15 7.35%
Concrete framed 3 1.47%
Unspecified 1 0.49%
Selection of Selected competition 164 80.39%
Chapter 2: Construction cost estimation techniques 33
contractor Open competition 10 4.90%
Design and build - competitive 12 5.88%
Two stage tendering 12 5.88%
Unspecified 6 2.94%
Table 2.2 Sample descriptions - part 2
Category Description
Gross floor area (m2) Range from 64 to 19670; M=1471.64; SD=2209.57
Stories Range from 1 to 4; M=1.41; SD=0.60
Schedule (months) Unspecified: 87
Remaining: range from 5 to 32; M=10.83; SD=4.27
Ground condition Unspecified: 23
Bad(1)-Moderate(3)-Good(5); M=3.74; SD=1.42
Work space Unspecified: 21
Highly restricted(=1)-Restricted(=3)-Unrestricted(=5); M=3.92; SD=1.24
Site access Unspecified: 17
Highly restricted(1)-Restricted(3)-Unrestricted(5); M=3.71; SD=1.20
Market condition
Unspecified: 49
Low competitive(1)-Less competitive(2)-Average(3)-Competitive(4)-Highly
competitive(5); M=3.99; SD=0.79
Air Conditioning Yes=1; No=0; 26 cases are 1; 178 cases are 0
Preliminaries (£) Range from 0 (1 case) to 3,391,713; M=318,448; SD=443,345
Preliminaries/GFA Range from 0 (1 case) to 702; M=251; SD=116
Elemental cost items framework and sample descriptions
The first edition of an Elemental Standard Form of Cost Analysis was released
in 1961 by the Royal Institution of Chartered Surveyors (RICS). The 2012 edition of
this standard is used as the reference to construct the analysis framework. Elemental
cost items in the first column of Table 2.3 refer to variables used for model
development in the MLR process, and those in the second code column refer to the
selected variables in the PCR process.
Table 2.3 Elemental cost items framework
Elemental cost items Code Elemental cost items Code
1 Substructure x1 5F Space heating and air treatment x19
2A Frame x2 5G Ventilating systems x20
2B Upper floors x3 5H Electrical installations x21
2C Roof x4 5I Gas installations x22
2D Stairs x5 5J Lift and conveyor installations x23
2E External walls x6 5K Protective installations x24
2F Windows and external
doors
x7 5L Communications installations x25
34 Chapter 2: Construction cost estimation techniques
2G Internal walls and
partitions
x8 5M Special installations x26
2H Internal doors x9 5N Builder's work in connection x27
2 Superstructure 5O Builder's profit and attendance x28
3A Wall finishes x10 5 Services
3B Floor finishes x11 6A Site works x29
3C Ceiling finishes x12 6B Drainage x30
3 Internal finishes 6C External services x31
4 Fittings x13 6D Minor building works x32
5A Sanitary appliances x14 6 External works
5B Services equipment x15 7 Contingencies x33
5C Disposal installations x16 8 Preliminaries y
5D Water installations x17
5E Heat source x18
Experimental results
In the model development phase four models are developed by stepwise
regression modelling under the criteria SSE (model 1), adjusted R square (model 2),
AIC (model 3) and by directly entering all predictor variables (model 4), as presented
in Table 2.4. The diagnoses for overfitting and collinearity follow.
In this case, using LOOCV to detect overfitting involves developing
204x4=816 sub-models to evaluate their predictive ability. The MSELOOCV values
for each model are presented in Table 2.4, in parentheses. Although all four models
are comparable in MSE values, the model overfitting varies greatly. For example,
Model 2 has the lowest MSE and highest adjusted R2 but its predictive ability is
weaker. Model 3, developed under the AIC, has the lowest MSELOOCV. As the
collinearity diagnoses for the four models indicate, collinearity is a common problem
- with more than one VIF larger than 10, CI larger than 10 and CVP larger than 0.5.
That is to say, even the overfitting-reduced Model 3 suffers collinearity. Although
both MATLAB and SPSS can help handle such diagnoses, MATLAB programming
is selected for its ability to visualize the collinearity diagnostics (Friendly and Kwan,
2009). Figure 2.1 provides a visual illustration of the diagnostics for Model 3. For
clarity, only the principal components with CI values larger than five are shown.
Table 2.4 Developed regression models Models Equations MSE (1010)
Chapter 2: Construction cost estimation techniques 35
1
(SSE) �� = 25220.000 + 0.396x1 − 0.373x6 − 0.915x8 − 1.381x10 + 2.391x12
+ 1.069x14 + 2.621x17 + 0.604x19 + 0.760x21
− 4.892x22 + 0.690x24 + 1.497x25 − 1.586x27
− 1.299x31 + 0.135x32
1.410
(2.160)
2 �� = 27296.000 + 0.322x1 + 0.276x2 + 0.103x4 − 0.390x6 − 0.920x8
− 0.602x9 − 1.558x10 + 0.292x11 + 2.413x12
+ 0.837x14 − 0.511x15 + 3.050x17 + 0.528x19
+ 0.725x21 − 7.454x22 + 0.382x24 + 1.336x25
− 1.692x27 − 3.903x28 + 0.082x29 − 1.354x31
+ 0.118x32 + 0.186x33
1.355
(2.643)
3 �� = 22722.000 + 0.418x1 − 0.385x6 − 0.885x8 − 1.364x10 + 2.407x12
+ 0.937x14 + 2.748x17 + 0.576x19 + 0.738x21
− 5.893x22 + 0.568x24 + 1.453x25 − 1.665x27
− 1.396x31 + 0.136x32 + 0.200x33
1.391
(2.118)
4 �� = 36079.000 + 0.239x1 + 0.313x2 + 0.397x3 + 0.132x4 − 0.148x5
− 0.402x6 − 0.069x7 − 0.943x8 − 0.587x9 − 1.468x10
+ 0.245x11 + 2.210x12 − 0.051x13 + 1.151x14
− 0.613x15 + 0.088x16 + 2.908x17 − 0.440x18
+ 0.539x19 + 0.205x20 + 0.743x21 − 7.191x22
+ 0.665x23 + 0.375x24 + 1.239x25 + 0.186x26
− 1.129x27 − 4.189x28 + 0.081x29 + 0.015x30
− 1.400x31 + 0.119x32 + 0.159x33
1.414
(4.090)
PCR �� = 28152.762 + 0.366x1 − 0.390x6 − 0.804x8 − 1.531x10 + 2.439x12
+ 1.320x14 + 2.463x17 + 0.607x18 + 0.753x21
− 4.395x22 + 0.444x24 + 1.612x25 − 1.764x27
− 1.241x31 + 0.110x32
1.324
AIC-
PCR �� = 26801.134 + 0.430 x1 − 0.387 x6 − 0.857 x8 − 1.314 x10 + 2.426 x12
+ 1.010 x14 + 2.643 x17 + 0.587 x19 + 0.718 x21
− 6.337 x22 + 0.473 x24 + 1.700 x25 − 1.881 X27
− 1.419 x31 + 0.102 x32 + 0.132 x33
1.301
Figure 2.1 Collinearity diagnostics for Model 3
According to the diagnoses of overfitting and collinearity, this dataset is
suitable for testing the AIC-PCR approach as it displays both features. Following the
AIC-PCR approach, Model 3 with the lowest overfitting and best predictability is
obtained by applying the PCR under AIC. Table 2.4 gives the MSE results of four
models for comparisons with Model 1 representing: a stepwise regression under SSE;
PCR, representing the traditional PCR approach developed under SSE; Model 3
36 Chapter 2: Construction cost estimation techniques
representing stepwise regression under AIC; and AIC-PCR representing the proposed
approach. It is found that the AIC-PCR approach not only avoids overfitting (by
applying the AIC criterion) and collinearity (by applying PCR) but also improves
predictability (with 7.73% less MSE than the default stepwise regression model) and
is more accurate (with 1.74% less MSE than the traditional PCR approach using the
SSE criteria).
Comparisons with other methods
The AIC-PCR approach is compared with other data mining techniques of
ANN, PCA-SVR and K-Nearest Neighbour (KNN) as a basic type of CBR, to see
how well it performs. Two absolute evaluation criteria are the root mean squared
error (RMSE) and mean absolute error (MAE). A scaled criterion of mean absolute
percentage error (MAPE) is also used, where:
RMSE = √1
n∑ (Yi-Yi)
2n1 (2.18)
MAE =1
n∑ | Yi-Yi
n1 | (2.19)
MAPE =1
n∑ |
Yi-Y
Yi|n
1 (2.20)
Table 2.5 Comparison of results for Application 1, price estimating
Models RMSE MAE MAPE
AIC-PCR 114061.725 74954.314 0.419
PCR 115064.347 76398.030 0.425
ANN 194183.576 89780.485 0.623
KNN 260417.245 128660.219 0.419
PCA-SVR 211449.878 103757.670 0.621
The comparison of results presented in Table 2.5 confirms the effectiveness of
AIC-PCR, its error rate being the same or lower than the other four methods
whichever criteria are used.
2.1.5 Conclusions
The main aim of this study was to build an alternative approach to deal with
the ubiquitous overfitting and collinearity problems that occur in construction
research. A hybrid AIC-PCR method is developed and tested using construction cost
Chapter 2: Construction cost estimation techniques 37
data on 204 construction projects. The study has found that the hybrid approach not
only reduces the risk of overfitting and collinearity, but also results in better
predictability compared with the commonly used stepwise regression models and
traditional PCR approach under the SSE criterion. The study also validates its
applicability by comparison with other conventional methods including ANN, CBR
and SVM. The approach is a promising alternative to be recommended for equivalent
situations where overfitting and collinearity can be problematic especially when the
linear form is approximate to describe relationships between independent variables
and dependent variables..
Some limitations should be acknowledged. Firstly, the sample tested in the
study is from construction projects. Additionally, there is not yet a standard cut-off
study on determining a certain linear level for applying the proposed approach or
other approaches like SVM. These two limitations are sufficient to prevent this
approach from global generalization. However, the new model offers fertile ground
for further research and practice. Whilst the experimental application presented
should be taken cautiously in wider generalization, it does demonstrate the capability
of the hybrid approach in avoiding overfitting and collinearity problems and gaining
accurate estimates. Future studies could benefit from testing it applicability in other
contexts and improving the proposed procedure.
38 Chapter 2: Construction cost estimation techniques
2.2 IMPACTS OF EARLY COST DRIVERS
Statement of contribution
The authors listed below have certified that:
1. They meet the criteria for authorship in that they have participated in the
conception, execution, or interpretation, of at least that part of the publication in their
field of expertise;
2. They take public responsibility for their part of the publication, except for
the responsible author who accepts overall responsibility for the publication;
3. There are no other authors of the publication according to these criteria;
4. Potential conflicts of interest have been disclosed to (a) granting bodies, (b)
the editor or publisher of journals or other publications, and (c) the head of the
responsible academic unit, and
5. They agree to the use of the publication in the student’s thesis and its
publication on the Australasian Research Online database consistent with any
limitations set by publisher requirements.
In the case of this chapter:
Impacts of early cost drivers
Bo Xiong*, Bo Xia, Examining the impacts of early cost drivers on contingencies
with path analyses, 2014 ASCE Construction Research Congress, Atlanta, USA,
May 19-21, 2014, pp. 1518-1527.
Contributor Statement of contribution
Bo Xiong
Conducted a literature review, designed the research, wrote the
manuscript and acted as the oral presenter.
07/03/2016
Bo Xia Assisted with manuscript revision.
Principal Supervisor Confirmation
I have sighted email or other correspondence from all Co-authors confirming their
certifying authorship.
Chapter 2: Construction cost estimation techniques 39
Martin Skitmore
___________________ _____________________ _________________
Name Signature Date
40 Chapter 2: Construction cost estimation techniques
2.2.1 Introduction
An accurate cost estimate cannot be achieved without a clear scope by the
client, a completed design by the designer and a sound effort estimate by the cost
engineer. However, time, money and professional skills limit the performance of all
the stakeholders - a reality that motivates the setting of contract contingencies, which
aim to provide sufficient reserve money to cover the cost of mistakes and risks in
future. Other than flexibility in specification, float schedule, and some other
arrangements, the contingencies discussed in this research specifically refer to the
money put aside by the client for “known” “unknown” risks and “unknown”
“unknown” risks.
The determination of suitable contingencies is challenging to the client or
consultant. If the amount of contingencies is set too high, the unused money will be
wasted. If this amount is set too low, it will be insufficient for compensating
potential risks and may even hinder construction progress. The ideal amount of
contingencies should be close to the amount of project cost overrun. To obtain
accurate estimates of typical contingency values, a number of estimating techniques
such as the floor area method, percentage method, regression, artificial neural
networks, and Monte Carlo simulation are available (Mak and Picken 2000; Idrus et
al. 2011). Many early cost drivers can be considered in these methods to generate
representative values. However, the inconsistency of these “input” variables
undermines the possibility of understanding their impact.
This research aims to identify the early cost drivers of contingency values and
explore their impact via path analysis modelling. 133 UK school building contracts
with contingency values are used as empirical cases. Gross floor area (GFA),
proposed schedule and presence of air conditioning (AC) are used as independent
variables. The Soble test shows the proposed schedule have a mediating effect. The
total effects of GFA, schedule and AC are then calculated. This squared multiple
correlation (SMC) is 0.624, indicating that the identified three variables explain
62.4% variance of contingencies - a comparatively satisfactory result considering the
unknown estimating techniques and the different projects involved - and suggests
that contingency setters to be quite homogeneous.
Chapter 2: Construction cost estimation techniques 41
2.2.2 Early cost drivers
Estimating in early pre-design stage is often inaccurate due to the limited and
vague information available. It is estimated that inaccuracy at this time is around -
40% to +20% (Barnes 1974; Skitmore 1987). Flyvbjerg et al (2003) analysed data
from 258 transportation infrastructure projects worth US$90 billion and found that
cost overrun in nine out of ten projects were caused by inaccurate estimation in the
early stage. Similarly, it is reported that inaccuracy of such estimates is around 30%
in Germany, and it is mainly caused by simply multiplying floor area with an certain
ratio, neglecting other cost drivers (Stoy and Schalcher 2007). A quick and accurate
early estimate is important to the decision-making of clients. To achieve this, a
review of previous studies on relevant building cost drivers at early stage is
necessary. The impacts of these drivers on building cost components should be also
examined, which motivates this research. Contingencies are selected as the
dependent variable in this research for its sensitivity to early project information.
Skitmore (1987) proposed that building prices should be seen as a result of a
series of interdependent causal mechanisms and identified building type, size,
complexity and quality, type of client, contractor selection, contractual arrangements,
location, and economic, legal environment of project location as primary cost
drivers. By analysing empirical cases from Singapore, Gunner and Skitmore (1999)
found three variables, i.e. floor area, number of stories above ground and contract
period, to have comparatively high correlations with contract sums. Li et al. (2005)
constructed regression models to estimate the building cost for office buildings in
Hong Kong using seven variables in modelling, i.e. average floor area, total floor
area, average storey height, number of above-ground stories, total building height,
number of basements and completion year. Total floor area, total building height, and
average floor area were found to be the most important ones. For many cases where
it is hard to know building heights at the early stage, Stoy and Schalcher (2007)
recommended the situation with or without air conditioning systems is an indicator
(Stoy and Schalcher 2007). Client type (public or private) also significantly affects
bid decisions of contractors according to a wide questionnaire survey conducted in
China (Ye et al.).
According to the findings from literature review and information accessible at
early project stage, nine early cost drivers are used to establish the initial model as
42 Chapter 2: Construction cost estimation techniques
presented in Figure 2.2. Differences due to location, contract period and inflation rate
are compensated by rebasing all 133 cases with corresponding cost indices to make
them comparable.
Figure 2.2 The initial model
2.2.3 Research method
This section describes the advantages, application and measurement indices of
path analysis modelling. Then descriptions of selected cost drivers and sample data
are presented.
Path analysis
Path analysis (PA) is the original structural equation modelling (SEM)
technique, which is widely used to explore and test causal relationships in social
science, such as in psychology, education and health (Kline 2010). SEM normally
describes the relationships between two kinds of variables, i.e. latent and observed.
Latent variables cannot be observed directly due to their abstract character. In
contrast, observed variables contain objective facts or use an item rating scale in a
questionnaire. Several observed variables can reflect one latent variable. Compared
with other multivariate analysis methods, SEM has the ability to estimate multiple
Chapter 2: Construction cost estimation techniques 43
and interrelated relationships and define a model to explain these relationships well
(Kline 2010; Xiong et al. 2013).
PA is a comparatively simple SEM form for analysing structural models with
observed variables only. Due to resource limitations and situation constraints, it is
not always possible for several variables to reflect one and this is why PA is still
widely used (Kline 2010). For example, path analysis accounts for 25% of around
500 applications of SEM published in 16 psychology journals from 1993-1997
(MacCallum and Austin 2000). There are also increasing uses of path analysis to
explore construction-related issues. Brown et al (2007) construct a path analysis
model to explore the relationship between human capital and time performance in
project management and found that performance improves with increase investment
in human capital (Brown et al. 2007). Zhang and Fang (2013) apply path analysis to
explore the cognitive reasons of Chinese scaffolders’ unwillingness to use harnesses
in work and built a path analysis model to explain data collected from questionnaire.
The proposed cost drivers in this paper can be observed directly or transformed in
understandable ways. Therefore, path analysis is suitable to explore the impact of
cost drivers on contingency values. The software AMOS is used to do the modelling.
An ideal model should be theoretically sensible and fit the sample data well.
Measuring goodness of fit is an essential task preceding path analyses. Many criteria
have been generated for this purpose. The overall measurement is the probability
level of the Chi-square test, if the p value is 0.05 or less, the departure of the data
from the proposed model is significant at the 0.05 level, i.e. the proposed model is
not significantly consistent with the observed data. However, the Chi-square test has
some severe flaws, such as sensitivity to violations of the assumption of multivariate
normality, model complexity and sample size (Finney and DiStefano 2006; Lei and
Wu 2007). Other fit indices thus have been developed to judge models from other
three perspectives. These include absolute fit, incremental fit (comparative fit) and
parsimonious fit (Xiong et al. 2013). The commonly used indices are presented in
Table 6.10. It is worth mentioning that there are no commonly agreed thresholds for
the listed parsimony indices, which are used mainly to choose the most parsimonious
of several acceptable, but similar, models. For this research, they are used to choose
between the next to last model (Figure 6.2) and the final model (Figure 6.3).
44 Chapter 2: Construction cost estimation techniques
Variables and Data
Ten variables (see Table 2.6) are used in the analysis. Due to the lack of
information at in the early design stage, Schedules are those stipulated in invitations-
to-bid by clients. These may be different from schedules proposed by contractors
and agreed schedules in contracts. Ground condition, work space and site access
reflect the quality of site conditions. Market condition describes the demand and
supply condition or market competition of the local construction industry and at a
certain time. Public/Private refers to the client type. To be comparable, the
contingencies of 133 cases are all rebased to a common level by applying cost
indices. All the variables except air conditioning and client type (public/private) are
zero-mean normalized (Z score) to be comparable in path analysis (Brown et al.
2007). For the sample size, there is a rule of thumb to measure that the ratio of
sample size to number of items tested should be more than 5, and the higher the
better in the range 5-20 (Kline 2010; Lei and Wu 2007). Therefore the sample size is
satisfactory. The contractor selection method of selected cases is limited to the open
competition/selected competition other than negotiation, since contract selection was
identified as a factor affecting construction time and cost (Skitmore and Ng 2003).
Table 2.6 Description of variables
Gross floor area (m2) Range from 105-13835; mean=1522.29; SD=1980.83
Schedule (months) Range from 4-32; mean=11.00; SD=4.40
Stories Range from 1-4; mean=1.47; SD=0.61
Ground condition
Bad (=1), moderate (=3), good (=5); Mean=3.59;
SD=1.37
Work space
Highly restricted (=1), restricted (=3), unrestricted (=5);
mean=3.84; SD=1.26
Site access
Highly restricted (=1), restricted(=3), unrestricted(=5);
mean=3.67; SD=1.27
Market condition
Low competitive (=1), less competitive (=2), average
(=3), competitive (=4), highly competitive (=5);
mean=4.14; SD=0.68
Air Conditioning Yes=1; No=0; 115 cases are 0; 19 cases are 1
Public/Private Public=1; Private=0; 48 cases are 0; 85 cases are 1
Contingencies (£)
Range from 5605-530415; mean=91136.69;
SD=98833.71
Chapter 2: Construction cost estimation techniques 45
2.2.4 Path analysis modelling
An initial path model is developed as shown in Figure 6.1 and the model fit is
presented in Table 2.7. This model performs badly as many insignificant variables
are incorporated and some necessary relationships are ignored. After a series of
corrections, the next to last model is achieved (see Figure 2.4). In this model the
coefficient ( -0.015) of air conditioning (AC) schedule is not significant (at the
0.05 level). This relationship is thus deleted to obtain a more parsimonious model
and improvements are reflected in parsimony indices between the next to last model
(Figure 2.3) and the final model (Figure 26.4) as presented in Table 2.7. The
coefficients shown in Figure 2.2 and Figure 2.3 are the standardized estimates. **
means the p value of a highlighted coefficient is smaller than 0.01; *** means the p
value of a highlighted coefficient is smaller than 0.001. In the final model, the
squared multiple correlations (SMC) of contingencies (0.624) indicates the identified
three variables and relationships can explain 62.4% variance of contingencies, which
is regarded as satisfactory considering that all those settings of contingencies are
generated by different sources with unknown estimate techniques in different project
situations.
Figure 2.3 The next to last path analysis model
46 Chapter 2: Construction cost estimation techniques
Figure 2.4 Final path analysis model
Goodness of fit
In Table 2.7, both of the last two models show acceptable goodness of fit
according to the Chi-square test, absolute fit indices and incremental fit indices. The
parsimonious fit indices are the used for the final selection. Although all the three
indices used for testing parsimony do not have commonly agreed cutoffs, higher
PNFI, PGFI and smaller CAIC reflect better parsimony.
Table 2.7 Model fit indices
Goodness of fit
measure Criteria Initial model
The next to
last model
Final
model
Chi-square test
Probability level >0.05 0.000 0.628 0.862
Absolute fit
GFI >0.9 0.758 0.999 0.999
RMSEA <0.08 0.211 0.000 0.000
SRMR <0.05 0.185 0.019 0.023
Incremental fit
CFI >0.9 0.369 1.000 1.000
TLI >0.9 0.211 1.022 1.024
NFI >0.9 0.349 0.999 0.999
Parsimonious fit
PNFI Higher 0.279 0.166 0.333
PGFI Higher 0.496 0.100 0.200
CAIC Smaller 360.969 53.248 47.421
Chapter 2: Construction cost estimation techniques 47
Direct effects, indirect effects and total effects
As seen from Figure 2.3 and Figure 2.4, the GFA has an indirect impact on
contingencies through schedule (GFAScheduleContingencies). In order to test
whether the mediating role of schedule is significant, the Sobel test is used to verify
the significance of mediation effects (Sobel 1982; Xiong et al. 2013).
The Sobel test statistic is 3.172 (p=0.002), which means the mediation effect of
GFAScheduleContingencies is significant at 0.05 level. The standardized direct,
indirect and total effects are shown in Table 2.8.
Table 2.8 Standardized direct, indirect and total effects of variables
Effects on Contingencies Gross floor area Schedule Air condition
Direct effects 0.579 0.250 0.150
Indirect effects 0.000 0.177 0.000
Total effects 0.579 0.427 0.150
2.2.5 Findings and discussions
The main research finding of the path analysis is that gross floor area,
schedule, air conditioning are the three most influential variables that can be used to
predict contingency values, and their total effects are presented in Table 2.8.
Gross floor area (GFA)
GFA is the most influential cost driver on determination of contingencies with
the total effect of 0.756. It is interesting to see that GFA has both a direct effect
(=0.579) and an indirect effect (=0.177) via schedule on contingencies. GFA also has
a major impact (=0.705) on schedule determination. The powerful role of GFA is
consistent with previous research findings and the wide use of GFA method in cost
estimation (Gunner and Skitmore 1999; Skitmore 1987; Stoy and Schalcher 2007).
The defect of using GFA ratio method can also be identified in this research.
The SMC of schedule is 0.498, which indicates GFA can explain 49.8% variance of
schedule. The unexplained variance part of schedule makes it inaccurate to use GFA
solely to predict schedule. Additionally, the unexplained variance of contingencies
and the existence of air conditioning inevitably lead to inaccurate prediction when
using GFA as the solo predictor to estimate contingencies.
48 Chapter 2: Construction cost estimation techniques
Schedule
Schedule has a direct effect (=0.250) on contingencies. The schedule also acts
as a mediator between GFA and contingencies as confirmed by the Sobel test. Cost
estimation and schedule estimation are sometimes highly correlated and may be
determined by similar factors (Skitmore and Ng 2003). For example, the GFA is a
common factor for both schedule and contingencies. It needs to be mentioned that
the schedule used in this research refers to the proposed schedule by the client at pre-
tender stage.
Air conditioning (AC)
The effect of air conditioning has been little explored in previous research.
Installing air treatment systems requires high ceiling superstructures with
comparatively high median floor height (Stoy and Schalcher 2007). By investigating
290 properties, Stoy and Schalcher (2007) found that the average height of projects
with AC was 0.11m higher than that of projects without AC. Considering the general
mild climate in UK, installing AC possibly also indicates higher requirement for
quality and higher degree of risk. The use of air conditioning in mild climates is an
interesting socioeconomic issue in its own right.
2.2.6 Conclusion
Of the variables path analysed, the three most influential in their effects on
contingency values of 133 UK school building contracts are GFA, Schedule and AC,
with Schedule acting in a mediating role. In explaining 62.4% of the variance, it is
demonstrated that consultants involved are quite homogeneous in their contingency
valuations.
Some limitations of this analysis should be mentioned. First, the sample cases
are school buildings in UK, thus it is not possible at this stage to generalize the
conclusions to other types of buildings or other countries. Another limitation is that,
although the analysis provides some insights into the considerations taken into
account in setting contingency values, the lack of data of project cost overruns
disallows any examination of the accuracy of the contingency values involved. That
is, we have gained some impression of how contingency values of obtained but,
although the indications are that this quite common among contingency setters, this
Chapter 2: Construction cost estimation techniques 49
does not shed any light on the extent to which the values are appropriate (cover the
unexpected additional building costs involved). Of course, it can be argued with
some justification that assigning higher values to contingencies may well attract
higher extra costs due to contractors knowing the amount of money been set aside
and therefore being more vociferous in claiming the extra costs, etc.
Future research would benefit from examining more closely the effects of the
personal characteristics of the contingency setters and incorporate information on
cost overruns for measuring the appropriateness of the contingency values made.
This may reveal some statistical patterns that provide insight into the underlying
phenomena at work.
50 Chapter 2: Construction cost estimation techniques
Chapter 3: Conceptual framework and structural equation modelling 51
Chapter 3: Conceptual framework and
structural equation modelling
3.1 TOWARDS A CONCEPTUAL FRAMEWORK OF JOB
PERFORMANCE
3.1.1 Introduction
With individual expertise critical to the success of projects and the company,
the performance of construction professionals is a key concern for both academics
and practitioners alike (Ahadzie, Proverbs, & Olomolaiye, 2008a; Leung,
Olomolaiye, Chong, & Lam, 2005), since construction projects have become
increasingly complex in recent decades (Xia & Chan, 2012). Factors such as working
environment, individual personality, job knowledge, working experience and
psychological reactions have been identified as predictors of job performance in a
few explorative studies (Leung, Liu, & Wong, 2006; Pheng & Chuan, 2006).
However, a conceptual model to reveal the mechanism of job performance is needed
to link these developed concepts with theoretical foundations.
In the area of human resource management and organisational psychology, the
person-environment (P-E) fit is a concept widely used (Schneider, 2001). In
personnel selection, managers favour candidates who share similar values to those of
the company and have the specific skills needed to fit in well (Greguras &
Diefendorff, 2009). For current employees, a P-E misfit may result in increased staff
turnover (Westerman & Cyr, 2004). Although a few studies (such as Xiong,
Skitmore, and Xia, 2015b) point out that employee behaviour is affected by
psychological reactions encountered in specific situations, P-E fit theory has been
little used in research into the job performance of construction professionals. This
could be attributed to the criticism that directly measuring the discrepancy between
the commensurate constructs of P and E is not easy and faces several conceptual
barriers (Kristof, 1996; Schneider, 2001). After a thorough review of studies on P-E
fit (Caplan, 1987; Chuang, Hsu, Wang, & Judge, 2015), this paper applies indirect
measures of P-E fit comprising job satisfaction, work performance and organisational
commitment as extrinsic assessments.
52 Chapter 3: Conceptual framework and structural equation modelling
Being associated with the stimulus-organism-response (S-O-R) paradigm, the
two psychological reactions in terms of P-E assessments are used to connect
environmental stimulus and individual characteristics with job performance. The
model augments previous conceptual frameworks that link environmental factors or
psychological reactions with job performance without mediators. The mediation role
of P-E fit assessments identified in this model contributes to knowledge of the job
outcomes of construction professionals. Therefore, the proposed framework aims to
reveal a bigger picture for job performance and its antecedents, thereby further
increasing the effectiveness of human resource management practices. A further
objective is to propose a research agenda based on the proposed model for the benefit
of future studies.
3.1.2 Development of the conceptual framework
Many theories, such as human resource management, personality, competency,
motivation, self-determination, work adjustment and P-E fit, have been used to
explain the job performance of employees (Greguras & Diefendorff, 2009;
Schneider, 2001). This section intends to develop a framework for examining
organisational behaviour and human resource management by adjusting assessments
of person-environment fit within the stimulus-organism-response (S-O-R) paradigm
of human behaviour.
Person-environment fit theory
Behaviour was early considered as a function of person and environment by
Lewin (1935). P-E fit has been the dominant theory used to address various issues in
psychology, such as personnel selection, vocational psychology and social
psychology (Schneider, 2001). With the increasing awareness of psychological
illness caused by work stress in the 1980s, P-E fit theory was used by many
researchers (Caplan 1987); Edwards, 1996) to derive findings in related studies
(Edwards, Caplan, & Van Harrison, 1998; Xiong, et al., 2015b). Fit is also
emphasised in personnel selection, in that the knowledge, skills, ability and
personality of an individual should match the criteria of a specific job. As Bretz and
Judge (1994) point out, P-E fit is a direct predictor of career success, and recruiting
individuals with a better fit leads to a more satisfied work force. In addition to the
Chapter 3: Conceptual framework and structural equation modelling 53
effect on willingness to join organisations, perceptions of P-E fit are found to be
critical to turnover intentions through the mediation of attitudes such as job
satisfaction and organisational commitment (Westerman & Cyr, 2004). Work
performance and organisational outcomes are two long-term outcomes attributed to
P-E fit (Kristof, 1996)
Several distinctions are proposed to explore the multi-dimensional P-E fit
concept. Muchinsky and Monahan (1987), for example, propose the notion of
supplementary and complementary fit. Supplementary fit means that an individual
fits the environment by sharing similar characteristics with those that exist in the
environment. Based on the psychological paradigm of similarity-attraction, it is
common for a person to assess the similarity of his/her values and attitudes with the
organizational climate and values of a company (Kristof, 1996). Some sub-themes of
P-E fit, such as person-person fit and person-group fit, focus on supplementary fit
(Chuang et al., 2015). Complementary fit, in contrast, is developed based on the
psychological paradigm of needs-fulfilment, which occurs when a person can
provide something needed by the environment (e.g. organisation) or vice versa
(Kristof, 1996; Muchinsky & Monahan, 1987). “Current themes of PE fit that follow
the concept of complementary fit include demands-abilities (D-A) fit and needs-
supplies (N-S) fit” (Chuang et al., 2015, p. 482). As indicated in Kristof’s (1996)
literature review, empirical studies of work performance are mostly linked to
complementary fit. Therefore, the complementary fit is used to represent the P-E fit
in the development of the conceptual framework.
The measurement constructs for P-E fit have been a key concern, since
findings vary for different methods (Spokane, 1987). In the theory of work
adjustment, it is argued that independent commensurate measures of people and
environment are desirable (Spokane, 1987). Difference scores, such as the algebraic
form (X-Y), absolute form (|X-Y|) and squared differences (X-Y)2 are used as direct
measures in many studies, with the assumption that a lower discrepancy between P
and D results in better outcomes (Kristof, 1996; Rounds, Dawis, & Lofquist, 1987).
However, direct measures face several criticisms. One is that the independent effects
of P and E are hard to explore when two constructs are confounded (Kristof, 1996).
Direct measures are believed to be a violation of the Lewinian conceptualisation of
“constellation”, in the definition of behaviour as a constellation of person and
54 Chapter 3: Conceptual framework and structural equation modelling
environment, and also to be inappropriate for anthropomorphising environments
(Schneider, 2001). Therefore, alternative measures of assessing the person and
environment simultaneously across multiple dimensions of P-E fit are necessary. As
pointed out by Kristof (1996), the perception of P-E fit in organisational situations
may have a stronger influence on variables such as stress, satisfaction and
commitment, than does fit itself. Since P-E fit is the implicit key to understanding
human behaviour (Schneider, 2001), it is important to identify extrinsic measures of
P-E fit.
Psychological reactions including job satisfaction and work stress are used as
measurable assessments of two P-E fit types —needs-supplies (N-S) fit, demands-
abilities (D-A) fit— as presented in Figure 3.1 (Giauque, Resenterra, and Siggen,
2014; Kristof, 1996). Job satisfaction is a response to the discrepancy between ‘How
much is there?’ and ‘How much should there be?' (Nerkar, McGrath, & MacMillan,
1996; Wanous & Lawler, 1972a), and is demonstrated to be an assessment of "needs-
supplies" fit (Pervin, 1987; Rounds, et al., 1987). Work stress is a reaction to the
deviation between the requirements and actual abilities of employees in fulfilling job
tasks (Tennant, 2001) and therefore an assessment of “demands-abilities” fit. As
demonstrated by Edwards (1996), D-A fit is critically linked to tension and N-S fit to
satisfaction. Similarly, Leung, Chan, and Yuen (2010) used items of D-A
discrepancy of construction works to measure the level of work stress.
Figure 3.1 Main P-E fits and psychological reactions
Chapter 3: Conceptual framework and structural equation modelling 55
Conceptual framework
Unlike the Stimulus-Response mechanism dominating most animals, human
behaviours applying judgement and analytical ability are usually regarded as
following the mechanism of Stimulus-Organism-Response (S-O-R) (Mehrabian &
Russell, 1974). Although some studies on the job performance of construction
professionals explore the direct effects of environmental factors, few use P-E fit
assessments as mediators. Based on the S-O-R paradigm and P-E fit theory, an
adapted conceptual model described as Stimulus-Reactions-Performance is
developed for studying employee behaviour, as shown in Figure 2.2. The postulation
here is that environmental factors and individual factors affect job performance
(fully/partially) as mediated by P-E fit assessments. In the time perspective, it is
reasonable to assume job performance may affect environmental factors and P-E fit
assessments in future. As noted by Spokane (1987), reciprocal relationships may
exist between reinforcers in the work environment and the needs of the individual.
Figure 3.2 Proposed conceptual framework
Psychological reactions as the P-E fit assessments
As pointed out by Caplan (1987), there are two basic assessments of P-E fit
when exploring influences: “one involving the fit between environmental supplies
and personal motives, goals and values and the other involving the fit between
environmental demands and personal skills and abilities” (Caplan, 1987, pp. 295-
56 Chapter 3: Conceptual framework and structural equation modelling
296). Job satisfaction and work stress can be used to assess these two kinds of P-E
fit.
The assumption that “happier workers produce more” can be dated back to the
Hawthorne studies and the human relations movement of the 1930s (Brayfield &
Crockett, 1955), whereas work stress received little attention until there was
increased recognition and study of mental disorders in the 1980s (Tennant, 2001).
Since then, the nexus between work stress and employee behaviour has been
increasingly studied. Following the Hawthorne studies of job satisfaction among
employees, research into the possible connections between job satisfaction and job
performance comprises an appreciable portion of behaviour research in management
(Organ, 1988b). Three mainstream hypotheses of the job S-P linkage include: (1) job
satisfaction causes job performance; (2) job performance causes job satisfaction; (3)
another complex relationship exists that includes moderators, mediators or
antecedent variables. For potential antecedents, job satisfaction is positively related
to organisational learning climate (Egan, et al., 2004). As noted by Tett and Meyer
(1993), job satisfaction is a strong predictor of organisational commitment and
employee turnover. A meta-analysis of 55 studies of OCB supports job attitudes and
job satisfaction as robust predictors of OCB (Organ & Ryan, 1995).
Work stress, indicating the deviation between requirements and actual abilities
of people in fulfilling job tasks, has become an important concept in organisational
management for the prevalence of psychological disorders (Tennant, 2001). In
addition to health issues related to work stress such as diastolic blood pressure under
stressful working conditions (Matthews, Cottington, Talbott, Kuller, and Siegel
(1987), exploring the antecedents and influences of work stress in the managerial
context has practical and theoretical implications. For example, according to an
online survey of 306 nurses, including 263 American hospital nurses and 40 non-
American nurses, social support from co-workers decreases job stress and improves
job performance (AbuAlRub, 2004). Additionally, in another survey of 305 Chinese
employees in 48 service organisations, co-worker support is found to be a significant
moderator for the nexus between job stress and performance, in that higher stress
results in better performance if the level of co-worker support is high (Hon, 2013).
Leung, et al. (2005) use work stress to predict the estimated performance (task
performance) of construction cost engineers in Hong Kong.
Chapter 3: Conceptual framework and structural equation modelling 57
3.1.3 Discussion
The proposed framework links psychological reactions and job performance
developed in the S-O-R paradigm and P-E fit theory. This model provides a rich
framework for understanding the ‘big picture’ of employee behaviour research. The
framework could also be useful for identifying corrective approaches such as
developing organisational support to improve employee performance.
Consistent with the tradition of focusing on individual differences in the
selection of employees (Schneider, 2001), individual characteristics — especially
knowledge, skills and ability — are emphasised. For example, Hunter (1986)
reviewed hundreds of papers measuring relationships between general cognitive
ability and job performance in various jobs, and found that cognitive ability affects
job performance through daily-used job knowledge and skills. Wade and Parent
(2002) found that a deficiency of job skills leads to lower job performance of
webmasters, in their analyses of a worldwide survey with 232 responses. Dilchert,
Ones, Davis, and Rostow (2007), in an analysis of 3,021 applicants for the police
force in the US, found that individual cognitive ability negatively affects
counterproductive work behaviour (CWB) and that workers with higher cognitive
ability consider before engaging in counterproductive activities. Meier and Spector’s
(2013) longitudinal study of 663 employees in the US over an eight month period
found a reciprocal nexus between stressful working conditions and CWB.
Based on the organisational psychology tradition, especially of the Stimulus-
Organism-Response (S-O-R) paradigm (Mehrabian & Russell, 1974; Schneider,
2001), the proposed model includes the effects of environmental factors. As
Schneider (2001) points out, many studies of P-E fit are narrowly focused on
identifying commensurate measures for P and E. By using three assessments as
indirect measures to reflect P-E fit, therefore, the proposed model is able to examine
effects of organisational factors such as organisational support, organisational
politics and organisational learning climate. For instance, Smith et al. (1983) have
identified a positive relationship between job satisfaction and altruistic behaviour in a
survey of 422 employees and their supervisors in two banks in the US. Eisenberger et
al. (1986) incorporated commitment items into a US Survey of Perceived
Organizational Support, and their analysis of 361 responses found individual
58 Chapter 3: Conceptual framework and structural equation modelling
absenteeism to be negatively correlated with organisational support. Rhoades and
Eisenberger’s (2002) meta-analysis of 70 studies related to organisational support
shows that employees care for beneficial organisational supports such as fairness,
supervisor support, organisational rewards and enjoyable job conditions. Further, and
consistent with the norm of reciprocity, employees are likely to establish a long term
approach to social exchanges by paying back with hard work and job loyalty
(Wayne, Shore, & Liden, 1997).
According to Ferris and Kacmar (1992), the political nature of the working
environment is not a concept but a fact of life. A business company is a political
coalition where decisions are not totally decided by the market but also by bargaining
processes (March, 1962). Perceptions of organisational politics are caused by the
employees’ tendency to assign humanlike characteristics to organisations
(Eisenberger, Huntington, Hutchison, & Sowa, 1986; Rhoades & Eisenberger, 2002).
Although more political behaviour happens in the higher levels of organisations
(Ferris & Kacmar, 1992), lower level employees perceive more impact from their
lack of control of organisational processes, which in turn decreases their job
satisfaction (Gandz & Murray, 1980).
Consistent with the general definition of organisational climate (Hellriegel &
Slocum, 1974), organisational learning climate (OLC) can be regarded as involving a
set of attributes related to the learning of members in an organisation. The effects of
OLC on organisational performance have been largely acknowledged by both
academics and practitioners (Mikkelsen & Grønhaug, 1999). Additionally, OLC is
believed to improve organisational learning when an individual or group of
individuals in an organisation face problems and need help from ‘the organisation’
(Argyris & Schön, 1978). Egan et al.’s (2004) examination of the relationships
among OLC, job satisfaction and organisational performance also found that OLC is
positively related to job satisfaction and intentions to transfer knowledge among
employees, while turnover intention is negatively influenced by OLC and job
satisfaction.
In addition to the factors discussed above, potential moderators need to be
considered when solving complex and unsettled problems (Xiong et al., 2015a).
Other variables to be included in the model include gender, age, industry, country,
Chapter 3: Conceptual framework and structural equation modelling 59
culture, job alternatives in the market, reward contingency and individual learning
style.
Building on previous studies, the proposed model assumes that psychological
reactions act as variables in which the effects of environmental and individual factors
on job performance are fully or partially mediated. The relationships between
psychological reactions and job performance are major concerns in this thesis. Two
specific propositions are proposed: (1) job satisfaction, an indicator of N-S fit,
positively affects job performance; and (2) work stress, an indicator of D-A fit,
negatively affects job performance. The exact relationship might be in the form of a
reverse U shape.
3.1.4 Conclusions
The job performance of construction professionals is a product of the
interactions between person and environment. In addition to objective environmental
factors and individual differences, psychological reactions (defined as P-E fit
assessments in this study) are also critical. From a cross-sectional perspective, the
study assumes that stimulus factors in the environment and individual differences
affect job performance via the mediation effects of P-E fit assessments. When time
lags (seasons, years) are considered, job performance may influence future
perceptions of external stimulus factors and P-E fit assessments.
The proposed conceptual framework can be used both to understand previous
studies and to underpin future studies. Based on P-E fit theory and the S-O-R
paradigm, the framework can be used as a reference in avoiding pseudo-causation
conclusions. Although further refinements are inevitable, the proposed framework
also contributes to: expansion to other outcomes, such as turnover intention, and
adaptation with new concepts; identifying potential moderator in these relations..
60 Chapter 3: Conceptual framework and structural equation modelling
3.2 STRUCTURAL EQUATION MODELLING
Statement of contribution
The authors listed below have certified that:
1. They meet the criteria for authorship in that they have participated in the
conception, execution, or interpretation, of at least that part of the publication in their
field of expertise;
2. They take public responsibility for their part of the publication, except for
the responsible author who accepts overall responsibility for the publication;
3. There are no other authors of the publication according to these criteria;
4. Potential conflicts of interest have been disclosed to (a) granting bodies, (b)
the editor or publisher of journals or other publications, and (c) the head of the
responsible academic unit, and
5. They agree to the use of the publication in the student’s thesis and its
publication on the Australasian Research Online database consistent with any
limitations set by publisher requirements.
In the case of this chapter:
Bo Xiong*, Martin Skitmore, Bo Xia. A critical review of structural equation
modeling applications in construction research, Automation in Construction, 2015,
49 (Part A), 59-70.
Contributor Statement of contribution
Bo Xiong Searched previous studies which applied SEM in construction
research, conducted a critical review, made suggestions for future
research, wrote the manuscript and acted as the corresponding
author.
07/03/2016
Martin Skitmore Directed and guided this study, and proofread the manuscript.
Bo Xia Directed and guided this study.
Principal Supervisor Confirmation
I have sighted email or other correspondence from all Co-authors confirming their
certifying authorship.
Chapter 3: Conceptual framework and structural equation modelling 61
Martin Skitmore
___________________ _____________________ _________________
Name Signature Date
62 Chapter 3: Conceptual framework and structural equation modelling
3.2.1 Introduction
Since Bentler's appeal to apply the technique to handle latent variables (i.e.
unobserved variables) in psychological science Bentler (1980), structural equation
modelling (SEM) has become a quasi-routine and even indispensable statistical
analysis approach in the social sciences. Computer programs designed for
conducting SEM analyses have emerged and enabled the technique to be used in
even wider applications (Baumgartner & Homburg, 1996). Newly developed
graphical user interfaces have also made much easier for researchers and
practitioners to use (Kline, 1998).
On one hand, the utility of SEM in approximating reasonable results in
measurement and structural analyses has been widely acknowledged (Bagozzi & Yi,
2012; Bentler, 1980; Byrne, 2001; Hair, 2006; Jöreskog & Sörbom, 1996). On the
other hand, SEM has been criticized for generating implausible conclusions due to its
indiscriminate use (Baumgartner & Homburg, 1996). Some results obtained through
SEM are of doubtful authenticity, especially when both researchers and reviewers
have little experience with the method. The overall quality of SEM applications in
construction research is similarly affected. Many mistakes exist in current
publications and basic principles are often violated or ignored.
Despite the special care needed in SEM applications, no explicit body of
knowledge has been developed for their use in construction research to assess the
proposed models, and errors continue to be made over assumptions and
interpretations. The purpose of this paper, therefore, is to provide a comprehensive
and critical review of SEM applications in construction research to date, through the
evaluation of previous applications of SEM to solving related research problems
including, but not limited to, papers published in leading construction journals. The
review focuses on the practical use of the SEM technique and analyses the
applications in terms of model design, model development and model evaluation
issues for the benefit of future research.
Chapter 3: Conceptual framework and structural equation modelling 63
3.2.2 Methodology
Introduction to SEM
The emergence and development of SEM was regarded as an important
statistical development in social sciences in recent decades and this “second
generation” multivariate analysis method has been widely applied in theoretical
explorations and empirical validations in many disciplines (Fornell & Larcker,
1981a; Kline, 2005). Compared with other statistical tools such as factor analysis and
multivariate regression, SEM carries out factor analysis and path analysis
simultaneously (Xiong, Skitmore, Xia, et al., 2014), since it can (1) measure and
accommodate errors of manifest variables (i.e. observed variables); (2) represent
ambiguous constructs in the form of latent variables (i.e. unobserved variables) by
using several manifest variables; and (3) simultaneously estimate both causal
relationships among latent variables and manifest variables (Kline, 2005; Xiong,
Skitmore, Xia, et al., 2014). In addition, SEM can also provide group comparisons
with a holistic model, resulting in much more vivid impressions than traditional
ANOVA. SEM can also handle longitudinal designs when time lag variables are
involved (Gollob & Reichardt, 1987; MacCallum & Austin, 2000).
As introduced above, SEM describes and tests relationships between two kinds
of variables - latent variables (LVs) and manifest variables (MVs). Latent variables
cannot be observed directly due to their abstract character. In contrast, observed
variables contain objective facts and easier to measure. Several observed variables
can reflect one latent variable (Byrne, 2001). As presented in Fig.1, a structural
equation model usually consists of two main components, a structural model and
several measurement models. A simple measurement model includes a latent
variable, a few associated observed variables and their corresponding measurement
errors. The structural model consists of all LVs and their interrelationships. For
model development purposes, some researches aim to validate their assumptions of a
dimensional framework of one or several discriminant LVs (e.g. Ding and Ng
(2007)), while others aim to elicit the causal relationship between the LVs.
Confirmatory factor analysis (CFA) with correlating latent variables satisfies the
former purpose, while these correlations need to be replaced by directional
relationships for the latter (Kline, 2005; Xiong, Skitmore, Xia, et al., 2014).
64 Chapter 3: Conceptual framework and structural equation modelling
Figure 3.3 provides a simple example of a structural equation model
investigating the effect of LV Y1 on LV Y2, and where several MVs are used to
represent the LVs. The MVs are shown in rectangles, the LVs in ellipses,
measurement errors in circles and with arrows indicating the direction of the effects.
If directional arrow between Y1 and Y2 is replaced by a correlation two-way arrow,
the model is a CFA and its purpose is to test whether MVs can represent LVs well
(i.e. convergent validity) and whether Y1 and Y2 are different (i.e. discriminant
validity). The basic concepts and principles of SEM are now well established with
the help of early explorations by researchers in the 1980s (e.g. (Bagozzi & Yi, 1988;
Baron & Kenny, 1986; Bentler, 1980; Bentler & Chou, 1987; Fornell & Larcker,
1981a; Mulaik et al., 1989)), structured textbooks (e.g. Byrne (2001); Kline (2005)),
well developed soft programs (e.g. LISREL by Jöreskog (1970), EQS by Bentler
(1989) and AMOS by Arbuckle (1994)), and Structural Equation Modeling, the first
ranked journal for mathematical methods, in publication since 1994 (Golob, 2003).
These are rich sources for beginners to acquire the basic knowledge needed before
applying SEM.
Figure 3.3 Schematic diagram of a structural equation model
The use of SEM in construction research is relatively new, with the early work
by Sarkar et al, published in the Journal of International Management (Sarkar,
Aulakh, & Cavusgil, 1998), in their examination of the mediation effects of relational
bonding between variables such as role clarity and the collaborative behavioural
processes of global construction firms. Another early work is Molenaar et al's
examination of the effects of a range of factors on contract disputes between owners
and contractors (Molenaar, Washington, & Diekmann, 2000), published in the
Chapter 3: Conceptual framework and structural equation modelling 65
Journal of Construction Engineering and Management. In both cases, SEM helped to
deepen the understanding of traditional research topics. SEM has also proved to be a
helpful tool in some emerging research areas. Lee and Yu, for example use SEM to
examine the effects of three antecedent variables on the intention to use the Project
Management Information System and user satisfaction, and the effect on construction
management efficiency (Lee & Yu, 2012), while Yang et al apply SEM to assess the
impact of information technology on project success, finding that project
performance is not affected directly but through the mediation role of knowledge
management (Yang, Chen, & Wang, 2012). Son, Park, Kim, and Chou (2012)
applied SEM to measure the acceptance and usage of mobile computing devices
among construction professionals in South Korea and Park et al. investigated the
effects of selected antecedent variables such as organizational support for
construction professionals' acceptance of web-based training (Park, Son, & Kim,
2012).
Article selection
Many previous review papers (e.g.Baumgartner and Homburg (1996);
(MacCallum & Austin, 2000; Sunindijo & Zou, 2012)) focus on analysing
publications in leading journals in their specific research fields, such as marketing.
However, research in construction can be seen as a combination of multiple
disciplines covering both technical and managerial topics. Therefore, this review
provides a comprehensive search of quality SEM applications for solving problems
in construction. Although it is an obvious option to use academic databases, none of
these is fully inclusive. Elsevier’s Scopus, for example, while they publish
AUTCON, IJPM and B&E, JCEM and JME are from the ASCE library, CME from
Taylor& Francis, and ECAM from Emerald.
To achieve a comprehensive search, the Google Scholar was used as the first
stage. According to a recently published analysis in Science, Nicolás Robinson-
Garcia, a bibliographer at the University of Granada in Spain said that “Google
Scholar's compendium of articles is at least as comprehensive as the leading
commercial academic search databases Thomson Reuters’ Web of Science and
Elsevier's Scopus - and for many disciplines in the social sciences and humanities,
even better.” (Bohannon, 2014). Additionally, Harzing conducted a longitudinal
study of Google Scholar coverage between 2012 and 2013 of four disciplines in
66 Chapter 3: Conceptual framework and structural equation modelling
Chemistry and Physics concluded that Google Scholar has become suitable for
bibliometric research (Harzing, 2014). The oversell impression is that all leading
construction journals are included in a Google Scholar search.
Firstly, two key phrases “structural equation model” and “construction
industry” were used to search in Google Scholar. Admittedly, while the use of
“construction industry” rather than “construction” may exclude a few relevant
publications, the abstract and multiple meanings of “construction” make the search
results too broad. To reduce the risk of missing relevant publications, a series of
“research” searches without using the “construction industry” key phrase was
conducted directly in 31 journals. 532 records were initially found on 4 April 2013.
Each of these records were examined to identify articles where SEM was applied as
the main statistical tool, the problems targeted are construction related or involve
related subjects such as professionals/companies in the industry, and are from peer
reviewed journals to assure selection quality. The source journals of the articles
selected in this way were then searched directly.
Path analysis (PA) models are special cases of the SEM technique for
analysing structural models just with observed variables (Xiong & Xia, 2014).
Despite its comparatively simple form, PA still accounts for 25% of the roughly 500
applications of SEM published in 16 psychology journals between 1993 and 1997
(MacCallum & Austin, 2000). Partial least square path modelling, known as PLS-
SEM in some publications, is a “soft” and component-based modelling technique in
theoretical exploration involving less strict inherent model assumptions and biased
parameter estimates compared with traditional SEM (i.e. covariance-based SEM).
Their differences are similar to those of principal component analysis and factor
analysis. However, PLS path modelling is an appealing technique due to its
predictability with small sample sizes and non-normal data (Hair, Sarstedt, Pieper, &
Ringle, 2012). Although PA and PLS have their own uses as introduced above, the
traditional covariance-based and latent variables that contained SEM has had wide
applications and methodological advances over more than 30 years of development
(Ringle, Sarstedt, & Straub, 2012). Articles using PA and PLS are excluded in this
review - a common practice in similar reviews in other fields (e.g. Baumgartner and
Homburg (1996); (Hair, Sarstedt, et al., 2012)). Finally, 84 suitable articles published
Chapter 3: Conceptual framework and structural equation modelling 67
during 1998-2012 were identified as satisfying the selection criteria. The selection
process is illustrated in Figure 3.4.
Outline the research design (e.g., quantitative, qualitative). If quantitative, spell
out the independent, dependent and classificatory variables (and sometimes
formulate an operational statement of the research hypothesis in null form so as to set
the stage for an appropriate research design permitting statistical inferences). If
qualitative, explain and support the approach taken and briefly discuss the data
gathering procedures that were [will be] used (observations, interviews, etc.)
Figure 3.4 Article selection
Unit of analysis
In the situation where several models are presented in one article, the models
selected for analyses were based on similar criteria to those of Shah and Goldstein.
68 Chapter 3: Conceptual framework and structural equation modelling
That is: (1) when the initial model and other alternative models are evaluated
simultaneously, only the final model is included in the analysis; (2) when a single
model is evaluated by splitting a sample, only the model tested with the verification
sample is included (Shah & Goldstein, 2006); and (3) when parallel constructs are
evaluated separately as confirmatory factor analyses, only the model with best
goodness of fit is included. In this way, only one model was selected for analysis
from each article. This process resulted in 84 models, of which 7 are Confirmatory
Factor Analysis (CFA) models and 77 are SEM models. The CFA models were
mainly used for validation of existing or newly developed frameworks, while the
SEM models were mainly used for exploring the interrelationships among latent
variables. If the objective and main contributions of one article is validation with
CFA, only the final CFA model was selected for analysis, as is the case with Ding
and Ng, for example, in their testing of the reliability and validity of the Chinese
version of McAllister's trust scale (Ding & Ng, 2007).
Overview and trend
7 of the 31 journals are regarded as key journals in this review and specially
marked in Figure 3.5, which shows the increase in the frequency of SEM application-
based articles in 3-year periods. To assess the growth of SEM applications, the
number of construction management articles were regressed on an index of
publication years (yearly from 1998), considering both the linear and quadratic
effects of time. The regression model is highly significant (F(2,12) = 34.6,
p=1.04*10-5<0.0001) and, with R2 =0.852, explains 85.2% of the variance of SEM
applications. The linear trend (t= -2.61, p=0.02) and quadratic effect (t=2.62, p=0.02)
are both significant, simultaneously growing more negative linearly and accelerating
positively over time. In comparison, SEM applications in marketing and psychology
grew linearly over time without acceleration (Baumgartner & Homburg, 1996;
Hershberger, 2003), while applications in operations management did not grow
linearly but accelerated over time. This research aims to enhance the suitability of
future applications by taking a critical review of current applications.
Chapter 3: Conceptual framework and structural equation modelling 69
Figure 3.5 Number of SEM-based articles by journals and year
3.2.3 Critical issues in the application of SEM
Issues relating to research design
Research design: cross-sectional studies and longitudinal studies
An SEM cross-sectional study involves a system of variables and constructs at
a certain time point, while a longitudinal study is concerned with the
interrelationships between constructs over time (MacCallum & Austin, 2000). Cross-
sectional designs are common with SEM applications in psychology research
(MacCallum & Austin, 2000). Cross-sectional studies are often focused on
identifying directional relationships among variables. However, these “causal”
models may be not appropriate in situations where the variables involved are
continually changing, since they omit the values of the variables at prior times, the
effects of variables on themselves over time and time interval for these causal
relationships (Gollob & Reichardt, 1987). In such cases, therefore, it is necessary to
consider time lags in the research design. In other words, a longitudinal component is
needed.
70 Chapter 3: Conceptual framework and structural equation modelling
As MacCallum and Austin point out, there are two commonly applied
longitudinal designs in SEM with repeated data of the same observed variables. The
first type is sequential design, where different variables are measured on successive
occasions to explicate the interrelationships among variables over time. The second
type comprises what are known as ‘growth curve models’, where the interest is in
changes in the same variables over time. These two types of design are not mutually
exclusive (MacCallum & Austin, 2000).
Opportunities exist, therefore, for construction management SEM designs to be
enriched by the consideration of time lags. Longitudinal designs are also preferred to
cross-sectional designs in strict causal modelling in order to avoid potential halo
effects caused by neglected autoregressive influences. For example, the effects of
variable B at time 1 on itself at time 2 should be considered in investigating the
effect of variable A at time 1 on variable B at time 2 (Gollob & Reichardt, 1987).
83 of the 84 articles reviewed are cross-sectional designs. For example, Leung,
Zhang, et al. (2008), used a cross-sectional design in examining the effects of
organizational supports in cost estimation while Ahuja, Yang, Skitmore, and Shankar
(2010) used a cross-sectional design in examining the relationships between the
factors affecting the adoption of information communication technologies by small
and medium enterprises. 76 of the 83 cross-sectional studies reviewed are focused on
identifying directional relationships among variables. One article uses a combined
longitudinal design in describing the development of trust between cross-functional,
geographically distributed co-workers (Zolin, Hinds, Fruchter, & Levitt, 2004).
Model specifications: constructs, indicators and identification
An important and controversial issue that needs to be considered early in model
specification is the construct type of measurement models (Bagozzi & Yi, 2012).
There are two possible relationships between latent variables (LVs) and manifest
variables (MVs) in terms of reflective constructs and formative constructs in
measurement models. However, some studies have specification problems in that,
instead of correctly using formative constructs, they apply only reflective constructs
without considering any possible distinction between two model structures. For
example, Jarvis et al’s review of articles published in top-tier marketing journals
found 28% of constructs to be incorrectly specified. The main features of reflective
constructs are:
Chapter 3: Conceptual framework and structural equation modelling 71
1. the causal directions are from latent variables to manifest variables
2. changes in latent variables lead to changes in manifest variables
3. manifest variables can be exchanged or deleted without affecting theoretical
meaning of corresponding latent variables for covering same themes.
Formative constructs, however, have the corresponding features of:
1. the causal directions are from manifest variables to latent variables
2. changes in manifest variables lead to changes in latent variables
3. manifest variables cannot be exchanged or deleted without affecting
theoretical meaning of corresponding latent variables and are not necessary to share
common themes (Jarvis, MacKenzie, & Podsakoff, 2003).
Therefore, care is needed in specifying the constructs, since current covariance-
based SEM software such as LISREL, AMOS and EQS can only handle reflective
constructs. For dealing with formative constructs, a method such as partial least
square structural modelling is necessary (Hair, Sarstedt, et al., 2012).
Another issue, which concerns the research framework or questionnaire design
in some situations, is which manifest variables should be allocated to reflect a latent
variable. Allocating more manifest variables per latent variable leads to more distinct
sample moments for model identification but also more parameters to estimate,
increasing the required sample size. It is not necessary to have a larger MV:LV ratio
to achieve a better model fit. Adding more variables is inappropriate in some
situations, as less data for each variable leads to worse parameter estimates and away
from the “true model” (Posada & Buckley, 2004). Therefore, variable selection needs
to take into consideration the information available and the principle of parsimony. A
measurement model can only be identified with three or more manifest variables, and
Keline proposes a three-variable principle, where three manifest variables are used to
reflect a latent variable (Kline, 2005). However, many papers contain models with an
MV:LV ratio of less than 3. Shah and Goldstein’s review of operations management
applications found this to be the case for 33.6% (38 of 113) of the models
encountered (Shah & Goldstein, 2006).
Single indicator constructs using only one manifest variable to represent one
latent variable are only suitable when a manifest variable can perfectly represent a
72 Chapter 3: Conceptual framework and structural equation modelling
latent concept. As Ringle, et al. (2012) pointed out, using a single indicator is a risky
choice as it performs worse than multi-item scales in most situations. Model
identification is also important for successful modelling. An obvious inherent feature
of identification is that there must always be a positive difference between the
number of known equations and the number of parameter estimates needed. The
degree of freedom (d.f.) is a function of this difference. If the number of MVs is p,
the known equations representing the total number for variance-covariance matrix to
be analysed is the sum of variances of each MVs (=p) and covariance between MVs
(=p(p-1)/2) (Byrne, 2001). Therefore, d.f. = p(p+1)/2-q. where q is the number of
free parameters to estimate in the proposed model (Rigdon, 1994). Model
identification is a complex problem that cannot be explained thoroughly in one
paragraph, but low degrees of freedom generally indicate unreliable results. In
addition to the indication of model identification, larger values of degree of freedom
also indicate that a smaller sample size can be tolerated for a similar model fit
(MacCallum, Browne, & Sugawara, 1996).
In our review, 25% (21 of 84) models have a general MV:LV ratio of less than
3 and 55.4% (46 of 83, one unreported) models contain at least one measurement
model with less than 3 manifest variables. In many cases also, the identification
problems involved in some or all of the measurement components are not explained,
nor is any consideration made of adding additional constraints. 13.3% of the models
(11 of 83) contain at least one single indicator construct. However, many
applications do not meet the mentioned requirements of applying single indicator
constructs. For example, one article (Cheung, Chow, & Yiu, 2009) uses a single item
in asking if “the negotiating parties were forced to articulate and clarify their
positions” to reflect the latent variable “position clarification”, but the factor loading
is only 0.45 which means only 20.25% variance of the latent variable is explained by
the selected single item and 79.75% variance is explained by the error. Only 52.4%
(44 of 84) articles provided d.f. values, while some articles presented Chi square test
results with degree of freedom ratios but not the d.f. values.
Mediators and moderators
There are two important classifications of (latent) variables in SEM. The first
divides variables into endogenous variables (i.e. dependent variables in regression
models) and exogenous variables (i.e. independent variables). The second
Chapter 3: Conceptual framework and structural equation modelling 73
categorization is based on the “positions” of these variables, with antecedents,
dependent variables, mediators and moderators. Mediators and moderators are often
necessary in research design, especially for solving complex and unsettled problems
in theory development. Identifying and quantifying the mediation (moderation)
effects of variables is useful in making contributions to the body of knowledge and
both variables are the focus of research design in many situations (Baron & Kenny,
1986). Even mediated moderation and moderated mediation are necessary in more
complex situations (Muller, Judd, & Yzerbyt, 2005).
In our review, all the applications are restricted to covering only simple
mediation or moderation effects. 11.9% of the (10 of 84) articles examined mediation
effects, but few tested their significance. For example, Mostert et al compare
mediated models and alterative models and confirm the mediating effects of negative
WHI (Work–home Interference) in the relationship between job demands/job
resources and burnout, and the mediating effect of positive WHI in the relationship
between job resources and work engagement (Mostert, Peeters, & Rost, 2011). 3.6%
of the (3 of 84) articles examined the effects of moderators in detail. Yang et al tested
the moderating effect of team relationships and team size separately by conducting a
two-way ANOVA when examining the relationship between knowledge
management and project performance (Yang, et al., 2012). Such analyses rare
however.
Sample size issues
Establishing the sample size is enough for testing the proposed model is
another critical decision to be made before data collection and analysis. Bagozzi and
Yi (2012) advise having a sample size of at least 100 for the results to be reasonably
reliable and suggest 200 to be more appropriate since less than this increases the risk
of sample non-normality and hence the accuracy of results. Compared with the
arbitrary threshold values of sample size, another rule of thumb is to have a
minimum number of parameters to estimate ratio of 5:1, although a 10:1 ratio is also
recommended for assuring the distribution of variables (Bentler & Chou, 1987).
Kline also recommends bootstrapping analysis as a method of improving the
reliability of SEM results obtained from comparatively small samples (Kline, 2005).
Another caution for sample size is that if the aim is to identify differences
among different respondent groups (i.e. multiple group analysis is necessary), each
74 Chapter 3: Conceptual framework and structural equation modelling
group needs to have a large enough sample size. One advantage of using SEM is that
it is powerful in testing hypotheses across samples. The multiple group analyses
allows many interesting tests, such as identifying factor loadings across groups, path
coefficients between latent variables across groups and the means of factors across
groups (Bagozzi & Yi, 2012)
In the papers reviewed, 31.0% (26 of 84) of models are derived from sample
sizes less than 100, 77.4% (65 of 84) have a sample size less than 200, 10.8% (7 of
65) have a sample size of less than 200 after applying bootstrapping, 85.7% (72 of
84) have a sample size to free parameters ratio less than 5, and 94.0% (79 of 84) have
a sample size to free parameters with a ratio of less than 10. Three studies conducted
multiple group analysis - across gender (M. Goldenhar*, Williams, & G. Swanson,
2003), country (Mohamed, 2003) and parental status, job type and race (Mostert, et
al., 2011).
Software programs
SEM was popularized by the launch of the linear structural relationships
(LISREL) computer program as the first SEM program developed by Jöreskog
(1970), resulting in SEM being regarded as the same as LISREL for a few years
(Golob, 2003). Two other popular software programs are EQS by Bentler (1989) and
AMOS by Arbuckle (Arbuckle, 1994). Apart from the very early versions of
LISREL, all of these programs provide a graphical user interface platform as a
replacement or complement of previous programming platforms, which makes SEM
easier for researchers and practitioners to use. Kline’s detailed comparison of these
three programs, found them to be similarly powerful in analysing structural equation
models and that the choice should be based on user preference (Kline, 1998). For
example, AMOS has a very user friendly user interface platform and is good at
handling incomplete data. EQS, on the other hand, does well in data screening and
dealing with non-normal data, while LISREL has advantages in dealing with very
complex situations, such as where nonlinear constraints are needed. When the
correlation matrix is only available as the input matrix rather than the covariance
matrix and raw data, EQS and LISREL are recommended since current AMOS
versions cannot handle the correlation matrix (Shah & Goldstein, 2006). In our
review, 55.4% (46 of 83, one unknown) models were built in AMOS, 31.3% (26 of
83) models in LISREL and 13.3% (11 of 83) in EQS.
Chapter 3: Conceptual framework and structural equation modelling 75
Table 3.1 Issues related to research design Categories Tested items Total CFA (=7) SEM (=77)
Research
design
Cross-sectional designs 83 7 76
Longitudinal designs 1 0 1
Model
specification
Models with control variables 4 0 4
With second order CFA structure in SEM 8 / 8
Multi group analysis 3 0 3
Mediation effect tested 10 / 10
Moderator effect tested 3 / 3
Bootstrap 7 0 7
Latent variables N=84 N=7 N=77
Mean (SD) 7.13 (3.63) 5.71 (3.25) 7.25 (3.65)
Median 6 5 6
Range (2,28) (2,11) (3, 28)
Structural model relations N=83 N=7 N=76
Mean (SD) 9.84 (9.05) 6.71 (4.75) 10.13 (9.31)
Median 8 6 8
Range (1, 72) (1, 15) (2,72)
MVs in the smallest construct N=83 N=7 N=76
<3 46 (55.4%) 3 (42.9%) 43 (56.6%)
Single indicator construct 11 (13.3%) 1 (14.3%) 10 (13.2%)
Mean (SD) 2.63 (1.23) 2.57 (0.98) 2.63 (1.25)
Median 2 3 2
Range (1,6) (1, 4) (1,6)
Number of manifest variables N=84 N=7 N=77
Mean (SD) 28.65 (17.58) 17 (5.13) 29.7 (17.9)
Median 24 19 24
Range (8, 108) (8, 23) -8,108
MV: LV ratio N=84 N=7 N=77
<3 21 (25%) 2 (28.6%) 19 (24.7%)
Mean (SD) 4.19 (2.04) 3.41 (1.00) 4.26 (2.10)
Median 3.5 3.2 3.5
Range (1.9, 13.8) (2.1, 4.8) (1.9, 13.8)
Sample size
(N=84)
<100 26 (31.0%) 2 (28.6%) 24
Between 100 to 200 39 (46.4%) 2 (28.6%) 37
>200 19 (22.6%) 3 (42.8%) 16
Mean (SD) 162.4(122.6) 165.3 (76.1) 162.1 (126.3)
Median 125.5 196 116
Range (32, 831) (32, 232) (36, 831)
Sample size/
parameter
ratio (N=84)
<5 72 4 68
<10 79 6 73
Mean (SD) 3.13 (3.00) 5.09 (4.37) 2.95 (2.82)
Median 1.99 3.70 1.94
Range (0.4,14.3) (0.9, 13.6) (0.4, 14.3)
Software
programs
applied
(N=84)
AMOS 46 7 39
LISREL 26 0 26
EQS 11 0 11
Unknown 1 0 1
Issues relating to model development
Model development issues after collecting data comprise data screening,
reliability tests and validity tests of constructs. The normality of data should be
considered when choosing estimation methods in SEM. Many articles present the
76 Chapter 3: Conceptual framework and structural equation modelling
validity of constructs and model evaluation at the same step, but it is common for
models to have poor goodness of fit (GOF), often caused by the inadequate validity
of constructs. Additionally, the validity of constructs is critical for approximating
“true” models, which is the core of SEM design but can be questionable in practice.
Research design: cross-sectional studies and longitudinal studies
Before SEM model building, it is important to test the characteristics of the
data. Multivariate normality of data is an important assumption made when applying
the default estimation method of maximum likelihood in SEM. Violation of this
assumption, especially with small samples, may inflate the GOF statistic and
underestimate the standard errors (MacCallum, Roznowski, & Necowitz, 1992). The
normality of the data can usually be evaluated by observing the skewness and
kurtosis statistics. Skewness is the standardized third moment of the data and
measures the extent to which a variable’s distribution is asymmetrical (towards right
or left). Kurtosis is the standardised fourth moment of the data and measures a
distribution’s peakedness (narrow/heavy tailed) (Hair, 2014). Both statistics are
asymptotically zero for the normal distribution and values more extreme than ±1 are
often taken to indicate non-normality.
When dealing with non-normal data, the choice of suitable estimation methods
is important for achieving reliable SEM results. There are many estimation methods
available for model development, such as the commonly used maximum likelihood
(ML), generalized least square (GLS), unweighted least squares (ULS) and
asymptotically distribution-free (ADF) methods. While ML is comparatively robust
to moderate violations of normality, and some distribution-free methods such as ULS
and ADF can also be helpful in these situations, distribution-free methods are
generally less powerful (Shah & Goldstein, 2006). It is also recommended to use the
robust methodology available in EQS to handle non-normality issues (Kline, 1998).
Special care is needed in research design, data collection and related factors
affecting missing values (Bagozzi & Yi, 2012). Some traditional considerations such
as dealing with missing values, identifying suspicious responses and outliers are also
necessary. Since these are quite common problems, not specific to SEM but
mentioned in only a few of the articles reviewed, some suggestions for missing
values are: (1) mean value replacement is not a good option when there are more
than 5% missing values per indicator as this decreases the variability of data (Hair,
Chapter 3: Conceptual framework and structural equation modelling 77
2014); (2) a returned questionnaire with more than 15% missing values should be
treated as an invalid response (Hair, 2014); and (3) the full information maximum
likelihood (FIML) method is more efficient than list wise deletion, pairwise deletion
and similar response pattern imputation (Enders & Bandalos, 2001).
The reliability test discussed here refers to the widely used Cronbach’s α>0.7
coefficient (Cronbach, 1951). This is an acceptable indication of the internal
consistency of constructs. However, in SEM, the composite reliability statistics
indexed in Bagozzi and Yi (1988) are needed as an indicator of internal consistency
of indicators within a construct. Fornell and Larcker (1981a)’s average variance
extracted (AVE) method, however, can be used to retest the validity of constructs
instead. Composite reliability is preferred as informative statistics.
Of the articles reviewed, only 14.3% (12 of 84) provide multivariate normality
test results or qualitatively state that this requirement was met. In some cases, other
multivariate normality tests are applied instead. For example, a Chi-square Q-Q plot
of each variable was used to assess multi-normality (Ding & Ng, 2010). The
estimation methods used are rarely mentioned and often ignored. 65.5% (55 of 84)
present Cronbach’s α values, but only a few (e.g. Cegarra-Navarro and Sánchez-
Polo (2011); (Chou & Yang, 2012)) provide composite reliability statistics.
Validity of constructs
Construct validity is necessary for reliable model testing and theory
development. Related issues have been criticized for decades in many research fields
such as marketing (Jarvis, et al., 2003). It covers both “the degree of agreement of
indicators hypothesized to measure a construct and the distinction between those
indictors and indicators of a different construct(s)” (Bagozzi & Yi, 2012). The two
common tests are for convergent validity as mentioned above and discriminant
validity.
Convergent validity measures the degree of positive correlation of one MV and
other MVs within the same construct, since MVs within the same construct should
share a comparatively high proportion of commonality (Hair, 2014). This is done by
assessing factor loadings, in which standardized factor loadings of the MVs larger
than (≈0.7) are taken to indicate a sufficient latent variable contribution (Hair,
2014), while standardized factor loadings less than 0.5 are considered for deletion
5.0
78 Chapter 3: Conceptual framework and structural equation modelling
(Xiong, Skitmore, Xia, et al., 2014). On the construct level, AVE is usually used to
measure convergent validity and should be larger than 0.5 to indicate a satisfactory
convergent validity (Fornell & Larcker, 1981a)
Discriminant validity aims to test whether a construct is truly distinct from
other constructs, which is critical to model development. The Fornell and Larcker
(1981a) criterion is widely used for assessing discriminant validity. This insists that
the AVE of one construct should be higher than its highest squared correlation with
other constructs (i.e. the square root of each construct’s AVE should be larger than
its highest correlation with other constructs).
Only 19.0% (16 of 84) of the articles reviewed conducted related convergent
tests without evaluating their suitability at this stage. With the MV factor loadings
provided in 53 articles, we calculated the AVE values of each construct and found
64.2% articles to be of questionable convergent validity (i.e. having at least one
construct’s AVE less than 0.5). For articles that considered convergent validity, 25%
(4 of 16) are questionable, 62.5% (10 of 16) are satisfactory with AVEs of all
constructs larger than 0.5, and 12.5% (2 of 16) of the articles did not disclose the MV
standardized factor loadings. 19.0% (16 of 84) conducted related discriminant tests
without evaluating their suitability at this stage, with only 12 articles conducting both
convergent and discriminant validity tests. 25 articles reported the correlation matrix
among latent variables, with 17 of these also reporting the standardized factor
loadings. After retesting the Fornell-Lacker criterion in these 17 applications, 29.4%
(5 of 17) have questionable discriminant validity (i.e. at least one construct’s AVE <
its highest squared correlation with other constructs). In addition, discriminant
problems are possibly more serious, since some suspicious models did not report the
authentic correlation matrix between constructs. For example, in the final model
presented in (Wong, Cheung, & Fan, 2009), the paths from double-loop learning to
project efficiency and project effectiveness are 0.91 and 0.95 respectively. The AVE
values of the latter two constructs are 0.65 and 0.50 respectively, likely suggesting a
flawed discriminant validity assessment. 16.7% (14 of 84) conducted exploratory
factor analysis (EFA) including principal component analysis or factor analysis
before doing the confirmatory factor analysis (CFA). Table 3.2 provides a summary
of the main results of this section.
Table 3.2 Issues related to model development
Chapter 3: Conceptual framework and structural equation modelling 79
Categories Tested items Number Percentage
Procedure
details
EFA before CFA/SEM 14 16.67%
Internal consistency reliability reported 55 65.48%
Convergent reliability considered 16 (of 55) 19.05%
Discriminant validity considered 16 (of 55) 19.05%
Construct
validity
retested
Reported standardized factor loadings 53 63.10%
Reported correlations between latent variables 25 29.76%
Reported both 17 20.24%
Convergent validity questionable 34 (of 53) 64.15%
Discriminant validity questionable 5 (of 17) 29.41%
Issues relating to model evaluation and reporting of results
Assessing the goodness of fit (GOF) of developed models is important for
model improvement and the discussion of findings. Many criteria have been
developed for this purpose and can be grouped into three broad categories: absolute
indices, incremental fit indices and parsimonious fit indices. Since numerous
statistics have been developed to measure model fit, this review presents only those
that are most important and commonly used.
Absolute fit indices
The Chi-square (χ2) test is the traditional measure for assessing overall model
fit by analysing the discrepancy between the sample and the proposed model (Hu &
Bentler, 1999). A probability, p, larger than 0.05 (Hair, 2006) is conventionally taken
to indicate a sufficiently good fit. This is not to be confused with the p values in t-
tests, where p<0.05 is preferred. However, χ2 statistics have been criticized for
being sensitive to sample size and for only providing a dichotomous ‘accept or
reject’ result (Kline, 2005; McDonald & Ho, 2002). The comparative χ2 of the χ2
to degrees of freedom ratio can be used to minimise the impact of sample size
(Hooper, Coughlan, & Mullen, 2008). Values of this ratio less than 2 indicate a good
fit (Marsh & Hau, 1996; Reisinger & Turner, 1999). In practice, several criteria are
often used for measuring the same GOF index. Those mostly used are summarised in
Table 3.3. For example, Kline (2005) and Pesämaa, Eriksson, and Hair (2009)
suggest ratio values of 3 and 5 respectively for the comparative χ2 index. Other
statistics in this category are also well developed (Hooper, et al., 2008; Hu &
Bentler, 1999; Marsh & Hau, 1996).
80 Chapter 3: Conceptual framework and structural equation modelling
The absolute indices measure the fit between the tested model and the sample
data (McDonald & Ho, 2002) and are the most fundamental indication of how well
the proposed theory fits the real world (Hooper, et al., 2008). In addition to the χ2
test, the absolute indices include the root mean square error of approximation
(RMSEA), goodness-of-fit index (GFI), adjusted goodness-of-fit index (AGFI), root
mean square residual (RMR) and standardized root mean square residual (SRMR).
RMSEA, as a very informative statistic, measures how well the parameter estimates
generated in the proposed model fit the population matrix (Byrne, 2001). An
RMSEA<0.05 indicates an excellent fit (Marsh & Hau, 1996); 0.08>RMSEA>0.05
indicates an acceptable error of approximation (Browne, Cudeck, Bollen, & Long,
1993); and RMSEA>0.10 indicates poor fit (Byrne, 2001). In addition, a 0.06
RMSEA cut-off proposed by Hu and Bentler (1999) has some support (Hooper, et
al., 2008). There is no best criterion and current results can be evaluated separately
by each since most have well-developed theoretical support.
For the articles reviewed, 36% (9 of 25) reported p values of χ2 tests that
were confused with those of the t-tests; 48% (12 of 25) correctly stated or applied the
probability criterion level of the χ2 tests; the remaining four had unclear results.
Only 48% (12 of 25) have the recommended χ2 with p>0.05 (Hair, 2006; Marsh &
Hau, 1996). However, 83.7% (41 of 49) have a comparative χ2 ratio of less than
two, indicating a good fit. 86.9% (73 of 84) reported values of RMSEA, with 97.3%,
75.3%, 41.1%, and 27.4% of these having values less than 0.1, 0.08, 0.06 and 0.05
respectively. The results are presented in Table 3.4.
Incremental fit indices
The incremental fit indices, also known as relative fit indices, are a group
statistic obtained by comparison with a baseline model (Jöreskog & Sörbom, 1996;
McDonald & Ho, 2002). These indices include the normed fit index (NFI),
comparative fit index (CFI), Tucker-Lewis Index (TLI/NNFI), incremental fit index
(IFI) and relative fit index (RFI). NFI measures a model by comparing the χ2 test
value of the model to the χ2 value of the null model in which all of the MVs are
assumed to be uncorrelated (Hooper, et al., 2008). A NFI>0.9 is generally taken to
indicate a good fit (Hair, 2006; Marsh & Hau, 1996), although Hu and Bentler
propose a stricter cut-off value of 0.95 (Hu & Bentler, 1999). However, NFI is
Chapter 3: Conceptual framework and structural equation modelling 81
sensitive to sample size and is underestimated when the sample size is small
(Hooper, et al., 2008). Therefore, NFI is not recommended for sole use (Kline,
2005). CFI is an extension of NFI that takes into account sample size and performs
well in small sample situations. Definitions of other statistics are provided in Hooper
et al (Hooper, et al., 2008), Hu and Bentler (1999) and Marsh and Hau (1996).
Descriptions and criteria for incremental fit statistics are summarised in Table 3.3.
As shown in Table 3.4, CFI is the most widely reported statistic in this category, with
80.95% (68 of 84) of the reviewed articles reporting values of CFI and 72.1% and
38.2% of models having CFI>0.90 and CFI>0.95 respectively.
Parsimonious fit indices
The parsimonious fit indices aim to avoid models becoming overly complex in
the search for improved GOF without necessary theoretical considerations (Mulaik,
et al., 1989). These indices include the parsimony normed-fit index (PNFI),
parsimony comparative fit index (PCFI) and parsimony goodness-of-fit index
(PGFI). PNFI, for example, is a modified form of NFI obtained by adjusting the
degrees of freedom. Although PNFI >0.5 is usually accepted in practice (e.g. (Chen
& Fong, 2012)), Mulaik et al note that it is possible to obtain a good fit model with a
value less than 0.5 (Mulaik, et al., 1989).
Table 3.3 GOF evaluation criteria and practical results
Fit index Evaluation criteria No. Proportion
Chi-square test
Probability
Reported number 25
p>0.05 (Marsh & Hau, 1996) (Hair,
2006) 12 48.0%
p>0.01 (Pesämaa, et al., 2009) 12 48.0%
Chi-square/df
Reported or not 49
smaller than 2 (Marsh & Hau, 1996) (Reisinger & Turner, 1999)
41 83.7%
Smaller than 3 (Kline, 2005) 48 98.0%
Smaller than 5 (Pesämaa, et al., 2009) 49 100.0%
Absolute fit indices
RMSEA
Reported number 73
Smaller than 0.05 (Marsh & Hau,
1996) 20 27.4%
Smaller than 0.06 (Hu & Bentler,
1999) 30 41.1%
Smaller than 0.08 (Browne, et al.,
1993) 55 75.3%
Smaller than 0.1 (Byrne, 2001) 71 97.3%
GFI Reported number 53
Greater than 0.95(Hooper, et al., 2008) 9 17.0%
82 Chapter 3: Conceptual framework and structural equation modelling
Greater than 0.90 (Marsh & Hau,
1996) (Hair, 2006) 21 39.6%
AGFI
Reported number 25
Greater than 0.95 (Hooper, et al., 2008) 1 4.0%
Greater than 0.90 (Marsh & Hau,
1996) 5 20.0%
Greater than 0.80 (Gefen, 2000) 15 60.0%
RMR
Reported number 15
Smaller than 0.05 (Chen & Fong,
2012) 9 60.0%
Smaller than 0.08 (Hu & Bentler,
1999) 12 80.0%
SRMR
Reported number 5
Smaller than 0.05 (Xiong, Skitmore,
Xia, et al., 2014) 2 40.0%
Smaller than 0.08 (Hu & Bentler,
1999) 4 80.0%
Incremental fit indices
CFI
Reported number 68
Greater than 0.95 (Hu & Bentler, 1999) 26 38.2%
Greater than 0.90 (Marsh & Hau,
1996) (Hair, 2006) 49 72.1%
NFI
Reported number 33
Greater than 0.95(Hu & Bentler, 1999) 9 27.3%
Greater than 0.90 (Marsh & Hau,
1996) (Hair, 2006) 21 63.6%
TLI/NNFI
Reported number 43
Greater than 0.95 (Hu & Bentler, 1999) 11 25.6%
Greater (Hair, 2006) 24 55.8%
IFI
Reported number 25
Greater than 0.95 (Hu & Bentler, 1999) 11 44.0%
Greater than 0.90 (Marsh & Hau,
1996) 19 76.0%
RFI Reported number 7
Greater than 0.90 (Marsh & Hau,
1996) (Hair, 2006) 1 14.3%
Parsimonious fit
PNFI Reported number 6
Greater than 0.50 (Chen & Fong,
2012) 5 83.3%
PCFI Reported number 2
Greater than 0.50 (Chen & Fong, 2012) 2 100.0%
PGFI Reported number 2
Greater than 0.50 (Xiong, Skitmore,
Xia, et al., 2014) 2 100.0%
Table 3.4 Description of reported GOF indices
Fit index No. Proportion Mean SD Median Range
Chi-square test
Chi-square 50 59.52% /
Probability level 25 29.76% /
Chapter 3: Conceptual framework and structural equation modelling 83
Chi-square/d.f. 49 58.33% 1.76 0.49 1.68 (1.02, 3.5)
Absolute fit indices
RMSEA 73 86.90% 0.068 0.039 0.066 (0.000,0.329)
GFI 53 63.10% 0.856 0.086 0.846 (0.620, 0.983)
AGFI 25 29.76% 0.808 0.111 0.829 (0.530, 0.950)
RMR 15 17.86% 0.065 0.061 0.049 (0.013, 0.230)
SRMR 5 5.95% 0.071 0.045 0.057 (0.038, 0.150)
Incremental fit indices
CFI 68 80.95% 0.918 0.064 0.934 (0.744, 1.000)
NFI 33 39.29% 0.893 0.083 0.913 (0.690, 0.998)
TLI(NNFI) 42 50.00% 0.880 0.105 0.901 (0.428, 1.016)
IFI 25 29.76% 0.927 0.055 0.941 (0.941, 1.000)
RFI 7 8.33% 0.773 0.110 0.730 (0.670, 0.994)
Parsimonious fit indices
PNFI 6 7.14% 0.583 0.154 0.650 (0.277, 0.688)
PCFI 2 2.38% 0.748 0.027 0.748 (0.729, 0.767)
PGFI 2 2.38% 0.653 0.028 0.653 (0.633, 0.673)
3.2.4 Discussion and recommendations
SEM is a very useful and versatile technique for both theoretical research and
experimental studies, and applications in construction research continue to increase.
Every method of statistical analysis, however, has its strengths and limitations and it
is important to understand these properties and characteristics in order to make
suitable choices among available alternatives. This is especially the case with SEM,
where many pitfalls await the unwary researcher in terms of sample size, construct
validity assessment, goodness of fit measures, etc. Many of these are identified in
this review of all the 84 articles containing SEM in solving construction research
problems over the period 1998-2012, including questionable convergent and
discriminant validity, and misunderstood p values in Chi-square tests. These and
many other important issues such as longitudinal studies, mediation effects,
moderation effects and multi group analysis are discussed and recommendations for
selected issues are summarised in Table 3.5.
The three-step procedure can be helpful for researchers in organizing their
application of SEM. At the research design stage, researchers can evaluate if SEM is
suitable and how to design their models and hypotheses. In the model development
stage, researchers can evaluate whether it is possible to solve the proposed models
accordingly. Many problems in model development are related to carelessness over
some critical issues in the research design stage. Therefore, researchers are
84 Chapter 3: Conceptual framework and structural equation modelling
encouraged to ensure a suitable MVs to LVs ratio, sample size and construct type as
early as possible. For example, it is inadvisable to apply single indicator constructs
without sufficient theoretical support, it being better to use manifest variables
directly if necessary (Ringle, et al., 2012). In the model development stage, a two-
step procedure is recommended: (1) the CFA phase: correlate all constructs together
firstly to test reliability and validity and refine or even change models accordingly;
and (2) the SEM phase: replace the correlations among constructs to the proposed
causal relations in the theoretical model and refine the models again.
It is also noticed that 16.7% (14 of 84) of the articles conducted an exploratory
factor analysis (EFA) before doing the confirmatory factor analysis (CFA). However,
its value and necessity are uncertain. Instead, the motivational differences between
EFA and CFA (see more in Thompson (2004)) should be considered, as should the
fact that CFA can handle MVs categorization and model refinements well. Since
model evaluations have been presented in detailed in Section 3.3, they are not
presented in Table 3.5. Additionally, it is recommended for researchers to present a
graphical form of the developed model for its clarity. It is a fact that few models are
perfectly correct and this can be a guide for researchers to assess and report their
models comprehensively (Shah & Goldstein, 2006). Since the principal of parsimony
is useful in selecting the best model from all candidate models especially when the
other two types of indices are comparable (Mulaik, et al., 1989), it is recommended
for further research to report more on parsimonious fit indices.
Table 3.5 Recommendations for selected issues in SEM
Issue Recommendation
Number of MVs per LV
Use three or more MVs per LV (Kline, 2005). The single indicator
construct is not recommended for its inadequate representation and
model deterioration; unless a single MV can present the LV perfectly
(Sarkar, et al., 1998; Shah & Goldstein, 2006).
Formative vs reflective
constructs
Check the causal directions between LVs and MVs as discussed in 3.1.2
section. Current SEM software (i.e. LISREL, AMOS and EQS) only
handles reflective constructs well. Solving formative constructs needs
additional constraints (Shah & Goldstein, 2006), however, and other
methods such as partial least square structural modelling are needed.
Model identification Calculate the d.f. values before data collection to make sure that it is
possible solve the original model and the alternatives.
Sample size issues Try to have a sample size larger than 100 (Bagozzi & Yi, 2012) or the
sample size to unknown parameters ratio should be larger than 5:1
(Bentler & Chou, 1987). Use bootstrapping to confirm the reliability of
results. Report GOF indices adjustments for small samples, such as
NNFI and Chi-square/d.f (Hooper, et al., 2008; Shah & Goldstein,
2006).
Multivariate normality Multivariate normality of data is an inherent assumption when applying
the ML and violations of this will cause problems such as inflated
Chapter 3: Conceptual framework and structural equation modelling 85
goodness of fit (MacCallum, et al., 1992).. It is recommended to use
estimation methods such as “ML, Robust” in EQS (Shah & Goldstein,
2006) and normal ML available in AMOS and LISREL as they are
robust to moderate violations of normality (Shah & Goldstein, 2006).
Some other distribution-free methods such as ULS and ADF can be
used (Shah & Goldstein, 2006).
Convergent validity
Assessing construct validity is necessary for making reliable
conclusions. The AVE of constructs should be larger than 0.5. Factor
loadings less than 0.5 should be considered for deletion (Hair, Sarstedt,
et al., 2012).
Discriminant validity The AVE of one construct should be higher than its highest squared
correlation with other constructs (Fornell & Larcker, 1981a).
3.2.5 Conclusions
Since it is hard to discuss everything important in SEM, the discussion and
recommendations section is organized to cover the common drawbacks of current
applications in our field. In doing this review of current SEM applications in solving
construction related problems, therefore, the goal has not been to cast doubts on the
SEM results to date. Rather, it has been to provide suggestions, recommendations
and guidelines for future SEM from research design to model development and
evaluation. It is hoped, therefore, that this review will be helpful for researchers to
enrich the body of knowledge. Other advanced techniques such as measurement
invariance and multitrait-multimethod studies are well developed in psychology, but
have seen little use in our field to date. Readers interested in applying these are
advised to consult the appropriate literature. Meanwhile, the intention of this paper
has been to contribute to the acceleration of research development in the construction
field by helping to create more technically informed researchers in the basic
application of structural equation modelling.
SEM can not only be a powerful tool for handling complex research problems
in traditional research topics, it can also be a helpful tool for construction academics
and technicians to assess the acceptance, usage and success of newly developed
technologies (e.g. Lee and Yu (2012); (Park, et al., 2012; Son, Park, et al., 2012;
Yang, et al., 2012)). This review will help them to design and apply SEM
applications in a more logical and efficient way.
86 Chapter 3: Conceptual framework and structural equation modelling
Chapter 4: Work stress 87
Chapter 4: Work stress
STATEMENT OF CONTRIBUTION
The authors listed below have certified that:
1. They meet the criteria for authorship in that they have participated in the
conception, execution, or interpretation, of at least that part of the publication in their
field of expertise;
2. They take public responsibility for their part of the publication, except for
the responsible author who accepts overall responsibility for the publication;
3. There are no other authors of the publication according to these criteria;
4. Potential conflicts of interest have been disclosed to (a) granting bodies, (b)
the editor or publisher of journals or other publications, and (c) the head of the
responsible academic unit, and
5. They agree to the use of the publication in the student’s thesis and its
publication on the Australasian Research Online database consistent with any
limitations set by publisher requirements.
In the case of this chapter:
Work stress
Bo Xiong*, Martin Skitmore, Bo Xia. Exploring and validating the internal
dimensions of occupational stress: Evidence from construction cost estimators in
China, Construction Management and Economics, 2015. 33(5-6), 495-507.
Contributor Statement of contribution
Bo Xiong
Conducted a literature review, designed the questionnaire, collected
data, wrote the manuscript and acted as the corresponding author.
07/03/2016
Martin Skitmore Directed and guided this study, and proofread the manuscript.
Bo Xia Directed and guided this study, and assisted with questionnaire
design.
88 Chapter 4: Work stress
Principal Supervisor Confirmation
I have sighted email or other correspondence from all Co-authors confirming their
certifying authorship.
Martin Skitmore
___________________ _____________________ _________________
Name Signature Date
Chapter 4: Work stress 89
4.1 INTRODUCTION
Occupational stress has been an important concept in organizational
management since the increased awareness of the prevalence of mental disorders
such as depression in the 1980s Tennant (2001). In the construction industry, because
of the complexity and dynamic uncertainty of its projects and often, workers and
professionals are frequently expected to confront and cope with stressful situations
(Leung, Ng, et al., 2005; Love, Edwards, & Irani, 2010). This, together with the
heavy workloads involved, can lead to serious occupational stress (Bowen, Edwards,
& Lingard, 2012). In addition to concerns of the wellbeing of those affected, such as
occupational illness and injuries (Lundstrom, Pugliese, Bartley, Cox, & Guither,
2002), the study of occupational stress is especially important in organisational terms
for the effects on organizational commitment, production performance and even
intentions to leave (Boyas, Wind, & Kang, 2012; Jamal, 1990; Leiter & Maslach,
1988).
Although identification and categorization studies of the stressors involved (i.e.
working conditions causing stress) are not uncommon in construction management
research (e.g. (Leung, Zhang, et al., 2008; Leung, Ng, et al., 2005; Richmond &
Skitmore, 2006)), the sub-dimensions of occupational stress (i.e. divisibility of
emotional reactions caused by work conditions) have received little treatment to date.
Additionally, occupational stress is widely regarded as a holistic concept with little
regard for the dimensions involved. Hurrell and McLaney (1988), for example, use
the general term "strain" to describe the emotional reaction to stressful conditions.
However, such reactions should not only include negative ones but also the joy of
stress (Hanson, 1987).
Realizing a similar situation in psychosomatic research, Levenstein et al.
(1993) developed a 30-question perceived-stress questionnaire (PSQ), validated with
responses from 230 medical subjects comprising in-patients, out-patients, students
and health workers in Italy. Fliege et al. (2005) later used an adapted version with
650 German subjects also in the medical context to conduct a principal component
analysis (PCA); identifying four underlying dimensions, comprising one stressor –
demand - and three emotional reactions in terms of worry, tension and joy (Fliege, et
al., 2005). However, this remarkable finding of exactly four dimensions has yet to be
90 Chapter 4: Work stress
confirmed empirically in other contexts, the construct consisting of three emotional
reactions validated and their effects determined.
This paper therefore firstly tests the applicability of an adapted Chinese PSQ
by applying exploratory factor analysis, and then validates the three dimensional
emotional reaction framework by conducting confirmatory factor analysis and
structural equation modelling. Two hypotheses are firstly tested:
Hypothesis 1 Fliege et al.’s (2005) four categories, comprising demand, worry,
tension and lack of joy are identifiable in the Chinese PSQ version.
Hypothesis 2 The three dimensional occupation stress framework of worry,
tension and lack of joy is reliable and valid in this context.
These two hypotheses are derived from the adapted and translated
measurement scale. Hypothesis 3 is developed to assess nomological validity in
exploring relationships between these internal dimensions and a related construct of
organizational commitment. The term organizational commitment is introduced for
this purpose, since some researchers (e.g. Leiter and Maslach, 1988; Jamal, 1990)
found a negative effect of occupational stress on organizational commitment.
Additionally, organizational commitment is an important concept highly correlated
with task performance and much other organizational behaviour, including
organizational citizen behaviour and turnover of employees (Chun, Shin, Choi, &
Kim, 2013; Porter, et al., 1976; Porter, et al., 1974). This leads to
Hypothesis 3 The three internal dimensions of occupational stress negatively
affect organizational commitment.
4.2 LITERATURE REVIEW
4.2.1 Occupational stress and its effects
Occupational stress can be regarded as adverse subjective emotions
experienced by employees when facing an imbalance between requirements and
ability and other working conditions (Bowen, Edwards, Lingard, & Cattell, 2014;
Leung, Zhang, et al., 2008). Worry, tension and lack of joy are found to be important
components of such emotions (Fliege et al., 2005). Occupational stress has become
an important topic in construction during recent decades since the industry has a high
Chapter 4: Work stress 91
exposure to uncertainties and many professionals experience high levels of stress.
According to a survey on stress conducted by the Chartered Institute of Building
(CIOB) in 2006, 68.2% of the 847 respondents admitted suffering stress and 26.6%
of these sought medical help. It is also revealed that only 15% of respondents thought
the industry had become less stressful and mental health was coped with well in
workplace (Campbell, 2006). A more recent survey by Bowen (2014a) indicates 55%
respondents of construction professionals in South Africa face high stress.
Occupational stress can also result in poor mental health according to Love et al.’s
(2010) survey among construction professionals in Australia. Occupational stress,
which is described as emotional stress in Leung, et al. (2010)’s survey of stress
among construction workers in Hong Kong, rather than demand-ability imbalance
causing accidents and injuries.
The influence of occupational stress on people is less easy to understand.
Evidence from a survey of 306 mainly American nurses indicates that perceived
social support from co-workers improves reported job performance and reduces
reported job stress (AbuAlRub, 2004), while Hon (2013), with evidence from 305
employees in 48 hotels and service organizations in China, finds co-worker support
is a significant moderator between working-creativity-caused stress and service
performance. Interestingly, AbuAlrub (2004) found job stress and job performance
had a U shape relationship, with mainly American nurses reporting moderate job
stress and believing their performance is worse than those reporting low/high job
stresses - which is consistent with Hanson’s (1987) statement that a medium stress
level is needed for more efficient work output (Gunning & Cooke, 1996). Jamal
(1984), on the other hand, in analysing sample data from 440 nurses working in
Canada, proposes employee professional and organizational commitment as
moderators in the stress-performance link, although this is only partially supported
by the data. In construction research, Bowen, Edwards, et al. (2014) examine four
categories of effects of occupational stress in terms of psychological effects,
physiological effects, sociological effects and substance usage (including alcohol,
cigarettes and even illegal drugs) in a survey of construction professionals in South
Africa.
In this study, the effect of occupational stress on organizational commitment is
targeted to validate the divisibility of occupational stress. The term “commitment” is
92 Chapter 4: Work stress
widely used in analysing both organizational and individual behaviours but no
commonly acknowledged definition has been developed (Becker, 1960). For
example, the activities of a committed person may be a result of considering
“generalized cultural expectations” (e.g. a trustworthy person do not change jobs
frequently) and “impersonal bureaucratic arrangements” (e.g. the economic loss of
quitting the current job) (Becker, 1960). However, the term “organizational
commitment” discussed here is not concerned with various definitions of general
commitment but the commitment related behaviours of employees characterized by:
(1) acceptance and appreciation of goals and values of the organization; (2)
willingness to make extra efforts for the success of the organization; and (3) a strong
desire to stay in the organization (Mowday, et al., 1979). Some academics (e.g.
Porter et al., 1974) point out that organizational commitment takes longer to build
but is more stable. Chun et al. (2013) analyse data from 3821 employees of 130
Korean companies and find organizational commitment positively affects the
financial performance of these companies via a mediation effect of organizational
citizen behaviour. Porter et al. (1974) conducted a longitudinal study to examine the
relationships between organizational commitment, job satisfaction and turnover with
evidence from psychiatric technician trainees, and found that general attitudes
concerning organizational commitment are important in deciding whether to stay or
leave (Porter et al., 1974) and that the level of organizational commitment declines
before leaving the current job (see Porter et al., 1976).
For the relationship between occupational stress and organizational
commitment, Leiter and Maslach (1988) found a significant negative effect of
occupational stress on organization commitment with empirical cases from 52 nurses
and support staff in a small hospital. Similarly, Jamal (1990) found low
organizational commitment and high turnover intention when employees face high
occupational stress and stressors, with empirical observations of a large hospital in
Canada with around 350 nurses. A recent survey in Iran conducted by Aghdasi,
Kiamanesh, and Ebrahim (2011) also found a significantly negative effect of
occupation stress on organizational commitment when exploring the relationship
between emotional intelligence and organizational commitment and the mediating
effect of occupation stress.
Chapter 4: Work stress 93
4.2.2 Stressors and coping strategies
Hurrell and McLaney (1988) point out that job stressors lead to psychological,
physiological and behavioural reactions of employees. In stress related research, the
importance of identifying internal and external stressors has been widely
acknowledged (Levenstein et al., 1993). For example, the 15-stressor inventory
developed in Jamal’s (1984) study measuring the relationship between occupational
stress and job performance of Canadian nurses categorize the 15 stressors into four
types in terms of role ambiguity, role conflict, overload and resource inadequacy.
Another psychometric instrument developed by Hurrell and Mclaney (1988)
categorizes 13 occupational stressors into workload, responsibility, role demands and
conflict. In construction, Leung et al. (2005a) find work overload, role conflict, job
ambiguity and working environment to be the ones most affecting stress levels of
consultant cost engineers (quantity surveyors) in Hong Kong. Organizational support
factors are also regarded as antecedents of stressors with this group, with stressors
such as lack of autonomy acting as mediators between organizational support and
employee stress (Leung et al., 2008b). Gunning and Cooke (1996) surveyed 39
construction professionals and 22 lecturers active in the Northern Ireland
construction industry and found "working to impossible deadlines", "client demands"
and "hiring/firing staff" to be three main causes of stress. Love et al. (2010) find
work-support to be an important predictor of occupational stress of consultants in the
Australian construction industry too, with lack of support resulting in the generally
poor mental health status of those affected (Love et al., 2010). Bowen et al. (2012)
evaluate the status of some stressors including job demands, job control, job support,
job certainty and opportunities, and the general work environment among South
African construction professionals. Ng, Skitmore, and Leung (2005) research on
measuring the manageability of stress in relation to Hong Kong construction projects
categorized 33 stressors into seven groups of work-nature related stressors, work-
time related stressors, organisation policy related stressors, organisation position
related stressors, situational/environmental stressors, relationship related stressors
and personal stressors.
Negative effects of occupational stress occur when insufficient resources are
available to cope with stressors (Cohen, Kamarck, & Mermelstein, 1983). The act of
coping, describes the situation when people defend themselves from threats to
94 Chapter 4: Work stress
current psychological conditions, such as integrity, and has gained in popularity since
1960s (Lazarus, 1993). In the Ways of Coping Questionnaire developed by Folkman
and Lazarus (1988), eight coping categories are presented: confrontational coping,
distancing, self-controlling, seeking social support, accepting responsibility, escape-
avoidance, planned problem solving and positive reappraisal. As mentioned earlier,
only 26.6% of the stressed people in the 2006 CIOB survey sought medical advice
and the mostly dependent coping mechanism is the support from other colleagues
(Campbell, 2006). Yip and Rowlinson (2006) exploratory factor analysis of the stress
coping behaviours of construction professionals in Hong Kong identified four main
categories of rational problem solving, resigned distancing, support seeking and
passive wishful thoughts. Adverse coping behaviours such as the consumption of
alcohol, cigarettes and illegal drugs have already been mentioned Africa (Bowen et
al., 2014a). The study of appropriate coping strategies is therefore an associated
common topic. Aiming to help project participants better cope with stresses, Ng et al.
(2005), for example, conducted a questionnaire survey to measure the manageability
of the stressors most confronted by Hong Kong construction project participants.
Richmond and Skitmore (2006) also provide 14 stress coping strategies such seeking
social support, improving communication and taking exercise for 50 identified
potential stressors by conducting interviews with project managers in the Australian
IT industry. In coping strategy selection, Haynes and Love (2004) found that active
coping is more useful than other strategies such as social coping and self-control in
their survey of male project managers in Australia.
4.2.3 Measures of occupation stress and divisibility
Hurrell et al. (1998) divide occupational stress research into two main types:
research studies on occupational stressors faced by employees in their working
environment, and studies of employees' emotional reactions (e.g. stain) to working
conditions. Although it is debatable whether the measurement of stress should
concentrate on stressors or stress reactions, it is acknowledged that both approaches
provide useful different perspectives (Hurrell et al., 1998; Fliege et al., 2005).
Therefore, some basic questions such as what is "occupational stress" and "what
dimensions should be included?" are also worth exploring.
Chapter 4: Work stress 95
The divisibility of occupational stressors has been widely acknowledged and
applied (Kahn, Wolfe, Quinn, Snoek, & Rosenthal, 1964; Leung, Ng, et al., 2005;
Porter, et al., 1974) as well as the difference between stressors and stress, but
occupational stress is still seen as a holistic concept in most studies. For example, in
Jamal’s (1990) study examining the effects of stress and stressors on employee job
satisfaction, organizational commitment and turnover intention among nurses in
Canada, total scores of eight stress-related items are used in further analyses. Since
occupational stress can be regarded as a result of the imbalance between job demands
and actual ability of employees (Bowen et al. 2014a), the total score of ten items
measuring such imbalance is used to describe occupational stress in (Leung, Zhang,
et al., 2008) and (Leung, Chan, Chong, & Sham, 2008) study of construction cost
engineers. However, such imbalance is more similar to a stressor rather than an
emotional reaction. In Bowen et al.’s (2014a) study of the stress effects of
construction professionals in South Africa, only a single 10-point scale was applied
to measure stress levels. Therefore, it is critical to contribute to the body of
knowledge to provide more attention to measurement-related issues of occupational
stress (Hurrell et al., 1998).
Levenstein et al. (1993) developed a perceived stress questionnaire (PSQ) to
explore the divisibility of occupational stress, but which has been criticised by Fliege
et al. (2005), however, for overlapping occupational stress and stressors.
Nevertheless, the developed PSQ reveals the existence of internal dimensions of
occupational stress. Additionally, Fliege et al. (2005) admit that their PSQ version
includes one occupational stressor, demands, and three other emotional stress
reactions. This study develops a Chinese PSQ version based on their work and aims
to demonstrate the divisibility of occupation stress.
4.3 RESEARCH METHOD
An adapted questionnaire was developed based on the PSQ constructed by
Levenstein et al. (1993) and Fliege et al. (2005). To assure content validity in the
Chinese context, a translation and back translation technique was applied. Principal
component analysis, confirmatory factor analysis and structural equation modelling
were used for testing construct validity.
96 Chapter 4: Work stress
4.3.1 Perceived stress questionnaire
The PSQ developed by Levenstein et al. (1993) and Fliege et al. (2005) was
used as the main instrument in the study. After analysing the results of Fliege et al’s
factor analysis and considering the likely drawbacks and suitability of these items in
the Chinese context, a modified 4x4 (four dimensions of stress, with each containing
four items) version was conjectured for the Chinese cost engineer context. Also,
while Fliege et al.’s PSQ refer to the respondent as “you”, the respondent were
address as “I” in this study make it easier for Chinese respondents to report more
personal emotional reactions. Additionally, the four-scale questionnaire response
format used in Levenstein et al. (1993) was changed to a seven-point Likert scale
format to elicit more finely grained information. Furthermore, a “don’t know”
option, omitted from Levenstein et al.’s and Fliege et al’s versions, was offered in the
questionnaire as a standard procedure for those unable to answer corresponding
questions.
The main part of the questionnaire is presented in Table 4.1. According to
Fliege et al.‘s (2005) categorization, Q1-Q4 belongs to demands, Q5-Q8 to worry,
Q9-Q12 to tension and Q13-Q16 belongs to joy. Q13-Q16 was reversed in the
analysis and named as AQ13-AQ16 indicating the lack of joy dimension to be
consistent with Levenstein et al.’s categorization.
Table 4.1 Perceived stress questionnaire No. Occupational stress 1-not at all to 7
very intensive Don't know
Q1 I have too many things to do 1 2 3 4 5 6 7 □
Q2 I do not have enough time for myself 1 2 3 4 5 6 7 □
Q3 I feel under pressure from deadlines 1 2 3 4 5 6 7 □
Q4 I feel I am in a hurry 1 2 3 4 5 6 7 □
Q5 I have many worries 1 2 3 4 5 6 7 □
Q6 My problems seem to be piling up 1 2 3 4 5 6 7 □
Q7 I fear I may not manage to attain my goals 1 2 3 4 5 6 7 □
Q8 I feel frustrated 1 2 3 4 5 6 7 □
Q9 I feel tense 1 2 3 4 5 6 7 □
Q10 I feel mentally exhausted 1 2 3 4 5 6 7 □
Q11 I have trouble relaxing 1 2 3 4 5 6 7 □
Chapter 4: Work stress 97
Q12 I find it hard to feel calm 1 2 3 4 5 6 7 □
Q13 I feel I am doing things I really like (R) 1 2 3 4 5 6 7 □
Q14 I am light hearted (R) 1 2 3 4 5 6 7 □
Q15 I feel safe and protected (R) 1 2 3 4 5 6 7 □
Q16 I am full of energy (R) 1 2 3 4 5 6 7 □
Note: Adapted from Fliege et al. (2005).
4.3.2 Translation and back translation
Translation and back translation is a widely accepted technique in cross-
cultural research since translation quality and equivalence between source and target
versions are critical (Brislin, 1970). Despites of its importance, this technique was
not yet widely acknowledged and used in construction research. Siu, Phillips, and
Leung (2003), for example, apply the technique in a safety attitudes questionnaire
used in some European studies when measuring the role of age on safety attitudes
and performance among Hong Kong construction workers. Ding and Ng (2007) also
apply the technique in validating their translated Chinese version of McAllister’s
trust scale. Because of differences in cultural backgrounds and languages, the
translation of questionnaires from English to Chinese needs be carried out with care
in this research. A two-stage translation and back translation technique was therefore
adopted. For the first stage, a Chinese version of the questionnaire was translated
from the English version by a bilingual PhD candidate with knowledge of PSQ, with
the preliminary Chinese draft emerging after several rounds of discussions with a
bilingual member of university academic staff. For the second stage, another pair of
bilingual assistants (i.e. PhD student and academic staff) without prior knowledge of
the PSQ English version of the questionnaire, translated the Chinese questions back
to English. The two English versions were then compared for significant inaccuracies
(Table 4.2). The discrepancies found were then corrected to produce the final
version.
Table 4.2 Translation and back translations No. Final Chinese version Back translation-1 Back translation-2
Q1 我有太多事情要做 I have a lot of things to do. I have too many works to do
Q2
我感到留给自己的时间
不够 I feel that I have limited time
to myself.
I feel not enough time for
myself
98 Chapter 4: Work stress
Q3
我感到来自截止日期的
压力 I feel the pressure from
deadlines. I feel deadline pressure
Q4 我感觉自己很着急 I feel that I am in a hurry. I feel I am in a hurry
Q5 我有很多担心 I have many worries. I have a lot of concerns
Q6 我的问题似乎越堆越多 It seems that my problems are
increasing.
My problems seem to be
accumulating
Q7
我担心我不能实现我的
目标(们) I am afraid that I cannot
achieve my goals.
I am concerned about not
realising my objective(s)
Q8 我感到受挫与沮丧 I feel frustrated and depressed. I feel frustrated and depressed
Q9 我感到紧张 I feel nervous. I feel nervous
Q10 我感觉到精神上的疲惫 I feel mentally exhausted. I feel mentally exhausted
Q11
我在放松身心上存在问
题 I have some problems on
relaxing my body and mind.
I have problem in physical and
psychological relaxation
Q12 我很难冷静 It is hard for me to keep calm.
I have difficulty in calming
down
Q13
我感觉我在做自己真正
喜欢的事情 I feel I am doing the things
that I like.
I think I am doing the work
that I truly like
Q14 我很轻松 I feel relaxed. I am very relaxed
Q15 我有安全感 I feel a sense of security. I feel secure
Q16 我感觉充满能量 I feel that I am full of energy. I feel energetic
4.3.3 Data collection and demographics
To validate the developed PSQ with empirical evidence from China, a
questionnaire applying the snowball sampling technique was used as recommended
in Shi, Ye, Lu, and Hu (2014) to approach potential participants rather than direct
delivery to companies, due to the sensitive questions asked in the PSQ. Young cost
engineers are targeted to validate the PSQ as it is impossible to cover all construction
populations in China. As indicated in some surveys including Love et al. (2010) and
Bowen et al. (2014a), different construction professionals differ largely in stress
levels and related effects. Construction cost engineers, with huge responsibilities and
high stress levels in construction projects, have been targeted as subjects in several
previous studies (e.g. (Bowen, et al., 2012; Leung, Chan, et al., 2008; Leung, Zhang,
et al., 2008; Leung, Ng, et al., 2005; Leung, Olomolaiye, et al., 2005)). Haynes and
Love (2004) found that less working experience is a significant predictor of high
occupational stress, while such a negative effect is not significant in (Bowen,
Govender, & Edwards, 2014). It is reasonable, therefore, to assume that young (i.e.
inexperienced) cost engineers have higher risks of becoming stressful at work due to
Chapter 4: Work stress 99
the imbalance between lack of experience and high demands. In Winefield and
Anstey (1991) survey of the occupational stress of general practitioners, emotional
exhaustion and depression of respondents younger than 40 are much higher than
those elder. Boyas et al. (2012) also found among child protection workers that
occupational stress levels and coping mechanisms differed greatly by age groups,
attributing this to differing social capital. Similarly, young doctors familiar with a
doctor's daily work find their job to be less stressful, emphasizing the effect of
experience (Bolanowski, 2005). Since five years’ experience is generally
acknowledged as the necessary time for practitioners to master construction cost
estimation skills (Skitmore, et al., 1990), potential respondents were restricted to
having no more than five years’ working experience. 144 valid responses were used
for further analyses. Of these, 74 (51.4%) are male and 69 (47.9%) are female (1
missing data); 42 (29.2%) are younger than 25, 100 (69.4%) range from 25 to 34 and
1 (0.7%) from 35 to 44 (1 missing data); and for their highest educational level, 13
(9%) possessed diplomas, 109 (75.7%) a bachelor’s degree and 22 (15.3%) a
master’s degree.
4.3.4 Data reliability
Cronbach's alpha is used to evaluate the internal consistency of the
questionnaire items. The overall value is 0.885, with 0.845, 0.834, 0.790 and 0.753
for the demands (Q1-Q4), worry (Q5-Q8), tension (Q9-Q12) and lack of joy (AQ13-
AQ16) dimensions respectively. Since all these values are larger than the 0.7 cut-off
value (Cronbach, 1951), the whole and the parts of the questionnaire are considered
to be acceptably consistent. Since the Cronbach alpha value is affected by length of
scale, the matrix of correlations of individual items is also examined for confirming
scale reliability (Ding and Ng, 2007). With a mean of the absolute values of item-
item correlations of 0.329 (SD=0.182), the results indicate an acceptable level of
reliability.
Although PCA deals well with non-normal distribution situations (Wang & Du,
2000)tests on sample distributions are still useful to reflect information concerning
the population distribution. Additionally, multivariate normality is an inherent
assumption when using the default maximum likelihood estimation method in
structural equation modelling (SEM) (MacCallum, et al., 1992; Xiong, et al., 2015a).
100 Chapter 4: Work stress
The sample skewness and kurtosis statistics can be used to test the normality of
distribution of variables and both should lie within the [-1, +1] interval (Hair, 2006).
Here, the skewness and kurtosis values of all 16 variables are within the range of -
0.86 to 0.45 and -0.54 to 0.46 respectively, which indicates the normal distribution
assumption to be satisfied.
4.4 DATA ANALYSIS AND DISCUSSION
Tests of these hypotheses are carried out with a sample of 144 predominately
young cost engineers working in the Chinese construction industry. As will be seen,
the first two hypotheses are supported and hypothesis 3 is supported partially as a
structural equation model indicates that only the lack of joy has a significantly
negative effect on organizational commitment.
4.4.1 Principal component analysis
Consistent with Levenstein et al. (1993) and Fliege et al.’s (2005) exploratory
study using PCA, the PCA confirms the hypothesized four-dimensional structure of
the PSQ, with a 0.840 Kaiser-Mayer-Olkin measure of sampling adequacy higher
than the cut-off value of 0.5 (Hair, 2006) and a highly significant p<0.0001 for
Bartlett’s test for sphericity, indicating that the items are suitable for factor analyses.
The forced 4-factor solution applying the varimax rotation, a widely applied
orthogonal rotation method maximizing the sum of the variances of the squared
loadings (Abdi, 2003) and used in Leventein et al. (1993) and Fliege et al. (2005),
explains 70.1% of the overall variance, with components 1, 2, 3 and 4 accounting for
38.1%, 16.6%, 10.0% and 5.4% respectively. The allocated components, means (M),
standard deviations (SD) and communalities (h2) of the items are summarised in
Table 4.3. For clarity, the largest factor loadings of each item are shown in bold.
Table 4.3 PCA with varimax rotation
Items Components Item parameters
1 2 3 4 M SD h2
Q1 0.009 0.856 0.019 -0.109 5.72 1.19 0.75
Q2 0.092 0.879 0.063 0.108 5.59 1.44 0.80
Q3 0.042 0.74 -0.106 0.366 5.49 1.45 0.70
Chapter 4: Work stress 101
Q4 0.256 0.643 0.108 0.524 5.22 1.43 0.77
Q5 0.327 0.47 0.077 0.645 5.26 1.41 0.75
Q6 0.379 0.297 0.077 0.699 4.60 1.50 0.73
Q7 0.402 -0.044 0.244 0.706 4.76 1.71 0.72
Q8 0.789 -0.045 0.2 0.344 3.78 1.67 0.78
Q9 0.789 0.149 0.187 0.258 4.08 1.63 0.75
Q10 0.663 0.304 0.384 0.184 4.44 1.58 0.71
Q11 0.695 0.13 0.345 0.022 3.86 1.64 0.62
Q12 0.683 -0.023 -0.072 0.245 3.32 1.71 0.53
AQ13 0.182 -0.102 0.625 -0.007 3.42 1.38 0.43
AQ14 0.147 0.371 0.776 -0.038 3.99 1.40 0.76
AQ15 0.108 0.068 0.822 0.171 3.55 1.40 0.72
AQ16 0.155 -0.279 0.667 0.395 3.38 1.27 0.70
4.4.2 Discussion-PCA results
With the exception of Q8 – “I feel frustrated and depressed” – the PCA
supports the hypothesised 4x4 structure. This anomaly is discussed below in terms of
the four dimensions involved, together with the relationship of the results with the
findings of previous studies of stress emotional reactions.
The tension dimension, comprising Q8-Q12, explains the largest proportion of
variance (38.1%) in the data, which is consistent with Jamal’s (1984) view of job-
related tension being regarded as occupational stress. According to Fliege et al.’s
(2005) original categorization, Q8 (“I feel frustrated”) is not included in this
dimension but in the worry dimension. This may be due to Fliege et al’s selection of
5 items from Levenstein et al.’s (1993) original 13 items for this dimension. If we
carry out a semantic analysis between Q5-Q7 and Q8, however, it is easy to see that
there are no words of “worry”, “afraid” or “fear” in Q8. Additionally, two Chinese
words are used to represent “frustrated” exactly and they are back translated as
“frustrated and depressed”. Therefore, it is reasonable to include Q8 in the tension
dimension. Also worth mentioning is the slightly low communality value (0.53) of
Q12 and a slight increase (0.009) of Cronbach's alpha value if deleted. This indicates
an inconsistent understanding of “calm” by the respondents, possibly related to the
fact that “calm” refers not only to “not excited or nervous” but also to “reasonable
and wise” in the Chinese culture, which is significantly influenced by Confucius’
wisdom . Therefore, some minor changes may be necessary for future applications of
Q12.
102 Chapter 4: Work stress
The demands dimension, comprising Q1-Q4, explains 16.6% of the variance in
the data. Cronbach's alpha value is rather high (0.845) but would not increase if any
item is deleted. According to Fliege et al.’s (2005) explanation, this dimension is
actually an extra stressor dimension that is similar to the term “overload” mentioned
in many stressor studies (e.g. Jamal, 1984; Leung et al., 2005a) and different in
nature to the other three dimensions.
The lack of joy dimension, comprising AQ13-AQ16, explains 10.0% of the
variance, and has an acceptable Cronbach alpha value of 0.753, but would be
increased a little (by 0.007) if AQ13 is deleted. Additionally, the communality of
AQ13 is comparatively low (0.43), indicating some confusion among respondents
when answering Q13 (“I feel I am doing the things that I like”), which is similarly
reflected in Levenstein et al.’s (1993) factor analysis results where the factor loading
on this item in the lack of joy factor is also comparatively low.
The worry dimension, comprising Q5-Q7, explains 5.4% of the variance, and
has a high Cronbach alpha value (0.803) that would not increase if any item was
deleted. The issue of Q8 is discussed above. To remain in the worry dimension, the
wording of Q8 needs to be changed to such as “I am afraid of/fear frustration” with a
greater emphasis on “worry”.
Investigating the differences among variables is a very informative way to
understand the multi-attributes of the sample. As shown in Table 4.3, items under the
demands sub-dimension among participants have comparatively high mean values,
indicating the young cost engineers experience a general “overload” feeling. The
average value of this sub-dimension (5.52) is higher than that (4.13) of the “work
overload” feeling among their counterparts in Hong Kong according to a 7 point
Likert scale survey by Leung et al. (2005a). Additionally, Leung et al. (2005a) found
that the “work overload” factor is the most predictive stressor of stress of
construction cost engineers in Hong Kong. This difference may be related to the
extensive construction work needed to cope with Mainland China’s rapid
urbanization, where the sub-sector of construction cost consultancy reached CNY
80.685 billion and 237,100 employees in 2011 after a 10% annual increase rate for
several years (Shi et al., 2014). With such a fast increasing market and following
needs to recruit new employees, therefore, it is not surprising to find that current
employees experience high “demands”. According to the results shown in Table 4.3,
Chapter 4: Work stress 103
young construction cost engineers also experience intense worry but with a little less
tension and less still lack of joy.
4.4.3 Validation with SEM
Although the PSQ developed by Levenstein et al. (1993) and Fliege et al.
(2005) helps in a obtaining a deeper understanding of occupational stress, the internal
dimensions of occupational stress need further construct validation. Since the PCA
results confirm the applicability of the PSQ, confirmatory factor analysis (CFA) is
applied for testing hypothesis 2 and hypothesis 3. In order to test nomological
validity and understand the potential different effects of the three different emotional
reactions, organizational commitment was introduced as the dependent variable in
SEM to test hypothesis 3. Five items were used, such as Mowday et al’s (1979)
organization commitment measure of "I really care about the fate of this
organization".
A CFA model is firstly developed to test the reliability and validity of a
construct consisting of three sub-dimensions of occupational stress in terms of worry,
tension and lack of joy. Since the CFA model, as presented in Table 4.4, is a good fit,
a further SEM model is developed to test hypothesis 3. Because the weightings of
manifest variables on latent variables in the CFA are quite similar to those in the
SEM, only the weightings in the SEM are presented in Table 4.4 for the sake of
clarity and simplicity.
Confirmatory factor analysis
Confirmatory factor analysis is a specific application of structural equation
modelling to validate established measurement constructs or model validation (Xiong
et al., 2015). For example, Molenaar et al. (2009) validated their five-dimensional
framework to measure corporate safety culture by applying CFA. (Wong, Cheung,
Yiu, & Pang, 2008) also applied CFA to validate a three-dimensional framework to
measure trust in construction contracting, while (Ding, Ng, Wang, & Zou, 2012)
validated a two-dimensional trust framework by CFA with empirical evidences in
construction. In this study, a three-dimensional framework for measuring
occupational stress is developed and tested with CFA. In such studies, model fit,
reliability and the validity of the constructs are critical for validating the developed
104 Chapter 4: Work stress
models (Xiong et al., 2015). The overall model fit as presented in Table 4.5 is
generally satisfactory. To assure the reliability of constructs, Cronbach's alpha is still
useful in determining the internal consistency of constructs. An alternative is
composite reliability, CR, where
𝐶𝑅 =(∑ 𝜆𝑖)2
(∑ 𝜆𝑖)2+∑ 𝑉𝑎𝑟(𝑒𝑖) (4.1)
as indexed in Bagozzi and Yi (1988) and recommended as a more informative
statistic in the SEM context for its ability to assess internal consistency of indicators
within a construct. A value larger than 0.7 indicates good quality (Bagozzi and Yi
1988). As Table 4.4 indicates, all CR values are acceptable.
Construct validity tests normally include convergent validity and discriminant
validity. Convergent validity measures the extent of positive correlations of one
manifest variable (MV) with other MVs within same constructs, since MVs should
share a comparatively high proportion within the same constructs (Hair, 2014). To
assess convergent validity, the standardized regression weights and squared multiple
correlations (SMCs) for each item are calculated. As presented in Table 4.4, all the
standardized regression weights are highly significant and above 0.5, ranging from
0.538 to 0.853, indicating acceptable validity (Xiong, Skitmore, Xia, et al., 2014). It
is worth mentioning that the standardized regression weights of Q12 and AQ13 are
close to the threshold, which is consistent with previous PCA results. They are still
kept to ensure the completeness of measurements as their deletion does not leading to
any improvement in the CFA and SEM results. Discriminant validity (that constructs
in the model are significantly different) can be confirmed by comparing the
unconstrained model and constrained alternatives. Since the unconstrained model is
significantly better than the model equally constrained correlations between
constructs (Chi-square (df=2)= 40.967, p=0.000), it is reasonable to regard these
constructs as different ones. Since this study aims to validate the divisibility of
occupational stress, nomological validity, another useful although little mentioned
construct validation, is recommended by applying structural equation modelling
(Ding and Ng, 2007).
Table 4.4 Standardized regression weights
Item Standardized regression weights SMC CR
Chapter 4: Work stress 105
Worry Tension Lack of joy OC
Q5 0.714
0.510
0.808 Q6 0.811
0.658
Q7 0.764
0.583
Q8
0.852
0.727
0.856
Q9
0.853
0.728
Q10
0.777
0.603
Q11
0.633
0.401
Q12
0.538
0.289
AQ13
0.538
0.289
0.766 AQ14
0.654
0.428
AQ15
0.786
0.618
AQ16
0.693
0.480
Q17
0.708 0.501
0.887
Q18
0.840 0.705
Q19
0.779 0.607
Q20
0.819 0.671
Q21 0.758 0.575
Structural equation modelling
In some studies, the CFA phase is usually undertaken as a first step before
placing directional relationships between constructs in the model (Xiong, Skitmore,
Xia, et al., 2014). Since a good model fit is achieved in the CFA phase, a further
SEM model is developed to test Hypothesis 3. As indicted in Table 4.5, the model fit
is acceptable. The final results are presented in Figure 4.1 and Table 4.4. The
correlations between emotional reactions are also tested and presented as broken
lines in Figure 4.1, where the observed variables such as Q5 are shown in rectangles;
latent variables such as worry are shown in ellipses; with directional arrows
reflecting effects of sub-dimensions of occupational stress on organizational
commitment. It is found that only lack of joy has a significantly negative effect on
organization commitment, with the other two emotional reactions having no
significant effects.
Table 4.5 Goodness of fit
Goodness of fit measure Criteria CFA SEM
χ2/df <5.0 2.255 2.078
Absolute fit
RMSEA <0.1 0.093 0.087
AGFI >0.8 0.822 0.785
SRMR <0.08 0.065 0.062
Incremental fit
CFI >0.9 0.918 0.902
106 Chapter 4: Work stress
IFI >0.9 0.919 0.904
Parsimonious fit
PNFI >0.5 0.667 0.690
PGFI >0.5 0.578 0.621
Figure 4.1 Effects of dimensions of stress on organizational commitment
4.4.4 Validation with SEM Discussion of the CFA and SEM results
The CFA results validate the three sub-dimension construct of occupational
stress, which supports Hypothesis 2. It is also noticed that worry and tension are
highly correlated but lack of joy is less correlated. The high correlation agrees with
many previous studies, since occupational stress is simply regarded as a mix of
tension and worry (Hurrell Jr, Nelson, & Simmons, 1998). As richer meanings have
been identified in occupational stress (Levenstein et al., 1993, Fliege et al., 2005), it
will be necessary to take into account its multi-dimensional nature in subsequent
research.
The SEM results reveal that the three emotional reactions have different effects
on organizational commitment. Lack of joy has a largely negative effect on
organizational commitment (as lack of joy increases, organizational commitment
decreases) but worry and tension do not. This is generally consistent with previous
Chapter 4: Work stress 107
studies of Leiter and Maslach (1988), Jamal (1990) and Aghdasi et al. (2011) in that
occupational stress negatively affects organizational commitment, while this study
finds that the main contributor is lack of joy other than worry and tension.
Acknowledging the differences between lack of joy discussed here and the term "job
satisfaction", these findings are consistent with (Tett & Meyer, 1993), where job
satisfaction was found to be highly correlated with organizational commitment and
turnover. Similarly, Currivan (1999) found greater intensity of job stressors
comprising role ambiguity, role conflict and workload leads to lower job satisfaction,
which also leads to weaker organizational commitment. It is surprising to find that
worry and tension, although highly correlated with workloads, do not have a
significant effect on organizational commitment, as greater workloads have been
found to lead to weaker organizational commitment in some previous studies (e.g.
(Currivan, 1999; De Cuyper & De Witte, 2006)).
This puzzling paradox demonstrates the necessity to understand the sub-
dimensions of occupational stress in terms of worry, tension and lack of joy, since
these emotional reactions may have different causes and effects. An additional
stepwise principal component regression analysis as used in Gan, Zuo, Ye, Skitmore,
and Xiong (2015) was conducted to predict organizational commitment not only with
worry, tension and lack of joy but also gender, experience and even demands in a
second test. The result is same in that only lack of joy is significant predictor. The
insignificant effects of worry and tension on organizational commitment may
possibly be attributed to a U-shape relationship between stress and job performance
and stress and organizational relationship (AbuAlrub, 2004; Leung et al., 2005b). As
Leung et al. (2005b) point out, a moderate stress level would result in better
performance among cost engineers, while stress measured as the imbalance between
actual ability and job expectations is more a proxy for job fit rather than stress level
(Lauver & Kristof-Brown, 2001).
4.5 CONCLUSION
The applicability of a revised PSQ based on Levenstein et al. (1993) and Fliege
et al.’s (2005) studies in China is demonstrated, which means Hypothesis 1 is
supported. A translation and back translation technique and principal component
108 Chapter 4: Work stress
analysis are used to firstly validate the questionnaire data. This confirms it can be
used in future studies, in contrast with many previous studies that suffer from
measurement deficiencies concerning occupational stress. In order to record distinct
stressor and emotional reactions, a further three-dimensional framework for
measuring occupational stress is developed.
The second contribution is that the divisibility of occupational stress is
demonstrated in this study, which means Hypothesis 2 is supported. The three sub-
dimensions in terms of worry, tension and lack of joy are developed and validated by
structural equation modelling. Since the model comprising three emotional reactions
is supported with CFA, further research would benefit from treating occupational
stress as a multi-dimensional concept. The measurement issue is always a most
critical in this kind of research and some previous studies use one or several stressors
such as workload and ability imbalance to indicate occupational stress. Such
practices could be reasonable in some situations, but would be problematic when
exploring the relationship between stressors and stress. Some research directly asks
for respondents’ general perception of stress, which may be defined differently from
person to person and hard to measure without the help of more observable measured
items. This new framework identifies the core characters and manifest variables of
occupational stress, which helps the standardization of occupational stress
measurement and provides a standard way to develop a measurement framework.
Another theoretical implication of this study is that sub-dimensions may act
differently in organizational contexts. As proposed in Hypothesis 3, three dimensions
are presumed to negatively affect organizational commitment, while only lack of joy
has significantly negative effect. Therefore, more exact descriptions should be used
when examining and describing the relationship between occupational stress and
occupational stressors or other organizational influences.
It should also be noted that the empirical work in this study is limited to the
specific context of construction professionals in China, although findings from this
research may also interest researchers outside the construction field. Further research
will benefit from applying the findings of this research in other settings and
exploring relationships between the sub-dimensions of occupational stress with other
managerial factors. In addition, much work remains to be done in identifying
uncovered dimensions of occupational stress and improve measurement accuracy.
Chapter 5: Job satisfaction 109
Chapter 5: Job satisfaction
5.1 THE NEXUS BETWEEN JOB SATISFACTION AND JOB
PERFORMANCE OF CONSTRUCTION COST ENGINEERS
5.1.1 Introduction
The relationship between job satisfaction and performance (S-P) has been an
important topic of study for academics and organisation managers for many decades
since the Hawthorne studies and the human relations movement in the 1930s (Judge
et al., 2001). It is initially proposed that “happier workers produce more” which
gains popularity as an argument because of its consistency with intuition. In the
1960s, some researchers (Lawler and Porter, 1967) argued that job satisfaction was
induced by performance for rewards, and that good performers gain more rewards
and are happier. Both opinions are supported by theory. The former opinion is
supported by the theory of reciprocity — that an employee has a natural intention to
respond reciprocally to perceived kindness and unkindness (Falk & Fischbacher,
2006). The latter opinion is based on motivation theory, which reasons that rewards,
led by the job performance of employees, result in satisfaction and even higher
subsequent performance in response to the effects of organisational commitment and
goal setting (Latham & Pinder, 2005). However, convincing empirical evidence for
both assumptions are still lacking. Some researchers (Fisher, (2003) describe the S-P
nexus as simple "folk wisdom".
Reviewing previous studies, weak and inconsistent empirical evidence for the
S-P nexus can be attributed to changing definitions of concepts and divisibility of
abstract terms. For example, satisfaction may have several facets, especially
economic satisfaction (ES) and production-related/noneconomic satisfaction (PS)
(Xiong et al., 2014). Similarly, dimensions of job performance include task
performance, organisational citizen behaviour and even anti-productive behaviours
(Viswesvaran & Ones, 2000). This study divides job satisfaction into economic
satisfaction and noneconomic satisfaction, and uses task performance (TP) as the
measure of job performance. It is proposed that PS increases TP and then TP
increases ES. A literature review is firstly conducted and then a conceptual
110 Chapter 5: Job satisfaction
framework is proposed. Statistical analyses are further applied to validate the
hypothesised model.
5.1.2 Literature review
Linkage between individual satisfaction and performance
Studies on the relationships between job satisfaction and job performance
comprise an appreciable portion of behaviour research in management (Organ,
1988b). Additionally, the discrepancy between the strong intuition among
practitioners that satisfaction has an obvious influence on productivity and low
correlations for these elements of performance obtained in empirical studies has
made this an appealing topic for researchers for decades (Judge et al., 2001). There
are three mainstream hypotheses on the S-P nexus: (1) job satisfaction causes job
performance; (2) job performance causes job satisfaction; (3) there are other complex
relationships between the two including moderators, mediators or antecedent
variables.
The first of these hypotheses again goes back to the Hawthorne studies and
human relations movement, when the idea that improvement in employee morale
leads to production improvement became widely accepted (Schwab & Cummings,
1970). Despite little supporting empirical evidence, the hypothesis that job attitudes
affect employee behaviour became accepted as logically reasonable (Judge et al.,
2001) and used as a common assumption in many studies. The second hypothesis
reverses the cause and the effect, with Lawler and Porter (1967), for example,
pointing out that rewards were not adequately considered in previous research, and it
was therefore reasonable to assume that satisfaction follows the rewards produced by
performance. Although there is some empirical evidence in favour of the second
hypothesis (Judge et al., 2001), it is still insufficient to be convincing and has been
criticised as containing a hidden and questionable presumption that performance and
rewards are closely linked for individual workers (Fisher, 2003).
Because of the weak empirical evidence relating to the first two hypotheses,
some researchers have turned to exploring common antecedent variables for
satisfaction and performance in terms of mediators and moderators in the job S-P
linkage (Judge et al., 2001; Schwab & Cummings, 1970). Some researchers such as
Chapter 5: Job satisfaction 111
Schwab and Cummings, 1970) argue that the unsatisfactory outcomes of S-P linkage
research have been mainly caused by the ambiguity of definitions of job satisfaction.
Although some measures of job satisfaction such as the Job Descriptive Index
(Smith, 1969) and Minnesota Satisfaction Questionnaire (Weiss, Dawis, & England,
1967) have been developed, job satisfaction is still seen as a holistic concept in
applications connecting satisfaction and job performance. It has been suggested that
researchers should explore the relationship between specific attitude measures and
specific job behaviours, rather than the link between general satisfaction and a
specific behaviour (Fisher, 2003). Lai (2007), for example, divided the job
satisfaction of dealers in the motor industry into social satisfaction and economic
satisfaction, and found that noneconomic satisfaction was much more important than
economic satisfaction in influencing performance. This dichotomy is also consistent
with Brown’s (2001) finding that economic satisfaction should be treated separately
for analysis, since it is highly related to pay factors like pay equity.
However, some previous studies (Janssen and Van Yperen, 2004) fail to
connect satisfaction with performance, while other studies (Lai, 2007; Nerkar et al.,
1996) assume that all disaggregated satisfaction facets share common unidirectional
relationships with performance; for instance, all satisfaction sub-dimensions lead to
performance. Therefore, the vital unanswered question is whether it is possible that
the low correlation observed in previous studies between overall satisfaction and
performance was caused by different or even conflicting causal relationships
between satisfaction sub-dimensions and performance. For example, economic
satisfaction (satisfaction with pay) generated by receiving rewards is caused by
performance rather than being a cause of performance, while some other satisfaction
dimensions (such as satisfaction with co-workers and supervisors) may enhance
performance.
Another explanation for unsatisfactory previous research results can also be
attributed to changes in the conceptualisation of job performance. In early
organisational studies such as the Hawthorne studies, job performance is considered
to be virtually the same as task performance, defined as
… the proficiency with which incumbents perform activities that are formally
recognized as part of their jobs; activities that contribute to the organization’s
technical core either directly by implementing a part of its technological
112 Chapter 5: Job satisfaction
process, or indirectly by providing it with needed materials or services.
(Borman & Motowidlo, 1993a, p73)
In recent decades, another category of employee behaviour, known as
organisational citizen behaviour (OCB), has been identified and accepted by both
academics and practitioners. This assumes that job responsibilities, expressed active
involvement in the organisation, and innovation for the benefit of the organisation
take place even without reward expectations (Eisenberger et al., 1990). Job
performance nowadays includes task performance, OCB and even counterproductive
behaviours in some situations (Viswesvaran & Ones, 2000). As an early stage
exploration, this study focuses on task performance (TP).
Conceptual model development
Many conceptual models describing job satisfaction and performance have
been proposed, as presented in Figure 5.1 adapted from Judge, et al. (2001). The first
three models assume there are causal relationships between job satisfaction and job
performance. Model 4 and Model 5 assume there are antecedents or moderators
affecting the S-P nexus.
Figure 5.1 Main conceptual models of the S-P nexus
Following the majority of previous studies (Judge et al., 2001; Organ, 1988b),
this paper assumes there is a positive correlation between overall job satisfaction and
Chapter 5: Job satisfaction 113
job performance. In addition to the overall S-P nexus, a fine-grained hypothesised
model is developed by using two broad dimensions of job satisfaction in terms of ES
and PS. The theory of reciprocity and motivation theory are used to develop the
conceptual model, as presented in Figure 5.2.
Figure 5.2 Proposed conceptual model for this study
5.1.3 Research method
Questionnaire survey
To explore the S-P nexus, related items of the questionnaire survey concerning
interactions between person and environment are used, as presented in Table 5.1.
Respondents are construction cost engineers, also known as quantity surveyors in
China. To measure job satisfaction, eight items as presented in Table 5.1 are used,
based on previous works (Smith, 1969; Cotton and Tuttle, 1986; Xiong, et al., 2014).
Following the previous works of Skitmore and Marston (1999b) and Leunget al.
(2005), five items such as “I estimate the budget of the project without overrunning”
are used to measure task performance of the professionals.
Table 5.1 Measures of Job satisfaction No. Job satisfaction measures 1-not at all to 7 very intensive Don't know
Q1 Satisfaction with pay 1 2 3 4 5 6 7 □
Q2 Satisfaction with promotional opportunities 1 2 3 4 5 6 7 □
Q3 Satisfaction with organisational welfare 1 2 3 4 5 6 7 □
Q4 Satisfaction with work itself 1 2 3 4 5 6 7 □
Q5 Satisfaction with supervision 1 2 3 4 5 6 7 □
Q6 Satisfaction with co-workers 1 2 3 4 5 6 7 □
Q7 Satisfaction with workload 1 2 3 4 5 6 7 □
Q8 Satisfaction with current tasks 1 2 3 4 5 6 7 □
Because of cultural and linguistic differences, the translation of questionnaires
from English to Chinese needed be carried out with care. To ensure content validity,
the translation and back translation technique (see detailed steps in Xiong, Skitmore,
and Xia, 2015b) was applied with the assistance of four bilingual researchers.
114 Chapter 5: Job satisfaction
Data collection and demographics
The snowball sampling technique is useful to gather sensitive information,
especially in a situation where random sampling is not available. Snowball sampling
allows researchers to access informants through contact information provided by
other informants, and has been the most widely employed sampling method in many
disciplines across the social sciences (Noy, 2008). Considering the study context, this
technique is appropriate to this study. 285 complete responses among 310 returned
ones were considered valid for further analysis in this study. The majority of
respondents have a bachelor degree or higher education level. Respondents are
almost evenly distributed across some characteristics, including gender
(male/female), working city/state, company type (property developer/construction
company/consulting company) and employment sector (public/private). To evaluate
the internal consistency of the questionnaire items, Cronbach's alpha is calculated in
SPSS 21.0, with the overall value equal to 0.868, indicating good consistency.
5.1.4 Results
Principal component analysis
The PCA confirms a two-dimensional structure of job satisfaction, with a 0.836
Kaiser-Mayer-Olkin measure of sampling adequacy higher than the cut-off value of
0.5, and a highly significant p<0.0001 for Bartlett’s test for sphericity indicating that
the items are suitable for factor analyses. The solution gained by applying varimax
rotation explains 65.7% of overall variance, with component 1 and component 2
accounting for 50.5% and 15.2% respectively. Loadings with components, means,
standard deviations and communities (h2) of items are summarised in Table 5.2.
Table 5.2 Principal component analysis with varimax rotation
Items Components Item parameters
1 2 Mean SD h2
Q1 0.130 0.857 3.860 1.325 0.751
Q2 0.263 0.813 3.912 1.328 0.730
Q3 0.242 0.817 3.891 1.391 0.726
Q4 0.652 0.387 4.488 1.165 0.576
Q5 0.628 0.383 4.656 1.439 0.540
Q6 0.792 -0.087 5.193 1.163 0.635
Chapter 5: Job satisfaction 115
Q7 0.680 0.347 4.284 1.327 0.583
Q8 0.797 0.287 4.442 1.254 0.717
Correlation and regression analysis
To investigate the necessity for distinguishing ES and PS, effects of sub-
dimensional satisfaction on task performance are explored by applying regression
analysis. Average values of ES, PS, and task performance are calculated. Overall
satisfaction is attained by calculating the average of ES and PS assuming equal
weight. Correlations of these factors are presented in Table 5.3. Regression analysis
is applied as Model A, presented in Figure 5.3. If these two dimensions share
consistency (Nerkar et al., 1996), their effects on task performance should be
consistent. However, it is found that only PS has a significant positive effect on task
performance. Model B is then developed as a conceptual model, presented in Figure
5.3. The attained results confirm that PS positively affects task performance, and TP
positively affects ES.
Table 5.3 Correlations between factors
Factors A B C
A. task Performance 1 B. ES_A 0.193** 1
C. PS_A 0.344** 0.558** 1 D. overall satisfaction 0.296** 0.905** 0.858**
Note: **. Correlation is significant at the 0.01 level (2-tailed).
Figure 5.3 Model evaluations by regression analysis
116 Chapter 5: Job satisfaction
The form of interaction
The above linear regression results reveal overall positive effects of PS on TP
and TP on ES. It is found in studies of stress that although work stress has an overall
negative effect on job performance, the relationship would be better seen as an n-
shaped or inverted U-shaped one, as there is a quadratic effect of stress on
performance (Leung et al., 2005; Xiong et al., 2015b). Considering the similarity
between stress and satisfaction, this study explores the relationship form of two
relationships in terms of PS-TP and TP-ES. Results are presented in Table 5.4.
Table 5.4 forms of effects Relationships Relationship form
PS-TP
TP-ES
5.1.5 Discussion and conclusions
In previous studies on the nexus between job satisfaction and job performance,
job satisfaction has been widely taken as a holistic term without investigating the
internal dimensions. In the literature review, a two-dimensional structure of job
satisfaction is proposed. As presented in Table 5.2, economic satisfaction (ES) and
production-related satisfaction (PS) are different components. Similarly, job
performance is a multi-attribute concept. This study focuses on task performance
only.
Chapter 5: Job satisfaction 117
To evaluate the validity of the proposed model in Figure 5.2, correlation and
regression analyses are applied. Comparing the modelling results of Model A and
Model B, it is necessary to distinguish ES and PS. Additionally, this study proposes a
new model to describe relationships between the sub-dimensions of job satisfaction
and performance. In addition to support from theories including reciprocity theory
and motivation theory, this model is demonstrated as valid by empirical evidence
gained from construction professionals in China.
In addition to the overall positive linear effects of PS-TP and TP-ES as
presented in Figure 5.3, it is found that an n-shaped relationship would be better to
describe the effect of TP on ES. Findings in this study will benefit further studies on
the nexus between job satisfaction and performance.
There are a few limitations worth mentioning. Following the stimulus-
organism-response paradigm in studying employee behaviour, the antecedents of job
satisfaction and performance include working environment factors like
organisational support and individual characteristics (Xiong, 2015). Without
considering these factors thoroughly, situations like Model 4 and Model 5 as
described by Judge et al. (2001) need further investigation in future research.
118 Chapter 5: Job satisfaction
5.2 EXAMINING THE INFLUENCE OF PARTICIPANT PERFORMANCE
FACTORS ON CONTRACTOR SATISFACTION: A STRUCTURAL
EQUATION MODEL
Statement of contribution
The authors listed below have certified that:
1. They meet the criteria for authorship in that they have participated in the
conception, execution, or interpretation, of at least that part of the publication in their
field of expertise;
2. They take public responsibility for their part of the publication, except for
the responsible author who accepts overall responsibility for the publication;
3. There are no other authors of the publication according to these criteria;
4. Potential conflicts of interest have been disclosed to (a) granting bodies, (b)
the editor or publisher of journals or other publications, and (c) the head of the
responsible academic unit, and
5. They agree to the use of the publication in the student’s thesis and its
publication on the Australasian Research Online database consistent with any
limitations set by publisher requirements.
In the case of this chapter:
Examining the influence of participant performance factors on contractor
satisfaction: A structural equation model
Bo Xiong,* Martin Skitmore, Bo Xia, Md Asrul Masrom, Kunhui Ye, Adrian
Bridge. International Journal of Project Management, 2014, 32(3), 482-491.
Contributor Statement of contribution
Bo Xiong
Conducted a literature review, designed the research, wrote the
manuscript and acted as the corresponding author.
07/03/2016
Martin Skitmore Directed and guided this study, and proofread the manuscript..
Bo Xia Assisted with the interpretation of results and manuscript
revisions.
Md Asrul Masrom provided data for validation.
Kunhui Ye Assisted with the interpretation of results and manuscript
revisions.
Adrian Bridge Assisted with questionnaire design.
Chapter 5: Job satisfaction 119
Principal Supervisor Confirmation
I have sighted email or other correspondence from all Co-authors confirming their
certifying authorship.
Martin Skitmore
___________________ _____________________ _________________
Name Signature Date
120 Chapter 5: Job satisfaction
5.2.1 Introduction
The construction industry plays an important role in providing employment
opportunities and enhancing economic development, especially in developing
countries such as China, India, and Malaysia (Doloi et al. 2012; Ye and Xiong 2011;
Yong and Mustaffa 2012). However, the industry has a poor record for project
success in terms of cost, time, quality, etc. Participant satisfaction is a crucial aspect
of this, as noted by Al-Tmeemy et al. (2011) and Leung et al. (2004), in addition to
qualified project completion.
Participant satisfaction describes the level of “happiness” of project
participants and slow decisions made by clients, poor labour productivity, and
architects' reluctance to change, for example, contribute to both reduced satisfaction
and unsuccessful projects (Doloi et al. 2012). Enhanced satisfaction, therefore, not
only helps to improve motivation and cooperation among participants but also
increases the likelihood of successful project completion, making its evaluation
important in judging the success or otherwise of a project.
Construction contractors are responsible for the actual production work
involved (cost management, schedule management, quality management etc.) in
projects and so their performance is critical to the success of projects. Furthermore,
replacing a contractor with another during project execution is very costly. It is
therefore important to understand the factors influencing contractor performance, and
measuring the degree of contractor satisfaction offers a means of achieving this as
well as providing an opportunity to enhance the effectiveness of cooperation between
contractors and other participants. That is to say, contractor satisfaction is central to
maintaining the cohesiveness and level of teamwork needed for a project (Chan et al
2002).
Previous satisfaction research in construction, however, is concerned much
more with the satisfaction of clients and customers than that of contractors. In
addition, current limited studies on measuring contractor satisfaction consider only
the effects of client behaviour and regard satisfaction holistically (Soetanto and
Proverbs 2002). A more detailed, multi-dimensional account of contractor
satisfaction will take into account the behaviour of the different participants
involved.
Chapter 5: Job satisfaction 121
Structural equation modelling (SEM) enables this to be done. Developed
from data collected by a postal survey of Malaysian construction contractors, a
structural equation model demonstrates that project participants appear to
fundamentally influence contractor satisfaction on two dimensions: economic-related
satisfaction and production-related satisfaction. Corresponding hypotheses are also
developed and tested by applying SEM, describing the causal relationships involved
in terms of satisfaction dimensions and associated participant performance factors.
5.2.2 Introduction
The concept of customer satisfaction emerged in the early 1980s in the USA
and subsequently widely used in the fields of psychology, business, marketing and
economics (Liu and Leung 2002). Defined as the response to the difference between
‘How much is there?’ and ‘How much should there be?' (Wanous and Lawler 1972),
satisfaction is particularly useful in the measurement of performance outcomes
(Nerkar et al. 1996).
In the construction industry, the term ‘satisfaction’ has become progressively
used over the past decade, its increased attention being taken to indicate a positive
change from a pure focus on business performance to a greater emphasis on
stakeholder performance (Love and Holt 2000). Therefore, in addition to the
traditional objective outcome measures of time, cost and quality, measuring
satisfaction has become another effective way of helping to improve project
performance, especially for large and complex projects (Cheng et al. 2006; Ling et
al. 2008; Toor and Ogunlana 2010). Furthermore, satisfaction can boost repeat
business and increase long-term profitability (Wirtz 2001).
There exist a variety of applications of satisfaction measurement in the
construction context. These comprise studies of client satisfaction levels associated
with contractor and consultant performance (Cheng et al. 2006; Mbachu and Nkado
2006); customer satisfaction with the products and services of the industry (Maloney
2002; Yang and Peng 2008); and home buyer and occupant satisfaction in terms of
comfort (Paul and Taylor 2008; Torbica and Stroh 2001). Leung et al. (2004) also
measures the degree of correlation between project participant satisfaction and
potential contributing factors.
122 Chapter 5: Job satisfaction
However, although there are studies measuring contractor performance,
contractor satisfaction has received much less attention. The sole example to date is
that of Soetanto and Proverbs (2002), who establish an overall contractor satisfaction
regression equation based on responses from 80 top UK contractors. However, this is
restricted to the measurement of contractor satisfaction exclusively in response to
client behaviour. Extending this to accommodate the influence of other participants
has yet to be undertaken.
Satisfaction in the construction industry is also viewed as a holistic entity in
current research on client satisfaction, homebuyer satisfaction and contractor
satisfaction (Cheng et al., 2006; Kärnä et al., 2009; Paul and Taylor 2008; Soetanto
and Proverbs 2002). However, research conducted in the manufacturing industry
demonstrates the importance of distinguishing economic satisfaction from non-
economic satisfaction in manufacturer-distributor relationships (del Bosque
Rodríguez et al., 2006). Although construction is uniquely different to manufacturing
in many ways, the role of manufacturers in the production and transfer of products to
the market via distributers has some similarity with the role of construction
contractors, who construct and transfer products to clients directly or via client to end
users. It is likely, therefore, that construction contractor satisfaction will benefit from
receiving a similar decomposition.
5.2.3 Research method
To examine the influence of participant performance factors on contractor
satisfaction, two main research methods are adopted: questionnaire survey and
structural equation modelling (SEM). Eighteen hypotheses are first proposed
according to the literature review. A conceptual model is then developed based on
these hypotheses by SEM. In the questionnaire design, Keline's (2005) principle,
which uses three measurement variables to reflect one latent variable, is applied in
order to obtain a stable equation structural model. 125 complete and reliable
responses collected from contractors in Malaysia comprise the basis for the data
analysis.
Hypotheses
One conceptualisation of satisfaction is in the form of an input-process-output
system where, although the internal process is still unknown, performance outcomes
Chapter 5: Job satisfaction 123
provide an input leading to satisfaction/dissatisfaction as the output (Soetanto and
Proverbs 2002). Performance outcomes are determined by different construction
project participants, with contractors, as performance assessors, having their own
psychological interpretation of the performance levels of others (Soetanto and
Proverbs 2002). Thus, the satisfaction of contractors is treated as being caused by
participant performance.
The Construction Industry Development Board (CIDB), which was
established by the Malaysian Federal Government in 1994 and is in charge of
planning direction of the industry, reported in its 2006-2015 construction industry
plan that project failures are not solely caused by contractors, but also by other
participants, such as the architect, engineer, subcontractors and suppliers (CIDB
2006). It is clear, therefore, that project success depends on the efforts of all
participants, as unsatisfactory work by any one participant can lead to the failure of a
whole project. In addition, delayed government projects in Malaysia are known to be
due not only to the poor performance of contractors, but also to a lack of
communication between participants, inadequate client finance and late provision of
construction drawings by consultants (Sambasivan and Soon 2007).
Adapting del Bosque Rodríguez et al. (2006), contractor satisfaction is
divided into two dimensions: economic-related satisfaction (ES) and production-
related satisfaction (PS). The former dimension refers to contractor satisfaction with
economic issues such as project cost, project profitability and potential business
opportunities arising from current projects. In contrast, production-related
satisfaction refers to contractor satisfaction with production quality, including project
quality, safety and timely completion.
The measurement of contractor satisfaction should therefore take into account
the effects of several participants. Perhaps the most important of these is the client,
who plays an important role in both project completion and contractor satisfaction.
Several infrastructure projects in Jordan, for example, have suffered in terms of
delays due to client-related factors, including finance, payments for completed work,
and slow decision making (Odeh and Battaineh 2002). Similarly, massive client-led
changes in project scope have caused up to 70% poor time performance in Saudi
Arabian projects (Assaf and Al-Hejji 2006). Also, Park's (2009) survey of 27
contractors found effective preplanning and client clarity of intention to be the most
124 Chapter 5: Job satisfaction
important factors affecting scope dimension and project success in South Korea. This
suggests corresponding hypotheses of:
• H1: The client's clarity of objectives (OC) has a positive influence on ES.
• H2: OC has a positive influence on PS.
• H3: OC has a positive influence on DC
• H4: OC has a positive influence on construction risk management (RM).
• H5: The client's promptness of payment (PP) has a positive influence on ES.
• H6: PP has a positive influence on PS.
Suitable design is another crucial factor to project success, with contractors
regarding defective design as a major risk in South Korea (Park 2009), for example,
while accounting for 50% of quality failures in Malaysia (CIDB 2006), leading to the
corresponding hypotheses:
• H7: DC has a positive influence on ES.
• H8: DC has a positive influence on PS.
An increasing number of project uncertainties have a fundamental effect on
project performance in the UK (Atkinson et al. 2006). These uncertainties lead to
negative relationships between parties, conflicts, mismatched objectives and
adversarial relationships (Harmon 2003). Construction risk management provides a
means of overcoming this to some extent and is therefore necessary to project
success, with the corresponding hypotheses being:
• H9: RM has a significant influence on ES.
• H10: RM has a significant influence on PS.
• H11: RM has a positive influence on PP.
• H12: RM has a positive influence on the effectiveness of other project
participants' work (EW)
Chapter 5: Job satisfaction 125
• H13: RM has a positive influence on respect and trust among project
participants (RT)
The ineffective contribution of other project participants is recognised as a
major cause of project failure, being attributed to poor schedule performance in
Saudi Arabia for example, particularly in public projects (Al-Kharashi and Skitmore
2009). Similarly, the performance of subcontractors and suppliers is also an
important factor contributing to the success and quality of construction projects in
Finland (Kärnä et al. 2009), giving rise to the corresponding hypotheses of:
• H14: EW has a positive influence on ES.
• H15: EW has a positive influence on PS.
Participant attitudes during the project are also very important in influencing
collaborative work and service quality (Ling and Chong 2005; Soetanto and Proverbs
2002). Similarly, enhancing understanding and trust among project participants is
beneficial in increasing the satisfaction levels of all participants (Lehtiranta et al.
2012). The corresponding hypotheses are:
• H16: RT has a positive influence on ES.
• H17: RT has a positive influence on PS.
• H18: RT has a positive influence on EW
All these hypotheses together comprise a conceptual model, which is also
regarded as the structural component in the perspective of SEM, as illustrated in
Figure 5.4.
126 Chapter 5: Job satisfaction
Figure 5.4 Structural component
Structural equation modelling
The structural equation modelling (SEM) technique is widely used to explore
and test causal relationships in the social sciences, such as in psychology, education
and health. SEM is a combination of factor analysis, multiple correlation, regression
and path analysis. Compared with other multivariate analysis methods, such as
multiple regression and neural networks, SEM has the ability to (1) estimate multiple
and interrelated dependence relationships; (2) represent unobserved concepts in these
relationships; (3) consider measurement errors in estimation; and (4) define a model
explaining an entire set of relationships (Keline 2005; Cho et al. 2009).
Because of these advantages, SEM is being increasingly used in construction-
related studies. For example, Islam and Faniran (2005) construct an SEM model to
investigate three factors influencing project-planning effectiveness; Cho et al. (2009)
use SEM to explore the effects of project characteristics on project performance;
while Anvuur and Kumaraswamy (2012) investigate the effects of four job cognition
variables on four cooperative behaviours. SEM is also recommended for increased
use in the construction industry due to its suitability in solving construction-related
problems (Oke et al. 2012). Likewise, SEM applied here aims to provide a way to
investigate the effects of participant performance factors on two contractor
satisfaction dimensions.
Chapter 5: Job satisfaction 127
SEM describes the relationships between two kinds of variables: latent and
observed. Latent variables cannot be observed directly due to their abstract character.
In contrast, observed variables contain objective facts or use an item rating scale in a
questionnaire. Several observed variables can reflect one latent variable (Byrne 2010;
Islam and Faniran 2005). One structural equation model divides into two
components: the measurement and the structural component. The measurement
component consists of the measurement errors of the measurement variables and the
relationships between observed variables and the represented latent variable. The
structural component expresses the relationships among latent variables. Thus, a
structural equation model consists of one structural component and several
measurement components (Washington et al. 2011). A two-step modelling method is
usually used to develop a structural equation model in preference to establishing the
model directly (Anvuur and Kumaraswamy 2012; Byrne 2010). This comprises, first,
a confirmatory factor analysis (CFA) followed by SEM. The aim of the CFA is to
test the validity of the measurement components and provide the foundation for the
next step. If the goodness of fit is satisfactory in the CFA phase, the next step is to
replace the correlations between the latent variable with hypothesized causal
relationships and then test the model.
Of course, as with all analyses of this kind, the existence of statistical
correlation or association does not prove causation or influence but simply lends
support to the logical or intuitive belief in their presence. Bearing this in mind, the
word 'influence' denotes "appears to influence" rather than to indicate any irrefutable
proof of such influence.
To apply SEM, many computer software systems, such as AMOS, EQS and
LISREL, have been developed (Jyh-Bin and Shen-Fen 2008). Of these, the SPSS
AMOS version 19 is used to construct and analyse the contractor satisfaction model.
Based on SEM, Figure 5.4 shows the structural component composed of all the
hypotheses that describe the direct relationships between two variables.
Structural equation modelling Questionnaire Survey
The hypotheses shown in Figure 5.4 are tested according to Keline’s three-
variable principle, where three observed variables are used to reflect a latent variable
(Keline 2005). To do this, the observed variables are extracted from Masrom's (2011)
larger questionnaire of Malaysian contractors, formerly used to construct a multiple
128 Chapter 5: Job satisfaction
regression contractor satisfaction model from 95 contributing factors. Bearing in
mind the requirement of high reliability and clear classification, both subjective
methods (e.g. brainstorming) and statistic methods (e.g. reliability testing) were used
to obtain the measurement framework as shown in Table 5.5.
Table 5.5 Constructs and measurement of SEM
Latent variables Abbr. No. Items
Performance variables: Which performance level would you rate? (1=very bad, 5=very good)
Client's clarity of
objectives OC Q1 Quality of project brief (e.g. needs and requirements)
Q2 Completeness of project brief
Q3 Certainty of project brief
Client's promptness of
payments PP Q4 Ease of final account settlement
Q5 Speed of final account settlement
Q6 Promptness of progress payment made by the client
Designer carefulness DC Q7 Design constructability
Q8 Comprehensiveness of design
Q9 Flexibility of design to accommodate changes
Construction risk
management RM Q10 Efficiency of risk control (e.g. identification, evaluation)
Q11 Effectiveness of conflict management
Q12 Appropriateness of sharing risks with other participants
Effictiveness of other
project participants EW Q13 Productivity of project manpower
Q14 Efficiency of subcontractor to undertake their work
Q15 Supplier effectiveness in material supply
Respect and trust among
project participants RT Q16 Participants’ respect and friendliness during the project
Q17 Trust between participants and project team
Q18 Understanding between participants and project team
Satisfaction variables: Which satisfaction level would you rate? (1=very dissatisfied, 5=very satisfied)
Economic-related
satisfaction ES Q19 Project cost management performance (actual vs budget)
Q20 Project profitability
Q21 Potential business development in future
Production-related
satisfaction PS Q22 Schedule performance (actual vs budget)
Q23 Construction product quality performance
Q24 Safety of worksite
Chapter 5: Job satisfaction 129
Data
The data comprise 125 responses from senior experienced personnel, with a
41.7% valid response rate. This is comparable with the previous SEM studies, e.g.
Islam and Faniran (2005) with 52 cases (61% response rate), Cho et al's (2009) 151
cases and Anvuur and Kumaraswamy's (2012) 153 cases (18% response rate), while
exceeding the minimum of 100 cases for SEM suggested by Gorsuch (1983) and
Bagozzi and Yi (2012). Of the respondents, 17.6% companies have been in business
for 1-5 years; 28% for 6-10 years; 20.8% for 11-15 years; 12% for 16-20 years; and
21.6 % companies for more than 20 years. Concerning company size, 53.6% are
large companies (G7), and 46.4% are small to medium companies (G1-G6)
according to the company size criteria and corresponding tendering capability in
Malaysia (CIBD 2006). Table 5.6 describes the basic characteristics of the
respondents, and further details are contained in Masrom (2011).
Table 5.6 Details of respondents
Respondent's information Groups Frequency Percent Cumulative Percent
Education level Certificate 11 8.8 8.8
Diploma 39 31.2 40
Bachelor degree 69 55.2 95.2
Master degree 6 4.8 100
PHD 0 0 100
Education background Architecture 9 7.2 7.2
Project management 32 25.6 32.8
Quantity surveying 31 24.8 57.6
Civil engineering 40 32 89.6
other 13 10.4 100
Management position Top level 61 48.8 48.8
Middle level 57 45.6 94.4
Low level 7 5.6 100
Experience 1-5 years 22 17.6 17.6
6-10 years 52 41.6 59.2
11-15 years 26 20.8 80
16+ years 25 20 100
Reliability test
Cronbach’s alpha value is used to test the reliability of the hypothesized
construct based on the data. If a Cronbach’s alpha value is above 0.7, the received
data is deemed to be acceptable for significant consistency (Cho et al. 2009; Doloi et
130 Chapter 5: Job satisfaction
al. 2012). As shown in Table 5.7, the items measured in eight variables and the
overall construct are sufficiently reliable.
Table 5.7 Reliability test of the questionnaire responses Variables All
24
Q1-3 Q4-6 Q7-9 Q10-
12
Q13-
15
Q16-
18
Q19-
21
Q22-
24
Chronbach’s
Alpha value
0.922 0.873 0.863 0.839 0.870 0.793 0.861 0.814 0.758
5.2.4 Results
A two-step method is used to develop the structural equation model.
Confirmatory factor analysis (CFA) provides the first step, and demonstrates a
satisfactory goodness of fit. Since the goodness of fit is satisfactory in the CFA
phase, the next step replaces the correlations between the latent variables with
hypothesized causal relationships as shown in Figure 5.4. Maximum likelihood
estimation is used to conduct both steps.
Confirmatory factor analysis
The measurement components are similar in structure. For example, as shown
in Figure 5.5, the client's clarity of objectives (OC) is reflected in three observed
items: Q1-Q3; and their measurement errors. The observed variables are shown in
rectangles, the latent variable in ellipses, measurement errors in circles and with
arrows indicating the direction of effects. To identify a measurement component, one
coefficient between the latent item and measurement items is given the value of unity
firstly before calculating the next step of standardization (Keline 2005). Likewise, a
starting value of unity is given between Q1 and OC. A dummy variable is used to
denote company size, with 0=small/medium and 1=large contractors.
Chapter 5: Job satisfaction 131
Figure 5.5 Measurement component
Table 5.8 presents the standardized regression weights and squared multiple
correlations (SMCs) for each observed item. Statistically significant standardized
regression weights of 0.5 or higher indicate good convergent validity (Anvuur and
Kumaraswamy 2012). In this case, all the regression weights (factor loadings) are
highly significant and range from 0.65 to 0.93 with the SMCs ranging from 0.42 to
0.86. For example, the SMC for ‘quality of project brief’ (Q1) is 0.67, indicating that
67% of the variance in ‘quality of project brief’ is explained by ‘client clarity of
objectives' (OC).
Table 5.8 Standardized regression weights and SMCs
Item Standardized regression weights
SMC OC PP DC RM EW RT ES PS
Q1 0.82a
0.67
Q2 0.91
0.82
Q3 0.81
0.65
Q4
0.93a
0.86
Q5
0.91
0.83
Q6
0.65
0.42
Q7
0.79
0.63
Q8
0.88
0.78
Q9
0.74a
0.55
Q10
0.81a
0.65
Q11
0.87
0.76
Q12
0.81
0.66
Q13
0.69a
0.47
Q14
0.87
0.75
Q15
0.70
0.48
Q16
0.74
0.54
Q17
0.91
0.82
Q18
0.84a
0.71
Q19
0.78a
0.61
Q20
0.77
0.60
Q21
0.76
0.58
Q22
0.73 0.53
Q23
0.71 0.51
Q24 0.72a 0.52
Note: All results are from analyses that included company size as a control variable. All
factors without superscript ‘a’ are significant at p<0.001; Factors with superscript ‘a’ are
fixed to 1.00 before estimation.
132 Chapter 5: Job satisfaction
Measuring the goodness of fit is an important part in developing structural
equation models and a large number of goodness of fit criteria has been developed
for this purpose (Washington et al., 2011). Generally, three types of model fit
measures are used to judge the fitness of the measurement components: absolute fit,
incremental fit and parsimonious fit (Ong and Musa 2012). Of these, Ong and
Musa’s criterion is used in this phase giving χ2 =311.391 (df =235, χ2/df = 1.325)
(Table 5.9). As the χ2/df value is between 1 and 2, this indicates an excellent fit
(Doloi et al. 2011).
Table 5.9 Results of goodness of fit (Adapted from Ong and Musa (2012))
Goodness of fit measure Index Criteria
χ2/df 1.325 <5.0
Absolute fit
RMSEA 0.051 <0.08
SRMR 0.045 <0.05
Incremental fit
CFI 0.957 >0.9
TLI 0.945 >0.9
Parsimonious fit
PNFI 0.665 >0.5
PGFI 0.610 >0.5
Structural equation modelling
As a 'good model' goodness of fit is obtained in the CFA phase, the correlations
between the latent variables are replaced by hypothesized causal relationship as
shown in Figure 5.4. The final model is shown in Figure 5.6, where the observed
variables Q1 to Q24 are shown in rectangles; latent variables such as OC are shown
in ellipses; with the arrows reflecting the hypothesized direction of effect. Figure 5.6
includes eight measurement components and the structural component which refers
to all latent variables and their interrelationships shown in Figure 5.4.The
measurement errors and factor loadings between the latent variables and
measurement variables are not shown as they are very similar to those in Table 5.8.
The company size variable continues to be dummy coded and is not shown. The
standardized coefficients of the hypothesized causal relationships are shown, with the
coefficients not significant at p<0.05 being shown in parentheses. The influence of
company size variable on ES and PS is quite weak.
Chapter 5: Job satisfaction 133
Figure 5.6 Final SEM model results
The SMC of ES in the model is 0.620, which indicates that 62% of the variance
in ES is explained by the six performance predictors and the dummy coded company
size variable. The SMC of PS is 0.713, indicating that 71.3% of the variance in ES is
explained by the six performance predictors and the dummy coded company size
variable. Both SMCs indicate usefulness in choosing contributing factors.
As can be seen in Figure 5.6, contractor satisfaction is significantly
influenced by the client's clarity of objectives (OC) and promptness of payment (PP),
designer carefulness (DC), construction risk management (RM) and effectiveness of
the other project participants (EW). Respect and trust among project participants
(RT) have no significant influence on economic-related satisfaction (ES) or on
production-related satisfaction (PS), but appears to affect ES and PS via EW. RT has
a positive effect on EW (r=0.414), which positively affects PS (r=0.466). However, a
significant test of indirect effects is needed to assess this fully.
The concept of indirect effects or mediation is invoked to investigate this
latter issue. In terms of the SEM model, if some variables act as mediators between
X and Y, then X has both a direct effect on Y and an indirect effect on Y via the
134 Chapter 5: Job satisfaction
mediating factors. Figure 5.6 already shows the direct effects between variables in
terms of the calculated coefficients. The Sobel test based on the work of Sobel
(1982) determines the significance of mediation effects. The probability column of
Table 5.10 summarises the results for the seven paths, together with the values of
indirect effects in the corresponding column. Clearly, since RT->EW->PS and PR-
>EW->PS are not significant, OC->RT->EW->PS and OC->PR->EW->PS are also
not significant.
Table 5.10 P values and indirect effects (Sobel test)
Paths Probability Indirect effect
RT ->EW->PS 0.068 0.193
RM->EW->PS 0.083 0.188
RM->PP ->ES 0.058 0.100
OC->RM->ES 0.022 -0.311*
OC->RM->PS 0.012 -0.360*
OC->DC->ES 0.002 0.367*
OC->DC->PS 0.004 0.353*
* indirect effects when p < 0.05
5.2.5 Findings and discussion
The main research finding is that DC, OC and PP positively influence (have a
positive influence on) ES while DC, EW and OC positively influence PS, with RM
negatively influencing ES and PS. Also, other than H6, H14, H16 and H17, 14
hypotheses shown in Figure 5.4 are supported.
The model development results shown in Figure 5.6 support the hidden
assumption that contractor satisfaction is caused by participant performance and that
satisfaction is divisible into economic-related satisfaction (ES) and production-
related satisfaction (PS). In addition, the performance variables have different effects
on the two satisfaction dimensions, with client prompt payments (PP) having a
positive effect on ES but no significant effect on PS. While the effectiveness of other
project participants (EW) has a positive effect on PS but no significant effect on ES.
Client clarity of objectives (OC)
That OC positively influences ES (β=0.314) and PS (β=0.342) supports the
importance of clear objectives and demonstrates the positive relationship between
clear objectives and contractor satisfaction. It confirms Park's (2009) finding that
effective preplanning, clarity of contract and understanding of project requirements
Chapter 5: Job satisfaction 135
rank highest in measuring critical success and Leung et al's (2004) assertion that goal
specificity is positively associated with goal commitment, which in turn is positively
associated with construction participant satisfaction. OC also has a positive effect on
DC (r=0.645) and RM (r=0.525). For indirect effects, OC influences ES (-0.311 and
0.367) by mediation of RM and DC respectively. Similarly, OC influences PS (-
0.360 and 0.353) by mediation of RM and DC respectively.
Client's promptness of payment (PP)
PP is characterized by ease and speed of final account settlement, and
promptness of progress payments made by the client. It has a positive effect on ES
(β=0.211), but its influence on PS is too weak to be significant. This confirms Yong
and Mustaffa's (2012) result in which the financial capability of client ranks the 1st
of 37 factors critical to project success in Malaysia, and Al-Kharashi and Skitmore's
(2009) finding that lack of finance and delay in progress payments are critical factors
for both clients and contractors in Saudi Arabian public projects. PP can ensure that
the contractor obtains sufficient cash flow during and after construction, and this is
probably why ES increases with PP. However, that the model does not show PP
having a significant influence on PS may be attributed to production issues such as
safety being influenced by government regulations etc.
Designer carefulness (DC)
DC refers to the quality of the designer's work, characterized by design
constructability, comprehensiveness of design, and the flexibility of the design to
accommodate changes in the measurement component. It is a key factor in the
model, with the strongest positive effect on both ES (β=0.569) and PS (β=0.548) and
supports H7 and H8 in which DC positively influences contractor satisfaction. This
confirms the widely acknowledged importance of design (Al-Kharashi and Skitmore
2009; Park 2009; Yong and Mustaffa 2012). OC's positive influence on DC supports
H3 and suggests that improving the clarity of project goals may also be beneficial in
improving the quality of design.
Construction risk management (RM)
RM is usually a duty of contractors, and it is characterized by the efficiency of
risk control, effectiveness of conflict management and appropriateness of sharing
risks with other participants in the measurement component. As can be seen in
136 Chapter 5: Job satisfaction
Figure 5.6, RM is positively influenced by OC (r=0.525), confirming that the level of
risk management, associated with level of project uncertainty as Masrom (2011)
notes, is largely related to project characteristics and client clarity of objectives. This
is consistent with Siang and Ali’s (2012) findings that systematic risk management is
not implemented actively by most of contractors in Malaysia and all three case
companies, which are publicly listed in Malaysia, rate “avoid unsatisfactory projects
and to enhance margins” as the least important of ten benefits of risk management
(two selected 10th and one 9th). In view of this, it is not surprising to find that the
model also indicates RM to have a strong negative effect on both ES (β=-0.592) and
PS (β=-0.686), suggesting that contractors are unhappy with the effectiveness of their
risk management despite it being critical to project success. This makes Soetanto and
Proverbs' (2002) finding (that contractor satisfaction is negatively influenced by the
perception that clients know exactly what they want) more understandable in that
higher OC is associated with the lower ES and PS when mediated by RM. However,
the model is more complex, with consideration of the direct effects of OC and
indirect effects via DC and RM. Also of note is that RM is positively related to RT
(r=0.721) and EW (r=0.404), both of which are critical to project success. That is to
say, although risk management does not appear to bring satisfaction to contractors in
Malaysia directly, it is already regarded as an important way of enhancing the
productivity and harmony of participants.
Effectiveness of other project participants (EW)
EW is characterized by the efficiency of subcontractors in undertaking their
work, supplier effectiveness in material supply and the productivity of project
manpower, and has a significant effect on PS (β=0.466). This confirms Yong and
Mustaffa's (2012) finding that the allocation of manpower ranks as the most
important project-related factor critical to project success. On the other hand, EW
does not have a strong effect on ES, which may be due to the price of work being
based on workload rather than efficiency considerations.
Respect and trust among project participants (RT)
Atkinson et al. (2006) state that trust can be used as a way of reducing
uncertainty, while enhancing trust is regarded as a better way to solve hidden
problems in the construction process, with shared authorities among participants
being a critical factor contributing to project success (Yong and Mustaffa 2012). This
Chapter 5: Job satisfaction 137
model confirms these findings in indicating that RT can enhance EW greatly
(r=0.414). Here, RT is characterized by the level of respect, understanding and trust
among participants. However, as Figure 5.6 shows, although RT can contribute to
project success, it does not have any significant effect on ES and PS. A similar
phenomenon occurs with Leung et al.'s (2004) finding that correlations do not exist
between the degree of participant satisfaction and their level of communication, or
the amount of authority clients and project managers have in setting project goals.
5.2.6 Conclusions
A framework is presented to measure construction contractor satisfaction,
which comprises two satisfaction dimensions: economic-related satisfaction (ES) and
production-related satisfaction (PS). This is used to develop a structural equation
model to investigate how project participants' performance affects contractor
satisfaction in terms of six factors: the client's clarity of objectives (OC) and
promptness of payments (PP), carefulness of the designer (DC), construction risk
management (RM), effectiveness of other project participants' work (EW) and
respect and trust among project participants (RT). The findings confirm 14
hypotheses and deny 4 hypotheses. In particular, the results support the view that
contractor satisfaction is a result of many participant effects and the six factors act
differently on ES and PS.
Three important implications can be concluded from these results. Firstly, it
is demonstrated that ES and PS provide a meaningful classification of contractor
satisfaction and that each is affected differently by the six predictors. Of special note
is that PP solely affects ES while EW solely affects PS. It is therefore necessary to
examine the internal dimensions of contractor satisfaction before their measurement,
as different types of satisfaction correlate differently with the different activities
involved.
Secondly, the developed model offers a potential means of improving
contractor satisfaction. For example, ES is influenced positively by OC, PP and DC,
and negatively by RM. Thus a possible way to improve ES and enhance project
success at the same time is for the client and designer to improve OC, PP and DC.
Reducing RM, on the other hand, is counter-productive as RM positively related to
138 Chapter 5: Job satisfaction
EW and RT, both important to project’s successful delivery. On the contrary, if
construction risk management level needs to improve for assuring project success, it
may be possible for the participants to combine to increase contractor satisfaction in
other ways, such as by improving OC, PP and DC or OC, DC and EW. This reflects
the delicate difference between ES and PS.
Thirdly, a theoretical foundation is provided for participants, especially the
client, to estimate the potential contractor satisfaction to be gained from the project
prior to selecting the project contractor. In previous studies and practices, the client
chooses the contractor by comparing bid prices without considering contractor
satisfaction. It could be that an unsatisfactory contractor with the lowest tender price
is much worse than a satisfactory contractor with a higher priced tender. For this
concern, it is necessary to figure out a way to compare contractor satisfaction at
tendering stage. Besides graphical way, SEM can also be expressed in regression
equation way and many such cases can be found in Keline (2005). Based on direct
significant effects showed in Figure 5.6 and indirect significant effects showed in
Table 5.10, two equations are proposed as follows to calculate changes of ES and PS
with participant performance factors’ change:
ΔES=0.569ΔDC+0.314ΔOC+ 0.211ΔPP-0.592ΔRM+(0.367-0.311)ΔOC and
ΔPS=0.548ΔDC+0.466ΔEW+ 0.342ΔOC-0.686ΔRM+(0.353-0.360)ΔOC.
For each equation, the first four components refer to direct effects from
participant performance factors and the last component refers to the indirect effects.
With these two equations, the client can effectively identify a more satisfied
contractor by evaluating and measuring the variation of these performance factors
among different bidding contractors with focusing on significant factors. Similarly,
the model provides the opportunity for contractors to estimate potential satisfaction
and choose a project with a higher likely level of satisfaction, especially in
circumstances where many bidding opportunities arise with similar profit
expectations. Alternatively, a contractor may decide to bid for projects only where
the expected satisfaction exceeds a specific threshold value. Further, as it is
reasonable to speculate that better contractors will have a higher satisfaction
threshold value, it would then benefit clients to attract good contractors to bid by
improving corresponding aspects such as OC, DC, EW and PP.
Chapter 5: Job satisfaction 139
It should be mentioned, however, that some potential limitations exist for
further development. The data are all from a sample of contractors in Malaysia and
therefore, although the conclusions are certainly valid for the sample, and probably
so for most Malaysian contractors, their applicability outside Malaysia is uncertain,
even in other developing countries. Differences in awareness and practices of risk
management should be considered particularly when applying similar research in
other countries. In addition, although the sample size of 125 used in the study meets
the requirements for conducting SEM generally, more data is needed for the
development of a complex model and improved model fit. For future research,
benefits are envisaged in further exploring the internal dimensions of contractor
satisfaction, a more detailed study of the relationship between contractor satisfaction
and project success, and the evaluation of satisfaction (for the client) to choose
contractors or (for contractors') decision to bid. The results also suggest that future
research in the Malaysia context may benefit from a more simplified data collection
instrument based on reduced number of hypotheses.
140 Chapter 5: Job satisfaction
5.3 THE NEXUS BETWEEN CONTRACTOR SATISFACTION AND
PROJECT MANAGEMENT PERFORMANCE
Statement of contribution
The authors listed below have certified that:
1. They meet the criteria for authorship in that they have participated in the
conception, execution, or interpretation, of at least that part of the publication in their
field of expertise;
2. They take public responsibility for their part of the publication, except for
the responsible author who accepts overall responsibility for the publication;
3. There are no other authors of the publication according to these criteria;
4. Potential conflicts of interest have been disclosed to (a) granting bodies, (b)
the editor or publisher of journals or other publications, and (c) the head of the
responsible academic unit, and
5. They agree to the use of the publication in the student’s thesis and its
publication on the Australasian Research Online database consistent with any
limitations set by publisher requirements.
In the case of this chapter:
The nexus between contractor satisfaction and project management
performance
Bo Xiong*, Martin Skitmore, Md Asrul Masrom, Bo Xia. A fine-grained analysis of
contractor satisfaction in promoting project management performance, Submitted to
Project Management Journal, under revision.
Contributor Statement of contribution
Bo Xiong
Conducted a literature review, designed the research, wrote the
manuscript and acted as the corresponding author.
07/03/2016
Martin Skitmore Directed and guided this study, and proofread the manuscript.
Md Asrul Masrom Provided data for validation and assisted in manuscript revision.
Bo Xia Assisted with the interpretation of results.
Chapter 5: Job satisfaction 141
Principal Supervisor Confirmation
I have sighted email or other correspondence from all Co-authors confirming their
certifying authorship.
Martin Skitmore
___________________ _____________________ _________________
Name Signature Date
142 Chapter 5: Job satisfaction
5.3.1 Introduction
Since the emergence of stakeholder theory (Freeman 1984), the notion of
stakeholder satisfaction (i.e. the satisfaction of organizations or groups of people) has
been widely accepted. With the increasing popularity of stakeholder management in
project-based industries such as construction, offshore engineering and software
development - accounting for almost 30% of the global economy (Turner 2008) -
studies of project participant satisfaction have become increasingly prominent over
the past decade, indicating a positive change in focus from individual performance to
a greater emphasis on stakeholder interests (Love and Holt, 2000; Toor and
Ogunlana, 2010). In addition to the traditional project management “iron triangle”
requirements in terms of measured cost, schedule and quality, satisfaction
measurement has been regarded as another effective means of improving
organisational performance (Cheng et al., 2006; Li et al., 2013; Toor and Ogunlana,
2010). Additionally, Project participant satisfaction becomes an early warning sign
of project outputs in complex projects, (Williams et al., 2012; Xiong et al., 2014).
Although the individual S-P nexus has been widely acknowledged and
explored, its counterpart nexus between organizational satisfaction and performance
has been little considered (Judge, et al., 2001). Pearsall and Ellis (2006), for example,
examine the effects of critical team member assertiveness on team performance and
team satisfaction, although without consideration of the S-P nexus. For the S-P nexus
of project participating companies, it could follow various studies at individual level
and assume that there is an overall S-P nexus or complex relationships among
performance and disaggregated satisfaction facets. A few extent studies on
contractor satisfaction like Xiong et al (2014) indicates that the need of satisfaction
disaggregation in revealing different effects of external factors on contractor
economic satisfaction and noneconomic satisfaction respectively. However, the
nexus between contractor satisfaction and contractor performance is still unknown.
This study aims to investigate the nature of the satisfaction-performance (S-P)
nexus of project participating companies with special cases from construction
contractors. A special concern is on dimensions of contractor satisfaction. In doing
this, two series of structural equation models are developed and compared. The first
group of models treats contractor satisfaction as a holistic concept in examining the
Chapter 5: Job satisfaction 143
S-P nexus. In conducting a fine-grained analysis of contractor satisfaction, the
second group of S-P models divides contractor satisfaction into two facets of
economic-related satisfaction (ES) and production-related satisfaction (PS) following
Xiong et al. (2014). Analyses are done through a survey of 117 construction
companies' satisfaction and performance levels relating to their recent project work.
5.3.2 Theoretical background and hypotheses development
Satisfaction is defined as a response function of the discrepancy between ‘How
much is there?’ and ‘How much should there be?' (Nerkar, et al., 1996; Wanous &
Lawler, 1972b). In early research in industrial settings, satisfaction is mainly referred
to as job satisfaction. This became a popular term with the Hawthorne studies in
emphasizing linkages between employees' job satisfaction and their performance - a
relationship described as the “Holy Grail” by industrial-organizational researchers
(Judge et al., 2001). For the weak empirical evidence relating to the individual S-P
nexus as prompted some researchers (e.g. Judge et al., 2001) argue that the
unsatisfactory outcomes in S-P linkage research have been mainly caused by the
ambiguity in defining satisfaction, although some measures of job satisfaction such
as the Job Descriptive Index (JDI) (Smith, 1969) and Minnesota Satisfaction
Questionnaire (MSQ) (Weiss, et al., 1967) have been developed. Economic
satisfaction should be treated separately for analysis since it is highly related to pay
factors such as pay equity (Brown, 2001). Lai (2007) divides dealer’s satisfaction
into social satisfaction and economic satisfaction when investigating the mediating
effects of satisfaction in the Taiwan motor industry, finding noneconomic
satisfaction to be much more important than economic satisfaction in influencing
performance.
A project is a natural and an ideal organizational form to develop products
facing increasing product complexity, changing markets, cross-functional expertise
cooperation, customer-oriented innovation and technological uncertainty (Hobday,
2000). For many cross-functional projects, the final product is not produced by
routine practices of workers in assembly lines but by the collaboration of
participating project organizations (Turner, 2008). In projects, the performance of
participating companies is not only affected by environmental factors, but also
affected by perceived satisfaction. As indicated in the stakeholder theory developed
by Freeman (1984), stakeholder satisfaction management is useful for solving
144 Chapter 5: Job satisfaction
corporate social-related issues effectively and generate good business performance
(Porter & Kramer, 2006). In many cases, perception of success by complex projects
has little to do whether they are completed on time, at cost and with the desired
quality, but achievement of the desired objectives of different stakeholders (Turner
and Zolin, 2012). Participant satisfaction has been accepted by many as a new
dimension of project success (Liu and Walker, 1998) and a new approach for early
warning sign (Xiong et al., 2014).
With the increasing popularity of project economics and stakeholder theory,
project participant satisfaction has become an emergent area (Orlitzky & Swanson,
2012). The organizational satisfaction levels of project participating companies in
the construction industry, such as clients and contractors, have also been studied by a
few researchers (e.g. Tang et al., 2003; Masrom et al. 2013). For example, client
satisfaction has especially been regarded by some researchers as a criterion for
defining success (De Wit, 1988; Munns and Bjeirmi, 1996). However, previous
studies focus on measuring and exploring the driving factors of participant
satisfaction (Leung, et al., 2004; Li, et al., 2013), contractor satisfaction (Masrom, et
al., 2013; Soetanto & Proverbs, 2002; Xiong, Skitmore, Xia, et al., 2014) and client
satisfaction (Cheng, et al., 2006; Mbachu & Nkado, 2006; Yang & Peng, 2008), all
of which have been studied in recent times. This study aims to explore the nexus
between satisfaction and performance of construction contractors by taking sub-
dimensions of contractor satisfaction into comparisons. Previous research indicates
that contractor satisfaction should include two dimensions: noneconomic satisfaction
and economic satisfaction. Geyskens, Steenkamp, and Kumar (1999)’s meta-analysis
of 71 studies of satisfaction in marketing relationships resonates with Lopez (1982)
results in demonstrating the necessity to distinguish between economic and
noneconomic satisfaction as distinct constructs with different causes and
consequences. Likewise, del Bosque Rodríguez, Agudo, and Gutiérrez (2006)
demonstrate the necessity of such a categorization by examining the determinants of
economic and noneconomic satisfaction in manufacturer-distributor relationships. In
addition to the supply-chain, such a separation of organizational satisfaction is also
applicable in construction projects. For example, a similar outcome was obtained by
Xiong et al (2014) in examining effects of other key project participants’
performance on the economic satisfaction (ES) and production satisfaction (PS) of
Chapter 5: Job satisfaction 145
construction companies. However, few studies combine this categorization with
studies on the S-P nexus.
5.3.3 Methodology
Conceptual models and hypotheses
Project outcomes are generated by the combined efforts of different project
participants, and contractors, as key performance assessors, have their own
psychological interpretation of other participant's performance (Soetanto & Proverbs,
2002). For the influences of other project participants, Wang and Huang's (2006)
survey of construction supervising engineers in China found client performance to be
significantly correlated with project success. A major contributing factor seems to be
client-led changes in project scope, which cause up to 70% of poor schedule
performance in Saudi Arabian projects for example (Assaf & Al-Hejji, 2006). Poor
performance of designers also contributes significantly to delayed government
projects in Malaysia, where defective designs account for most quality problems
(Sambasivan & Soon, 2007). In addition, Xiong et al (2014) identified clients' scope
clarity and designer performance as the most significant variables of contractor
satisfaction. Therefore, the influences of clients and designer are considered as
antecedent variables when establishing models to test the S-P nexus. nexus of
construction contractors.
This study disaggregated satisfaction into two parts and proposed different
directional relationships based on previous studies. Puzzled by the weak empirical
evidence concerning both directional S-P links at individual level, Schwab and
Cummings (1970) point out that the unsatisfactory outcome may be caused by the
ambiguity in defining job satisfaction. Although some researchers follow this
statement and explore the relationships between disaggregated satisfaction
components and performance (e.g. Nerkar et al., 1996; Lai, 2007), these studies
assume that all disaggregated satisfaction facets share the same unidirectional
relationship with performance. If such an inherent assumption is correct and these
studies obtain significant results, the link between holistic satisfaction and
performance should also be significant, which is not consistent with the conclusions
achieved to date. To test the S-P nexus of the Malaysian construction contractors, it
is proposed that the directions between satisfaction components and performance are
different.
146 Chapter 5: Job satisfaction
Two conceptual models describing the relationship between contractor
satisfaction (COS) and contractor project managenet performance (CPMP) are
developed.
Conceptual model 1
Conceptual model 1 is developed based on the assumption that COS is a
holistic concept comprising economic related satisfaction (ES) and production
related satisfaction (PS). ES refers to organisation satisfaction with economic issues
such as cost, profitability and potential business opportunities arising from current
business activities. In contrast, PS refers to organisation satisfaction with production
or service quality (Xiong et al. 2014). The contractor S-P nexus is explored by
including influences on the performance of clients and designers in terms of the
clarity of customer needs (OC) and carefulness in the design of products or services
to meet those needs (DP).
The following hypotheses are proposed: H1 concerns the relationships between
the performance of customers and designers separate from the performance and
satisfaction of the organisation. H2 concerns the linkage between organisation
satisfaction and performance, with H2A and H2B indicting that COS affects CPMP
and CPMP affects COS respectively (H2A and H2B are contained in two separate,
but otherwise identical models named Model 1A and Model 1B).
Hypothesis1A: OC has a positive direct effect on COS
Hypothesis1B: OC has a positive direct effect on CPMP
Hypothesis1C: OC has positive effects on DP.
Hypothesis1D: DP has a positive direct effect on COS
Hypothesis1E: DP has a positive direct effect on CPMP
Hypothesis2A: COS has a positive direct effect on CPMP
Hypothesis2B: CPMP has a positive direct effect on COS
Chapter 5: Job satisfaction 147
Figure 5.7 Conceptual model 1
Conceptual model 2
Similar to conceptual model 1, the influences of OC and DP are considered in
exploring the contractor S-P nexus. A difference here is that satisfaction is divided
into ES and PS, prompted by Geyskens, et al. (1999), Rodríguez, et al.(2006), Lai
(2007) and Xiong et al (2014) as leading examples. Unlike these previous studies, the
different directional relationships that ES and PS may have with performance are
included here. As suggested by the previous literature, this research proposes that PS
has a positive effect on CPMP and CPMP has a positive effect on ES. The
hypotheses are as follows:
H3 concerns relationships between performance of customers and designers
separately with performance and satisfaction of the organisation. H4 concerns the
linkage between the two facets of organisation satisfaction and its performance.
Hypothesis3A: OC has a positive direct effect on ES
Hypothesis3B: OC has a positive direct effect on CPMP
Hypothesis3C: OC has a positive direct effect on PS
148 Chapter 5: Job satisfaction
Hypothesis3D: OC has a positive direct effect on DP
Hypothesis3E: DP have positive direct effects on CPMP
Hypothesis3F: DP has positive effects on PS.
Hypothesis4A: PS has a positive direct effect on CPMP
Hypothesis4B: CPMP has a positive direct effect on ES
Figure 5.8 Conceptual model 2
Questionnaire survey
The data comprise questionnaire survey 117 usable responses from 136
received responses of Masrom (2012)’s survey of 300 Malaysian contractors
registered with Construction Industry Development Board (CIDB).Respondents are
mostly senior personnel and provided feedback based on the company’s most recent
construction project. The companies are evenly distributed in terms of size, with
53.0% being large companies (G7) and 47.0% small to medium companies (G1-G6)
categorized by company size and permitted tendering capability (see Masrom, 2012).
Company size is used as a control variable in model development, with 0 =
small/medium and 1 = large contractors. Of the company representatives, 93.2%
have a diploma or higher degree and 82.9% have more than 5 years’ working
experience. Table 5.11 presents basic information concerning the construction
projects involved.
Table 5.11 Description of projects
Chapter 5: Job satisfaction 149
Project characters Groups Frequency Percent Cumulative
percent
Client type Federal government 40 34.19% 34.19%
State or local authority 30 25.64% 59.83%
Private sector 38 32.48% 92.31%
other 9 7.69% 100.00%
Procurement method Traditional (DBB) 75 64.10% 64.10%
Management contract 12 10.26% 74.36%
Design and build 25 21.37% 95.73%
other 5 4.27% 100.00%
Project duration < 1 year 59 50.43% 50.43%
1-2 year 35 29.91% 80.34%
2-3 year 14 11.97% 92.31%
> 3 year 9 7.69% 100.00%
The sample size here exceeds the recommended number of 100 cases
suggested by Bagozzi and Yi (2012) for SEM, and is comparable with previous SEM
studies in the construction industry (Xiong, et al., 2015a). For example, Cheung and
Chow (2011) used 103 responses to explore the underlying factors contributing to
withdrawal in construction project dispute negotiation; and Wong and Lam (2011)
used 107 responses to investigate the effect of organization learning and unlearning
on the performance of construction organizations. In addition to the comparatively
simple model structure, each construct contains at least three variables, which assures
the model identification of each measurement construct and requires a smaller
sample size for fitting the model (MacCallum, et al., 1996; Xiong, et al., 2015a).
To test the conceptual models and corresponding hypotheses, the measurement
framework in Table 5.12 was built based on Keline (2005)’s three-variable principle,
where three observed variables are used to reflect a latent variable . To do this, the
observed variables are extracted from Masrom's (2012) larger questionnaire of
Malaysian general contractors, involving 95 relevant indicators.
Table 5.12 Measurement constructs and items
Constructs No. Items Main sources
What performance level would you rate your project? (1=very bad, 5=very good)
Client's clarity of
objectives (OC)
Q1
Quality of project brief (e.g. needs and
requirements)
Soetanto and Proverbs
(2002);Soetanto and
Proverbs (2004); Assaf
and Al-Hejji (2006);
Park (2009); Masrom
(2012)
Q2 Completeness of project brief
Q3 Certainty of project brief
150 Chapter 5: Job satisfaction
Designer
performance (DP)
Q4 Design constructability Tang, et al. (2003);
CIDB (2006); Yang and
Peng (2008);Park
(2009); Masrom (2012)
Q5 Comprehensiveness of design
Q6
Flexibility of design to accommodate
changes
Contractor project
management
performance (CPMP)
Q7 Productivity of project manpower Munns and Bjeirmi
(1996); Maloney (2002);
Soetanto and Proverbs
(2004); Tang, et al.
(2003); Cheng, et al.
(2006); Wang and Huang
(2006); Yang and Peng
(2008);Park (2009)
Q8
Two- way communication between
participants and your project team
Q9
Collaborative work between participants
and your project team
Q10
Quality of relationship between
subcontractors and your project team
Which satisfaction level would you rate? (1=very dissatisfied, 5=very satisfied)
Production-related
satisfaction (PS)
C11 Schedule performance (actual vs budget) Schwab and Cummings
(1970); Nerkar, et al.
(1996); Geyskens, et al.
(1999); del Bosque
Rodríguez, et al. (2006);
Lai (2007); Masrom
(2012); Masrom, et al.
(2013); Xiong, Skitmore,
Xia, et al. (2014)
C12 Construction product quality performance
C13 Safety of worksite
Economic-related
satisfaction (ES)
C14
Project cost management performance
(actual vs budget)
C15 Project profitability
C16 Potential business development in future
Chronbach’s alpha value is used to test the reliability of the hypothesized
construct based on the data, where a value exceeding 0.7 is taken as indicating the
received data is acceptable for meeting the consistency requirement (Cho, Hong, &
Hyun, 2009; Lai, 2007). As shown in Table 5.13, the items are measured in five
variables and the overall constructs are sufficiently satisfied.
Table 5.13 Reliability test
Variables All16 Q1-3 Q4-6 Q7-110 Q11-13 Q14-Q16
Cronbach’s Alpha value 0.914 0.874 0.85 0.805 0.753 0.806
Structural equation modelling: CB-SEM and PLS-SEM
Structural equation modelling (SEM) is widely accepted and used in exploring
and testing relationships among different constructs in the social science disciplines
and its evolution is regarded as the most important statistical progress in social
sciences in recent decades (Hair, Ringle, & Sarstedt, 2012). A structural equation
model includes observed variables and latent variables that are hard to observe
Chapter 5: Job satisfaction 151
directly due to their abstract character and are represented by using several observed
variables (Byrne, 2010). According to model structures, one structural equation
model generally comprises a structural component consisting of the relationships
among latent variables and several measurement components, which consist of the
measurement errors of the measurement variables and the relationships between
observed variables and the represented latent variable (Washington, Karlaftis, &
Mannering, 2010). Compared with first generation models such as principle
component analysis and linear regression, SEM is a second generation multivariate
analysis method (Fornell & Larcker, 1987). It has many strengths, such as enabling
the use of one model to explore an entire set of complex relationships, or the use of
several observable items to represent ambiguous constructs (Cho, et al., 2009;
Fornell & Larcker, 1987; Keline, 2005). Additionally, the popularity of SEM is
enhanced by the availability of many SEM software packages offering graphical
interfaces for model development (Xiong, et al., 2015a).
There are two types of SME approaches: CB-SEM and PLS-SEM. CB-SEM is
appropriate for confirming theoretical hypotheses as it focuses on minimizing the
difference between the model-implied covariance matrix and the sample covariance
matrix, and obtaining accurate parameter estimates. In contrast, PLS-SEM is
preferred in prediction as it focuses on maximizing the explained variance of targeted
constructs (Hair, Ringle, et al., 2012). CB-SEM is more popular in theoretical studies
for its stricter rules concerning data and sample size and accurate estimates of
parameters, while PLS-SEM has recently increased in popularity for its ability to
provide accurate predictions of target variables with a comparatively small sample
size. Additionally, PLS-SEM can handle both reflective measurement constructs and
formative measurement constructs, while CB-SEM can only handle the reflective
measurement construct (Ringle, et al., 2012). In reflective constructs, changes in
latent variables lead to changes in observed variables, while changes in observed
variables do not lead to changes in latent variables, which is important for explaining
the selected latent variable when deleting an observed variable. In formative
constructs, changes in latent variables do not lead to changes in observed variables,
while changes in observed variables lead to changes in latent variables, which
changes the theoretical meaning significantly when deleting an observed variable
(Jarvis, et al., 2003).
152 Chapter 5: Job satisfaction
In the current context, organisational satisfaction is represented by company
satisfaction (COS), which is composed of economic related satisfaction (ES) and
production related satisfaction (PS) with both ES and PS being reflected by several
observed variables, which means that COS in conceptual model 1 is a second order
reflective-formative model and therefore more suitable for PLS-SEM (Becker et al.,
2012). However, as Hair et al (2012) argue, CB-SEM and PLS-SEM have different
strengths and should be complementary rather than conflicting. In this case, both
methods are used - PLS-SEM for developing conceptual model 1 and CB-SEM for
developing conceptual model 2. The software SmartPLS 2.0 (Ringle, Wende, &
Will, 2005) and SPSS AMOS 21.0 are used for developing the models accordingly.
5.3.4 Results
Conceptual model 1
PLS-SEM has become a popular SEM method in recent years for its ability to
handle both reflective and formative measurement constructs (Becker et al., 2012;
Jarvis et al., 2003; Ringle et al., 2012). For this study, satisfaction is divided into
economic related satisfaction and non-economic satisfaction that is specific to
production related satisfaction. That is to say contractor satisfaction is a formative-
reflective construct. To test H2A and H2B separately, two models are constructed.
Each contains three latent variables (OC, DP, CPMP) with corresponding reflective
indicators, one second-order hierarchical latent variable (COS) and two first-order
constructs (reflective) that form the second-order construct (formative). The repeated
indicator approach is used for the conceptual model for its comparatively high
reliability and wide applications in reflective-formative construct problems (Becker,
Klein, & Wetzels, 2012; Chin, 2010; Ringle, et al., 2012). For this approach, a
higher-order latent variable comprising lower-order latent variables can be
constructed by representing all the observed variables belonging to the lower-order
latent variables. This approach can estimate the scores of latent variables
simultaneously instead of estimating different order constructs separately and then
uses the needed construct scores to test the proposed relationships as a separate step
(Hair Jr, Hult, Ringle, & Sarstedt, 2013; Ringle, et al., 2012).
Model fit evaluation
Chapter 5: Job satisfaction 153
To validate the measurement components, three types of validity are assessed:
internal consistency reliability, convergent validity and discriminant validity.
Compared with Cronbach’s alpha values presented in Table 5.8, values of composite
reliability (CR) are preferred in the use of internal consistency reliability as such
measures do not assume that the observed variables share the same out-loadings
(Hair Jr, et al., 2013). The average variance extracted (AVE) is used to test
convergent validity. As Table 5.8 indicates, the values of composite reliability and
AVE are greater than the required 0.7 and 0.5 respectively (Fornell & Larcker,
1981b; Hair Jr, et al., 2013); all factor loadings of the observed variables on latent
variables are significant at the level of 0.01; the discriminant validity is satisfactory
as the square root of the average variance extracted for each construct is more than
the maximum correlations with other constructs (Fornell & Larcker, 1981b); and the
loadings are greater than the cross loadings by 0.1 as required (Hair Jr, et al., 2013).
Table 5.14 Validity test results
Latent
variables CR AVE
Correlations between constructs
OC DP CPMP ES PS
OC 0.923 0.800 0.894
DP 0.909 0.769 0.531 0.877
CPMP 0.872 0.631 0.499 0.635 0.795
ES 0.887 0.724 0.500 0.530 0.411 0.851
PS 0.861 0.674 0.480 0.484 0.469 0.640 0.821
COS 0.889 0.572 N/A N/A N/A N/A N/A
Note: The bold numbers in the diagonal row are the square roots of the average variance extracted.
Although the results presented are generated in Model 1A, the values of CR and AVE in Model 1B are
the same to within 0.001 and the differences in correlations between the two models to within 0.005.
Model development
Having confirmed their validity, the results of developing conceptual model
1are presented in Figure 5.9 and Figure 5.10 respectively to test H2A and H2B. COS
is a second order latent variable consisting of ES and PS and the relationships among
the three variables is calculated by the repeated indicator approach. Since satisfactory
results are achieved, only the links between three latent variables of OC, DP and
CPMP, and the second order COS are presented. As the company size control
variable has weak and insignificant effect in both Model 1A and Model 1B, its
results are not presented.
154 Chapter 5: Job satisfaction
Figure 5.9 Model 1A testing H2A: COS causes CPMP
Figure 5.10 Model 1B testing H2B: CPMP causes COS Note for Fig. 5.9 and Fig.5.10: The values shown above the arrows are the of the path coefficients
validated by bootstrapping. t-statistics are shown in parentheses and their significance at the 1% (***)/
5% (**) level with values greater than 2.58 / 1.96. n.s below the arrows meaning the relationship is
not significant at the 5% level.
These show that both COSCPMP and CPMPCOS are highly significant at
the 0.1 level (t-statistic=1.64), but still far from significant at the 0.05 level (t-
Chapter 5: Job satisfaction 155
statistic=1.96). These results are generally consistent with many previous studies of
the S-P relationship in that they are positive but weak. The results of the hypothesis
tests are summarised in Table 5.16.
Table 5.15 Results of hypothesis tests
Hypotheses Model 1A Model 1B
H1A: OC has a positive direct effect on COS Supported Supported
H1B: OC has a positive direct effect on CPMP Not supported Not supported
H1C: OC has positive effects on DP. Supported Supported
H1D: DP have positive direct effects on COS Supported Supported
H1E: DP has a positive direct effect on CPMP Supported Supported
H2A: COS causes CPMP Not supported N/A
H2B: CPMP causes COS N/A Not supported
Conceptual model 2
The results gained in conceptual model 1 indicate the potential benefits of
satisfaction disaggregation. In developing model 2, although PLS-SEM is used to
obtain the significance of the links and coefficients, CB-SEM is preferred for its
ability to provide more accurate parameter estimates for the first-order reflective
constructs (Hair, Ringle, et al., 2012). When applying CB-SEM, a two-step
modelling method is often used (Anvuur & Kumaraswamy, 2011; Byrne, 2010;
Xiong, et al., 2015a). This involves firstly carrying out a confirmatory factor analysis
(CFA) from the correlations of all the latent variables and then, if the model fit
results of the CFA are acceptable, changing these to proposed directional
relationships for further analyses.
Confirmatory factor analysis
Table 5.16 Standardized regression weights and SMCs
Item
Standardized regression weights SMC
OC DP CPMP PS ES
Q1 0.894a 0.799
Q2 0.836 0.699
Q3 0.882 0.778
Q4 0.836 0.698
Q5 0.852 0.727
Q6 0.740a 0.548
156 Chapter 5: Job satisfaction
Q7 0.627a 0.393
Q8 0.721 0.519
Q9 0.748 0.560
Q10 0.762 0.581
Q11 0.718a 0.516
Q12 0.707 0.500
Q13 0.717 0.514
Q14 0.768 0.590
Q15 0.797 0.635
Q16 0.732a 0.536
Table 5.16 provides the standardized regression weights and squared multiple
correlations (SMCs) for each observed item. All the regression weights (factor
loadings) range from 0.627 to 0.894 and, being above 0.5, are therefore highly
significant (Anvuur & Kumaraswamy, 2011). The SMCs range from 0.393 to 0.799
(mean=0.600, sd=0.113). The average SMCs of the items in the measurement models
are the average variance extracted (AVE) of latent variables and, ranging from 0.510
to 0.759, are all greater than the 0.5 threshold. In terms of goodness of fit as
presented in Table 5.11, χ2 = 187.214 (df = 170, χ2/df = 1.101, p=0.174) for the CFA
phase, the χ2/df value of less than 2 indicating a good fit (Xiong et al., 2015).
Structural equation modelling
As a good model fit is obtained in the CFA phase, the correlations between the
latent variables are replaced by the hypothesized directional relationships of
conceptual model 2. The final model is shown in Figure 5.11. Company size is
excluded for being highly insignificant and weak. The observed variables Q1 to
Q16, measurement errors and connections of items are omitted due to space
limitations.
Chapter 5: Job satisfaction 157
Figure 5.11 Model 2
Table 5.17 Goodness of fit
Goodness of fit measure Criteria CFA SEM
χ2/df <5.0 1.101 1.294
Absolute fit
RMSEA <0.08 0.021 0.036
SRMR <0.08 0.0466 0.0619
RMR <0.05 0.031 0.044
Incremental fit
CFI >0.9 0.991 0.973
TLI >0.9 0.987 0.962
Parsimonious fit
PNFI >0.5 0.646 0.648
PGFI >0.5 0.573 0.574
Table 5.18 Hypothesis direct effects
Hypotheses Model 2
H3A: OC has a positive direct effect on ES Supported
H3B: OC has a positive direct effect on CPMP Not supported
H3C: OC has a positive direct effect on PS Supported
158 Chapter 5: Job satisfaction
H3D: OC has a positive direct effect on DP Supported
H3E: DP have positive direct effects on CPMP Supported
H3F: DP has positive effects on PS. Supported
H4A: PS has a positive direct effect on CPMP Supported
H4B: CPMP has a positive direct effect on ES Supported
Table 5.18 presents the results of testing the hypothesis direct effects. The
SMC of ES is 0.485, which indicates that 48.5% of the variance in ES is explained
by both direct effects from OC and CPMP and indirect effects from OC, DP and PS.
This is similar for CPMP and PS with SMCs of 0.685 and 0.462 respectively.
Following Anvuur and Kumaraswamy (2012), the bias-corrected bootstrap approach
is used with 500 resamples and maximum likelihood estimation to test the
significance of the indirect effects. This indicates the significance to range from
0.002 to 0.050. For example, the indirect effect of DP on CPMP via the mediation of
PS is 0.337×0.388=0.131, with a 95% confidence interval of (0.000, 0.322), p=0.045.
The total effects are the sum of the direct effects and indirect effects. The
standardized direct effects are shown in Figure 5.11, and Table 5.19 provides the
standardized indirect effects and total effects. To maintain consistency, and as the
effects of the insignificant link OCCPMP are small, Table 5.19 shows the
influencing effects on the dependent variables without deleting this link.
Table 5.19 Standardized direct/indirect/total effects
Effects Variables OC DP PS CPMP
Direct
effects
DP 0.626
PS 0.416 0.337
CPMP -0.091 0.601 0.388
ES 0.455 0 0 0.34
Indirect
effects
DP 0
PS 0.211 0
CPMP 0.62 0.131 0
ES 0.179 0.249 0.132 0
Total
effects
DP 0.626
PS 0.627 0.337
CPMP 0.528 0.732 0.388
ES 0.634 0.249 0.132 0.34
Chapter 5: Job satisfaction 159
5.3.5 Discussion
This paper explored the project participant satisfaction-performance nexus with
empirical evidence from Malaysian construction contractors. This is partially
inspired by the question proposed by Judge et al. (2001) for future research of
“whether the S-P relationship will be stronger at group or organization level?” With
this concern in mind, two satisfaction dimensions including ES and PS are
introduced in the context of construction projects. Two series models were developed
to test the S-P link of construction contractors. When satisfaction is seen as the usual
holistic concept, the relationship between satisfaction and performance is weak -
similar to previous studies at the individual level. The good fit of Model 2 model
demonstrates the benefit of satisfaction disaggregation and the new S-P relationships
proposed in this study.
The development of the first conceptual model gives a similar result to
previous research at the individual level in that the linkage of satisfaction and
performance is positive but weak. Although Organ (1988) found a high probability of
the existence of a functional S-P relationship, and Judge et al (2001) obtained a 0.30
mean true correlation by conducting a meta-analysis, significant evidence of both
directional models was still lacking. Our findings for this model to test S-P links of
construction contractors are similarly insignificant. However, these results are
sufficiently counterintuitive that research continues in the measurement and
estimation of project participant satisfaction in order to improve satisfaction (e.g.,
Soetanto and Proverbs, 2002; Leung, 2004; Li et al, 2013; Masrom, 2013; Xiong et
al., 2014). This suggests the solution to the S-P problem to be more complex than
realized. Therefore, unlike previous work that assumes all satisfaction facets share
common unidirectional relationships with performance (e.g. Nerkar et al 1996; Lai
2007), the model 2 presented here hypothesises that PS has a positive effect on
CPMP and CPMP positively affects ES. In addition to providing a satisfactory
explanatory ability in the form of R2 of model 2, the hypothesized paths are
significant. The conflicting relationships of the two satisfaction dimensions on
performance validated here may also explain some of the previous findings at the
individual level. For example, Lai (2007) found that noneconomic satisfaction is
much stronger than economic satisfaction in determining the influence of decision
160 Chapter 5: Job satisfaction
strategies on performance. This suggests that a key problem of the debate on the S-P
nexus is the conflicting dimensions of satisfaction, which has important implications
for future research at both individual and organizational levels.
5.3.6 Conclusions
The nature of the nexus between satisfaction and performance has been
debated for decades without satisfactory resolution. This paper provides results of the
empirical validation of hypothesised conceptual models among construction project
participants. When applied to case study data of 117 Malaysian construction
companies, model 1 indicates that, as do previous individual level studies, only a
weak relationship exists between satisfaction and performance. Model 2, on the other
hand, indicates a substantial effect of non-economic satisfaction on performance
which, in turn, has a substantial effect on economic satisfaction. This result is of
fundamental theoretical importance, with significant implications for future research
and practice.
A further contribution is the comprehensive use of CB-SEM and PLS-SEM in
solving the research problem. Although CB-SEM and PLS-SEM are increasingly
used in social science studies, they are rarely used together in same research (Hair et
al., 2012). This study combines the strengths of both approaches to provide a suitable
model development procedure for testing the formative-reflective construct at the
first step and then disaggregating the formative component at the second step. This
can be seen as a good demonstration of using both methods simultaneously and
provides a reference for similar future research.
Several limitations of the research should be noted. First, as a pioneer study of
the organizational satisfaction-performance nexus, the measures for satisfaction and
performance are less well established compared with widely used scales such as JDI
and MSQ in measuring individual satisfaction. Building a comprehensive analysis
framework will be beneficial for further studies. Otherwise, the inconsistency of
measures may lead to a significant bias and difficulty in conducting comprehensive
analyses (e.g. meta-analysis). In addition, the framework needs to be sufficiently
flexible since the items to measure noneconomic satisfaction may be different for
different industries and working roles. This study contributes to this aspect by
Chapter 5: Job satisfaction 161
demonstrating the necessity to distinguish economic and non-economic dimensions.
A second limitation of the study is that, although two antecedent variables and one
control variable are considered here, some potential moderators may exist between
satisfaction and performance. For example, for trust among project participants, one
possible situation is that when trust is high, the positive link between noneconomic
satisfaction and performance will be larger, and vice versa. Additionally, industry
differences, culture differences, and country differences are also likely. Future
research will benefit by identifying the influence of potential antecedents, mediators
and moderators. Thirdly, although our hypotheses are verified by evidence from the
construction industry in Malaysia, the applicability of our results to other industries
and other countries is uncertain. More empirical studies of different industries and
contexts are necessary, therefore, to understand the bigger picture. Additionally, it
would be beneficial to test the hypothesis that noneconomic satisfaction positively
affects performance and performance positively affects economic satisfaction at the
individual level.
Finally, this study only focuses on the relationships between satisfaction and
performance of project participating companies. Further study may explore cross-
level influences, such as the impact of perceived organizational
economic/noneconomic satisfaction on individual performance and organizational
performance on individual satisfaction, and vice versa. Additional explorations on
how these findings could be beneficial to create harmonious relationships among
participants are necessary. Despite this study is derived from the context of
construction industry in Malaysia, models developed and findings achieved in this
study probably are useful for other industries and countries to improve project
planning, implementation and outcome assessments.
162 Chapter 5: Job satisfaction
Chapter 5: Conclusions 163
Chapter 6: Conclusions
Job performance is one of the most important topics in the research area of
organisational and professional management. This research have a specific concern
for construction cost engineers, since they face uncertain works and their job
performance is critical to the success of construction projects. Therefore, this thesis
by publication contributes to developing efficient prediction techniques, providing
new understandings of person-environment interactions, and revealing relationships
between psychological reactions and performance.
The first major concern of this research was to developing an efficient
prediction technique by dealing with problems frequently occurred in previous
studies and practices. This research question was addressed by figuring out
overfitting and multicollinearity problems, developing the hybrid Akaike information
criterion-principal component regression (AIC-PCR) approach, and evaluating its
efficiency with an application of construction cost estimation. Another concern was
the role of psychological reactions in promoting job performance of cost engineers.
This question was addressed by developing a conceptual framework based on the P-
E fit theory, examining dimensions of work stress and job satisfaction, and
investigating relationships between these psychological reactions and job
performance.
In the next section, the main findings of this thesis by publication will be
summarised in detail. The following section describes limitations and offers
recommendations for future research.
6.1 SUMMARY AND DISCUSSION
In Chapter 2, the hybrid AIC-PCR approach is developed to deal with
overfitting and collinearity problems. AIC-PCR procedure and steps are also
introduced. and their usefulness for showing better predictive performance than
alternative methods, including MLR, ANN and SVR, is demonstrated in an
application of construction cost estimation. In section 6.2, path analysis is applied to
164 Chapter 6: Conclusions
examine effects of early cost drivers on the determination of construction
contingencies.
In Chapter 3, we firstly developed a conceptual model from a comprehensive
literature review, which can be used to understand the job performance of
construction professionals. Based on P-E fit theory and the S-O-R paradigm, a new
conceptual framework is achieved to be used as a reference for understanding the
role of psychological reactions. Another section on data analysis method, structural
equation modelling (SEM), is followed. SEM has been increasingly used in
construction research since the late 1990s, but previous SEM applications are not yet
satisfactory. Critical issues and suggestions for research design, model development
and model evaluation are introduced and discussed together with a review of
previous studies.
In Chapter 4, dimensions of work stress are explored and validated with
empirical evidence from construction cost engineers. The applicability of the adapted
perceived-stress questionnaire (PSQ) developed by Levenstein et al. (1993) and
Fliege et al. (2005) is tested by conducting a principal component analysis. SEM is
used to further validate the sub-dimensional model and examine differential effects
of three sub-dimensions on organisational commitment. These findings highlight the
necessity to bear in mind that work stress is a multi-dimensional concept which has
relationships with antecedents and outcomes.
In Chapter 5, a new model is proposed to describe the nexus between job
satisfaction and performance. Puzzled by the weak empirical evidence concerning S-
P links in previous studies assuming job satisfaction as the usual holistic concept,
this research followed Schwab and Cummings (1970)’s argument that the
unsatisfactory outcome may be caused by the ambiguity in definitions of satisfaction.
Additionally, this research proposes that previous unsatisfactory findings may be
because of the inconsistent causal directions between satisfaction components and
performance. Relationships between disaggregated satisfaction components and
performance are explored with empirical evidences from construction cost engineers.
The results obtained by dividing satisfaction into ES and PS demonstrate the benefit
of disaggregation, and reveal the true causal relationships involved. Therefore, our
findings on the relationship between satisfaction and performance can to some extent
help to mediate the decades of debate on the S-P link.
Chapter 5: Conclusions 165
A specific contribution of section 5.1 is the investigation into the form of
effects for PS-TP and TP-ES. Although generally positively linear relationships are
found for both links, an inverted U-shaped relationship is found to be the better one
to explain the TP-ES. A specific contribution of section 5.2 is the comprehensive use
of Covariance based SEM (CB-SEM) and PLS-SEM in solving the S-P link problem.
Although CB-SEM and PLS-SEM are increasingly used in social science studies,
they are rarely used together in the same research (Hair et al., 2012). Combining the
strengths of both approaches provides a suitable procedure for development of a
model to test the formative-reflective construct at the first step and then disaggregate
the formative component at the second step. This approach provides a good
demonstration of using both methods simultaneously and a reference for similar
future research.
6.2 LIMITATIONS AND RECOMMENDATIONS
Besides of specific limitations discussed in several chapters, a few general
limitations of the thesis should be noted. As a, there are some drawbacks related to
the nature of the thesis by publication: some literature review parts might be
overlapping; several datasets are used; and relationships among chapters are not quite
consistent. Additionally, empirical evidences are generated from the construction
industry. Findings may be applicable in other industries with further validation.
The research presented in this thesis has several implications for future studies.
With regard to the first major objective of developing prediction technique in
construction cost estimation, the following studies could be undertaken in the future:
Estimation technique innovation through adaptation of AIC-PCR.
The AIC-PCR approach proposed in this thesis could benefit future research
and practice. In particular, predictors used in construction estimation easily
encounter collinearity and overfitting problems. As well as multiple linear
regression, case based reasoning (CBR) is a widely used approach in cost
estimation. The AIC-PCR approach could assist in determining similarly
important weights and improve estimation performance for CBR.
The impact of sustainability requirements on construction cost.
166 Chapter 6: Conclusions
Sustainable development (SD) has been emphasised in twentieth-century
works such as Limits to Growth (1972) and Silent Spring (1962), and in some
late twentieth-century governmental reports such as Our Common Future
(1987), in which it is defined as “development that meets the needs of the
present without compromising the ability of future generations to meet their
own needs.” In recent years, sustainable development has been accepted as a
must. For example, more than 95% of major companies in Europe and the
USA accept its importance, and many are members of the World Business
Council for Sustainable Development (Giddings, Hopwood, & O'Brien,
2002). Because the construction industry has such large environmental
impacts (Hill & Bowen, 1997; Ofori, 2000), increased awareness of
sustainable building construction is considered to be key to reducing
environmental impacts and finding best practice (Pitt, Tucker, Riley, &
Longden, 2009). In this process, clients act as a key force and are paying
more attention to owning sustainable buildings (Gan et al., 2015). Buildings
labelled by related certifications such as Building Research Establishment
Environmental Assessment Methodology (BREEAM), Leadership in Energy
and Environmental Design (LEED) and Green Star are increasing. Although a
few academic articles argue that increased capital cost is the biggest barrier to
achieving sustainability in construction (Häkkinen & Belloni, 2011), potential
benefits gained by energy-saving are overlooked. Although some industry
reports (such as BRE and Cyril, 2005) argue that sustainability is a significant
factor in the life cycle costs of a building, these reports can be criticised for
using small sample sizes and neglecting interactions with other factors like
project size. Future research will benefit from examining the impact of
sustainability requirements on the construction costs in terms of both capital
costs and life cycle costs of a building.
With regard to the second major objective of exploring P-E interactions, the
following studies could be undertaken in the future:
Examining effects of organisational support on job performance via the
mediation of psychological reactions.
Organizational support and job performance have been important issues in
organizational management. Based on the conceptual model developed in this
Chapter 5: Conclusions 167
thesis, psychological reactions in terms of job satisfaction and work stress
could mediate the impact of organisational support. Future research can
benefit from testing these direct and indirect effects.
Impact of personal characteristics on the relationships between organisational
support factors and job performance mediated by psychological reactions.
As indicated in many previous studies, moderating or direct effects of
personal characteristics on job performance of construction professionals are
worth exploring in the future. These characteristics may include age,
education level, personal traits, learning style, and so on.
Extensions and refinements of the stimulus-reactions-performance conceptual
model.
The conceptual model proposed in Section 2.1 can be used to understand
previous studies and to underpin future studies. Future research can benefit
from extending the model with new concepts and making necessary
adaptions. Longitudinal studies are also relevant to simulate P-E interactions
as a dynamic process.
168 Chapter 6: Conclusions
Bibliography 169
Bibliography
Aamodt, A., & Plaza, E. (1994). Case-based reasoning: Foundational issues,
methodological variations, and system approaches. AI communications, 7(1),
39-59.
AbuAlRub, R. F. (2004). Job stress, job performance, and social support among
hospital nurses. Journal of nursing scholarship, 36(1), 73-78.
Aghdasi, S., Kiamanesh, A. R., & Ebrahim, A. N. (2011). Emotional Intelligence and
organizational commitment: testing the mediatory role of occupational stress
and job satisfaction. Procedia-Social and Behavioral Sciences, 29, 1965-
1976.
Ahadzie, D., Proverbs, D., & Olomolaiye, P. (2008a). Model for predicting the
performance of project managers at the construction phase of mass house
building projects. Journal of Construction Engineering and Management.
Ahadzie, D. K., Proverbs, D. G., & Olomolaiye, P. (2008b). Towards developing
competency-based measures for construction project managers: Should
contextual behaviours be distinguished from task behaviours? International
Journal of Project Management, 26(6), 631-645.
Ahuja, V., Yang, J., Skitmore, M., & Shankar, R. (2010). An empirical test of causal
relationships of factors affecting ICT adoption for building project
management: An Indian SME case study. Construction Innovation:
information, process, management, 10(2), 164-180.
Akaike, H. (1974). A new look at the statistical model identification. Automatic
Control, IEEE Transactions on, 19(6), 716-723.
Akinci, B., & Fischer, M. (1998). Factors affecting contractors' risk of cost
overburden. Journal of Management in Engineering, 14(1), 67-76.
Akintoye, A. (2000). Analysis of factors influencing project cost estimating practice.
Construction Management & Economics, 18(1), 77-89.
Al-Kharashi, A., & Skitmore, R.M., (2009). Causes of delays in Saudi Arabian
public sector construction projects. Construction Management and
Economics, 27 (1), 3-23.
170 Bibliography
Al-Tmeemy, S.M.H.M., Abdul-Rahman, H., & Harun, Z. (2011). Future criteria for
success of building projects in Malaysia. International Journal of Project
Management 29 (3), 337-348.
An, S.-H., Kim, G.-H., & Kang, K.-I. (2007). A case-based reasoning cost estimating
model using experience by analytic hierarchy process. Building and
Environment, 42(7), 2573-2579.
Andersen, C. M., & Bro, R. (2010). Variable selection in regression—a tutorial.
Journal of Chemometrics, 24(11‐12), 728-737.
Anvuur, A. M., & Kumaraswamy, M. M. (2011). Measurement and antecedents of
cooperation in construction. Journal of Construction Engineering and
Management.
Arbuckle, J. L. (1994). Computer announcement amos: Analysis of moment
structures. Psychometrika, 59(1), 135-137.
Argyris, C., & Schön, D. A. (1978). Organizational learning: A theory of action
perspective (Vol. 173): Addison-Wesley Reading, MA.
Assaf, S. A., & Al-Hejji, S. (2006). Causes of delay in large construction projects.
International journal of project management, 24(4), 349-357.
Atkinson, R., Crawford, L., & Ward, S. (2006). Fundamental uncertainties in
projects and the scope of project management. International Journal of
Project Management. 24 (8), 687-698.
Attar, A. M., Khanzadi, M., Dabirian, S., & Kalhor, E. (2013). Forecasting
contractor's deviation from the client objectives in prequalification model
using support vector regression. International Journal of Project
Management, 31(6), 924-936.
Bagozzi, R. P., & Yi, Y. (1988). On the evaluation of structural equation models.
Journal of the academy of marketing science, 16(1), 74-94.
Bagozzi, R. P., & Yi, Y. (2012). Specification, evaluation, and interpretation of
structural equation models. Journal of the Academy of Marketing Science,
40(1), 8-34.
Barnes, N. M. L. (1974). Finanial Control of Construction In W. S.H. (Ed.), Control
of Engineering Porjects: Edward Arnold.
Bibliography 171
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction
in social psychological research: Conceptual, strategic, and statistical
considerations. Journal of personality and social psychology, 51(6), 1173.
Baumgartner, H., & Homburg, C. (1996). Applications of structural equation
modeling in marketing and consumer research: a review. International
Journal of Research in Marketing, 13(2), 139-161.
Becker, H. S. (1960). Notes on the concept of commitment. American journal of
Sociology, 32-40.
Becker, J.-M., Klein, K., & Wetzels, M. (2012). Hierarchical latent variable models
in PLS-SEM: guidelines for using reflective-formative type models. Long
Range Planning, 45(5), 359-394.
Belsley, D. A., Kuh, E., & Welsch, R. E. (2005). Regression diagnostics: Identifying
influential data and sources of collinearity (Vol. 571): John Wiley & Sons.
Bentler, P. M. (1980). Multivariate analysis with latent variables: Causal modeling.
Annual review of psychology, 31(1), 419-456.
Bentler, P. M. (1989). EQS Structural Equations Program Manual Pages: BMDP
Statistical Software.
Bentler, P. M., & Chou, C.-P. (1987). Practical issues in structural modeling.
Sociological Methods & Research, 16(1), 78-117.
Bergeron, D. M., Shipp, A. J., Rosen, B., & Furst, S. A. (2011). Organizational
citizenship behavior and career outcomes the cost of being a good citizen.
Journal of Management, 0149206311407508.
Bohannon, J. (2014). Google Scholar Wins Raves—But Can It Be Trusted? Science,
343(6166), 14-14.
Bolanowski, W. (2005). Anxiety about professional future among young doctors.
International journal of occupational medicine and environmental health,
18(4), 367-374.
Bonanno, A., & Constance, D. H. (2001). Globalization, Fordism, and Post‐Fordism in Agriculture and Food: A Critical Review of the Literature.
Culture & Agriculture, 23(2), 1-18.
172 Bibliography
Borman, W. C., & Motowidlo, S. (1993a). Expanding the criterion domain to include
elements of contextual performance. Personnel Selection in Organizations;
San Francisco: Jossey-Bass, 71.
Borman, W. C., & Motowidlo, S. J. (1993b). Expanding the criterion domain to
include elements of contextual performance. Personnel selection in
organizations, 71(1993), 98.
Bowen, P., Edwards, P., & Lingard, H. (2012). Workplace stress experienced by
construction professionals in South Africa. Journal of Construction
Engineering and Management, 139(4), 393-403.
Bowen, P., Edwards, P., Lingard, H., & Cattell, K. (2014). Workplace Stress, Stress
Effects, and Coping Mechanisms in the Construction Industry. Journal of
Construction Engineering and Management, 140(3), 04013059.
Bowen, P., Govender, R., & Edwards, P. (2014). Structural equation modeling of
occupational stress in the construction industry. Journal of Construction
Engineering and Management, 140(9), 04014042.
Bowling, N. A., & Eschleman, K. J. (2010). Employee personality as a moderator of
the relationships between work stressors and counterproductive work
behavior. Journal of Occupational Health Psychology, 15(1), 91.
Boyas, J., Wind, L. H., & Kang, S.-Y. (2012). Exploring the relationship between
employment-based social capital, job stress, burnout, and intent to leave
among child protection workers: An age-based path analysis model. Children
and Youth Services Review, 34(1), 50-62.
Brayfield, A. H., & Crockett, W. H. (1955). Employee attitudes and employee
performance. Psychological bulletin, 52(5), 396.
BRE, & Cyril. (2005). Putting a price on sustainability: Building Research
Establishment.
Bretz, J. R. D., & Judge, T. A. (1994). Person–Organization Fit and the Theory of
Work Adjustment: Implications for Satisfaction, Tenure, and Career Success.
Journal of Vocational Behavior, 44(1), 32-54.
Brislin, R. W. (1970). Back-Translation for Cross-Cultural Research. Journal of
Cross-Cultural Psychology, 1(3), 185-216.
Bibliography 173
Brown, M. (2001). Unequal pay, unequal responses? Pay referents and their
implications for pay level satisfaction. Journal of Management Studies, 38(6),
879-886.
Browne, M. W., Cudeck, R., Bollen, K. A., & Long, J. S. (1993). Alternative ways of
assessing model fit. Sage Focus Editions, 154, 230-258.
Brundtland, G. H. (1987). Report of the World Commission on environment and
development:" our common future.": United Nations.
Burnham, K. P., & Anderson, D. R. (2002). Model selection and multimodel
inference: a practical information-theoretic approach: Springer.
Byrne, B. M. (2001). Structural equation modeling with AMOS: basic concepts,
applications, and programming. London: Lawrence Erlbaum Associates.
Byrne, B. M. (2010). Structural equation modeling with AMOS: Basic concepts,
applications, and programming. New York: Routledge.
Campbell, F. (2006). Occupational stress in the construction industry. Berkshire, UK:
Chartered Institute of Building.
Caplan, R. D. (1987). Person-environment fit theory and organizations:
Commensurate dimensions, time perspectives, and mechanisms. Journal of
Vocational behavior, 31(3), 248-267.
Carr, P. G., De La Garza, J. M., & Vorster, M. C. (2002). Relationship between
personality traits and performance for engineering and architectural
professionals providing design services. Journal of Management in
Engineering, 18(4), 158-166.
Cegarra-Navarro, J.-G., & Sánchez-Polo, M. T. (2011). Influence of the open-
mindedness culture on organizational memory: an empirical investigation of
Spanish SMEs. The International Journal of Human Resource Management,
22(1), 1-18.
Chau, K. W. (1997). Monte Carlo simulation of construction costs using subjective
data: response. Construction Management & Economics, 15(1), 109-115.
Chan, A., Scott, D., and Lam, E. (2002). Framework of success criteria for
design/build projects. Journal of Management in Engineering. 18 (3), 120–
128.
174 Bibliography
Chen, L., & Fong, P. S. (2012). Revealing performance heterogeneity through
knowledge management maturity evaluation: A capability-based approach.
Expert Systems with Applications.
Cheng, J., Proverbs, D. G., & Oduoza, C. F. (2006). The satisfaction levels of UK
construction clients based on the performance of consultants: Results of a
case study. Engineering, Construction and Architectural Management, 13(6),
567-583.
Cheung, F. K., & Skitmore, M. (2006). Application of cross validation techniques for
modelling construction costs during the very early design stage. Building and
environment, 41(12), 1973-1990.
Cheung, S., Chow, P., & Yiu, T. (2009). Contingent Use of Negotiators’ Tactics in
Construction Dispute Negotiation. Journal of Construction Engineering and
Management, 135(6), 466-476.
Cheung, S. O., & Chow, P. T. (2011). Withdrawal in construction project dispute
negotiation. Journal of Construction Engineering and Management, 137(12),
1071-1079.
Cheung, S. O., Wong, P. S. P., Fung, A. S., & Coffey, W. (2006). Predicting project
performance through neural networks. International Journal of Project
Management, 24(3), 207-215.
Chin, W. W. (2010). How to write up and report PLS analyses. In Handbook of
partial least squares (pp. 655-690): Springer.
Cho, K., Hong, T., & Hyun, C. (2009). Effect of project characteristics on project
performance in construction projects based on structural equation model.
Expert Systems with Applications, 36(7), 10461-10470.
Chou, J.-S., & Tseng, H.-C. (2011). Establishing expert system for prediction based
on the project-oriented data warehouse. Expert Systems with Applications,
38(1), 640-651.
Chou, J. S., & Yang, J. G. (2012). Project Management Knowledge and Effects on
Construction Project Outcomes: An Empirical Study. Project Management
Journal, 43(5), 47-67.
Chuang, A., Hsu, R. S., Wang, A.-C., & Judge, T. A. (2015). Does West “fit” with
East? In search of a Chinese model of person–environment fit. Academy of
Management Journal, 58(2), 480-510.
Bibliography 175
Chun, J. S., Shin, Y., Choi, J. N., & Kim, M. S. (2013). How does corporate ethics
contribute to firm financial performance? The mediating role of collective
organizational commitment and organizational citizenship behavior. Journal
of Management, 39(4), 853-877.
Christensen, P., & Dysert, L. R. (2005). Cost estimate classification system–as
applied in engineering, procurement, and construction for the process
industries. Morgantown, WV: AACE International. Chun-Te Lin, J., &
Livingston, A.(2007). Nanofiltration membrane cascade for continuous
solvent exchange. Chemical Engineering Science, 2728(2736), 22-30.
Cohen, S., Kamarck, T., & Mermelstein, R. (1983). A global measure of perceived
stress. Journal of Health and Social Behavior, 385-396.
Cotton, J. L., & Tuttle, J. M. (1986). Employee turnover: A meta-analysis and review
with implications for research. Academy of management Review, 11(1), 55-
70.
Cronbach, L. J. (1951). Coefficient alpha and the internal structure of tests.
Psychometrika, 16(3), 297-334.
Currivan, D. B. (1999). The Causal Order of Job Satisfaction and Organizational
Commitment in Models of Employee Turnover. Human Resource
Management Review, 9(4), 495-524.
Dainty, A. R., Cheng, M.-I., & Moore, D. R. (2005). Competency-based model for
predicting construction project managers’ performance. Journal of
Management in Engineering.
Dalal, R. S. (2005). A meta-analysis of the relationship between organizational
citizenship behavior and counterproductive work behavior. Journal of applied
psychology, 90(6), 1241.
De Cuyper, N., & De Witte, H. (2006). Autonomy and Workload Among Temporary
Workers: Their Effects on Job Satisfaction, Organizational Commitment, Life
Satisfaction, and Self-Rated Performance. International Journal of Stress
Management, 13(4), 441-459.
De Wit, A. (1988). Measurement of project success. International journal of project
management, 6(3), 164-170.
176 Bibliography
del Bosque Rodríguez, I. R., Agudo, J. C., & Gutiérrez, H. S. M. (2006).
Determinants of economic and social satisfaction in manufacturer–distributor
relationships. Industrial Marketing Management, 35(6), 666-675.
Dilchert, S., Ones, D. S., Davis, R. D., & Rostow, C. D. (2007). Cognitive ability
predicts objectively measured counterproductive work behaviors. Journal of
Applied Psychology, 92(3), 616.
Ding, Z., & Ng, F. (2007). Reliability and validity of the Chinese version of
McAllister's trust scale. Construction Management and Economics, 25(11),
1107-1117.
Ding, Z., & Ng, F. (2010). Personal Construct-Based Factors Affecting Interpersonal
Trust in a Project Design Team. Journal of Construction Engineering and
Management, 136(2), 227-234.
Ding, Z., Ng, F., Wang, J., & Zou, L. (2012). Distinction between team-based self-
esteem and company-based self-esteem in the construction industry. Journal
of Construction Engineering and Management, 138(10), 1212-1219.
Edwards, J. R. (1996). An examination of competing versions of the person-
environment fit approach to stress. Academy of management journal, 39(2),
292-339.
Edwards, J. R., Caplan, R. D., & Van Harrison, R. (1998). Person-environment fit
theory. Theories of organizational stress, 28, 67.
Egan, T. M., Yang, B., & Bartlett, K. R. (2004). The effects of organizational
learning culture and job satisfaction on motivation to transfer learning and
turnover intention. Human Resource Development Quarterly, 15(3), 279-301.
Eisenberger, R., Fasolo, P., & Davis-LaMastro, V. (1990). Perceived organizational
support and employee diligence, commitment, and innovation. Journal of
applied psychology, 75(1), 51.
Eisenberger, R., Huntington, R., Hutchison, S., & Sowa, D. (1986). Perceived
organizational support. Journal of Applied Psychology, 71(3), 500-507.
Elhag, T., Boussabaine, A., & Ballal, T. (2005). Critical determinants of construction
tendering costs: Quantity surveyors’ standpoint. International journal of
project management, 23(7), 538-545.
Bibliography 177
Enders, C. K., & Bandalos, D. L. (2001). The relative performance of full
information maximum likelihood estimation for missing data in structural
equation models. Structural Equation Modeling, 8(3), 430-457.
Falk, A., & Fischbacher, U. (2006). A theory of reciprocity. Games and economic
behavior, 54(2), 293-315.
Farrar, D. E., & Glauber, R. R. (1967). Multicollinearity in regression analysis: the
problem revisited. The Review of Economic and Statistics, 92-107.
Fellows, R. (1996). Monte Carlo simulation of construction costs using subjective
data: comment. Construction Management & Economics, 14(5), 457-460.
Ferris, G. R., & Kacmar, K. M. (1992). Perceptions of organizational politics.
Journal of management, 18(1), 93-116.
Fisher, C. D. (2003). Why do lay people believe that satisfaction and performance
are correlated? Possible sources of a commonsense theory. Journal of
Organizational Behavior, 24(6), 753-777.
Fliege, H., Rose, M., Arck, P., Walter, O. B., Kocalevent, R.-D., Weber, C., &
Klapp, B. F. (2005). The Perceived Stress Questionnaire (PSQ) reconsidered:
validation and reference values from different clinical and healthy adult
samples. Psychosomatic Medicine, 67(1), 78-88.
Flyvbjerg, B., Bruzelius, N., & Rothengatter, W. (2003). Megaprojects and risk: An
anatomy of ambition: Cambridge University Press.
Folkman, S., & Lazarus, R. S. (1988). Manual for the ways of coping questionnaire:
Consulting Psychologists Press.
Fornell, C., & Larcker, D. (1987). A second generation of multivariate analysis:
Classification of methods and implications for marketing research. Review of
marketing, 1, 407-450.
Fornell, C., & Larcker, D. F. (1981a). Evaluating structural equation models with
unobservable variables and measurement error. Journal of Marketing
Research (JMR), 18(1).
Fornell, C., & Larcker, D. F. (1981b). Evaluating structural equation models with
unobservable variables and measurement error. Journal of marketing
research, 39-50.
178 Bibliography
Fox, S., Spector, P. E., & Miles, D. (2001). Counterproductive Work Behavior
(CWB) in Response to Job Stressors and Organizational Justice: Some
Mediator and Moderator Tests for Autonomy and Emotions. Journal of
Vocational Behavior, 59(3), 291-309.
Freeman, R. E. (1984). Strategic management: A stakeholder approach. Boston:
Pitman/Ballinger.
Friendly, M., & Kwan, E. (2009). Where's Waldo? Visualizing collinearity
diagnostics. The American Statistician, 63(1).
Gan, X., Zuo, J., Ye, K., Skitmore, M., & Xiong, B. (2015). Why sustainable
construction? Why not? An owner's perspective. Habitat International, 47,
61-68.
Gandz, J., & Murray, V. V. (1980). The experience of workplace politics. Academy
of Management Journal, 23(2), 237-251.
Gefen, D. (2000). E-commerce: the role of familiarity and trust. Omega, 28(6), 725-
737.
Geyskens, I., Steenkamp, J.-B. E., & Kumar, N. (1999). A meta-analysis of
satisfaction in marketing channel relationships. Journal of marketing
Research, 223-238.
Giauque, D., Resenterra, F., & Siggen, M. (2014). Antecedents of job satisfaction,
organizational commitment and stress in a public hospital: A PE Fit
perspective. Public Organization Review, 14(2), 201-228.
Giddings, B., Hopwood, B., & O'brien, G. (2002). Environment, economy and
society: fitting them together into sustainable development. Sustainable
Development, 10(4), 187-196.
Gollob, H. F., & Reichardt, C. S. (1987). Taking account of time lags in causal
models. Child Development, 80-92.
Golob, T. F. (2003). Structural equation modeling for travel behavior research.
Transportation Research Part B: Methodological, 37(1), 1-25.
Gorsuch, R. L. (1983). Factor Analysis, Lawrence Erlbaum, Hillsdale, New Jersey
Bibliography 179
Greguras, G. J., & Diefendorff, J. M. (2009). Different fits satisfy different needs:
linking person-environment fit to employee commitment and performance
using self-determination theory. Journal of Applied Psychology, 94(2), 465.
Guerrero, M. A., Villacampa, Y., & Montoyo, A. (2014). Modeling construction time
in Spanish building projects. International Journal of Project Management,
32(5), 861-873.
Gunner, J., & Skitmore, M. (1999). Pre‐bid building price forecasting accuracy:
price intensity theory. Engineering Construction and Architectural
Management, 6(3), 267-275.
Gunning, J. G., & Cooke, E. (1996). The influence of occupational stress on
construction professionals. Building Research & Information, 24(4), 213-221.
Hair, J. F. (2006). Multivariate data analysis. Upper Saddle River, N.J: Pearson
Prentice Hall.
Hair, J. F. (2014). A primer on partial least squares structural equation modeling
(PLS-SEM). Thousand Oaks, Calif: SAGE Publications.
Hair, J. F., Ringle, C. M., & Sarstedt, M. (2012). Editorial-Partial Least Squares: The
Better Approach to Structural Equation Modeling? Long Range Planning,
45(5-6), 312-319.
Hair, J. F., Sarstedt, M., Pieper, T. M., & Ringle, C. M. (2012). The use of partial
least squares structural equation modeling in strategic management research:
a review of past practices and recommendations for future applications. Long
range planning, 45(5), 320-340.
Hair Jr, J. F., Hult, G. T. M., Ringle, C., & Sarstedt, M. (2013). A primer on partial
least squares structural equation modeling (PLS-SEM). Thousand Oaks: Sage
Publications.
Häkkinen, T., & Belloni, K. (2011). Barriers and drivers for sustainable building.
Building Research & Information, 39(3), 239-255.
Hanson, P. (1987). The joy of stress: A Pan Original.
Harmon, K.M.J. (2003). Conflicts between owner and contractors: proposed
intervention process. Journal of Management in Engineering, 19 (3), 121-
125.
180 Bibliography
Harzing, A.-W. (2014). A longitudinal study of Google Scholar coverage between
2012 and 2013. Scientometrics, 98(1), 565-575.
Haynes, N. S., & Love, P. E. (2004). Psychological adjustment and coping among
construction project managers. Construction Management and Economics,
22(2), 129-140.
Hellriegel, D., & Slocum, J. W. (1974). Organizational climate: Measures, research
and contingencies. Academy of management Journal, 17(2), 255-280.
Hershberger, S. L. (2003). The growth of structural equation modeling: 1994-2001.
Structural Equation Modeling, 10(1), 35-46.
Hill, R. C., & Bowen, P. A. (1997). Sustainable construction: principles and a
framework for attainment. Construction Management & Economics, 15(3),
223-239.
Hobday, M. (2000). The project-based organisation: an ideal form for managing
complex products and systems? Research policy, 29(7), 871-893.
Hon, A. H. Y. (2013). Does job creativity requirement improve service performance?
A multilevel analysis of work stress and service environment. International
Journal of Hospitality Management, 35(0), 161-170.
Hooper, D., Coughlan, J., & Mullen, M. R. (2008). Structural equation modelling:
Guidelines for determining model fit. Electronic Journal of Business
Research Methods, 6(1).
Hu, L. t., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance
structure analysis: Conventional criteria versus new alternatives. Structural
Equation Modeling: A Multidisciplinary Journal, 6(1), 1-55.
Hunter, J. E. (1986). Cognitive ability, cognitive aptitudes, job knowledge, and job
performance. Journal of Vocational Behavior, 29(3), 340-362.
Hurrell, J. J. J., & McLaney, M. A. (1988). Exposure to job stress--a new
psychometric instrument. Scandinavian journal of Work, Environment &
Health, 14 Suppl 1, 27.
Bibliography 181
Hurrell Jr, J. J., Nelson, D. L., & Simmons, B. L. (1998). Measuring job stressors and
strains: where we have been, where we are, and where we need to go. Journal
of occupational health psychology, 3(4), 368.
Islam, M. M., Faniran, O. O. (2005). Structural equation model of project planning
effectiveness. Construction Management and Economics, 23 (2), 215-223.
Jamal, M. (1990). Relationship of job stress and Type-A behavior to employees' job
satisfaction, organizational commitment, psychosomatic health problems, and
turnover motivation. Human Relations, 43(8), 727-738.
Janssen, O., & Van Yperen, N. W. (2004). Employees' goal orientations, the quality
of leader-member exchange, and the outcomes of job performance and job
satisfaction. Academy of management journal, 47(3), 368-384.
Jarvis, C. B., MacKenzie, S. B., & Podsakoff, P. M. (2003). A critical review of
construct indicators and measurement model misspecification in marketing
and consumer research. Journal of consumer research, 30(2), 199-218.
Johnson, J. B., & Omland, K. S. (2004). Model selection in ecology and evolution.
Trends in Ecology & Evolution, 19(2), 101-108.
Jöreskog, K. G. (1970). A general method for analysis of covariance structures.
Biometrika, 57(2), 239-251.
Jöreskog, K. G., & Sörbom, D. (1996). LISREL 8 user's reference guide: Scientific
Software International.
Judge, T. A., Thoresen, C. J., Bono, J. E., & Patton, G. K. (2001). The job
satisfaction–job performance relationship: A qualitative and quantitative
review. Psychological bulletin, 127(3), 376.
Jyh-Bin, Y., Shen-Fen, O. (2008). Using structural equation modeling to analyze
relationships among key causes of delay in construction. Canadian Journal of
Civil Engineering, 35 (4), 321-332.
Kahn, R. L., Wolfe, D. M., Quinn, R. P., Snoek, J. D., & Rosenthal, R. A. (1964).
Organizational stress: Studies in role conflict and ambiguity.
Kaka, A. P. (1996). Towards more flexible and accurate cash flow forecasting.
Construction Management and Economics, 14(1), 35-44.
182 Bibliography
Kärnä, S., Sorvala, V.M., Junnonen, J.M., 2009. Classifying and clustering
construction projects by customer satisfaction. Facilities, 27 (9/10), 387-398.
Keline, R. B. (2005). Principles and Practice of Structural Equation Modeling. New
York: Guilford Press.
Kim, G.-H., An, S.-H., & Kang, K.-I. (2004). Comparison of construction cost
estimating models based on regression analysis, neural networks, and case-
based reasoning. Building and Environment, 39(10), 1235-1242.
Kim, K. J., & Kim, K. (2010). Preliminary cost estimation model using case-based
reasoning and genetic algorithms. Journal of Computing in Civil Engineering,
24(6), 499-505.
Kim, S.-Y., Choi, J.-W., Kim, G.-H., & Kang, K.-I. (2005). Comparing cost
prediction methods for apartment housing projects: CBR versus ANN.
Journal of Asian Architecture and Building Engineering, 4(1), 113-120.
Kline, R. B. (1998). Software review: Software programs for structural equation
modeling: Amos, EQS, and LISREL. Journal of psychoeducational
assessment, 16(4), 343-364.
Kline, R. B. (2005). Principles and practice of structural equation modeling. New
York: Guilford Press.
Kristof, A. L. (1996). Person ‐ organization fit: an integrative review of its
conceptualizations, measurement, and implications. Personnel psychology,
49(1), 1-49.
Kullback, S., & Leibler, R. A. (1951). On information and sufficiency. The Annals of
Mathematical Statistics, 79-86.
Lai, C.-S. (2007). The effects of influence strategies on dealer satisfaction and
performance in Taiwan's motor industry. Industrial Marketing Management,
36(4), 518-527.
Latham, G. P., & Pinder, C. C. (2005). Work motivation theory and research at the
dawn of the twenty-first century. Annu. Rev. Psychol., 56, 485-516.
Bibliography 183
Lauver, K. J., & Kristof-Brown, A. (2001). Distinguishing between employees'
perceptions of person–job and person–organization fit. Journal of Vocational
Behavior, 59(3), 454-470.
Lawler, E. E., & Porter, L. W. (1967). The effect of performance on job satisfaction.
Industrial relations: A journal of Economy and Society, 7(1), 20-28.
Lazarus, R. S. (1993). Coping theory and research: past, present, and future.
Psychosomatic medicine, 55(3), 234-247.
Ledford Jr, G. E. (1999). Happiness and productivity revisited.
Lee, S.-K., & Yu, J.-H. (2012). Success model of project management information
system in construction. Automation in Construction, 25, 82-93.
Lehtiranta, L., Kärnä, S., Junnonen, J.-M., & Julin, P. (2012). The role of multi-firm
satisfaction in construction project success. Construction Management and
Economics, 30 (6), 463-475.
Leiter, M. P., & Maslach, C. (1988). The impact of interpersonal environment on
burnout and organizational commitment. Journal of organizational behavior,
9(4), 297-308.
LePine, J. A., Erez, A., & Johnson, D. E. (2002). The nature and dimensionality of
organizational citizenship behavior: A critical review and meta-analysis.
Journal of Applied Psychology, 87(1), 52-65.
Leung, M.-y., Chan, Y.-s., Chong, A., & Sham, J. F.-C. (2008). Developing
structural integrated stressor–stress models for clients’ and contractors’ cost
engineers. Journal of Construction Engineering and Management, 134(8),
635-643.
Leung, M.-y., Chan, Y.-S., & Yuen, K.-W. (2010). Impacts of stressors and stress on
the injury incidents of construction workers in Hong Kong. Journal of
Construction Engineering and Management.
Leung, M.-y., Shan Isabelle Chan, Y., & Dongyu, C. (2011). Structural linear
relationships between job stress, burnout, physiological stress, and
performance of construction project managers. Engineering, Construction
and Architectural Management, 18(3), 312-328.
184 Bibliography
Leung, M.-Y., Yu, J., & Chong, M. L. A. (2015). Effects of Stress and Commitment
on the Performance of Construction Estimation Participants in Hong Kong.
Journal of Construction Engineering and Management, 04015081.
Leung, M.-y., Zhang, H., & Skitmore, M. (2008). Effects of organizational supports
on the stress of construction estimation participants. Journal of Construction
Engineering and Management, 134(2), 84-93.
Leung, M. Y., Liu, A. M., & Wong, M. M. k. (2006). Impact of stress‐coping
behaviour on estimation performance. Construction Management and
Economics, 24(1), 55-67.
Leung, M. Y., Ng, S. T., & Cheung, S. O. (2004). Measuring construction project
participant satisfaction. Construction Management and Economics, 22(3),
319-331.
Leung, M. Y., Ng, S. T., Skitmore, M., & Cheung, S. O. (2005). Critical stressors
influencing construction estimators in Hong Kong. Construction
Management and Economics, 23(1), 33-44.
Leung, M. Y., Olomolaiye, P., Chong, A., & Lam, C. C. (2005). Impacts of stress on
estimation performance in Hong Kong. Construction Management and
Economics, 23(9), 891-903.
Levenstein, S., Prantera, C., Varvo, V., Scribano, M. L., Berto, E., Luzi, C., &
Andreoli, A. (1993). Development of the Perceived Stress Questionnaire: a
new tool for psychosomatic research. Journal of Psychosomatic Research,
37(1), 19-32.
Lewin, K. (1935). A dynamic theory of personality.
Li, H., Shen, L., & Love, P. (1999). ANN-based mark-up estimation system with
self-explanatory capacities. Journal of construction engineering and
management, 125(3), 185-189.
Li, H., Shen, Q., & Love, P. E. (2005). Cost modelling of office buildings in Hong
Kong: an exploratory study. Facilities, 23(9/10), 438-452.
Li, T. H., Ng, S. T., & Skitmore, M. (2013). Evaluating stakeholder satisfaction
during public participation in major infrastructure and construction projects: a
fuzzy approach. Automation in construction, 29, 123-135.
Bibliography 185
Ling, F. Y. Y., & Loo, C. M. (2013). Characteristics of jobs and jobholders that
affect job satisfaction and work performance of project managers. Journal of
Management in Engineering.
Ling, F.Y.Y., & Chong, C.L.K. (2005). Design-and-build contractors' service quality
in public projects in Singapore. Building and Environment, 40 (6), 815-823.
Ling, F.Y.Y., Low, S.P., Wang, S.Q., Egbelakin, T. (2008). Models for predicting
project performance in China using project management practices adopted by
foreign AEC firms. Journal of Construction Engineering and Management,
134 (12), 983.
Ling, Y. Y. (2002). Model for predicting performance of architects and engineers.
Journal of construction engineering and management, 128(5), 446-455.
Liu, A. M., & Fellows, R. (2008). Behaviour of quantity surveyors as organizational
citizens. Construction Management and Economics, 26(12), 1271-1282.
Liu, A.M.M., & Leung, M. (2002). Developing a soft value management model.
International Journal of Project Management, 20 (5), 341-349.
Liu, A. M., & Walker, A. (1998). Evaluation of project outcomes. Construction
Management & Economics, 16(2), 209-219.
Liu, M., & Ling, Y. Y. (2003). Using fuzzy neural network approach to estimate
contractors’ markup. Building and environment, 38(11), 1303-1308.
Liu, R., Kuang, J., Gong, Q., & Hou, X. (2003). Principal component regression
analysis with SPSS. Computer methods and programs in biomedicine, 71(2),
141-147.
Lopez, E. M. (1982). A test of the self-consistency theory of the job performance-job
satisfaction relationship. Academy of Management Journal, 25(2), 335-348.
Love, P. E., Edwards, D. J., & Irani, Z. (2009). Work stress, support, and mental
health in construction. Journal of construction Engineering and management,
136(6), 650-658.
Love, P. E., Edwards, D. J., & Irani, Z. (2010). Work stress, support, and mental
health in construction. Journal of construction Engineering and management,
136(6), 650-658.
Love, P. E., & Holt, G. D. (2000). Construction business performance measurement:
the SPM alternative. Business process management journal, 6(5), 408-416.
186 Bibliography
Love, P. E., & Li, H. (2000). Quantifying the causes and costs of rework in
construction. Construction Management & Economics, 18(4), 479-490.
Lowe, D., & Skitmore, M. (1994). Experiential learning in cost estimating.
Construction Management and Economics, 12(5), 423-431.
Lowe, D. J., Emsley, M. W., & Harding, A. (2006). Predicting construction cost
using multiple regression techniques. Journal of construction engineering
and management, 132(7), 750-758.
Lowe, D. J., & Skitmore, M. (2007). The learning climate of an organisation and
practitioner competence. Journal of financial management of property and
construction, 12(3), 151-164.
Lundstrom, T., Pugliese, G., Bartley, J., Cox, J., & Guither, C. (2002).
Organizational and environmental factors that affect worker health and safety
and patient outcomes. American journal of infection control, 30(2), 93-106.
M. Goldenhar*, L., Williams, L. J., & G. Swanson, N. (2003). Modelling
relationships between job stressors and injury and near-miss outcomes for
construction labourers. Work & Stress, 17(3), 218-240.
MacCallum, R. C., & Austin, J. T. (2000). Applications of structural equation
modeling in psychological research. Annual review of psychology, 51(1),
201-226.
MacCallum, R. C., Browne, M. W., & Sugawara, H. M. (1996). Power analysis and
determination of sample size for covariance structure modeling.
Psychological methods, 1(2), 130.
MacCallum, R. C., Roznowski, M., & Necowitz, L. B. (1992). Model modifications
in covariance structure analysis: the problem of capitalization on chance.
Psychological bulletin, 111(3), 490.
Maloney, W. F. (2002). Construction product/service and customer satisfaction.
Journal of construction engineering and management, 128(6), 522-529.
Maloney, W. F., & McFillen, J. M. (1983). Research needs in construction worker
performance. Journal of Construction Engineering and Management, 109(2),
245-254.
Bibliography 187
March, J. G. (1962). The Business Firm as A Political Coalition. The Journal of
Politics, 24(4), 662.
Marsh, H. W., & Hau, K.-T. (1996). Assessing goodness of fit: Is parsimony always
desirable? The Journal of Experimental Education, 64(4), 364-390.
Masrom, M. A., Skitmore, M., & Bridge, A. (2013). Determinants of contractor
satisfaction. Construction Management and Economics, 31(7), 761-779.
Masrom, M. A. N. (2012). Developing a predictive contractor satisfaction model
(CoSMo) for construction projects.
Matthews, K. A., Cottington, E. M., Talbott, E., Kuller, L. H., & Siegel, J. M.
(1987). Stressful work conditions and diastolic blood pressure among blue
collar factory workers. American Journal of Epidemiology, 126(2), 280-291.
Mbachu, J., & Nkado, R. (2006). Conceptual framework for assessment of client
needs and satisfaction in the building development process. Construction
Management and Economics, 24(1), 31-44.
McDonald, R. P., & Ho, M.-H. R. (2002). Principles and practice in reporting
structural equation analyses. Psychological methods, 7(1), 64.
Mehrabian, A., & Russell, J. A. (1974). An approach to environmental psychology:
the MIT Press.
Meier, L. L., & Spector, P. E. (2013). Reciprocal effects of work stressors and
counterproductive work behavior: A five-wave longitudinal study. Journal of
Applied Psychology, 98(3), 529.
Merrow, E. W., Chapel, S. W., & Worthing, C. (1979). Review of cost estimation in
new technologies: implications for energy process plants. Retrieved from
Mikkelsen, A., & Grønhaug, K. (1999). Measuring Organizational Learning Climate
A Cross-National Replication and Instrument Validation Study Among
Public Sector Employees. Review of public personnel administration, 19(4),
31-44.
Min, J. H., & Lee, Y. C. (2005). Bankruptcy prediction using support vector machine
with optimal choice of kernel function parameters. Expert systems with
applications, 28(4), 603-614.
188 Bibliography
Mohamed, S. (2003). Performance in international construction joint ventures:
Modeling perspective. Journal of Construction Engineering and
Management, 129(6), 619-626.
Molenaar, K., Washington, S., & Diekmann, J. (2000). Structural Equation Model of
Construction Contract Dispute Potential. Journal of Construction
Engineering and Management, 126(4), 268-277.
Moore, A. W. (2001). Cross-validation for detecting and preventing overfitting.
School of Computer Science Carneigie Mellon University.
Mostert, K., Peeters, M., & Rost, I. (2011). Work–home interference and the
relationship with job characteristics and well‐being: a South African study
among employees in the construction industry. Stress and Health, 27(3),
e238-e251.
Mowday, R. T., Steers, R. M., & Porter, L. W. (1979). The measurement of
organizational commitment. Journal of vocational behavior, 14(2), 224-247.
Muchinsky, P. M., & Monahan, C. J. (1987). What is person-environment
congruence? Supplementary versus complementary models of fit. Journal of
Vocational Behavior, 31(3), 268-277.
Mulaik, S. A., James, L. R., Van Alstine, J., Bennett, N., Lind, S., & Stilwell, C. D.
(1989). Evaluation of goodness-of-fit indices for structural equation models.
Psychological bulletin, 105(3), 430.
Muller, D., Judd, C. M., & Yzerbyt, V. Y. (2005). When moderation is mediated and
mediation is moderated. Journal of personality and social psychology, 89(6),
852.
Munns, A., & Bjeirmi, B. F. (1996). The role of project management in achieving
project success. International journal of project management, 14(2), 81-87.
Næs, T., & Martens, H. (1988). Principal component regression in NIR analysis:
viewpoints, background details and selection of components. Journal of
chemometrics, 2(2), 155-167.
Nerkar, A. A., McGrath, R. G., & MacMillan, I. C. (1996). Three facets of
satisfaction and their influence on the performance of innovation teams.
Journal of Business Venturing, 11(3), 167-188.
Bibliography 189
Newton, S. (1991). An agenda for cost modelling research. Construction
Management and Economics, 9(2), 97-112.
Ng, S. T., Skitmore, R. M., & Leung, T. K. (2005). Manageability of stress among
construction project participants. Engineering, Construction and
Architectural Management, 12(3), 264-282.
Noy, C. (2008). Sampling knowledge: The hermeneutics of snowball sampling in
qualitative research. International Journal of social research methodology,
11(4), 327-344.
O’brien, R. M. (2007). A caution regarding rules of thumb for variance inflation
factors. Quality & Quantity, 41(5), 673-690.
Oberlender, G. D., & Trost, S. M. (2001). Predicting accuracy of early cost estimates
based on estimate quality. Journal of Construction Engineering and
Management, 127(3), 173-182.
Odeh, A.M., & Battaineh, H.T. (2002). Causes of construction delay: traditional
contracts. International Journal of Project Management, 20 (1), 67-73.
Ofori, G. (2000). Greening the construction supply chain in Singapore. European
Journal of Purchasing & Supply Management, 6(3), 195-206.
Oke, A. E., Ogunsami, D. R., Ogunlana, S. (2012). Establishing a common ground
for the use of structural equation modelling for construction related research
studies. Australasian Journal of Construction Economics and Building, 12
(3), 89-94.
Ong, T.F., & Musa, G. (2012). Examining the influences of experience, personality
and attitude on SCUBA divers' underwater behaviour: A structural equation
model. Tourism Management, 33 (6), 1521-1534.
Organ, D. W. (1988a). Organizational citizenship behavior: The good soldier
syndrome: Lexington Books/DC Heath and Com.
Organ, D. W. (1988b). A restatement of the satisfaction-performance hypothesis.
Journal of management, 14(4), 547-557.
190 Bibliography
Organ, D. W., & Ryan, K. (1995). A meta‐analytic review of attitudinal and
dispositional predictors of organizational citizenship behavior. Personnel
psychology, 48(4), 775-802.
Orlitzky, M., & Swanson, D. L. (2012). Assessing stakeholder satisfaction: Toward a
supplemental measure of corporate social performance as reputation.
Corporate Reputation Review, 15(2), 119-137.
Ostroff, C. (1992). The relationship between satisfaction, attitudes, and performance:
An organizational level analysis. Journal of applied psychology, 77(6), 963.
Park, S. H. (2009). Whole life performance assessment: critical success factors.
Journal of construction engineering and management, 135(11), 1146-1161.
Park, W. R., & Chapin, W. B. (1992). Construction bidding: Strategic pricing for
profit: Wiley New York.
Park, Y., Son, H., & Kim, C. (2012). Investigating the determinants of construction
professionals' acceptance of web-based training: An extension of the
technology acceptance model. Automation in Construction, 22, 377-386.
Paul, W.L., & Taylor, P.A. (2008). A comparison of occupant comfort and
satisfaction between a green building and a conventional building. Building
and Environment, 43 (11), 1858-1870.
Pearsall, M. J., & Ellis, A. P. (2006). The effects of critical team member
assertiveness on team performance and satisfaction. Journal of Management,
32(4), 575-594.
Penney, L. M., & Spector, P. E. (2005). Job stress, incivility, and counterproductive
work behavior (CWB): The moderating role of negative affectivity. Journal
of Organizational Behavior, 26(7), 777-796.
Pervin, L. A. (1987). Person-environment congruence in the light of the person-
situation controversy. Journal of Vocational Behavior, 31(3), 222-230.
Pesämaa, O., Eriksson, P. E., & Hair, J. F. (2009). Validating a model of cooperative
procurement in the construction industry. International Journal of Project
Management, 27(6), 552-559.
Bibliography 191
Pheng, L. S., & Chuan, Q. T. (2006). Environmental factors and work performance
of project managers in the construction industry. International Journal of
Project Management, 24(1), 24-37.
Pitt, M., Tucker, M., Riley, M., & Longden, J. (2009). Towards sustainable
construction: promotion and best practices. Construction Innovation:
Information, Process, Management, 9(2), 201-224.
Podsakoff, P. M., Ahearne, M., & MacKenzie, S. B. (1997). Organizational
citizenship behavior and the quantity and quality of work group performance.
Journal of applied psychology, 82(2), 262.
Porter, L. W., Crampon, W. J., & Smith, F. J. (1976). Organizational commitment
and managerial turnover: A longitudinal study. Organizational Behavior and
Human Performance, 15(1), 87-98.
Porter, L. W., Steers, R. M., Mowday, R. T., & Boulian, P. V. (1974). Organizational
commitment, job satisfaction, and turnover among psychiatric technicians.
Journal of applied psychology, 59(5), 603.
Porter, M. E., & Kramer, M. R. (2006). The link between competitive advantage and
corporate social responsibility. Harvard business review, 84(12), 78-92.
Posada, D., & Buckley, T. R. (2004). Model selection and model averaging in
phylogenetics: advantages of Akaike information criterion and Bayesian
approaches over likelihood ratio tests. Systematic biology, 53(5), 793-808.
Raftery, J. (1987). The state of cost/price modelling in the construction industry: a
multicriteria approach. Building Cost Modelling and Computers. E & FN
Spon, 49-71.
Reisinger, Y., & Turner, L. (1999). Structural equation modeling with Lisrel:
application in tourism. Tourism Management, 20(1), 71-88.
Rhoades, L., & Eisenberger, R. (2002). Perceived organizational support: a review of
the literature. Journal of applied psychology, 87(4), 698.
Richmond, A., & Skitmore, M. (2006). Stress and coping: a study of project
managers in a large ICT organisation. Project Management Journal, 37(5), 5-
16.
Rigdon, E. E. (1994). Calculating degrees of freedom for a structural equation model.
Structural Equation Modeling: A Multidisciplinary Journal, 1(3), 274-278.
192 Bibliography
Ringle, C. M., Sarstedt, M., & Straub, D. W. (2012). Editor's comments: a critical
look at the use of PLS-SEM in MIS quarterly. MIS quarterly, 36(1), iii-xiv.
Ringle, C. M., Wende, S., & Will, A. (Singer-songwriters). (2005). SmartPLS 2.0
(beta). On: Hamburg.
Rivals, I., & Personnaz, L. (1999). On cross validation for model selection. Neural
Computation, 11(4), 863-870.
Rounds, J. B., Dawis, R., & Lofquist, L. H. (1987). Measurement of person-
environment fit and prediction of satisfaction in the theory of work
adjustment. Journal of Vocational Behavior, 31(3), 297-318.
Runeson, G., & Skitmore, M. (1999). Tendering theory revisited. Construction
Management & Economics, 17(3), 285-296.
Sackett, P. R., & Wanek, J. E. (1996). New developments in the use of measures of
honesty integrity, conscientiousness, dependability trustworthiness, and
reliability for personnel selection. Personnel Psychology, 49(4), 787-829.
Sambasivan, M., & Soon, Y. W. (2007). Causes and effects of delays in Malaysian
construction industry. International Journal of project management, 25(5),
517-526.
Sarkar, M., Aulakh, P. S., & Cavusgil, S. T. (1998). The strategic role of relational
bonding in interorganizational collaborations: an empirical study of the global
construction industry. Journal of International Management, 4(2), 85-107.
Schneider, B. (2001). Fits about fit. Applied psychology, 50(1), 141-152.
Schwab, D. P., & Cummings, L. L. (1970). Theories of performance and satisfaction:
A review. Industrial Relations: A Journal of Economy and Society, 9(4), 408-
430.
Shah, R., & Goldstein, S. M. (2006). Use of structural equation modeling in
operations management research: Looking back and forward. Journal of
Operations Management, 24(2), 148-169.
Shi, L., Ye, K., Lu, W., & Hu, X. (2014). Improving the competence of construction
management consultants to underpin sustainable construction in China.
Habitat International, 41, 236-242.
Bibliography 193
Siang, L.C., & Ali, A.S. (2012). Implementation of risk management in the
Malaysian construction industry. Journal of Surveying, Construction and
Property, 3, 1-15.
Siu, O.-l., Phillips, D. R., & Leung, T.-w. (2003). Age differences in safety attitudes
and safety performance in Hong Kong construction workers. Journal of
Safety Research, 34(2), 199-205.
Skitmore, M. (1985). The influence of professional expertise in construction price
forecast: Salford : University of Salford Department of Civil Engineering.
Skitmore, M. (1987a). Construction Prices : the Market Effect. Salford,United
Kingdom: University of Salford Environmental Resources Unit
Skitmore, M. (1987b). The effect of project information on the accuracy of building
price forecasts. Building Cost Modelling and Computers, 327-336.
Skitmore, M., & Marston, V. (1999a). Cost modelling: Taylor & Francis.
Skitmore, M., & Ng, S. T. (2003). Forecast models for actual construction time and
cost. Building and environment, 38(8), 1075-1083.
Skitmore, M., & Patchell, B. (1990). Developments in contract price forecasting and
bidding techniques. Quantity surveying techniques: New directions, 75-120.
Skitmore, M., Stradling, S., Tuohy, A., & Mkwezalamba, H. (1990). The accuracy of
construction price forecasts: University of Salford.
Skitmore, R. M., & Marston, V. (1999b). Cost modelling: Taylor & Francis.
Smith, C., Organ, D. W., & Near, J. P. (1983). Organizational citizenship behavior:
Its nature and antecedents. Journal of applied psychology, 68(4), 653.
Smith, P. C. (1969). The measurement of satisfaction in work and retirement: A
strategy for the study of attitudes. Chicago: Rand McNally.
Sobel, M E. (1982). Asymptotic confidence intervals for indirect effects in structural
equation models. Sociological Methodology, 13, 290–312.
194 Bibliography
Soetanto, R., & Proverbs, D. G. (2002). Modelling the satisfaction of contractors: the
impact of client performance. Engineering Construction and Architectural
Management, 9(5‐6), 453-465.
Soetanto, R., & Proverbs, D. G. (2004). Intelligent models for predicting levels of
client satisfaction. Journal of construction Research, 5(02), 233-253.
Son, H., Kim, C., & Kim, C. (2012). Hybrid principal component analysis and
support vector machine model for predicting the cost performance of
commercial building projects using pre-project planning variables.
Automation in Construction, 27(0), 60-66.
Son, H., Park, Y., Kim, C., & Chou, J.-S. (2012). Toward an understanding of
construction professionals' acceptance of mobile computing devices in South
Korea: An extension of the technology acceptance model. Automation in
Construction, 28, 82-90.
Spokane, A. R. (1987). Conceptual and methodological issues in person-environment
fit research. Journal of Vocational Behavior, 31(3), 217-221.
Stoy, C., & Schalcher, H.-R. (2007). Residential building projects: building cost
indicators and drivers. Journal of construction engineering and management,
133(2), 139-145.
Sullivan, J., & Joyce, P. (2005). Model selection in phylogenetics. Annual Review of
Ecology, Evolution, and Systematics, 445-466.
Sunindijo, R., & Zou, P. (2012). Political Skill for Developing Construction Safety
Climate. Journal of Construction Engineering and Management, 138(5), 605-
612.
Tah, J. H. M., Thorpe, A., & McCaffer, R. (1994). A survey of indirect cost
estimating in practice. Construction Management and Economics, 12(1), 31-
36.
Tang, S., Lu, M., & Chan, Y. (2003). Achieving client satisfaction for engineering
consulting firms. Journal of Management in Engineering, 19(4), 166-172.
Tennant, C. (2001). Work-related stress and depressive disorders. Journal of
Psychosomatic Research, 51(5), 697-704.
Bibliography 195
Tett, R. P., & Meyer, J. P. (1993). Job satisfaction, organizational commitment,
turnover intention, and turnover: path analyses based on meta‐analytic
findings. Personnel psychology, 46(2), 259-293.
Thompson, B. (2004). Exploratory and confirmatory factor analysis: Understanding
concepts and applications: American Psychological Association.
Toor, S.-u.-R., & Ogunlana, S. O. (2010). Beyond the ‘iron triangle’: Stakeholder
perception of key performance indicators (KPIs) for large-scale public sector
development projects. International Journal of Project Management, 28(3),
228-236.
Torbica, Z.M., & Stroh, R.C. (2001). Customer satisfaction in home building.
Journal of Construction Engineering and Management, 127 (1), 82-86.
Turner, J. R. (2008). The handbook of project-based management. US: McGraw-Hill
Professional.
Turner, R., & Zolin, R. (2012). Forecasting success on large projects: developing
reliable scales to predict multiple perspectives by multiple stakeholders over
multiple time frames. Project Management Journal, 43(5), 87-99.
Tuuli, M. M., & Rowlinson, S. (2009). Performance consequences of psychological
empowerment. Journal of Construction Engineering and Management,
135(12), 1334-1347.
Vapnik, V. (2000). The nature of statistical learning theory: Springer Science &
Business Media.
Vigneau, E., Devaux, M., Qannari, E., & Robert, P. (1997). Principal component
regression, ridge regression and ridge principal component regression in
spectroscopy calibration. Journal of chemometrics, 11(3), 239-249.
Viswesvaran, C., & Ones, D. S. (2000). Perspectives on models of job performance.
International Journal of Selection and Assessment, 8(4), 216-226.
Wade, M. R., & Parent, M. (2002). Relationships between job skills and
performance: A study of Webmasters. Journal of Management Information
Systems, 18(3), 71-96.
Wang, F., & Du, T. (2000). Using principal component analysis in process
performance for multivariate data. Omega, 28(2), 185-194.
196 Bibliography
Wang, X., & Huang, J. (2006). The relationships between key stakeholders’ project
performance and project success: Perceptions of Chinese construction
supervising engineers. International Journal of Project Management, 24(3),
253-260.
Wanous, J. P., & Lawler, E. E. (1972a). Measurement and meaning of job
satisfaction. Journal of Applied Psychology, 56(2), 95-105.
Wanous, J. P., & Lawler, E. E. (1972b). Measurement and meaning of job
satisfaction. Journal of applied psychology, 56(2), 95.
Washington, S. P., Karlaftis, M. G., & Mannering, F. L. (2010). Statistical and
econometric methods for transportation data analysis: CRC press.
Wayne, S. J., Shore, L. M., & Liden, R. C. (1997). Perceived Organizational Support
and Leader-Member Exchange: A Social Exchange Perspective. The
Academy of Management Journal, 40(1), 82-111.
Weiss, D. J., Dawis, R. V., & England, G. W. (1967). Manual for the Minnesota
Satisfaction Questionnaire. Minnesota studies in vocational rehabilitation.
Westerman, J. W., & Cyr, L. A. (2004). An integrative analysis of person-
organization fit theories. International Journal of Selection and Assessment,
12(3), 252-261.
Williams, T., Jonny Klakegg, O., Walker, D. H., Andersen, B., & Morten
Magnussen, O. (2012). Identifying and acting on early warning signs in
complex projects. Project Management Journal, 43(2), 37-53.
Williams, T., Lakshminarayanan, S., & Sackrowitz, H. (2005). Analyzing bidding
statistics to predict completed project cost. In Proceedings of the
International Conference on Computing in Civil Engineering.
Williams, T. P. (2003). Predicting final cost for competitively bid construction
projects using regression models. International Journal of Project
Management, 21(8), 593-599.
Winefield, H. R., & Anstey, T. J. (1991). Job stress in general practice: practitioner
age, sex and attitudes as predictors. Family practice, 8(2), 140-144.
Bibliography 197
Wong, P. S., Cheung, S. O., & Fan, K. L. (2009). Examining the relationship
between organizational learning styles and project performance. Journal of
Construction Engineering and Management, 135(6), 497-507.
Wong, P. S. P., & Lam, K. Y. (2011). Facing turbulence: Driving force for
construction organizations to regain unlearning and learning traction. Journal
of Construction Engineering and Management, 138(10), 1202-1211.
Wong, W. K., Cheung, S. O., Yiu, T. W., & Pang, H. Y. (2008). A framework for
trust in construction contracting. International Journal of Project
Management, 26(8), 821-829.
Xia, B., & Chan, A. P. (2012). Measuring complexity for building projects: a Delphi
study. Engineering, Construction and Architectural Management, 19(1), 7-
24.
Xia, B., Xiong, B., Skitmore, M., Wu, P., & Hu, F. (2015). Investigating the Impact
of Project Definition Clarity on Project Performance: Structural Equation
Modeling Study. Journal of Management in Engineering, 04015022.
Xiong, B. (2015). The Role of Person-Environment Fit in Promoting Job
Performance: Towards a Conceptual Model and a Research Agenda. Paper
presented at 6th International Conference on Engineering, Project, and
Production Management (EPPM2015), Gold Coast, Australia.
Xiong, B., Skitmore, M., & Xia, B. (2014). Exploring the internal dimensions of
work stress: evidence from construction cost estimators in China. In
Proceedings of the 30th Annual ARCOM Conference (pp. 321-329):
Association of Researchers in Construction Management (ARCOM).
Xiong, B., Skitmore, M., & Xia, B. (2015a). A critical review of structural equation
modeling applications in construction research. Automation in Construction,
49, 59-70.
Xiong, B., Skitmore, M., & Xia, B. (2015b). Exploring and validating the internal
dimensions of occupational stress: evidence from construction cost estimators
in China. Construction Management and Economics, 33(5-6), 495-507.
Xiong, B., Skitmore, M., Xia, B., Masrom, M. A., Ye, K., & Bridge, A. (2014).
Examining the influence of participant performance factors on contractor
satisfaction: A structural equation model. International Journal of Project
Management, 32(3), 482-491.
198 Bibliography
Xiong, B., & Xia, B. (2014). Examining the Effects of Early Cost Drivers on
Contingencies. In Construction Research Congress 2014 (pp. 1518-1527):
ASCE.
Xu, L. D. (1994). Case based reasoning. Potentials, IEEE, 13(5), 10-13.
Yang, J.-B., & Peng, S.-C. (2008). Development of a customer satisfaction
evaluation model for construction project management. Building and
Environment, 43(4), 458-468.
Yang, L.-R., Chen, J.-H., & Wang, H.-W. (2012). Assessing impacts of information
technology on project success through knowledge management practice.
Automation in Construction, 22, 182-191.
Ye, K.H., & Xiong, B. (2011). Corporate social performance of construction
contractors in China: Evidences from major firms. Proceedings of the 16th
International Symposium on Advancement of Construction Management and
Real Estate, Hong Kong Polytechnic Univ, 125-130.
Yip, B., & Rowlinson, S. (2006). Coping strategies among construction
professionals: cognitive and behavioural efforts to manage job stressors.
Journal for Education in the Built Environment, 1(2), 70-79.
Yong, Y.C., Mustaffa, N.E., 2012. Analysis of factors critical to construction project
success in Malaysia. Engineering, Construction and Architectural
Management, 19 (5), 543 - 556.
Yuan, X.-X. (2011). A correlated bidding model for markup size decisions.
Construction Management and Economics, 29(11), 1101-1119.
Zolin, R., Hinds, P. J., Fruchter, R., & Levitt, R. E. (2004). Interpersonal trust in
cross-functional, geographically distributed work: A longitudinal study.
Information and organization, 14(1), 1-26.