The 2013 national Dutch value of time study Gerard de Jong – Significance and ITS Leeds

29 October 2015

The 2013 national Dutch value of time studyGerard de Jong – Significance and ITS Leeds

29 October 2015

and reliability

3

Contents

1. What’s the question?

2. Data collection:I. The 2009 SP data: internet panelII. The 2011 SP data: en-route recruitment

3. Model estimation

4. Impact of recruitment method

5. The recommended values

6. A fair comparison of the 1997 and 2009/2011 VTT

4

Why do we need a VTT?

In many countries, transport projects (e.g. new road or railway line) are evaluated ex ante using cost-benefit analysis (CBA)

In CBA project effects are expressed in money units

Costs include construction, maintenance and external cost

Main benefit often is travel time saved□ There could also be journey time reliability benefits (often still ignored)

This is in hours or minutes, so we need a conversion factor to money□ This factor is called the value of travel time VTT (e.g. in euros per hour)

5

The context: CBA of transport projects

Costs BenefitsConstruction costs Time benefits: Pt*Qt

Pt: Value of travel time VTTQt: from transport model

Change in maintenance costs

Reliability benefits: Pr*Qr

Pr: Value of travel time variability VTTVQr: forecasting model or surcharge

Change in external costs Other transport cost savingsFrom transport model

… …… …

6

The new national study

The objective of this project, for the Dutch Ministry of Infrastructure and the Environment, was:to provide values of time (update) and travel time reliability (first Dutch empirically-based values) for passenger and freight transport by mode that can be used in cost-benefit analysis (CBA) of transport projects

The project was completed and the report was officially released in June 2013) (weblink at the end); the values are now official: used in all national transport projects

In The Netherlands the VTT and VTTV are specifically for use in CBA, not for inputs into transport forecasting models

The Netherlands also had national VTT studies (passengers) in 1988-1990 and 1997-1998

7

What’s the question?

This presentation is about the passenger transport component of the study

Values of time (VTTS) in passenger transport nowadays mainly come from Stated Preference (SP) surveys

(see international meta-analysis by Wardman et al., 2012)

Different interview methods:□ Mailback (pen and paper/cards)□ CAPI (used for freight transport)□ CATI□ Internet Recruitment method:

- Project-specific recruitment (e.g. en-route)- From existing internet panel

8

Initial choice of interview and recruitment method (2009 data)

The SP surveys required considerable customisation□ Mailback can only provide this through extensive two-step procedures

CAPI and CATI were considered too expensive for a large survey (labour cost)

Initial choice: internet survey using an existing internet panel

9

2009 survey procedure (1)

5,760 members of an existing on-line panel were interviewed using computerised stated preference interviews in November 2009

Specific target numbers of interviews were set (and reached) for different segments:□ Transport mode used (car, train/metro, bus/tram, airplane and

recreational navigation)□ Travel purpose (commuting, business, other)□ Time-of-day (peak, off-peak)□ Presence of transfers (public transport only)

All respondents were asked which modes they had used in the past three months, etc. □ This was used to allocate respondents to questionnaires for specific

segments

10

2009 survey procedure (2)

All respondents were drawn from the largest on-line panel of The Netherlands (240.000 participants)

The survey could be started by clicking on a weblink

The members received a reward for successfully completing the interview (equivalent to €1.50)

The interviews on average took 20 minutes

11

Example of an SP choice screen (exp. 1)

12

Example of an SP choice screen (exp. 2a)

13

Initial results (2009 data)

VOTs implausibly low □ About € 4 per hour for car and public transport□ Substantially lower than the official values (about € 9 per hour) and the

international literature

Checked for possible explanations:□ socio-economic composition of sample□ travel time distribution of sample□ changes in the statistical design of the SP□ Including reliability in the SP□ Increased use of mobile phones, smartphones□ Impact of economic crisis□ Increase in congestion

These only explained part of the differences with the official values

14

But there could be another explanation … The sample of respondents obtained from this internet panel might

be biased with respect to their value of time

Within each segment (socio-economic, trip purpose, trip length, mode), the respondents that participate in such an online panel (which takes time, for a rather low monetary reward) might have a lower VOT than a non-participant

This is a self-selection problem

Even after expansion, the resulting values of time would then be lower than the true values of time

To investigate this hypothesis, another data set was collected in the first half of 2011

15

The 2011 SP data: en-route recruitment

Almost 1500 respondents recruited at petrol stations, parking garages, train stations, bus stops, airports and ports

This is the same recruitment method as in earlier national value of time surveys of 1988/1990 and 1997/1998

Persons willing to participate were asked to answer an internet questionnaire on the intercepted trip: □ Almost identical to the questionnaire used in 2009□ We only asked one additional question to determine whether they were a

member of an internet panel (and whether this was “our” internet panel)

16

2011 Models distinguishing members/non-members of internet panels

MNL models

Advanced MNL models that:□ yield a higher VTT for higher base time and cost levels, and□ smaller VTTs for smaller changes offered in time and cost

Advanced MNL with socio-economic interaction terms

Advanced MNL with socio-economic interaction terms plus latent VTT classes (LC model)

17

Linear versus non-linear time and cost effects in the utility functionUtility function 1997:

Utility function 2009/2011:

18

4. Results for MNL model

Mode PurposeRelative VTT for panel member

(non-member=1)

MNL Adv MNL Adv MNL + socio

Latent class

Car/train/BTM

Commuting 0.61

Business 0.93

Other 0.90

Airplane All 1.07

Recr. navigation Other (0.96)

19

4. Results for MNL models

Mode PurposeRelative VTT for panel member

(non-member=1)

MNL Advanced MNL

Car/train/BTM

Commuting 0.61 0.80

Business 0.93 0.88

Other 0.90 0.93

Airplane All 1.07 0.82

Recr. navigation Other (0.96) (0.98)

20

4. Results for MNL models

Mode Purpose

Relative VTT for panel member (non-member=1)

MNL Advanced MNL

Adv.MNL w.

interaction

Car/train/BTM

Commuting 0.61 0.80 0.80

Business 0.93 0.88 0.87

Other 0.90 0.93 (1.01)

Airplane All 1.07 0.82 0.82

Recr. navigation Other (0.96) (0.98) (0.98)

21

4. Results for MNL and LC models

Mode Purpose

Relative VTT for panel member (non-member=1)

MNL Advanced MNL

Adv.MNL w.

interaction

Latent class

Car/train/BTM

Commuting 0.61 0.80 0.80 0.80

Business 0.93 0.88 0.87 (0.81)

Other 0.90 0.93 (1.01) (0.89)

Airplane All 1.07 0.82 0.82 0.82

Recr. navigation Other (0.96) (0.98) (0.98) (0.95)

22

4. Results for panel members 2009 and 2011

Mode Purpose

Relative VTT for panel member

(non-member=1)Advanced MNL

2009 (‘our’ panel)Advanced MNL

2011 (all panels)

Car/train/BTM

Commuting 0.64 0.80

Business 0.66 0.88

Other 0.74 0.93

Airplane All 0.70 0.82

Recr. navigation Other (1.06) (0.98)

23

Discussion of results: does it matter/help? (1) Especially for commuting (car, train, bus, tram, metro): significant

lower values for panel members, □ even after correcting for the different distributions for the travel time

and travel cost, and after inclusion of the socio-economic interactions

Similar findings for the business and for airplane segment

Other purposes and recreational navigation: no significant difference between panel and non-panel

We conclude that in the 2009 survey there was a bias towards low-VTT persons, who are willing to give up time to participate in an internet panel and to fill out web questionnaires for a rather small reward

24

Discussion of results: does it matter/help? (2)

The resulting VTTs from the 2011 survey are much more in line with the values found in 1988/1990 and 1997/1998, □ which have always been regarded as very plausible by the various

transport sectors, □ and are not considered to be particularly high in an international

perspective

Our conclusion is that the most likely explanation is that the 2011 values are correct and that the 2009 values are biased downwards

25

The final VTT results are based on a combination of the 2009 and 2011 data

The base VTT and VTTV levels come from estimates on the 2011 data

Socio-economic interaction effects and the effect of the base time and cost levels as well as of changes in time and cost offered in the SP are estimated on 2009 and 2011

Also: latent class models used here, and expansion of the estimation results to the population (in hours travelled) using the 2010 national travel survey (OViN)

This yields the recommended values for use in CBA

26

Recommended VTTs in euros per person per hour

  Car TrainBus, tram, metro

All surface modes

Air Recr. navigation

Commute 9.25 11.50 7.75 9.75    

Business employee 12.75 15.50 10.50 13.50 85.75  

Business employer 13.50 4.25 8.50 10.50 -  

Business 26.25 19.75 19.00 24.00 85.75  

Other 7.50 7.00 6.00 7.00 47.00 8.25

All purposes 9.00 9.25 6.75 8.75 51.75 8.25

Note: all values are rounded off to the nearest multiple of € 0.25

27

Recommended reliability ratios

Reliability ratio (RR) = value of standard deviation of travel time/VTT

Car, train, bus, tram and metro:□ Commuting 0.4□ Business 1.1□ Other 0.6

Air:□ Business 0.7□ Other 0.7

28

It’s just not fair!

29

It’s just not fair!

Fair comparison: comparing like with like

Otherwise conclusions will be incorrect: Not based on real differences but on differences in methodology

30

The seven differences (methodological)

1997 2009/2011

Estimation space Interactions with cost and time

Interactions with VTT

Cost and time terms Linear Linear & non-

linearDependence of VTT on travel time itself No Yes

Socio-economic interaction factors No education

Different set, including education

Expansion procedure

Weights per segment

Sample enumeration

Expansion totals OVG 1995 OViN 2010Type of model MNL panel latent class Therefore the VTT in the 1998 report and the 2013 report cannot

be compared

31

Expected changes 1997-2010

Consumer prices rose by 32% -> VTT +32%

Real income (over and above the price change) went up by 30%. The Dutch guidelines adopted an income elasticity of the VTT of 0.5

-> VTT +15% More congestion, more crowded trains, lower compensation of

cost, crisis New ICT has become much more common in this period:

□ Mobiles (including handsfree, car kit), smartphones, iPads, laptops□ Easier to use travel time in a more productive/enjoyable way

-> VTT

In the period 1988-1997 VTT did not change much:□ Gunn (2001): effect of real income growth more or less balanced by the

technology effects

32

Methodology to obtain a fair comparison

Applying the 1997 methods on the 2009/2011 data is not so interesting (no benefits from methodological improvements)

So we redid the analysis of the 1997 data using the 2009/2011 methods

We did this re-analysis step-by-step to see the impact of each of the seven differences (similar steps for the analysis of 2009/2011 data)

This gives two results:□ Which VTT would we have obtained in the nineties if we could had used

modern methods (and future population data)?□ The real evolution of VTT by mode and purpose between 1997 and

2009/2011

33

Detailed comparison for commute

34

Outcomes: impact of methodological differences on 1997 or 2009/2011 VTT

1997 2009/2011 VTT

Estimation spaceInteractions

with cost and time

Interactions with VTT -4%

Cost and time terms Linear Linear & non-linear 0%

Dependence of VTT on travel time itself

No Yes -2% Differs by mode

Socio-economic interaction factors No education Different set,

including education +1%

Expansion procedure

Weights per segment

Sample enumeration +4% Differs by

mode

Expansion totals OVG 1995 OViN 2010 0% Differs by mode

Type of model MNL panel latent class

35

Outcomes: impact of methodological differences on 1997 or 2009/2011 VTT

1997 2009/2011 VTT

Estimation spaceInteractions

with cost and time

Interactions with VTT -4%

Cost and time terms Linear Linear & non-linear 0%

Dependence of VTT on travel time itself

No Yes -2% Differs by mode

Socio-economic interaction factors No education Different set,

including education +1%

Expansion procedure

Weights per segment

Sample enumeration +4% Differs by

mode

Expansion totals OVG 1995 OViN 2010 0% Differs by mode

Type of model MNL panel latent class +31%General range: 20-40%

Unobserved heterogeneity leads to a large downward bias in the VTT

36

Fair comparison 1997 – 2010 (no inflation correction; comparison of p-LC models)

  Car TrainBus, tram, metro

All surface modes

Commute +13% +49% +9% +23%

Business +13% +60% +98% +26%

Other +71% +59% +52% +65%

37

Discussion of results of the comparison

All VTTs by mode and purpose have increased

Overall, the increase is slightly below the expected increase of +47% □ small overall impact of ICT changes?□ Income elasticity of 0.5 seems about right?

Relatively small changes for commute and business: ICT developments more important than for other travel?

Relatively small increases for car (relative to train): □ important ICT developments for train had already entered the market in

1997?□ trains more crowded than in 1997?

38

What do we conclude?

Beware of internet panels in VTT research!

Allowing for unobserved heterogeneity (using a panel Latent Class model) increases VTT considerably□ Much more than the other six differences

In the period 1997-2009/2011 the average VTT went up by about price change plus 0.5 times the real income change□ But differences between purposes and modes that could be related to

ICT developments

39

For more information

email: dejong@significance.nl or g.c.dejong@its.leeds.ac.uk

Final report and papers:

http://www.kimnet.nl/sites/kimnet.nl/files/filemanager/bijlagen/Bijlage_Value_of_time_and_reliability_in_passenger_and_freight_transport_in_the_Netherlands_reprint.pdf

Kouwenhoven, M., G.C. de Jong, P. Koster, V.A.C. van den Berg, E.T. Verhoef, J.J. Bates and P. Warffemius (2014) New values of time and reliability in passenger transport in The Netherlands, Research in Transportation Economics, 47, 37-49.

Jong, G.C. de, M. Kouwenhoven, J. Bates, P. Koster, E. Verhoef. L. Tavasszy en P. Warffemius (2014) New SP-values of time and reliability for freight transport in the Netherlands, Transportation Research Part E, 64, 71-87.