An empirical investigation of values of travel time savings from stated preference data and revealed preference data

ABSTRACT

A number of studies have found that the willingness-to-pay (WTP) results estimated from revealed preference (RP) and stated preference (SP) data tend to be different. In this paper, we empirically estimate values of travel time savings from an SP data set and an RP data set and compare the findings within this study and between studies. The evidence shows that the design of a stated choice experiment has a significant impact on the ratio of SP and RP WTP values and reveals that presenting a full distribution of travel time to address random travel time variation in the choice scenarios, along with using a real market reference alternative as a pivot in the SP design, significantly reduces the gap between values of travel time savings estimated from SP data and RP data.

KEYWORDS:

• Travel behavior
• stated preference
• revealed preference
• hypothetical bias
• stated choice
• value of travel time savings
• commuting

Acknowledgments

We thank NSW (New South Wales) Bureau of Transport Statistics for allowing us using the RP data set. We also thank Professor John Rose for his innovative stated choice design. This paper is funded by Xi'an Eurasia University Research Group Foundation. The valuable comments of the editor and reviewers significantly improve this paper.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. For this study, ‘travel time on average’ in the choice task should be better represented as ‘normal travel time.’ For example, for car travel, we use normal travel time in modeling, for example, 51 min as shown in Figure 2, not the calculated average.

2. In reality, the travel time for rail would vary; however, this variation is much less significant compared to road transport such as bus and car.

3. Skim refers to a set of outputs (time, cost, distance, toll, transfer, etc.) generated by Strategic Travel Models when traffic assignment models have converged (equilibrium has been obtained).

4. For long trips, the shares of walk-only and cycling trips are very small; walk-only and cycling trips were removed from the analysis. Walk and cycling as an access or egress mode to PT services are included as a trip legs on PT trips. For long trips, the shares of walk-only and cycling trips are very small.

5. Both the SP and RP models were estimated using Nlogit5, with 500 Halton draws. Starting values for mixed logit are MNL values. The convergence criteria is the gradient g’Hg < ε_g where g is the current derivative vector and H is the inverse of the current Hessian.

6. The original formulation of RPL was made much earlier Ortúzar and Willumsen (2011).

7. For example, under the normal distributions, the VTTS range includes negative values, while under the lognormal distributions, the VTTS range for PT travel is also behaviorally implausible (i.e. AU$0.1–7075.3 per person hour).

8. Let c be the center and s be the spread (i.e. half the range). The density starts at c-s, rises linearly to c, and then drops linearly to c+s. It is zero below c-s and above c+s. The mean and mode are c. The standard deviation is the spread divided by $\sqrt{\sigma }$; hence, the spread is the standard deviation times $\sqrt{\sigma }$. The height of the tent at c is 1/s (such that each side of the tent has area s × (1/s) × (1/2) = 1/2, and both sides have area 1/2 + 1/2 = 1, as required for a density). The slope is 1/s². For a constrained distribution, the mean parameter is constrained to equal its spread (i.e. β_jk = β_k + |β_k| T_j, and T_j is a triangular distribution ranging between −1 and +1), and the density of the distribution rises linearly to the mean from zero before declining to zero again at twice the mean. Therefore, the distribution must lie between zero and some estimated value (i.e. the β_jk). When a constrained triangular distribution is used, the reported standard deviation parameter is the spread parameter. The mean and spread are the same under a constrained triangular distribution.

9. There are statistically significant parameter estimates for each mode as alternative specific or as generic parameters. They also have a meaningful sign. What is comforting is the significant absolute higher marginal disutility for waiting time compared to in-vehicle travel time. Also, the VTTS for PT in-vehicle time is in the range that makes sense.

10. Wardman (1988) used a survey conducted in 1983 in which the SP response (873 respondents in total) was based on a five-point scale: definitely prefer coach, probably prefer coach, no preference between coach and train, probably prefer coach, and definitely prefer coach. Wardman compared values of time estimated from the SP and RP models and found that there was no significant difference between RP and SP values of main mode in-vehicle time, but an SP/RP VTTS ratio of 2.24 for other mode in-vehicle time.

11. ‘Flexible work time, PT fare provided, Free parking provided, Fuel cost provided’ are a set of dummy variables describing the fringe benefits provided to the worker by their employers. For model identification, these variables should be included in either PT or car utility, not both (i.e. they do not vary by alternative mode). We could specify these variables in the car utility, and the sign of the variables will be reversed. The negative parameters associate with ‘free parking provided’ suggests that when workers are provided with free parking, they are less likely to commute by PT, which is expected. Similarly, license holders are less likely to be PT commuters because its parameter is negative.