International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
A pooled RP/SP mode, route and destination choice modeling approach to capture the heterogeneity of mode and user type effects in Austria
Basil Christian Schmid, Florian Aschauer, Inka Roesel, Stefanie Peer, Reinhard Hoessinger, Regine Gerike, Sergio Jara-Diaz, Kay Werner Axhausen

Last modified: 28 March 2017

Abstract


Motivation and objectives

Mode choice models have been extensively used to evaluate policy implications and level-of-service changes, providing a powerful tool in transport planning for developing effective travel demand forecasts (Bhat, 1998). As a key valuation indicator, the value of travel time has always been subject to extensive debate in both academia and politics, because savings in travel time have accounted for the biggest share of user benefits in most cost-benefit analysis.

Following Jara-Diaz and Guevara (2003), the value of travel time is the sum of two components: 1) the subjective value of travel time savings (VTTS), representing the willingness to reduce travel time in favor of other activities that generate more utility and 2) the direct (dis)-utility derived from the time assigned to travel (VTAT). The sum of these components equals the value of time as a resource (VOR), which is the marginal utility of an additional unit of leisure. VTTS and VTAT both may differ according to characteristics of the trip, travel mode and the user (Mackie et al., 2001): Typically, mode effects describe differences in VTTS due to heterogeneity in time sensitivities (comfort, productive time use, etc.), while user type effects are driven by differences in socio-economic and other (unobserved) characteristics. Disentangling these two effects is difficult due to self-selection issues (e.g. people with high income choose more pleasant modes but still have a higher VTTS; Mackie et al., 2001), but highly relevant e.g. for agent-based modeling, which is also addressed in this paper.

A shift of focus from the VTTS to VTAT in cost-benefit analyses would shed new light in policy appraisals: Instead of prioritizing the option with the largest reduction of travel time savings, one could prioritize the most convenient option with the largest gain of utility of time assigned to travel. However, estimating VTAT heavily relies on proper estimates of VTTS, which is the main focus of this paper: Given the large heterogeneity in respondents and trips in our data set, the aim is to provide VTTS estimates capturing mode (car, walk, bike, public transport; PT) and user type (income, season ticket ownership, working hours and residential location) effects for different situations (trip purposes, time) and distances, applying a joint RP/SP modeling approach.

Data

Data was collected for a representative sample of 748 respondents in Austria between 2015 and 2016, using a seven-day mobility, activity and expenditure diary (MAED) to get information about time use, expenditure allocation and travel behavior. From this, 17’052 RP mode choice observations were generated. In addition, respondents received personalized SP experiments with attribute reference values depending on one selected RP trip, leading to additional 12’428 choice observations. Finally, six different data sets were combined: RP mode choice, SP mode choice, SP car and SP PT route choice, SP car and SP PT destination choice.

Methodology

We present a methodological framework how to qualitatively assess two different approaches to optimally model mode and user type heterogeneity regarding estimation efficiency, behavioral plausibility and feasibility: We compare 1) multivariate segmentation by a-priori dividing the sample into discrete subgroups of respondents, for each of them estimating a separate choice model (e.g. Jara-Diaz and Guevara, 2003) and 2) a pooled approach using advanced (discrete and continuous) and efficient interaction techniques for the segments in 1), where we will use a Heckman type simultaneous equation system to include endogenous variables such as season ticket ownership. Both approaches incorporate flexible disturbance terms and random-effects specifications, accounting for the panel structure of the data (Hensher, 2001).

At a later stage, user and mode-specific VTTS are combined with the respective VOR outcomes from the continuous time use and expenditure allocation choice model to calculate all components (VTAT) of the complete Jara-Diaz and Guevara (2003) model formulation.

Preliminary results

Pooling different sources of data requires to account for the heterogeneity in variances, which is captured by the scale parameters. In both approaches, i.e. segmentation (A1) and interaction (A2), a similar pattern occurs: All unlabeled SP experiments, with exception of SP mode choice, exhibit a lower variance than the RP mode choice data.

In both approaches and all segments (here using univariate segmentation), estimated coefficients show significant and expected effects, with time sensitivities differing substantially by mode: In A2, the average VTTS is about 14 Euro/hour for car, 5 Euro/hour for PT, 12 Euro/h for bike and 20 Euro/h for walk, differing substantially between different trip purposes. Income elasticity of travel cost and distance elasticity of travel time are significant and negative: Higher income and distance strongly decrease cost and time sensitivity, respectively. The same VTTS ranking occurs in A1 with segment-specific values: Low income respondents exhibit a 50% lower VTTS of about 11 Euro/hour for car and 4 Euro/hour for PT than high income respondents, with the latter group exhibiting a significantly lower probability of choosing PT. Interestingly, respondents living in rural areas exhibit a larger gap, showing a 20% higher car VTTS than respondents living in urban areas, but a 20% lower PT VTTS: Especially in rural areas, PT is mainly used for longer distance trips. Also, respondents that own some kind of season ticket show a more than three times lower PT VTTS of about 4 Euro/hour than respondents without season ticket, disentangling user type and mode effects by showing almost identical car and PT VTTS for people without season ticket.

Literature

Bhat, C. R. (1998) Accommodating variations in responsiveness to level-of-service measures in travel mode choice modeling, Transportation Research Part A: Policy and Practice, 32 (7) 495-507.

Hensher, D. A. (2001) The sensitivity of the valuation of travel time savings to the specification of unobserved effects, Transportation Research Part E: Logistics and Transportation Review37 (2) 129-142.

Jara-Diaz, S. and C. A. Guevara (2003) Behind the subjective value of travel time savings, Journal of Transport Economics and Policy, 37 (1) 29-46.

Mackie, P. J., S. Jara-Diaz and A. S. Fowkes (2001) The value of travel time savings in evaluation, Transportation Research Part E: Logistics and Transportation Review, 37 (2) 91-106.