Analysing the Effect of Latent Variables on Willingness to Pay in Mode Choice Models
R Espino, C Roman, Universidad de Las Palmas de Gran Canaria, ES; J de D Ortuzar, Pontificia Universidad Catolica de Chile, CL
The joint use of Revealed Preference (RP) and Stated Preference (SP) data has become recommended practice in transport demand analysis. RP data are based on individual choices and allow the researcher to characterize actual travel behaviour. SP data are based on individuals? stated behaviour under hypothetical scenarios and are useful when the problem is to analyse the demand for new alternatives or measure the effect of latent variables and their interactions with other attributes. The estimation of choice models combining RP and SP data exploits the advantages and overcomes the limitations that each type of data has separately.
Most applied research concerning the value of travel time savings (VTTS) or other willingness to pay (WTP) measures is based on the formulation of simplified models that impose strong restrictions on the distribution of random error terms (e.g. IIA property of the MNL model) or specify systematic utility not considering the interaction of other variables with the main policy attributes. Thus the WTP measures, which are simply derived as the ratio between the marginal utility of the corresponding attribute and the marginal utility of income (which is equal to minus the marginal utility of cost), are in most cases represented by a fixed value equal to the time (or other attribute) parameter divided by the cost parameter.
In this paper a SP choice experiment is designed to analyse how the interaction of a latent variable (comfort) with other policy attributes influence mode choice in an urban/interurban context and explores how the different (WTP) measures are affected by the levels considered for this variable. A previous RP survey of 950 interviews was carried out to obtain information about actual trip behaviour in the two main interurban corridors in Gran Canaria Island and to adapt the SP experiment to each individuals? experience. A focus group, recruiting public transport users and car users, helped us to define a final set of five attributes: travel cost, travel time, parking fee, frequency of the service and comfort. A fractional factorial design of 27 scenarios that allow us to measure main effects and two factor interactions among three attributes was divided into three blocks in order to reduce respondent burden. Out of an original sample of 372 individuals 97 answered the SP survey yielding a total of 871 choice observations. A detailed analysis allowed us to detect captive, lexicographic and inconsistent individuals and to examine how the removal of these observations affected the estimation results.
Several linear-in-parameter specifications of multinomial logit models considering only main effects and the interaction of comfort with the rest of the attributes have been estimated including only the SP data set. From these, consistent VTTS and other WTP measures were obtained. The estimation results show that the VTTS decrease from 18.9 ?/h to 9.03 ?/h as comfort is improved from the lowest to the highest level considered in the experiment; it was also found that the WTP for improvements in comfort range from 12.05? to 4.44? per minute of trip length. Models including pooled SP and RP data are currently being analysed and the estimation results from both cases will be compared in the paper.
Finally, more flexible mixed logit model specifications relaxing the hypotheses of independence and homoscedasticity will be studied. Different estimation techniques such as simulated maximum likelihood and hierarchical Bayesian procedures are being applied. The WTP results derived from these estimations will be contrasted with those obtained from the simpler models; previous experience in this sense has shown results to vary little whilst the estimation of WTP is considerably more involved. We will also touch on these issues.
Association for European Transport