Effect of Distributional Assumption of Random Parameters of Mixed Logit Model on Willingness-to-Pay Values

For rational estimation of users’ benefit, it is necessary to understand users’ willingness-to-pay (WTP). In several WTP studies, stated preference data have been analyzed using Mixed Logit (ML) model specification. In ML models, it is necessary to make an assumption regarding the distribution of random parameters. Researchers have developed ML models with different distributional assumptions of random parameters. However, the effect of distributional assumptions of random parameters in ML model on goodness-of-fit and WTP values has not been studied adequately. In the present work, an investigation is carried out in this regard taking reference to a case study of feeder service to bus stop in rural India. Various Mixed Logit models were attempted with different distributional assumptions of random parameters such as normal, log normal, triangular, uniform, constrained normal, constrained triangular, constrained uniform, etc. Variation of goodness of fit statistics and WTP values are observed across different ML models. The work indicates the importance of distributional assumption while developing ML model in WTP studies. The work also indicates that it is desirable to develop several ML models with different distributional assumptions and then select the superior one based on goodness of fit statistics.