Analysis of Discrete Choice Data with Repeated Observations: Comparison of Three Techniques in Intercity Travel Case

With the acceptance and widespread application of stated preference methods in transportation analysis, there has been a need for the analysis of choice data with repeated observations. Analysis of repeated choice data is complicated by correlation of responses across the choices made by a single individual. In the probit framework, repeated choice can easily be accommodated by way of correlation in the utilities associated with the errors. However, the probit framework has computational limitations; subsequently, researchers have looked to modeling repeated choice by using extensions of the logit model. Random parameter logit models estimated by simulation methods provide a promising modeling framework. The estimation results of some stated preference intercity business travel data are compared by using three alternative estimation techniques to account for correlations across the choices made by a single individual. It is believed that two of the alternative methods, which use random-parameter logit models (developed by researchers for use in an energy application), have not been used before in the transportation context. A comparison of the three techniques provides documentation for researchers and practitioners who plan to analyze such data.