Mode Choice Models Based on Stated Preference Panel Data Including Refreshed Samples

This paper deals with mode choice models based on panel data including refreshed samples at each wave. Four stated preference data sets concerning mode choice were used for this study, which were collected during 1987 to 93 in Hiroshima. It was first statistically confirmed that sample distribution with respect to socio-demographic characteristics for these data sets, represents the distribution in the population by examining the share of male respondents as a case study. Generalized Durbin-Watson test was used to check the null hypothesis on temporal independence for aggregate mode choice models of logit type. This test indicated that serial correlation for disturbance terms of the models constructed at each time point is significantly large. Therefore, models introducing the serial correlation were developed using Seemingly Unrelated Regression procedure. This could improved the statistical significance of estimated parameters, even though they did not provide higher correlation coefficients