This paper demonstrates the effectiveness of a correcting method of attrition bias in model building based on stated preference (SP) panel data. SP panel data for the New Transit in Hiroshima were obtained at two different points in time. The effect of attrition bias on mode choice model parameters based on SP data was empirically clarified. The attrition models which determine whether the respondents of first wave participate into the second survey or not were developed with the first wave data. Besides, a mode choice model was estimated based on the stayers at the second wave. The attrition bias of this model was corrected by sequential step on the assumption that the error correlates with that of the attrition model. It was found that this correcting could effectively censor out the biased share of each travel mode.
