Comparative Analysis of Alternate Econometric Structures for Trip Generation Models

Abstract

Traditionally trip generation models have been estimated with linear-regression structures even though this methodology does not recognize the nonnegativity and integer nature of the trips. Although the theoretical superiority of count-data models as an alternative approach is well recognized, the empirical benefits of such models have not been well established. In that context, the intent of this study is to undertake a comparative analysis of four different econometric structures for trip generation models. The structures are compared across three different trip purposes with significantly different distribution patterns. The models are estimated by using 2001 U.S. National Household Travel Survey data and are applied to samples from the 2009 National Household Travel Survey data. Predictive validations indicate that the ordered probit models are able to replicate the trip generation patterns better than linear-regression, log-linear, and negative-binomial models for all three trip purposes. The negative-binomial model performs reasonably well in the case of the non-home-based trips, which have a monotonically decreasing distribution pattern. The negative-binomial and the log-linear models have comparable mean errors for disaggregate predictions. Overall, this study recommends the use of ordered probit models as a substitute for the traditional linear-regression models.