Systematic Calibration of Multinomial Logit Models

Abstract

This paper presents a systematic procedure for calibrating a multinomial logit model through the use of the UTPS computer package developed by the Urban Mass Transportation Administration (UMTA) for the purpose of travel demand forecasting. A brief examination of the mathematical basis of the multinomial logit model is also presented. The proposed procedure incorporates the use of a number of UTPS program modules, and lends itself to the use of a data base commonly available to most regional and local planning agencies. A comprehensive case study is presented, demonstrating the applicability of the procedure in a medium‐sized urban area in southeast Michigan. The paper is written on the premise that the diagnostic tests contained in the ULOGIT module (the software developed by UMTA for calibrating logit models) constitute a necessary step towards model calibration, but are not sufficient for the identification of the overall best model. A number of other critical tests are recommended in the paper that require the use of the candidate models in actual multimodal allocation of the person trips, and the identification of the best model based upon the overall goodness of fit between the observed data and estimated data. The case study shows that the diagnostic tests contained in ULOGIT can be used to develop models with similar predictive qualities. The writers recommend the use of this procedure since it will aid the planner in identifying the model with the overall best predictive quality.