Jointly Estimated Cross-Sectional Mode Choice Models: Specification and Forecast Performance

This paper investigates a number of issues associated with jointly estimating disaggregate logit mode choice models for two periods using data collected at two points in time in independent cross-sectional travel surveys in a given urban area. These include: (1) the effect socio-economic characteristics of travelers have on the predictive and forecast performance of jointly estimated models; and (2) the effect of allowing the variance of the random utilities in the different time periods to differ and, more broadly, the impact transfer-bias scale parameters could have on joint-model predictive performance. The results show that well-specified jointly estimated models using data from two time periods yield comparable disaggregate and aggregate forecasts to those obtained from conventional forecasting models, estimated with data from a single cross-sectional survey. Socio-economic variables and transfer-bias scale parameters are found to enhance model fit to estimation data as well as precision of predictions. The shorter the intervening period between when the two cross-sectional data sets used in joint estimation are collected, the better the jointly estimated models are able to predict the travel choices in each of the survey years.