Estimating Household Trip Rates for Cross-Classification Cells with No Data: Alternative Methods and Their Performance in Prediction of Travel

Abstract

This paper investigates a number of alternative methods for addressing the empty-cell problem of traditional cross-classification analysis. Data used in the study were collected in the Toronto region in 1986, 1996, 2001, and 2006. Alternative models, developed on each year’s data, were assessed for how well they predicted travel at the disaggregate household level and at the aggregate traffic analysis zone level in the respective years. In addition, the alternative models estimated on the 1986 data set were assessed for their ability to replicate travel in 1996 and 2006. The results show that a method proposed by Mandel and a model developed in this research, which estimates the household trip rate for an empty cell through a linear combination of the predictions yielded by row and column models, overall give the best forecast performance of travel. They perform better than multiple classification analysis, which is the current industry standard for addressing this shortcoming of traditional cross-classification analysis. The combined categories model also performed very well, particularly in predicting travel at the aggregate level of planning interest.