Application of the logistic regression models for transportation data

Abstract

The multinomial model was fitted to the stated preference data of the transportation problem by commuters in Mamelodi, east of Pretoria. The data analyzed in the study was collected in 2001 among 151 less literate (highest level of education up to Standard 5 or Grade 7) and 194 literate (highest level of education from standard 6 or Grade 8 to Standard 10 or Grade 12) commuters in the CBD (Central Business District) of Pretoria. Seventeen (17) variables have been analyzed. The objective of the study is to determine if there are differences when three types of codings (dichotomous, binary and effect) are applied to the same data. The final interest is to determine those factors that affect commuters in choosing their mode of transport to work in the CBD of Pretoria. All the logistic regression models and multinomial logit models tested in the study were found to be statistically significant for the three different codings. Due the limitations that SAS has, the logistic regression models were fitted and used to carryout the analyses. When variables were selected by the stepwise procedure, and only those explanatory variables that were significant fitted in the model, the three models were all statistically significant for the Hosmer-Lemeshow goodness-of-fit statistics, but not for the Pearson and Deviance goodness-of-fit statistics.