COMPARISON OF PROBABILISTIC MODAL-CHOICE MODELS: ESTIMATION METHODS AND SYSTEM INPUTS

Abstract

TWELVE MODELS WERE FORMULATED BY SEGMENTING THE TOTAL TRAVEL TIME AND TOTAL TRAVEL COST BY RAPID TRANSIT AND BY AUTOMOBILE IN DIFFERENT WAYS OR BY LEAVING THEM OUT COMPLETELY AND INCLUDING ONLY SOCIOECONOMIC VARIABLES IN THE MODEL. THESE MODELS WERE THEN ESTIMATED BY USING LOGIT, PROBIT, AND DISCRIMINANT ANALYSES. THE RESULTS WERE EVALUATED IN 2 RESPECTS: ARE THERE DIFFERENCES IN PERFORMANCE AMONG THE METHODS OF ESTIMATION? AND ARE THERE DIFFERENCES IN PERFORMANCE AMONG THE 12 MODEL SPECIFICATIONS? THE RESULTS INDICATE THAT THERE ARE NO STATISTICALLY SIGNIFICANT DIFFERENCES EITHER AMONG THE METHODS OF ESTIMATION OR AMONG THE MODEL SPECIFICATIONS THEMSELVES. A MODEL THAT USES ONLY 2 USER CHARACTERISTICS, INCOME AND THE NUMBER OF WORKING HOUSEHOLD MEMBERS, AND 1 SYSTEM-RELATED VARIABLE, A DUMMY VARIABLE FOR WALK ACCESS TO THE TRANSIT STATION, PERFORMS NO WORSE THAN A MODEL THAT USES A WHOLE SET OF SYSTEM CHARACTERISTICS IN ADDITION TO THOSE 3 VARIABLES. VALUES OF TIME SIGNIFICANTLY LOWER THAN THOSE PREVIOUSLY REPORTED WERE FOUND; THE BEST ESTIMATE IN THIS STUDY IS ONLY 12 PERCENT OF THE WAGE RATE. /AUTHOR/