A MODEL OF WORK TRIP MODE CHOICE WAS DEVELOPED ON A SAMPLE OF WORKERS TAKEN BEFORE BAY AREA RAPID TRANSIT (BART) OPENED FOR SERVICE. VALIDATION TESTS OF THE MODEL WERE PERFORMED ON A SAMPLE OF WORKERS TAKEN AFTER BART SERVICE BEGAN. TWO VALIDATION METHODS WERE USED: (1) THE ACTUAL MODE SHARES IN THE POST-BART SAMPLE WERE COMPARED TO THE MODE SHARES PREDICTED BY THE MODELS ESTIMATED ON THE PRE-BART SAMPLE, AND (2) THE PARAMETERS OF MODELS ESTIMATED ON THE POST-BART SAMPLE WERE COMPARED WITH THE PARAMETERS OF THE MODELS ESTIMATED PRE-BART. THREE POSSIBLE REASONS WERE EXPLORED FOR THE DIFFERENCES IN ACTUAL AND PREDICTED SHARES AND IN THE PRE- AND POST-BART MODEL PARAMETERS: (1) FAILURE OF THE INDEPENDENCE FROM IRRELEVANT ALTERNATIVES (IIA) PROPERTY OF THE MULTINOMIAL LOGIT MODEL, (2) NON-GENERICITY IN THE ATTRIBUTES OF ONE OF THE ALTERNATIVES, AND (3) INCORRECT DATA FOR WALK TIMES. IT WAS FOUND THAT NON-GENERICITY AND INCORRECT DATA CONTRIBUTED SUBSTANTIALLY TO THE MISPREDICTIONS, WHILE FAILURE OF THE IIA PROPERTY CONTRIBUTED LESS. THE PRESENT STUDY CONCERNS ONLY ONE MODEL AND ONE TRANSPORTATION ENVIRONMENT. THE RESULTS OF THIS TEST, HOWEVER, CAN BE VIEWED ALONG WITH THE RESULTS OF OTHER VALIDATION STUDIES TO OBTAIN A SENSE OF THE PREDICTIVE ABILITY OF DISAGGREGATE MODE CHOICE MODELS.(Author/publisher).
Abstract