University of California Transportation Center

With modern multivariate statistical methods and activity-diary (time-use) data sets, it is possible to model household mobility decisions as being derived from decisions to participate in activities at various locations. We show how this can be accomplished by specifying activity participation by activity type and location as endogenous variables, with a simple locational distinction of “at home” versus “out of home.” The activity participation variables are then combined in a model system of simultaneous equations with variables that measure mobility demand: travel times by mode, household vehicle ownership and household vehicle utilization. We specify the model in terms of latent, multivariate normally distributed choice variables, and this treatment solves estimation problems associated with censored and ordinal observed endogenous variables. The estimation method provides accurate goodness-of-fit model evaluation and hypothesis testing. Results are shown from a model estimated using two-day activity diary data for male and female household heads and associated accessibility data collected in the Portland, Oregon, U.S.A. Metropolitan Area in 1994. The model system can be used in conjunction with conventional travel demand models, to provide forecasts of the effects of factors such as accessibility and in-home work, on travel demand by mode, car ownership, and car vehicle miles of travel. This type of model system has the potential of replacing some existing demand forecasting models.