This paper reports work done for an FHWA project to empirically test and demonstrate the transferability of activity-based (AB) models between regions. Transferability would allow regions to borrow and adapt activity-based models from other regions, avoiding the cost of a large household survey and estimating entirely new models.
Six regions in two states were included in the tests: Sacramento, northern San Joaquin Valley, Fresno and San Diego in California, and Jacksonville and Tampa in Florida. These regions participated in the 2008-2009 National Household Travel Survey “add-on” program, so that reasonably large samples of highly comparable survey data were available. The spatial and network data were region-specific, but substantial efforts were made to derive and use attributes that were as comparable as possible.
The data were combined and used to estimate AB models simultaneously for the six regions. A common base specification was used, and region-specific and state-specific parameters were estimated to identify regional and state differences in the model coefficients.
This is the first time that an estimation-based approach has been used to test transferability of an AB model, and it has advantages over an approach in which a model estimated for one region is applied in another region and checked to see how well it predicts. The approach allows explicit statistical tests of differences in parameters, can address a variety of hypotheses regarding transferability, and can test the transferability of coefficients for specific types of variables, rather than only test the transferability of the model as a whole.
The presentation and paper will:
• Describe the methods used to develop and combine data sets from the six regions.
• Discuss the strengths and weaknesses of NHTS data for estimating AB models and conducting transferability tests.
• Discuss issues related to comparability of the spatial and network impedance data sets.
• Present the basic hypotheses that were tested and the models for which they were tested.
• Describe the methods used to conduct the model estimation and hypothesis tests.
• Discuss issues encountered in conducting the tests.
• Present key results of the transferability tests.