The paper describes the recent experience with using multiple Household Travel Surveys (HTSs) for a coordinated development of an advanced Activity-Based travel Model (ABM) for four major metropolitan regions of Ohio (Columbus, Cincinnati, Cleveland, and Dayton). The same advanced ABM design and software were applied for each region. However, the structural details of each sub-model and coefficient were subject to transferability analysis. The data consisted of four different HTSs conducted for Columbus (5,555 households in 1999), Dayton (1,950 households in 2001), Cincinnati (2,050 households in 2009-10) and Cleveland (4,540 households in 2012-13) regions.
The data from the four HTSs were consolidated in a pooled dataset for the ABM estimation and validation by combining all HTSs and other complementary data sources (land-use data, transportation level-of-service, etc). This ensured a rich behavioral basis for statistical analysis and estimation of an advanced ABM, which is difficult to obtain for a single region.
The main principle was to develop all behavioral sub-models of the ABM in the most generic way such that they would be fully transferable between the regions. If the first estimation effort indicated significant statistical differences between the regions, the model specifications were revised to try to explain the differences by additional region-specific variables rather than to simply accept differences in coefficients between the regions. In the end, most of the sub-models proved to be generic and transferable. However, some sub-models proved to be not transferable, thus a region-specific segmentation of coefficients (full or partial) was applied as well as some region-specific modifications in the sub-model specification. The destination choice model was the most appealing example where regional specifics required some modifications in the model structure.
The paper discusses the estimation results for the most important sub-models such as the work location choice model, daily activity pattern, time of day choice model, tour mode choice, etc with an emphasis on transferability between regions. It provides evidence and insights on what sub-models can be transferred vs. what sub-models should be re-estimated for each region, and what sub-models may even require a region-specific specification. The overall conclusion of this project is that a coordinated development of an ABM of the same design for several regions and based on the pooled dataset is a worthwhile effort compared to an independent ABM development for each region.