Forecast Performance of Metropolitan Trip Generation Models Statistically Updated with US National Household Travel Survey Data

An investigation is conducted into whether trip generation models, estimated on travel survey data collected in the 1998 Nashville household travel behavior survey, can be enhanced in forecast performance by combining the 1998 Nashville survey data with data collected in metropolitan regions in either the 2009 National Household Travel Survey (NHTS) or in the 2009 State of Tennessee Add-On NHTS. Data for assessing the forecast performance of the alternative developed models were collected in the 2012 Nashville transportation and health survey. Linear regression analysis and Bayesian statistics were used to develop alternative trip generation models whose performances in forecasting travel were assessed. Major findings of the study include the following: (a) metropolitan trip generation models for the mandatory purposes of work and school did not benefit from update with national or state-level data for these trip purposes; (b) metropolitan trip generation models for nonwork nonschool purposes (discretionary purposes) and for trips across all travel purposes improved in forecast performance with update with national-level data collected in the 2009 NHTS only. This finding suggests that metro-level data collected in the regular NHTS rather than data collected in a specific US state's Add-on Survey should be the data source for updating metropolitan models of discretionary travel in medium-range forecasting; and (c) development of updated models by Bayesian updating, though relatively simple, had the best forecast performance of the tested models and hence is recommended as the preferred method for model updating in planning practice in the state.