Overview
This study is an extension to the SHRP 2 C10A project, Partnership to Develop an Integrated, Advanced Travel Demand Model and a Fine-Grained Time-Sensitive Network: Jacksonville-Area Application. The extension was used to develop regional activity-based modeling systems for the Tampa Bay and Jacksonville regions in Florida and to test the concept of transferability. Transferability was an important finding from the original C10A project. If transferability of parameters could be demonstrated to produce reasonable results, it could save metropolitan planning organizations (MPOs) millions of dollars in data collection and model estimation costs and make activity-based models practical for a wider market. This study used a travel demand model specification borrowed from the Sacramento Area Council of Governments and the DaySim activity-based modeling platform. To test the transferability of the modeling system and specifications, the Sacramento parameters were applied directly in each study region, then the models were calibrated using Tampa- and Jacksonville-specific (hereafter “local”) target values and distributions, which were derived from the National Household Travel Survey (NHTS) add-on samples for each region. In addition, the same data sources were used to estimate new local parameters for each region’s model system to look for significant differences in regional travel behavior, again following the variables specified in the original Sacramento implementation.
The model estimation tests were hampered by small sample sizes for many of the model components. Nevertheless, the study team was able to identify statistically significant differences in enough model components to begin to characterize travel patterns in the Tampa region as being dominated by different lifestyle considerations. Looking at pairings of regional models in which the same parameter was significant in both regions, there were proportionally far more differences in the Tampa-Sacramento pair than either Jacksonville-Sacramento or Tampa-Jacksonville. These differences pointed to the influence of the Tampa region’s large population of retirees as evidenced by significant effects of retiree-household and single-driver-household variables, single-auto households, and a reduced consideration of the presence of children on escort tour destination choices. In addition, the models estimated for the Tampa region had significantly higher propensities toward leisure tours and lower propensities toward work tours and shared rides involving more than two persons. The study team found that the NHTS sample size was insufficient to reestimate many of the model components found in the original Sacramento specification.
The study team concluded that for purposes of delivering production-ready versions of model systems to both regions, it would be better to start with the Sacramento specification (which was at least a holistic description of variation in regional travel behavior across a representative population) rather than to piece together versions of models that were a partial blend of estimated parameters from multiple regions. Specifically, the study team found that the NHTS samples lacked adequate representation of certain submarkets, such as young children, and underreported evening and non-work travel and non-auto modes.
In practice, model transfer typically does not involve reestimation of a complete model system. Sometimes, though, parts are reestimated, such as mode and destination choice, subject as they are to differences in characteristics of highway and transit networks as well as urban spatial structure. This study found sufficient evidence to recommend that when transferring a model, consideration should be given to reestimation of singly constrained destination choice models, data permitting. In addition, models that are reasonably consistent between regions, such as time-of-day choice models, may be good candidates for transfer and calibration rather than reestimation if the data available for estimation lack sufficient variation in observed multidimensional choices.
Using the Sacramento specification as a base, the Jacksonville and Tampa model systems were calibrated to benchmark values for the distribution of model outcomes using the NHTS data, which were more robust at the aggregate level. Those data were supplemented by other sources such as the Census American Community Survey (ACS), the Census Transportation Planning Package (CTPP), regional traffic counts, and transit boarding counts. The order in which individual model components were calibrated followed the order in which they are applied hierarchically in the DaySim model stream. Household auto ownership models and models related to predicting daily activity patterns and tour frequencies required more calibration effort than downstream models. This extra effort resulted in calibrating fewer parameters at the end of the model stream— namely, trip-level mode and destination and time-of-day choice models—which were conditioned by and benefited from the upstream tour-level calibrations. Recognizing the importance of the retiree market segment, the study team added retiree-bias constants in the daily-pattern choice model during calibration of the Tampa regional model, parameters that were not transferred over from the Sacramento specification because they did not exist in it. Similarly, other household and person segmentation variables were added to some of the Jacksonville models during calibration.
An important outcome of this study was the decision to use the richer, more complete variable and parameter specification from Sacramento rather than a local specification of variables and parameters; the latter is less robust in terms of explaining variation in the population due to the limitations of the estimation sample data. An important lesson learned in this study is that when reestimating an activity-based model specification to evaluate transferability of parameters, it is necessary to start with a model specification that is sufficiently parsimonious to allow estimation using both the regions’ household interview survey data and other supporting data. In other words, the complexity of the model system should be supported by the available data in the transfer-recipient’s region.