Making Advanced Travel Forecasting Models Affordable Through Model Transferability

Abstract

The research objective is to empirically test the transferability of activity-based (AB) models between regions. Using 2008-2009 National Household Travel Survey data, AB models are estimated simultaneously for six regions, in California and Florida. Statistical tests are applied to identify regional differences in the models. Tests reject the hypothesis that models based on data from different regions are identical, but most individual coefficients are not significantly different from region to region. There is evidence of model similarity among regions, but small sample sizes prevent strong conclusions. Estimability improves with larger survey samples; 6,000 households provide much better information for estimating coefficients than samples of size 2,500 or less. Thus, although estimation of models using a large local sample is best, it is better to transfer models built using a large sample from a similar region than to estimate new models using a much smaller local sample. However, local survey data are needed for calibrating components of the model. Four specific transferability hypotheses are tested, with results as follows: (1) Supported: Variables that apply to population segments defined by characteristics of individuals or their situational context (segment-specific variables) are more transferable than variables that apply to all individuals. (2) Not supported: Segment-specific variables are more transferable than alternative-specific constants. (3) Strongly supported: Models that deal with activity generation and scheduling are more transferable than models that deal with mode choice and location choice. (4) Inconclusive: Models for different regions within the same state are more transferable than models from different states.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 20 citations exist upstream.