TRANSFERABILITY ANALYSIS OF DISAGGREGATE CHOICE MODELS

Abstract

The transferability of disaggregate choice models is widely assumed in travel demand analysis and forecasting. However, research results are mixed in their assessment of transferability. This paper considers transferability from the perspective of the usefulness of information provided by a model that predicts in a context different from that in which it is estimated. Issues that influence transferability are discussed and methods are formulated to evaluate transferability. These methods are demonstrated by application to spatial transfer between pairs of geographic sectors within a single urban area by using identically specified models. This application provides useful insights into the transferability of choice models. We observe an inconsistency between general measures of error that indicate that transferability in this context is appropriate and statistical analyses that reject hypotheses that support transferability. Model transferability is a property of the estimation and application contexts as well as the specification of the model. Transferability is substantially improved by adjustment of alternative-specific constants. These results indicate the need for additional research to identify the conditions of model specification and context characteristics for which model transfer is effective. Directions for such research are identified. (Author)