The idea of model transferability is to use previously estimatedmodel parameters from a different area for model estimation. The combined transfer estimator is based on the mean squares error criterion and extends the bayesian procedure to explicitly account for the presence of a transfer bias. The suggested estimator is easy to apply because it is expressed as a linear combination of the direct estimation results and the previously estimated parameters. The combinedestimator is shown to have superior accuracy in a mean square errorsense to a direct (unbiased nontransfer) estimator whenever the transfer bias is relatively small. Numerical examples of the transfer region--where the combined estimator is superior to the direct estimator--are provided. This paper appeared in transportation research record no. 1139, Urban travel forecasting. For covering abstract see IRRD no 817822.
Abstract