An Assessment of the Calibration of Spatial Interaction Models

Abstract

<p>This paper is concerned with assessing the procedures used in calibrating spatial interaction models. It critically reviews calibration methodologies which have been proposed in the literature and determines that the statistical estimation techniques of maximum likelihood and least-squares are particularly suited to this estimation problem.</p> <p>The calibration statistics from the maximum likelihood and least-squares estimators are developed from first principles and special note is made of the behavioral assumption implicit in each.</p> <p>Two issues are then reviewed: the reliability of the random sample in representing the mean distribution of trips, and the definition of variable in calibration statistics. A hypothetical framework is proposed, within which an examination of these issues is made.</p> <p>The study results indicate that the sample reasonably represents the mean distribution and also that the incorporation of implicit behavioral assumptions does not necessarily result in better model predictions.</p>