A new method using medians to calibrate single-parameter spatial interaction models

Abstract

I present a method for calibrating the impedance parameter of a gravity spatial interaction model using only the median travel time as a measure of observed traveler behavior. Complete information about the spatial structure of origins, destinations, and travel times between origins and destinations is also required. Using Monte Carlo simulation techniques on stylized cities, I attempt to recover true (a priori known) impedance values with this method for a range of impedance values for both negative exponential and power impedance functions. The results are compared with estimates obtained by other fast methods. The proposed method proves to provide a fairly accurate estimate of the impedance parameter, with a mean percent error typically below 20% and often below 10% for common impedance values. The proposed method is an improvement over existing calibration methods in several respects. First, it allows for the estimation of the impedance parameter directly without lengthy iterative calculations. Second, because it only requires median travel times, it can be calibrated with smaller samples (n~200), allowing the construction of gravity models for specific modes and/or travel purposes. And third, the method does not require expensive travel demand software and so can be implemented more widely in practice.