Gravity Parameter Estimation

Abstract

The question of whether ordinary least‐squares (OLS) regression provides unbiased estimates for parameters of the gravity model is investigated. Holding attractiveness constant and looking only at the location of destinations, OLS is shown to give unbiased estimates if underlying behavior conforms to the gravity hypothesis. Johnston's conjectures on this are examined and found to explain only partially observed variations in distance exponents. Allowing attractiveness to vary, spatial autocorrelation in attractiveness is shown to lead to bias in OLS estimation due to misspecification, even where behavior conforms to the gravity hypothesis. Deterministic and stochastic illustrations are provided. 1979 The Ohio State University