This study presents an alternative method for estimating gravity models by multiple linear regression that is based on proxy variables, thus circumventing the endogeneity problems arising when least-squares estimators areused. The proxy variable approach generates consistent estimators for a gravity model without endogeneity bias. The presence of endogeneity is tested for using statistical tests developed specifically for our application.Proxy variables eliminate the endogeneity and produce consistent estimators in gravity models estimated using least squares. We also find, however,that endogeneity bias has no significant impact either on gravity model prediction or on urban transportation system planning processes based on such models. (A) Reprinted with permission from Elsevier.
Abstract