UC Irvine Institute of Transportation Studies

Estimation of generalized extreme value (GEV) models of discrete choice is hampered by computational complexity and convergence problems. However, the much simpler estimation routine for multinonial logit can be applied in a two-step procedure so as to test the null hypothesis of multinomial logit against any particular GEV model as an alternative hypothesis. The procedure also produces an approximate estimate of the GEV model. Monte Carlo data, generated alternatively by logit and by three different GEV models, provide evidence that both the test statistics and the approximate estimator have small-sample properties superior in important respects to maximum-likelihood estimation of the GEV model.