Resumen de Information matrix tests for multinomial logit models
Dante Amengual, Gabriele Fiorentini, Enrique Sentana Iváñez
We show that the influence functions of the information matrix test for the multinomial logit model are the Kronecker product of the outer product of the generalised residuals minus their covariance matrix conditional on the explanatory variables times the outer product of those variables. Thus, it resembles a multivariate heteroskedasticity test à la White (1980), which confirms Chesher’s (1984) unobserved heterogeneity interpretation. Our simulation experiments indicate that using theoretical expressions for the conditional covariance matrices involved substantially reduces size distortions, while the parametric bootstrap practically eliminates them. We also show that the test has good power against several relevant alternatives.
