TEST FOR LINK MISSPECIFICATION IN GENERALIZED LINEAR MODELS FOR BINARY DATA

This paper addresses the issue of check the correct specification of the link function in generalized linear models for binary data. To perform the RESET test we consider the likelihood ratio, Wald and score traditional statistics and we propose the use of the emerging gradient statistic. The performance evaluation of misspecification tests were performed using Monte Carlo simulations. The finite sample performance of the tests were evaluated in terms of size and power tests. It can be seen that the performance of the tests are influenced by the used link function and the sample size. The gradient statistic outperforms the traditional statistics, especially in smaller sample sizes. An empirical application to a real data set is considered for illustrative purposes.