Effects of misspecification of the link functions in beta regression model with varying dispersion

The beta regression model with varying dispersion is used to model continuous data in the interval (0; 1), assuming beta distribution for the variable of interest. In this model, regression structures are considered for the mean and dispersion parameters involving covariates, unknown parameters and link functions. This paper addresses the problem of the misspecification in the dispersion sub model of beta regression. Coverage rates and balancing of the regression parameters in the mean sub model were evaluated by Monte Carlo simulations, considering different link functions. Coverage rate and maximum and minimum average balance of the predicted mean and dispersion values were also assessed. The misspecification of the link function of the dispersion sub model influences on the inferences of the model. An application to real data is also presented and discussed.