Bootstrap-based testing inference in beta regressions (Q2180254)

Consider hypothesis testing for parameter values in a beta regression model. Using the likelihood ratio test, the test statistic is known to have a chi-squared distribution asymptotically. With small sample sizes, however, this approximation can be inaccurate. The authors consider two bootstrap procedures to overcome this problem. The first of these techniques uses bootstrap resampling to estimate the Bartlett correction factor for the test statistic. The second uses two levels of nested resampling. A simulation study and an application to real data illustrate these procedures compared to both the usual bootstrap and employing the likelihood ratio test without any bootstrapping.