Hi,
I have a question regarding Generalized Linear Mixed Models (GLMMs) in JMP pro. Most statisticians consider it inappropriate to use F-tests or t-tests to evaluate predictor contributions in GLMs or GLMMs with binomial errors. This is because F- and t-tests are based on sums of squares, which assume normally distributed residual errors.
While using a logit link function helps linearize the relationship between predictors and a binomial response variable, it does not normalize the residuals. Binomial mixed-effects models are typically optimized by minimizing the negative log-likelihood rather than the sums of squares, making likelihood ratio tests generally more appropriate than F-tests.
Given this, why does JMP allow the calculation of F-tests for GLMMs with a binomial distribution? Does JMP provide an option to perform likelihood ratio tests instead of F-tests in such models? Additionally, how can be verified the reliability of outputs that report F-tests?
I appreciate your insights on this matter.
Best regards,
Barbara