Empirical performance of estimation methods in Beta mixed models with application to ecological data

Abstract

This study uses a Monte Carlo simulation design to assess the performance of Beta and linear mixed models on bounded response  variables through comparison of four estimation methods. Four factors affecting the performance of the estimation methods were considered: the number of groups, the number of observations per group, the variance and distribution of the random effects. Our results showed that, for small number of groups (less than 30), the Beta mixed model outperformed the linear mixed model whatever the size of the groups. In the case of a large number of groups (superior or equal to 30), both approaches showed relatively close performance. The results from the simulation study have been illustrated with real life data.

Key words: Beta distribution; continuous proportion; transformations; hierarchical modelling; performance; application.