Testing a subset of random effects in linear mixed models A comparative study

Abstract In linear mixed-effects models, random effects are necessary, especially if the data has two sources of variation. As a result of the importance of random effects, various studies focused on testing the need for random effects. This topic is challenging, and several studies have been proposed for testing a subset of random effects using various approaches to tackling challenges with this problem. Existing tests involve the exact F-test, and other resampling-based tests. In this article, we conduct simulation studies to examine the size and the power of many of the recent tests for a subset of variance components in linear mixed-effects models. The study highlights the tests with the highest empirical power under some commonly used configurations.