A Exact Test for a Variance Ratio in an Unbalanced Mixed Linear Model

摘要

Statistical inference for a variance ratio in mixed linear models has many useful applications in practice. In this paper, we develop a new generalized p-value approach to derive a test for a variance ratio. It is shown that the resulted generalized p-values are exact p-values in frequency sense. Numerical results suggest that the power of the generalized p-value test is generally higher than that of the Wald test for very unbalanced designs when the variance ratio is smaller.