Author(s)

Yanyan Wu¹, Augustine Wong², Georges Monette², Laurent Briollais³

¹Department of Public Health Sciences, University of Hawaii at Manoa, Honolulu, HI, USA.
²Department of Mathematics and Statistics, York University, Toronto, ON, Canada.
³Lunenfeld-Taunenbaum Research Institute, Mount Sinai Hospital, Toronto, ON, Canada.

Mixed models provide a wide range of applications including hierarchical modeling and longitudinal studies. The tests of variance component in mixed models have long been a methodological challenge because of its boundary conditions. It is well documented in literature that the traditional first-order methods: likelihood ratio statistic, Wald statistic and score statistic, provide an excessively conservative approximation to the null distribution. However, the magnitude of the conservativeness has not been thoroughly explored. In this paper, we propose a likelihood-based third-order method to the mixed models for testing the null hypothesis of zero and non-zero variance component. The proposed method dramatically improved the accuracy of the tests. Extensive simulations were carried out to demonstrate the accuracy of the proposed method in comparison with the standard first-order methods. The results show the conservativeness of the first order methods and the accuracy of the proposed method in approximating the p-values and confidence intervals even when the sample size is small.

Family Data, Genetic Variant, Likelihood Ratio Test, Random Effects, Third-Order Method, Variance Component

Wu, Y. , Wong, A. , Monette, G. and Briollais, L. (2015) Evaluation of Third-Order Method for the Tests of Variance Component in Linear Mixed Models. Open Journal of Statistics, 5, 233-244. doi: 10.4236/ojs.2015.54025.