COVARIATE SCREENING IN MIXED LINEAR-MODELS

We address the important practical problem of selecting covariates in mixed linear models when the covariance structure is known from the da ta collection process and there are a possibly large number of covaria tes available. In particular, we consider procedures which can be cons idered extensions of the analysis of deviance to mixed linear models. This approach provides an alternative to likelihood ratio test methodo logy which can be applied in the case that the components of variance are estimated by restricted maximum likelihood (REML), thus resolving the open question of how to proceed in this context. Moreover, it is s imple to robustify and allows us to consider a wider class of procedur es than those which fit into the simple likelihood ratio test framewor k. The key insights are that the deviance should be specified by the p rocedure used to estimate the fixed effects and that the estimated cov ariance matrix should be field lived across different models for the f ixed effects. (C) 1996 Academic Press, Inc.