Decomposition of Explained Variation in the Linear Mixed Model

TL;DR

This paper introduces a decomposition method for explained variation in linear mixed models, enabling clearer interpretation and comparison of covariate effects, with applications in diverse fields like genomics and social sciences.

Contribution

It provides a proper decomposition of explained variation in LMMs, extending the adjusted R-squared concept and allowing covariate relevance ranking within a single model.

Findings

Decomposition of explained variation into unbiased estimators.

Extension of adjusted R-squared to LMMs.

Application to genomics and social science data.

Abstract

In the linear mixed model (LMM), the simultaneous assessment and comparison of dispersion relevance of explanatory variables associated with fixed and random effects remains an important open practical problem. Based on the restricted maximum likelihood equations in the variance components form of the LMM, we prove a proper decomposition of the sum of squares of the dependent variable into unbiased estimators of interpretable estimands of explained variation. This result leads to a natural extension of the well-known adjusted coefficient of determination to the LMM. Further, we allocate the novel unbiased estimators of explained variation to specific contributions of covariates associated with fixed and random effects within a single model fit. These parameter-wise explained variations constitute easily interpretable quantities, assessing dispersion relevance of covariates associated with both fixed and random effects on a common scale, thus allowing for a covariate ranking. For illustration, we contrast the variation explained by subjects and time in the longitudinal sleep deprivation study. By comparing the dispersion relevance of population characteristics and spatial levels, we determine literacy as a major driver of income inequality in Burkina Faso. Finally, we develop a novel relevance plot to visualize the dispersion relevance of high-dimensional genomic markers in Arabidopsis thaliana.