VARCOMP

VARCOMP is a term used in statistics to denote the variance components of a statistical model. In linear mixed models and related designs, the total observed variability of a response is partitioned into several sources corresponding to random effects (such as genetic, environmental, or experimental factors) and residual error. Estimating these variance components helps researchers understand the relative contributions to phenotypic variation and to derive measures such as heritability.

Estimation is typically performed under maximum likelihood (ML) or restricted maximum likelihood (REML). REML is often preferred because it reduces bias in the estimated variance components, especially with unbalanced data. The accuracy of estimates depends on model specification, the number and structure of random effects, and distributional assumptions (usually normality).

In SAS, PROC VARCOMP is a dedicated procedure for estimating components of variance in linear models with one or more random effects. It accepts ML or REML estimation options, supports balanced and unbalanced designs, and provides estimates of variance components, their standard errors, and tests for the significance of random effects. It can also output derived quantities such as heritability in genetic studies.

Applications of VARCOMP include animal and plant breeding, genetics, psychology, and agronomy, where partitioning variance informs selection decisions, experimental design, and interpretation of environmental versus genetic influences. Limitations include sensitivity to model misspecification and non-normality, potential negative variance estimates, and convergence issues in complex models. Researchers often compare models using information criteria or likelihood ratio tests to select an appropriate variance component structure.