Overview: VARCOMP Procedure

The VARCOMP procedure handles general linear models that have random effects. Random effects are classification effects with levels that are assumed to be randomly selected from an infinite population of possible levels. PROC VARCOMP estimates the contribution of each of the random effects to the variance of the dependent variable.

A single MODEL statement specifies the dependent variables and the effects: main effects, interactions, and nested effects. The effects must be composed of classification variables; no continuous variables are allowed on the right side of the equal sign.

You can specify certain effects as fixed (nonrandom) by putting them first in the MODEL statement and indicating the number of fixed effects with the FIXED= option. An intercept is always fitted and assumed fixed. Except for the effects specified as fixed, all other effects are assumed to be random. Their contribution to the model can be thought of as an observation from a distribution that is normally and independently distributed.

The dependent variables are grouped based on the similarity of their missing values. Each group of dependent variables is then analyzed separately. The columns of the design matrix are formed in the same order in which the effects are specified in the MODEL statement. A singular parameterization involving just 0–1 dummy variables is used, as in the GLM procedure.

You can specify four general methods of estimation in the PROC VARCOMP statement by using the METHOD= option. They are TYPE1 (based on computation of Type I sum of squares for each effect), MIVQUE0, maximum likelihood (METHOD=ML), and restricted maximum likelihood (METHOD=REML). A fifth method, METHOD=GRR, provides a specialized analysis for gauge repeatability and reproducibility (R&R) studies. See the section Gauge Repeatability and Reproducibility Analysis for further details. Note that this method, along with the CL option in the MODEL statement for confidence limits, applies only to certain designs, namely balanced one-way or two-way designs. The other four general methods apply to any random-effects model and design.

Other procedures, such as PROC GLM, PROC MIXED, and PROC GLIMMIX, fit similar random effects models. The VARCOMP procedure is usually more computationally efficient for certain special designs and models. See the section Relationship to PROC MIXED for a more precise comparison with the MIXED procedure in particular.

The GAUGE application in SAS/QC software provides a graphical interface for computing many of the same statistics as METHOD=GRR in PROC VARCOMP.