UCLA Academic Technology Services HomeServicesClassesContactJobs
Search

SAS Library
A Brief Comparison of PROC GLM and 
PROC MIXED of SAS


This page was adapted from a page created by Professor Oliver Schabenberger.  We thank Professor Schabenberger for permission to adapt and distribute this page via our web site.


Selected References

Criteria

PROC GLM

PROC MIXED

Nutshell

Designed for fixed effects model with allowance for certain adjustments in the presence of random terms. For many designs special attention needs to be given to least square means and contrasts since their standard errors are not necessarily correct. This is true, for example, for split-plot designs. Designed for mixed effects models. Random terms are incorporated into inference from the outset. Contrasts, least square means and estimates of linear combinations are reported with correct standard errors. Attention needs to be paid to the selection of an appropriate inference space.

Danger

If analysis is driven by accounting for degrees of freedoms and tests, p-values, contrasts, least square means etc. are taken for granted. Depending on the design, some results reported by PROC GLM are actually incorrect. In order to access the powerful inferential features of PROC MIXED, effects are treated as random when they are actually fixed. Prime example: block effects.

Estimation of fixed effects

Based on Ordinary Least Squares Based on Generalized Least Squares in a Gaussian error model (estimates are Maximum Likelihood Estimates under normality).

Estimation of variance components

Method-of-Moments estimation (ANOVA method) of solving expected mean squares for the variance components Maximum Likelihood or Restricted Maximum Likelihood estimation.

MODEL statement

All effects are listed, whether fixed or random Only the fixed effects are listed. Random effects are listed in the RANDOM statement.

LSMEANS statement

Interprets all effects as fixed, even the random effects. While least square means are correct, their standard errors are not necessarily correct. This is true, even if a RANDOM statement is used. Only effects in the MODEL statement are assumed fixed. The standard error estimates for least square means account for the random effects.

ESTIMATE statement

Same as LSMEANS statements Same as LSMEANS statements

Sums of Squares

They are everywhere Since it is not based on expected mean squares, there are no sums of squares in the standard output.

RANDOM statement

Invokes the calculation of expected mean squares for the listed effects and appropriate test using the /TEST option. The randomness of the effect is not incorporated into the tests of main effects, lsmeans, contrasts, etc.  Signals incorporation of the listed effects in all aspects of inference. Options allow selection of various correlation models describing the dependencies of multiple random effects.

TEST statement

Important statement to utilize correct error terms in testing model effects. Prime examples: Subsampling designs, split-plot designs Gone.

 


How to cite this page

Report an error on this page

UCLA Researchers are invited to our Statistical Consulting Services
We recommend others to our list of Other Resources for Statistical Computing Help
These pages are Copyrighted (c) by UCLA Academic Technology Services


The content of this web site should not be construed as an endorsement of any particular web site, book, or software product by the University of California