topic Re: PROC MIXED and "Tests of Hypotheses for Mixed Model Analysis of Variance" in Statistical Procedures

PROC MIXED and "Tests of Hypotheses for Mixed Model Analysis of Variance"

AnalytX — Wed, 02 Jul 2014 10:16:36 GMT

Hi everybody,

How can we obtain using PROC MIXED the "Tests of Hypotheses for Mixed Model Analysis of Variance" table like the one which is obtained using PROC GLM (TEST option in RANDOM statement)? We have by default only the "Type 3 Tests of Fixed Effects".

Thanking you in advance for your answers.

Best

Re: PROC MIXED and "Tests of Hypotheses for Mixed Model Analysis of Variance"

Funda_SAS — Wed, 02 Jul 2014 18:24:09 GMT

To get variance component estimates use COVTEST option of the PROC MIXED statement: SAS/STAT(R) 13.1 User's Guide

proc mixed data=yourdata covtest;

Note that COVTEST option produces the asymptotic standard errors and Wald Z-tests for the covariance parameter estimates.

Hope this woud be helpful.

Funda

Re: PROC MIXED and "Tests of Hypotheses for Mixed Model Analysis of Variance"

SteveDenham — Wed, 02 Jul 2014 18:39:56 GMT

You cannot get tests like those as GLM depends on partitioning sums of squares and expected mean squares for its tests using random effects. MIXED uses a (RE)ML method to construct quadratic forms that depend on maximizing the likelihood of the observed values, given the fixed effects of the model and conditional on the random effect estimates. 's response will help if you are interested in the covariance parameter estimates only. The Wald tests of significance for the parameters are woefully inadequate in most cases due to lack of sample size. If you have sufficient data, using PROC GLIMMIX with Gauss-Hermitian quadrature methods will improve the accuracy of the estimates and the (still) asymptotic errors.

Steve Denham

Re: PROC MIXED and "Tests of Hypotheses for Mixed Model Analysis of Variance"

AnalytX — Mon, 07 Jul 2014 07:51:44 GMT

Thanks a lot Funda and Steve for your answers!