topic Proc logistic, contrast , one level vs average of all levels in Statistical Procedures

Proc logistic, contrast , one level vs average of all levels

Ivy — Wed, 24 May 2023 15:31:38 GMT

Hello,

I am trying to write the contrast statement for race in reference coding schema:

black vs the average effect of the three levels (black vs (black + other + white)/3)

contrast 'race black vs (white black other)/3' race 2 -1 / estimate=all ;

The result seems not right.

May you please help with the statement?

Thanks!

Re: Proc logistic, contrast , one level vs average of all levels

sbxkoenk — Wed, 24 May 2023 17:30:03 GMT

Hello,

I have moved your post to the "Statistical Procedures" board.

Why don't you use effect coding (aka deviation-from-the-mean coding) instead of reference-cell coding?
This deviation effect coding system compares the mean of the dependent variable for a given level to the grand mean of the dependent variable.

Otherwise, see here:

Usage Note 24447: Examples of writing CONTRAST and ESTIMATE statements
https://support.sas.com/kb/24/447.html

Cheers,

Koen

Re: Proc logistic, contrast , one level vs average of all levels

SteveDenham — Fri, 26 May 2023 14:44:19 GMT

If you shift to effect coding (as in PROC GLM), you can use the following LSMEAN statement to compare each level to the average of all levels.

lsmeans race/diff=ANOM;

The p values associated with this option are for the difference on the logit scale. If you want to get the difference on the original scale, you will likely need to use the %NLEst and %NLMeans macros, which will draw on @sbxkoenk comments about ESTIMATE statements.

SteveDenham