topic Re: PROC LOGISTIC CONTRAST STATEMENTS in Statistical Procedures

PROC LOGISTIC CONTRAST STATEMENTS

tka726 — Thu, 28 Mar 2019 17:20:36 GMT

Hello!

I'm trying to write some contrast statements in proc logistic. The dependent variable is event (0,1) and the independent variable is quintile, which has 5 categories (0, 1, 2, 3, 4). I want the value of 2 to be the reference, but also want to compare other quintiles.

PROC LOGISTIC DATA=origs descending;
CLASS quint(ref="2")/param=ref;
model event= quint;

contrast '0vs2' quint 1 0 0 0 -1 /estimate=both;
contrast '1vs2' quint 0 1 0 0 -1 /estimate=both;
contrast '3vs2' quint 0 0 1 0 -1 /estimate=both;
contrast '4vs2' quint 0 0 0 1 -1 /estimate=both;

I found that the contrast statements above yield the same results as the model output (with 2 as the reference), but I can not figure out how to write contrast statements to compare, say quintile 3 vs 4. Any thoughts?

Re: PROC LOGISTIC CONTRAST STATEMENTS

StatDave — Thu, 28 Mar 2019 20:20:59 GMT

It's always best to avoid the CONTRAST and ESTIMATE statements when other statements that don't require specifying contrast coefficients can do what you want. In this case, you probably just need a single LSMEANS statement. Note the change of PARAM= to GLM.

PROC LOGISTIC DATA=origs descending;
CLASS quint(ref="2")/param=glm;
model event= quint;

lsmeans quint / diff oddsratio;

run;

Re: PROC LOGISTIC CONTRAST STATEMENTS

tka726 — Thu, 28 Mar 2019 21:31:39 GMT

This is perfect, thank you!