topic Re: Exact logistic regression in Statistical Procedures

Exact logistic regression

_Rose_ — Fri, 27 Oct 2017 08:03:33 GMT

Hi all,

I have a problem in interpreting the result of my exact logistic regression.

Here is my sas program:

PROC LOGISTIC EXACTONLY DATA=COMP_SER1 DESCENDING;
CLASS CL_inj(ref="1")/PARAM=ref;
MODEL SERO_HCV= CL_inj ;
EXACT CL_inj/joint estimate=both;
RUN;

Find attached my output

In the part: "Exact Conditional tests" I can see that my variable: Cinj that the effect of variable Cinj is significant with all the test.

But when I look at the part: "Exact parameter estimates" I can see that all the parameter of the variable are not significant.

How can I interprete this?am I wrong with the interpretation of this two part?

How do you interprete this two part?

Thanks for your help!!

Rose

Re: Exact logistic regression

Kurt_Bremser — Fri, 27 Oct 2017 08:39:05 GMT

Proc logistic is part of SAS/STAT, so I moved your question here.

Re: Exact logistic regression

_Rose_ — Fri, 27 Oct 2017 09:04:48 GMT

Thanks!!!

Re: Exact logistic regression

Rick_SAS — Fri, 27 Oct 2017 15:12:54 GMT

The parameter estimates show the difference in the effect of each nonreference level compared to the effect of the reference level. So the effect is important in the model, but the levels '2' and '3' are not significantly different than the reference level '1'.

Re: Exact logistic regression

StatDave — Sat, 28 Oct 2017 16:53:32 GMT

Following up on Rick's comment... If you just change the reference level of your predictor, then one of the parameters will compare levels 2 and 3 and probably will be significant:

CLASS CL_inj(ref="3")/PARAM=ref;

Re: Exact logistic regression

_Rose_ — Thu, 02 Nov 2017 15:36:48 GMT

Thanks a lot Rick!

Re: Exact logistic regression

_Rose_ — Thu, 02 Nov 2017 15:37:28 GMT

That's right!

Thanks