topic Re: Exact Logistic Regression after Inverse Probability Weighting in SAS Programming

Exact Logistic Regression after Inverse Probability Weighting

JonKetchup — Thu, 06 Oct 2022 12:49:20 GMT

I am trying to run several exact logistic regressions after using inverse probability weighting; however, it does not seem like I am able to run an EXACT statement and a WEIGHT statement simultaneously. Is there a work around? I am using exact logistic regression instead of logistic regression due to sample size concerns.

Here is my code:

proc psmatch data=DATA region=allobs;
	class TREATMENT VAR1;
	psmodel TREATMENT(Treated='1')= VAR1 VAR2 VAR3 VAR4 VAR5;
	ASSESS LPS VAR=(VAR2 VAR3 VAR4 VAR5)/VARINFO PLOTS=(BOXPLOT BARCHART)
		WEIGHT=ATEWGT;
output out(obs=ALL)=PSDATA lps=Lps5 ATEWGT=ATEWGT5 PS=PS5;
run;

PROC LOGISTIC DATA=PSDATA;
CLASS VAR6 TREATMENT(REF='0');
MODEL VAR6=TREATMENT;
EXACT TREATMENT;
WEIGHT ATEWGT2;
RUN;

The log warning I get is: "WARNING: The EXACT statement is ignored when the WEIGHT statement is specified."

Appreciate your help.

Re: Exact Logistic Regression after Inverse Probability Weighting

ballardw — Thu, 06 Oct 2022 14:45:39 GMT

From the documentation for Proc Logistic and the Exact statement:

Exact analyses are not performed when you specify a WEIGHT statement, a link other than LINK=LOGIT or LINK=GLOGIT, an offset variable, the NOFIT option, or a model selection method. Exact estimation is not available for ordinal response models.

What exactly are your sample size concerns?

Re: Exact Logistic Regression after Inverse Probability Weighting

JonKetchup — Thu, 06 Oct 2022 14:52:50 GMT

If I were to make contingency tables for treatment and each outcome, several cell frequencies and expected cell frequencies would be small (some are less than 1). I would normally use Fisher's exact test, but with non-integer weights from the IPW, that is not an option, hence my pursuit to use exact logistic regression. I know this is possible in StatXact, but I do not have a license for it.