Using "proc logistic" with some subjects having more than one case/row

[Manhort]

Hi everyone,

I have the following code for a logistic regression, 

proc logistic data=have descending;
  class var1(ref='3') var2/ param=reference;
  model treatment = var1 var2;
run;

In my model, the treatment one receives IS the outcome variable. 

Now, some people in the data have more than one occasion of receiving treatment for the medical condition in question. There are 20,000 people and 20,500 cases. It is a very small proportion of people with a "cluster" of cases, but I would like to account for it anyway, regarding standard error estimates. Can anyone point me in the direction of the proper code?

Thanks

[svh]

I think you are looking for the PROC GLIMMIX procedure, which allows you to build mixed effects logit models to account for the fact that some of your observations are not independent. It does seem like you should ask yourself why some of these subjects have more than one observation in your data and how that could influence the outcomes. E.g., if there was some kind of cumulative benefit of receiving additional treatments, that could make your sub-set of respondents a biased sample and not just a sample with random error.

[PaigeMiller · Accepted solution]

Lots of good information here: Logistic regression when there are more than one o... - SAS Support Communities