Using regression coefficients in a subset of data

[Ahmed Magdy]

I have data on 170 counties; the data is collected for 8 time periods (4 years divided into 2) with a total of 1,360 county-period observations. I am planning to conduct a case-control analysis using conditional logistic regression to identify key predictors that can be used to estimate probability of the outcome among the 170 counties. What is the appropriate way of applying the coefficients resulting from -clogit- (applied to a subset of the 1,360 observations) to estimate probability of the outcome for the 170 counties using predictor data (data restricted to one of the time periods in the same data)? I’m thinking to use -estimates store- & -estimates restore-, or -parmest-. Another option I'm considering is to use -if inrange- to run -clogit- and then -predict- without -if inrange-? Does this make sense? Is there another way?

Thanks in advance.

[Phil Bromiley]

As you thought, do the logit with a if statement to limit the observations, then do the predict without the limitation. However, red about prediction from conditional logit - it is a little trickier than with regression.

[Ahmed Magdy]

Thank you, Phil for your response. Yes, I later came across other posts about that and I will likely use mixed effects logistic regression so I can obtain appropriate predicted probabilities.