11-03-2016 12:49 PM
in my analyses i came across many limitations regarding binary dependent variables. My data is clustered so at first i simply used proc surveylogistic to get odds-ratios with robust standard errors for clustering. Then i read, that there are great concerns about interaction effects in logit models (and im using those). So now my plan is to switch to a linear probability model instead. Regarding those there is always the problem of heteroskedasticity. So as far as i understand, the standard errors have to be robust in two aspects: clustering and heteroskedasticity. Is there a way to include both in proc surveyreg?
Thank you in advance and best regards!