04-30-2013 12:33 AM
I have a survey sample with complex sampling design. So suppose I have to use commands like "proc surveyfreq" to get estimates. However, the dependent variable of interest is a dichotomous indicator reflecting rare events (e.g. smoking). For example, only 20 of 20000 students are smokers. I want to see how other factors (e.g. gender) influence the likelihood of being a smoker. Is there any method to deal with this kind of data? If yes, any special issue I should pay attention to? Thank you!
04-30-2013 04:42 PM
Surveylogistic or Surveyreg are probably good starting points. Assuming predominately class type data for most of the independent variables, gender, ethnicity, race then mostly Surveylogistic.