I am looking for an Fitting model for some repeated measure data and think that GEE or GENLINMIXED would work well.
The dependent variable is a 0,1 dummy variable called "attribute" (yes/no) and there are some independent variables such as gender, profession (nominal variable with 4 possible values), age (at the moment of the interview), and another dummy variable focussing on prices (yes/no that are awarded every year).
The repeated measurements took place various times a year (ie various times per award period) but are unfortunately not identically distributed within those years.
Now I thought of an syntax like this:
GENLIN attitude (REFERENCE=FIRST) BY profession gender award award_period WITH age_in_days
/MODEL profession gender age_in_days award INTERCEPT=YES DISTRIBUTION=BINOMIAL LINK=LOGIT
/REPEATED SUBJECT=respondent_ID WITHINSUBJECT=interview_number.
Interview number is uniquely identifying all interviews (repeated measurements), respondent_ID uniquely identifies all interviewees. The data are in the long format, so each respondent x measurement combination is one case.
However, I am not absolutely sure whether this is really the best way to deal with the data. By the way, we assume that there might be some profession and year specific "clustering" effects. Shall I include them in the "REPEATED" subcommand? Moreover, some interviewees that were included in the first measurements cannot be found in the last and vice versa.
Thanks for helping,