Hi everyone,
I have a longitudinal panel data with the person-spell format where a spell is defined as the time span between waves (2 years). I also have pweight, strata and psu variables. When I use pweight or respondent weight then the results become really unusual and inconsistent with previous studies. In contrast, when I drop the pweight and run the model only with strata and psu variables then the results are completely reasonable. Is there something wrong with using respondent weight to fit a biprobit model when one is using person-spell data format.
Thanks for your advice!
Best,
Nader