Is there some sort of arbitrary number of timepoints needed in order to estimate person fixed effects?