I am conducting an analysis using CPS-TUS data from 2006-2014/2015, and I would like to restrict the analysis to self-respondents. Is it preferable for me to delete non-self-respondents from my sample or to exclude them using the svy subpop option, considering that I’m using the TSRWT variable to account for weighting?

Relatedly, there is a small portion of respondents who appear to be self-respondents based on the TRESP variable, but who have a self-response weight (TSRWT) of zero. Should I assume these individuals are in fact self-respondents?

If you are aiming to calculate representative statistics with your analysis, then I’d suggest using the svy subpop option using the TSRWT variable. This will help ensure that any non-random non-response of self-respondent survey answers do not bias your calculations. On the related note, it seems that the TSRWT variable sets a small number of self-respondent survey respondents to zero. I checked to see if this was caused by the universe definition for these questions, but even after cleaning the data by the appropriate AGE and MISH restrictions cases with TRESP==1 and TSRWT==0 still persist. Therefore, it seems like these cases are given a weight of zero to adjust for the response rates of these self-respondent survey questions.