I have a methodological question regarding the calculation of weights for a survey-based study. For our sample of respondents, we would like to adjust for oversampling and nonresponse bias using demographics collected on both respondents and nonrespondents, so as to facilitate the projection of results to the larger population. However, for our particular research question we are only using a sub-set of the respondents (i.e., just those who reported being in a relationship). Before analyzing the data in Stata, would it be more appropriate to calculate the weights on the full sample of respondents, and then apply them only to the sub-sample in the analyses? Or should we calculate the weights based on our sub-sample of respondents who reported being in a relationship? If we do the latter, wouldn’t the selection of respondents based on relationship status introduce bias in the estimation of nonresponse bias weights?
Thanks in advance. Brian
Thanks in advance. Brian
Comment