Hi! I’m using the ASEC data from 1972 to 2017 where incwage has valid answers. In the middle of creating a sampling weight using ASECWT, I found that there are a number cases with ASECWT<=0. Specially, there are 162 with ASECWT=0 and 257 with ASECWT<0. These of course wouldn’t work for generating a weight variable. Why are there these values, and how to deal with them for creating a sampling weight variables? Many thanks.

The existence of these negative weights is mentioned in the ASECWT comparability statement. The negative weights are considered valid, and are designed to account for oversampling. However, since most stats packages can not accommodate them, and since the number of affected cases is small, some researchers simply chose to exclude those cases. Similarly, you may wish to also exclude those with a weight == 0.

Thanks, Michelle, for the answer, a version of that (about exluding the negative weights) I also found from a search. So far I’ve excluded the negative weighted cases because I created my own proportional samplng weight using ASECWT but kept the 0 weight cases because the 0 weight cases wouldn’t mess up the computation and just wouldn’t contribute to the overall analysis.

A post was split to a new topic: How to use weights?