Another question about weights

I see a few related questions about weighting here, but I don’t see my specific question yet. I attended the orientation webinar to using NHIS in IPUMS in October, and I believe the guidance was, if you are doing an analysis where some variables use PERWEIGHT and some use SAMPWEIGHT, you would always use the more restrictive weight, i.e. SAMPWEIGHT.

However, what about if your overall dataset contains some variables using PERWEIGHT and some using SAMPWEIGHT, but the specific crosstab you are running only includes variables using PERWEIGHT? Would you use PERWEIGHT in that case? Or should you always use SAMPWEIGHT whenever you are working with a dataset that merges the person file with the sample adult and sample child files?

Thank you!

The answer depends on what sub-sample you are aiming to estimate statistics on (e.g., the sample-adult/child sub-sample or the entire NHIS sample). For example, if you want to estimate the racial status categories for the entire NHIS sample, you should use the PERWEIGHT variable. If, on the other hand, you want to estimate this same statistic for the sample-adult/child sub-sample, you should use the SAMPWEIGHT variable. In both cases, implementing both weights are technically feasible and will calculate a statistic, what matters is which sub-sample you intend to study.

Thank you very much for the prompt reply! That guidance makes sense.