errickson.net
Back to Blog

Subsets and subpopulations in complex survey designs

#statistics

Generally you should not just subset complex survey data to do subpopulation analysis, instead using the built in tools, e.g. Stata’s svy, subpop(z): ... or R’s subset.survey.design

However, in the following two scenarios, subsetting versus formal subpopulations will give identical results:

  1. The subpopulation are strata. For example, say you stratify by race where race can take on categories “white”, “black” and “other.. If you want to look at one race (“black”) (or several races [“black” and “other”]), these subpopulations capture one (or several) entire strata.
  2. Every cluster has at least one member of the subpopulation.

Source: https://notstatschat.rbind.io/2021/07/22/subsets-and-subpopulations-in-survey-inference/


Back to Blog