CPS COUNTY variable for ASEC analysis

Hi all,

I’m hoping to calculate retirement coverage (using the PENSION variable from the ASEC) by county. Previously, I’d only been doing state-level analysis because I thought that was as low a geographic location as the CPS could accomodate. But then, I saw that there is a COUNTY variable in the CPS. In particular, I’m hoping to find retirement coverage for a few counties (ex. Lancaster County, PA [COUNTY code 42071]).

  1. Is it possible to do county-level analysis reliably? (I’m having troubling just replicating Lancaster County’s total population as a gut check).
  2. If so, what weight should be applied (WTFINL or HWTFINL).
  3. And lastly, can I analyze ASEC data (e.g., the PENSION variable) at the county level? And if so, what is the appropriate weight to use: WTFINL/HWTFINL or ASECWT/ASECHWT?

Thanks so much!

There are some limitations to using CPS data at the county level that are likely causing your observations; I will address these while answering your questions in-turn.

  1. The CPS is stratified at the state level and is therefore not designed to generate sub-state level estimates, it is primarily designed to calculate reliable national estimates. See the section titled “Geographic Limitations” on page 2-6 of the 2020 March CPS Codebook for more discussion. The CPS cautions against estimates for smaller metropolitan areas but notes that larger metropolitan areas like NYC or LA should generally be fine. Analysts may determine whether it is appropriate to use estimates at the county level for their specific use case or if they pool enough adjacent months or years of data together in order to increase the sample size (see this forum post for more information about pooling CPS samples). In addition, as is noted on the description tab for COUNTY, to preserve respondent confidentiality, not all counties are identifiable in all samples. Specifically, only about 45% of all households reside in an identifiable county in any given sample. Despite documentation cautioning against county-level analyses, many people use the data in this way; it may be helpful to consult the relevant literature for any guidance on how other researchers do sub-state analyses.
  2. If you are using ASEC data, ASECWT (person-level) and ASECHWT (household-level) are the appropriate weights to use. WTFINL and HWTFINL are the appropriate person-level and household-level, respectively, weights to calculate nationally-representative population using CPS basic monthly data. See this page summarizing available weights for the CPS.
  3. It is appropriate to use ASECWT/ASECHWT for analyses using ASEC data. You might also find this information useful on replicate weights.