I am trying to match average duration of unemployment (NSA) posted for each month for subpopulations by age on the BLS site with the durunemp variable. But, the averages I calculate based on the IPUMS CPS data typically show an average duration that is about one week shorter. Also, the difference tends to be higher for older workers than for younger ones. I am using the wtfinl weight for these calculations. What I am missing?
I wouldn’t expect differences of this size. However, without more information about your analysis, it is difficult to provide more information on the possible source of the discrepancy. Specifically, the exact table/data source you are trying to replicate, the years/months you are looking at, and the age and employment restrictions you are applying to your data would be helpful.
Hi Grace, I am trying to recreate the following three series, among several others, from the BLS: LNU03008297 – average weeks of unemployment for people 45 to 54
LNU03008301 – average weeks of unemployment for people 55 to 64
LNU03008305 – average weeks of unemployment for people 65 and older
For the IPUMS data extract, I am using the following Stata command:
table month age10year [pw=wtfinl] if inlist(empstat, 21, 22) & year==2020 & durunemp<999, c(mean durunemp)
age10year is my 10-year age grouping.
Thanks for any suggestions on where I may have gone wrong.
Thank you for providing more information. I replicated these tables myself and my results are similar to yours; average weeks unemployed are about one week shorter than BLS estimates. Public use files may differ somewhat from the data used to generate published estimates in order to protect respondent privacy. While top side estimates for major demographic categories and major labor force status will agree with published totals, small groups and/or combinations of these characteristics will differ due to the fact that steps are taken during the creation of CPS public use files in order to protect confidentiality of respondents. You can find more details on comparability between microdata and published estimates in the codebooks for the relevant samples; see Attachment 2 (Overview) under the “Comparability of CPS Microdata files with Published Sources” header.
Hi Grace, Thanks for taking the time to replicate and respond. Your explanation makes sense. I will be careful in using data for smaller groups. Take care and happy holidays.