I am trying to use the IPUMS CPS microdata to estimate flows from unemployment into employment for June - August 2021.
gen employed = 0
replace employed = 1 if inlist(empstat,10,12)
gen labforce = 0
replace labforce = 1 if inlabforce == 2
xtset cpsidp month, monthly
gen UEtoE = 1 if L.employed == 0 & L.inlabforce == 1 & employed == 1
egen UEtoEpop = total(panlwt) if UEtoE == 1
when I calculate the total number of people I get:
June: 1.963 million
July: 2.515 million
August: 2.360 million
Looking at the unadjusted series from FRED I see:
June 2021: 1.923 million
July 2021: 2.615 million
August 2021: 2.370 million
Is there some other weight I should be using to more accurately replicate flows?
I see from some of the linking exercises that replicating flows precisely requires additional adjustments.
From what I can tell, the FRED series you linked uses data from the CPS gross flows dataset (see the data source here). Although it says “unadjusted”, that means “not seasonally adjusted”, but it is still adjusted to be consistent with employment status stocks. As mentioned in the gross flows exercise from the CPS linking workshop, “The purpose of this analysis is NOT to replicate actual gross flows produced by the Bureau of Labor Statistics. Replicating published statistics requires adjustments to the data that are described in “Estimating gross flows consistent with stocks in the CPS” by Harley J. Frazis, Edwin L. Robison, Thomas D. Evans and Martha A. Duff (Monthly Labor Review, September 2005, pp. 3-9.”. In order to replicate those numbers, you need to adjust your estimates using an iterative raking procedure described in that paper, and it also requires mortality data that is not available in the CPS microdata. That said, I think the procedure you outlined (using PANLWT) is the best you can achieve using the public-use microdata and without the elaborate adjustment procedure detailed in that paper.
Thank you for clarifying this.