Descriptive Statistics with CPS Outgoing Rotation Group Data


Hi All,

I have two questions about using the CPS Outgoing Rotation Group data.

  1. I downloaded the Outgoing Rotation Group data by going to IPUMS-CPS and selected all of the person-level variables under “Outgoing Rotation Groups (Earner Study).” I dropped all observations whose month-in-sample value was not 4 or 8. However, when I try to replicate the union membership numbers on, which also uses the CPS-MORG data to run descriptive statistics by year, I find that the number of observations by year available through IPUMS is around 1/5 the number used to produce the numbers on When I weight by earnwt/12, I also find the the number of union members in the U.S. by year is 20-50% lower than the estimates available through the BLS. Here is the Stata code I used:

gen annualwt = earnwt / 12
sum empl [iw=annualwt] if year== 1984
display r(sum)
sum unmem [iw=annualwt] if year== 1984
display r(sum)

Does anyone know why I can’t get accurate population estimates using the CPS-MORG data available through IPUMS-CPS with the above weighting?

  1. Which CPS weight should i use to find the % female workers by year in my CPS-MORG dataset?

Many thanks!



I’ll answer each question individually.

(1) It seems like that the tables available on are using only the basic monthly samples (and not the ASEC sample) for each year. When I pull all of the basic monthly samples (for a given year) and restrict the sample to only those in the Outgoing Rotation Group (MISH==4 | 8), I find observation counts close to what is reported in the online tables. I’ve also noticed that in your data extracts, you’ve only included ASEC samples. This will be a smaller sample, in terms of number of observations, compared to simply pooling all of the basic monthly samples. Additionally, once this detail is addressed, I calculate percentages that match the tables available online.

(2) To calculate the % female workers by year, I’d actually recommend not simply using your Outgoing Rotation Group dataset. The reason being is that this data is available for a much larger sample of the population and you’ll be able to calculate much more precise estimates if you use a larger sample. I’d suggest using the variable WORKLY, which is available in ASEC samples dating back to 1988. Then you can tabulate by SEX, while using the WTSUPP sample weight.

I hope this is helpful.



Thanks, Jeff. You’re completely right that I had incorrectly selected the ASEC samples rather than the basic monthly samples. Many thanks for spotting that.

Your point about the using a larger sample than the MORG sample is a good one. If I did want to look at % female in the MORG sample though, would I again weight by earnwt / 12 for the pooled annual sample?