Matching BLS Union Membership Estimates


I’m currently attempting to match BLS estimates for overall union membership between 2000 and 2020 using IPUMS CPS microdata. I’ve been able to match the BLS estimates spot on for 2003 to 2020.

However, I’m running into an issue with the estimates pre-2003. When I produce the estimates for these years, there’s about a 1% different between my estimate and the BLS estimate for each year. Below you can find my estimate (first) versus the BLS estimate (second) each in the thousands.

2000: 16433 vs. 16258
2001: 16428 vs. 16275
2002: 16203 vs. 16145

The process I engage in to obtain my estimates involves a loop for each year that first filters the data to the outgoing rotation group for that year using the exact instructions from this IPUMS forum thread, summing EARNWT while grouping by MONTH and UNION, and then taking the mean of the result from the previous step. Again, this process appears to work for 2003 through 2020.

I’m aware that there can be some difference between estimates produced using IPUMS and official estimates, but I just want to make sure there isn’t anything that I’m doing wrong or a comparability issue with a variable I’m using that might throw off my pre-2003 estimates. Thank you for your help on this!!

You are indeed correct that there is a <1% difference between BLS and IPUMS estimates for union membership in the 2000, 2001, and 2002 CPS. This discrepancy has to do with CPS weights that were re-calculated after the 2000 Census (the weights that are currently available through IPUMS CPS). Unfortunately, we have not been able to replicate the BLS estimates with any other versions of the weights that we are aware of.

I wish I had better news, but this is the best explanation we can offer at this time.

1 Like