CPS Union Data vs BLS data

Hello. I am using the variable “union.” I limited to Louisiana and then I eliminated all cases labeled “NIU.” I combined 2 (member) and 3 (represented by a union but not a member). Then I tabulated the percentage of Louisiana workers who were represented by or members of unions from 1990-2019. However, the data I retrieved is different than the data I get from the BLS website (for some years it is the same or close and for other years it’s quite off). Below is what I put into the BLS search. Does anyone know why these would be different?

Series Id: LUU0204899722
Series title: (unadj)- Percent of employed, Represented by unions, (LA) Louisiana
Unadjusted series
Union: Represented by unions
Industry: All Industries
Occupation: All Occupations
Sex: Both Sexes
Race: All Races
Ethnic Origin: All Origins
Age: 16 years and over
Earnings: Person counts (number in thousands)
Class of Worker: Wage and salary workers, excluding incorporated self employed
Labor force status: Employed

If you could share 1) a link to the BLS data/interactive query tool you are using and 2) more information about the discrepancies you are seeing, that would be helpful. Without that information, I can share a couple of ideas about what might be happening. Please let me know if I can provide clarification on any of these comments.

My immediate suspicion is that this might be about how representation by unions is defined; because Louisiana is a “Right-to-Work” state, I wonder if the BLS numbers are including both membership and representation without membership (which is it sounds like your code using IPUMS CPS does). I also don’t see any restrictions based on class of worker in the IPUMS criteria that reflect the restriction you apply in the BLS query (while I don’t think this is likely, I am mentioning it as one possibility). It seems odd to me that the numbers match in some years but not others–I will try to provide a more thorough response if you link to the data source and describe the discrepancies you are seeing.

Thank you. This is the BLS tool I am using: https://www.bls.gov/webapps/legacy/cpslutab5.htm
Then I click only on “represented by unions - percent of employed” (the last column) just for Louisiana.

Then in the CPS data I downloaded, I created a new variable combining both ‘member of a union’ and ‘represented by a union’. I limited cases only to Louisiana. I changed NIU to missing. Then I just did a tab by year. Here are the BLS results vs mine. Mine are in parentheses.

2000: 9.3 (9.38)
2001: 10.2 (10.0)
2002: 10.3 (9.82)
2003: 7.9 (8.89)
2004: 9.3 (9.78)
2005: 7.4 (8.27)
2006: 7.2 (9.77)
2007: 6.5 (7.69)
2008: 5.6 (5.86)

When I look at the raw numbers they’re significantly different. Seems like the BLS sample is much larger than what I’m looking at. The numbers below show the # of people represented by a union by year in Louisiana.

2000: 158 (33)
2001: 176 (43)
2002: 170 (38)
2003: 132 (32)
2004: 121 (36)
2005: 108 (23)
2006: 97 (25)
2007: 110 (16)
2008: 96 (14)

Thanks again for your help.
Jesse

Thanks for sharing the additional detail. I have a few guesses about what might be causing the differences you are seeing.

First, I suspect you aren’t using all months of data for each year, but that the BLS tool is (to your point about them using a larger sample). Is it possible you are just using the ASEC instead of all basic monthly data for each year? It is unclear to me if you are interpreting the “raw numbers” as an unweighted count from BLS or not–to make sure we are on the same page, I will note that my understanding is that these are the estimates for the number of persons represented by unions in each year. I would not use the ASEC at all for making these estimates; I recommend using all 12 months of basic monthly data for each year.

Second, it doesn’t sound like you are applying weight to your analyses; given the complex sample design of CPS, weighting estimates is necessary for them to be accurate. Variables from the Outgoing Rotation Group (ORG), like UNION, should be weighted with the IPUMS CPS variable EARNWT. As noted on the IPUMS CPS page about ORGs, you will also need to divide EARNWT by 12 if you are pooling all months for each year to get accurate point estimates (as described above).

Using all 12 months of basic monthly survey data for 2000-2009 (using March basic monthly and not ASEC) and weighting with a EARNWT/12, I was able to pretty closely replicate the BLS numbers (both percentages and weighted counts–though my weighted counts aren’t a great match in all years, specifically 2004-2006). I wouldn’t expect to be able to replicate BLS estimates exactly using the public use files, but you should be able to get in the general ballpark of these. For reference, my results are below.