Cannot reconcile Occ1990 or Occ2010 CPS data with actual BLS statistics

I’m trying to analyze occupational mix shift over time, using CPS data (I’ve used both AESEC and monthly samples). When using harmonized occupational categories like Occ1990 and Occ2010, I’m getting some weird results in the 2002 and 2003 period. There is a big (~two million people) decrease in the employed people in management and business and financial operations (categories 3-37 for Occ1990; 10-950 in Occ2010), which is strange and inconsistent with official BLS occupational mix data (from their own website), which doesn’t show this kind of decline. Would love some insight from those who are much more steeped into the data… thank you

As I mentioned in response to another post of yours inquiring about general comparisons with harmonized occupation variables from IPUMS, these measures are designed to facilitate comparison across changes in the underlying Census occupation coding schemes that are updated about every ten years. The 1992-2002 CPS data use the 1990 Census classification scheme, while the 2003-2011 data use the 2002 scheme. There are some substantial differences between the two schemes and I am not surprised that there is some discontinuity between 2002 and 2003. The updates to the occupation coding schemes are intended to capture changes in the job market, which means some jobs may not translate well across the differences. That being said, the discrepancy you report is sizeable, particularly if not present in the BLS estimates. While I can’t say anything about the BLS estimates, I can share information about how to think about this in your analysis of the microdata.

Using the breakdowns suggested on OCC2010 (which are slightly narrower than your suggested category), I can see that the decreases in your broader category are stemming from the Management in Business, Science, and Arts = 10-430 and Business Operations Specialists = 500-730 groupings, while the number of persons in the Financial Specialists = 800-950 category increases across this time period. The “Management in Business, Science, and Arts” and “Business Operations Specialists” categories both increase considerably during the 1992-2002 period; it seems reasonable to me that these codes might be more affected by a change in occupation classification as the new scheme may further disaggregate among these to capture nuances in this growing group and assign them to a more appropriate/specific category.

I would begin with this technical paper on The Relationship between the 1990 Census and Census 2000 Industry and Occupation Classification Systems; Table 2 in that paper shows the crosswalk between the 1990 and 2000 occupation coding schemes (see * for info on the difference between the 2000 and the 2002 schemes). Because the IPUMS harmonized occupation & industry variables use modal assignment, the nuance of how the 1990 codes are divided under the 2000 scheme is muted (i.e., all cases of a 1990 code that is split under the 2000 scheme will be assigned to the 2000 code that has the largest share of the original 1990 occupation in the double-coded crosswalk). I would start by looking at the previously linked crosswalk, specifically case counts for 1990 codes of 3-37 that are classified as having a code above 95 in the 2000 scheme* (this is not the same as the 2010 scheme, but the two systems are fairly similar; I would be willing to use them for a back-of-the-envelope estimate like this). When I summed the “1990 CLF Redist to 2000” values for rows where 1990 codes 3-37 are assigned to 2000 codes above 95, this totals about 2.4 million, which seems in line with the decrease you reported seeing.

Other general resources for digging into this might be tabbing OCC against OCC2010 (or OCC1990) in the years surrounding 2002/2003 (or create a crosswalk using the directions outlined in this blog post) to explore if specific occupations seem to be implicated in this discrepancy. The crosswalks and codelists on the Census Bureau’s Industry & Occupation page may also be of interest.

I am sure you were hoping for a more definitive answer, but hope this gives you someplace to get started. While steeped in the data, I am certainly not as familiar as you are with this research area. If you have questions about specific codes or the process for creating the harmonized measures, please let me know and I will gladly provide more information.

*Note that the technical paper and table linked previously use the 2000 Census occupation codes, which are identical to the 2002 Census occupation codes other than omitting a trailing zero added to the 2002 version. The trailing zero made the codes 4 digits wide and streamlined the process of understanding the relationship between 2010 vintage codes that disaggregated codes from the 2000 system. This is why I think you will want to look for values above 95 and not above 950. The CPS ASEC data in these years uses the 2002 scheme with 4 digits and not the 2000 scheme with only 3 digits.

1 Like

Thank you so much for the very thorough response… I will dig into the technical paper and explore the solutions you’ve suggested. Thanks again!

1 Like

Thanks for your patience–it took longer to investigate this than I was anticipating!