Wonky employment estimates by OCC2010 across CPS ASEC Years

I was looking at employment by education, and noticed some really odd estimates in the CPS ASEC. From 2002 to 2003, the number of workers aged 25 and older (age>=25) with a high school diploma, but no bachelor’s degree (educ>=73 & educ <111) working as retail sales workers (occ2010=4760) doubled. Increasing from 720K to 1.5 million.

The 1.5 million number is what I would expect, and corresponds with what you see in the decennial Census for 2000.

Overall, employment among retail sales workers is about half of what I’d expect across all education groups in years 2000 to 2002. I haven’t looked at years prior to 2000, so not sure where the issue starts.

I also plan to contact the Census Bureau about this, but am curious if someone at IPUMS, or other researchers may have any insights?

UPDATE: This doesn’t seem to be an issue with occ2010, but with the occ code. The 1992 to 2002 scheme had 6 different occ codes classified retail sales workers by sector, and then starting in 2003, the codes were collapsed to a single “retail sales workers” category.

The CPS implemented a major change in the occupation coding system between 2002 and 2003. The codes of OCC2010 are not always completely consistent across such series breaks, though IPUMS CPS attempted to make it as comparable as possible. Here is the full list of the 2002 occupation codes; the codes that correspond to the OCC2010 category of retail sales workers are:

263 Sales workers, motor vehicles
264 Sales workers, apparel
265 Sales workers, shoes
266 Sales workers, furniture
267 Sales workers, radio, TV.
268 Sales workers, hardware

In 2003, a broader code of “retail salespersons” was used (see the full list here). You can see how the 2002 and earlier codes would be less inclusive, since there are many retail salespeople who do not sell cars, apparel, shoes, furniture, electronics, or hardware. Note that the occupation codes used in the 2000 Census weren’t implemented in CPS until 2003; so it makes sense that the 2000 Census numbers map more closely with the 2003 and later CPS numbers.