I’m trying to calculate the number of occupation transitions in the US that happen in a month and in a year. Using OCC2010 and PANLWT, I calculate that between Feb and March 2019 ~11 million workers switched occupation (~144 million remained in their occupation). I do this for all consecutive months and find that ~142 million people switch occupation every year. This seems like an unreasonable amount of occupation transitions, it would suggest almost all the labor force switches occupation in a year. I don’t know what I am doing wrong.
To give more details, I am using CPS data for 2019, variables OCC2010, YEAR, MONTH, PANLWT. I first drop all entries with “nan” PANLWT value, not in the labor force, or with no occupation. I use PANLWT and calculate a labor force of 158 million for February. When I do the yearly calculation I get a total count of ~1599 million monthly occupation remains and ~ 142 million occupation switches.
I am using PANLWT as suggested in this answer (Weights for linking CPS basic monthly data). I have also tried using WTFINL and results are similar.
In this post (Calculating Quarterly Unemployment Rates by Race for CA) they suggest to divide by number of pooling samples. Is this something I should consider? E.g. dividing by 12 month (or 11 possible monthly switches)? However, if I do this, I don’t know how I would interpret monthly job transitions, why would I divide those by 12?
Alternatively, I could perhaps divide by the number of months an individual is sample?
Any help would be much appreciated, thanks!
if this question of yours is still relevant, could you share your code?
Sorry for the very late reply - this post slipped through the cracks at the IPUMS User Support team. I just calculated the occupational transition rate between Feb-Mar 2019, and found numbers very close to yours, so I don’t think this is an error in your calculation. I also don’t think you should be adjusting your weights in this case, since the total switchers within the year should just be the monthly switchers added up over the year (this is for the gross number of occupational switchers). If you’re looking for an annual net rate (not counting people who switch out of their occupation and then back in), then you’d want to calculate it differently.
I think the reason your numbers seem excessively high is because the number of measured occupational transitions tends to be inflated in the CPS, especially when looking at the most detailed level (as captured in OCC2010). This is due to inaccurate proxy responses, miscoding, and imputation. For calculations of occupational employment in a single month, these errors tend to average out. In contrast, for month-to-month transitions, any change in either period due to error is recorded as an occupation transition. There is no averaging out. There is a paper (attached) by Kambourov and Manovskii (2013) that investigates this issue in detail.
CautionaryNote on using ASEC and PSID for occupational mobility.pdf (264.7 KB)
Here’s the code in Stata that I used, if it’s helpful (my data only included the Feb and Mar 2019 basic monthly samples):
*keep only records that were employed
keep if inlist(empstat,10,12)
sort cpsidp month
by cpsidp: gen occ_lag = occ2010[_n-1]
drop if mi(occ_lag)
gen changer = 0
replace changer = 1 if occ2010!=occ_lag
gen panlwt2 = round(panlwt)
tab changer [w=panlwt2],m