Possible issue in the construction of variable "ahrsworkt" for September 2020?


I am finding that the average of “ahrsworkt” suddenly drops in September 2020 in a manner that does not seem a priori reasonable (there is a sudden drop from August to September and then a sudden increase back to “average” in October; see table pasted below). Looking closely to the distribution of hours I found that the skewness of that variable increased a lot that month of September and that this is due to a large increase in the number of individuals that report working 32 hours for which I cannot find any economic explanation.

In case it helps, I am also finding that that same month (September) the proportion of full-time workers suddenly drops in September, and then by 15% or so and then increases back to “average” in October.

Is this variation in “ahrsworkt” and part-time work genuine or are they due to a specific issue related to the construction or compilation of those variables from the original Census source?

Any thoughts about this? All help would be appreciated.



. table modate if year==2020,c(mean ahrsworkt mean FT mean prop_h32)

modate | mean(ahrswo~t) mean(FT) mean(prop_h32)
2020m1 | 39.3083 .8332688 .0225601
2020m2 | 39.2335 .8280079 .0236424
2020m3 | 39.0264 .8478227 .0195566
2020m4 | 38.1601 .8821307 .0169106
2020m5 | 38.6207 .8619918 .0181029
2020m6 | 38.9371 .8619949 .0196313
2020m7 | 38.9844 .8758039 .0198097
2020m8 | 39.3177 .8650216 .0203174
2020m9 | 37.9057 .7303844 .0921946
2020m10 | 39.1605 .8179677 .0324198
2020m11 | 38.8835 .8093936 .040215

I suspect this is related to the late timing of Labor Day in 2020. Every several years the holiday falls on September 7, which is during the reference week for CPS respondents. A bit more explanation is available in this forum post about persons on holiday in September 2015 (the last year that Labor Day fell on September 7).

Hi Kari,

Nicely caught! Thanks a lot.


1 Like