Friday's Jobs Figures

10/06/2012 02:03:00 PM
Greg Mankiw has a nice post on the difficulties involved in economic data, in particular the household and establishment survey employment data:
"Total employment rose by 873,000 in September.
"Total nonfarm payroll employment increased by 114,000 in September.
"...The issue is that there are two surveys. The first estimate of employment comes from the survey of households; the second is from the survey of establishments." 
Of course, I think that this analysis might be a little too narrow. It misses other factors such as the roles that revisions and seasonal adjustment play.

On each month's jobs report for the establishment survey data, the BLS tells us not just the number of employees, but also revises previous months estimates. The reason for the revisions is that the first estimates, which come out less than a week after the month ends--hardly any time at all to complete the surveys--only have a tiny fraction of the total number of establishments to go by. Later revisions reflect the addition of more survey responders, which after about a year will comprise some 98% of all employers in the United States. That high long-run response rate suggests to me that the revised establishment data is pretty reliable--not that there aren't any problems with it--but unfortunately, that success breaks down when we look at the preliminary estimates, which often deviate wildly from reality. I suspect that part of the reason (but only part) that the household and establishment surveys gave such different numbers for job creation is that the preliminary estimates involve some naive extrapolation from previous months to fill in the missing data. Normally, this works quite well, but in periods such as economic recoveries with more variation in the data, the preliminary estimates--based on my casual observation--tend to be biased in the direction of the previous few months' estimates.  So maybe there really was a bit more job creation than the establishment data suggests--we won't know until the revisions are out.

Then there is also seasonal adjustment. The headline establishment data estimates are always seasonally adjusted, which basically means that we are subtracting from that figure the number of temporary jobs due to Christmas season hiring (and in January, seasonal adjustment means that we are adding back jobs lost in the annual post-Christmas recession). Seasonal adjustment, however, is pretty tricky to do real-time, since the trends in seasonal hiring do evolve over time. One of the recurring difficulties during this recovery in particular has been that the seasonal fluctuations have been out-of-step from what we normally have, leading to underestimates of job creation in the summer months and over-estimates in the early months of the year.