The household survey paints a clearer picture of the January employment report than does the nonfarm payroll
I’ll forward you to Spencer’s post on the January Employment report. As always, he sifts through this massive report and eloquently describes the state of the labor market. But I thought that I’d add a bit on the disparity between the household survey and the establishment survey.
The annual population revisions and weather distortions have confused some. The issue at hand is, that the BLS’ two surveys, CPS (Current Population Survey, from which the unemployment rate is derived and called the household survey) and CES (Current Employment Statistics, from which the nonfarm payroll is estimated and called the establishment survey), offer conflicting views on the strength of the headline report (i.e., just the statistics about the unemployment rate and the nonfarm payroll): the unemployment rate dropped 0.4% to 9.0%, while the nonfarm payroll increased a meager 36,000 when 146,00 was expected (by Bloomberg consensus).
The report is not conflicting, in my view – it’s just weather related stuff that impacts the CES, and to a much lesser extent the CPS. The drop in the unemployment rate, although usually the statistically less popular data point, is probably the best descriptor of the monthly shift in the labor market: strong. (A 9% unemployment rate cannot be described as strong by any measure out there; I digress.)
All of this information is stated in the BLS release, which you can find here. Below I describe (1) the revisions to the CPS and (2) the weather-related distortions that discount the establishment number.
Why the drop in the unemployment rate is credible. The summary statistics show the labor force falling by 504,000. The annual revisions dropped the labor force by 504,000, so the unrevised numbers show the labor force unchanged over the month.
The summary statistics show the number of employed increasing by 117,000 The annual revisions dropped the employed by 472,000, so the unrevised number of employed increased by 589k in the release. This is a big gain.
In fact, the revisions do not materially alter the 2010 unemployment rate nor its trend in any way. By my calculations – please correct me if I’m wrong – the unemployment rate would have been 9% with or without the CPS survey revisions.
For many reasons, the change in those ’employed’ in the household survey (+589k) does not match up to the change in the nonfarm payroll in the establishment survey (+36k); but the direction of the changes across both surveys are often similar. However, +589k is sizeable by any historical standard.
So why +36k in the establishment survey versus the +589k in the household survey? Weather. Nomura economists David Resler, Zach Pandl, and Aichi Amemiya did some research on weather-related months(no link available since this is proprietary paid research and bolded by me):
In one of the largest first reported declines on record, the BLS in its February 7, 1996 report calculated that non-farm payrolls FELL by 201,000 from the previous month. The outsized decline hit both manufacturing (-72,000) and services (141,000) but the construction industry registered a net job gain of 13,000. At the time, the BLS blamed the big winter storm for skewing the job loss and a month later reported that payrolls surged by 705,000 in February after a revised drop of 188,000 in January. Since then, subsequent benchmark and seasonal factor revisions have resulted in a history showing a drop of 19,000 in January 1996 followed by a 434,000 increase in February.
These observations suggest January’s severe winter storm could skew the measurement and estimation of payrolls this year as well. From those prior episodes, we calculate that the winter storms led to a .0007 to .0015 deviation from “normal” seasonal job change in those months. A similar deviation of employment from normal seasonal patterns implies that the change in non-farm payrolls would be likely to fall in a range of -50,000 to +56,000.
Ex post, they were right – they published this research before the January employment report was released – not only about the expected weather distortions, but also regarding the level (-50k to +56k). Accordingly, it’s very likely that next month we’ll see outsized gains, given the history of this type of distortion.
Therefore, until I see a weak February number (one month from now), I’m going to assume that the headline figures were strong and consistent with the unemployment rate dropping 0.4% to 9.0%.
Of course, there’s a slew of workers that are not in the labor force that may re-enter, which would undoubtedly drive the unemployment rate up (it probably will). And let’s remember this when talking about a “strong” employment report: 9% doesn’t represent the severity of the unemployment problem – the employment to population ratio does.