Anyone who’s been following recent discussion of the US economy will be aware that the Bureau of Labor Statistics produces employment statistics from two different surveys, and that the results have diverged radically since 2001. The BLS preferred numbers on employment growth come from a survey of employers (the Establishment Survey) while other numbers, including the unemployment rate are derived from a survey of households (Current Population Survey). As the BLS Commissioner’s latest statement notes (PDF file)
From the trough of the recession in November 2001 through January 2004, payroll employment decreased by 716,000. Over the same period, total employment as measured by the household survey increased by about 2.2 million (after accounting for the changes to that survey‚s population controls).
Not surprisingly supporters of the Administration have been pushing hard to discredit the Employment Survey in favour of the CPS. While noting some reasons for the discrepancy, the BLS seems to be sticking with the payroll survey, noting that there are a lot of problems in estimating employment growth from the CPS, and that the payroll data is consistent with data on new claims for unemployment benefits.
If that’s the case though, the implication appears to be that the CPS results are unreliable, and therefore that the unemployment rate (derived from the CPS) is an underestimate. Allowing for the fact that non-employed people are divided between unemployed and those not in the labour force, the discrepancy could easily be a full percentage point, implying that unemployment is now higher than when the recovery (as measured by output) began. This seems consistent with anecdotal impressions.