Evan Soltas
Jul 7, 2012

Household versus Establishment

Joe Weisenthal of Business Insider has a post up suggesting that the employment situation is, by the numbers, substantially better than consensus. Weisenthal is actually so surprised he asks for an explanation.

Here we go, then.The "employment level" data Weisenthal uses comes from the Bureau of Labor Statistics' "household" survey, formally known as the Current Population Survey. The data more commonly used, and I'll explain why in a moment -- nonfarm payrolls -- comes from the "establishment" survey, a.k.a. Current Employment Statistics. The BLS clarifies the distinction here.

In my experience, the household data tends to be significantly more noisy. That makes sense, because the household sample is much smaller than the establishment survey: 60,000 households versus 141,000 businesses and government agencies. The establishment payrolls survey tends to be preferred by news media and for official purposes.

There are other distinctions between the two surveys. Also importantly, the household survey counts the more marginal forms of employment, especially of new firms and self-employment, whereas the establishment numbers require the employed to show up on a payroll statement. (The two links there are to a detailed BLS study reviewing the source of discrepancies, and a discussion by Alan Meltzer arguing that household survey numbers are important to consider in recoveries, when firms are reorganizing and reconstituting themselves.) Moreover, the difference between the household survey numbers and the establishment payroll numbers is itself cyclical, Barry Ritholz writes.