What's Causing the Strange Jobs Data Discrepancies?

The Atlantic | February 4, 2011
By Daniel Indiviglio

There's a common question on the minds of many today who follow economic news: how did the unemployment rate decline by 0.4%, when only 36,000 jobs were created? I explained this in a fair amount of detail earlier. In short, it's complicated. But one contributing factor could be differences in the way these two statistics are measured.

The Bureau of Labor Statistics relies on two separate surveys for its monthly jobs report. The "Household Survey" provides the unemployment rate and the "Establishment Survey" provides the number of jobs added or lost overall and in each industry. For the past two months, the Household Survey had a relatively positive result, which caused the rate to decline a whopping 0.8%. Meanwhile, the Establishment Survey has been consistently lukewarm, showing just 157,000 jobs added.

Of course, this makes no sense, because for the unemployment rate to decline by 0.8%, you would need to add 1.2 million jobs, with all other variables remaining unchanged. For the past few months, other variables did change, however. The workforce shrunk as some unemployed Americans gave up looking for work. In January, the population also declined, which had an effect. Other workers went from being technically unemployed to being labeled marginally attached, because they weren't looking work, but still wanted a job.

But that doesn't provide the full explanation. According to the Household Survey, employment increased by 589,000 workers in January, while the Establishment Survey says it rose by just 36,000.* For December the difference was a little smaller -- 297,000 jobs added for the Household Survey and 121,000 for the Establishment Survey. What's going on here?

It's pretty hard to know for sure. But maybe we should start with how the BLS explains the ways in which these two surveys differ. We'll consider each:

The household survey includes people on unpaid leave among the employed. The establishment survey does not.
There's little reason to believe that this has had a significant effect recently. There hasn't been any sociological shock that would require many workers to take unpaid leave to inflate the Household Survey's numbers.
The household survey is limited to workers 16 years of age and older. The establishment survey is not limited by age.
This one can't explain it either. If anything, this would inflate the Establishment Survey's numbers, since it could include more employed workers under the age of 16. We're seeing better performance in the Household Survey instead.
The household survey has no duplication of individuals, because individuals are counted only once, even if they hold more than one job. In the establishment survey, employees working at more than one job and thus appearing on more than one payroll are counted separately for each appearance.
Again, this difference could only inflate the Establishment Survey, as it could double-count employment. So this can't explain any disparity either.
The household survey includes agricultural workers, the self-employed, unpaid family workers, and private household workers among the employed. These groups are excluded from the establishment survey.
This explanation might help to explain the differences. As Americans struggle to find firms hiring, more may be turning to self-employment or taking on the role of unpaid or private family/household workers. This implies that the Establishment Survey is underestimating employment, because it doesn't include those groups of workers.

Of course, you could just as easily argue that the Establishment Survey is more accurate, because the Household Survey is overestimating workers. For example, if 1000 unemployed people who were laid off in 2009 decide stop looking for work, rely on their spouses' income, and stay at home with the kids, did the economy just add 1000 jobs? It sure doesn't seem like it.

In fact, BLS provides the data for the categories of farm workers, self-employed, unpaid family workers, and private household workers. Unfortunately, it doesn't help much to reconcile the employment numbers for the Household and Establishment Surveys. In December, the Household Survey goes from predicting 176,000 more new jobs to predicting 245,000. Clearly, this is no silver bullet.Unfortunately the January change canno be tabulated, as BLS does not provide consistent data from December to January

These two surveys have suggested different rates of job growth since the recovery began. Here's the job growth since January 2010 for each survey and the corrected Establishment Survey, which includes the four categories of workers that the Household Survey includes:


And really, this is nothing new. These two surveys have diverged for some time. In fact, when you add the jobs that the Establishment survey overlooks to its total, the surveys' estimates have been moving close together since 2000.

So do these two surveys provide somewhat different statistics? They do. This is part of what contributed to the divergence we've been seeing over the past few months.

Differences in surveying do not account for the entire big decline in the unemployment rate over the past two months, but they do count for the difference in January. In December, the change was mostly to the labor market shrinking as more Americans stopped looking for work. That's something the Establishment Survey doesn't track, so there's no conflict between the two data sets regarding that statistic.

*This post was revised slightly to reflect a correction made to this post. See this other post for more detail. The January discrepancy was actually very large, and responsible for the unemployment rate decline.