Seemingly contradictory data in The U.S. Bureau of Labor Statistics  (BLS) report released yesterday morning indicates that the United States unemployment rate dropped from 8.1% in August to 7.8% in September, nonfarm payroll only increased by 114,000, and total employment went up by 873,000. How can this be? 

Why the Discrepancy?

The simple answer is that the employment data comes from two different methods that yield different information as detailed in an Employment Situation Technical Note linked to from the release.

The Current Population Survey (household survey)  is the source for the unemployment rate decline and indicated "total employment rose by 873,000 in September." The Establishment Survey indicated that "total nonfarm payroll employment  rose by 114,000." 

The "Technical Note" linked to above notes various reasons the two surveys can provide different results, including the fact that the household survey included agricultural employment and the self employed as well as unpaid family workers. Also, some businesses may be late sending in data for  the Establishment Survey resulting in it providing a low figure for jobs created.

Another reason for the discrepancy could be inaccuracies in either report due to sampling error. The household survey is a smaller sample so it would be more likely to have error in this case. It "is a sample survey of about 60,000 eligible households conducted by the U.S. Census Bureau for the U.S. Bureau of Labor Statistics." In contrast, The Establishment Survey involves "payroll records from a sample of nonagricultural business establishments" totalling about "141,000 business and government agencies, representing about 486,000 individual work sites. . . . The active sample includes approximately one third of all nonfarm payroll employees."

Both surveys only cover a portion of the month rather than the entire month. The household survey "reference period is generally the calendar week that includes the 12th of the month. In the establishment survey, the reference period is the pay period including the 12th, which may or may not correspond directly to the calendar week."

My Views

It seems an unusual coincidence that the September unemployment rate happened to look so much better than August's, just one month before the election after being above 8% for so long. Although at least a few argue that the data is manipulated for political purposes, I think it is more likely that the recovery accounts for the decline.

My guess (and it is only a guess) is that numerous employers were late reporting data. Therefore, future reports will indicate that the household survey was more accurate. But the household sample survey could be off instead.

I am already looking forward to the October employment data that will be released on November 2, a few days before the Presidential election. That report will likely be even more controversial if the two surveys provide substantially different data.