What’s behind the seemingly sudden drop in the unemployment rate?

While the economy stumbles along, no one would expect a sudden jump in employment. Job growth has averaged about 100,000 per month over the past six months, roughly consistent with other economic indicators suggesting slow growth. But the Labor Department reported today the unemployment rate fell from 8.1 percent in August to 7.8 percent in September, which makes a full half-point drop since July.

These results suggest something very strange is going on with the household survey. In this hyper-political season, some who should know better and some who know little at all are suggesting the Obama Administration is playing games with the numbers. This is almost certainly not the case. The professionals at the Bureau of Labor Statistics would never stand for it, and like all good bureaucrats, they have ways of getting the truth out. Something strange is going on, but not politics—rather, statistics caused this little rhubarb.

The federal government runs two main jobs surveys. The payroll survey queries employers and is more reliable mostly because it is a fairly large sample. The second survey queries households and is much smaller, but provides a different look at the employment picture. The household survey also generates the unemployment rate. Typically, the two surveys disagree somewhat month-to-month, but track fairly well with one another over longer periods of time, because over longer periods of time, the small size of the household survey matters less and less.

Because they are samples, there are margins of error around the reported figures, just as political polls are reported with margins of error. The larger the sample, the smaller the margin of error, but what this really means is that with a substantial probability, the true figure lies within the margin of error.

Probabilities are not certainties, however. If a survey reports a margin of error with a 90 percent probability, this also means that one time out of 10, the correct figure will be much lower or much higher. Even a 99 percent probability means that one time out of 100, the true figure will be much different than the reported figure. One time out of 100 means about once every eight years for a monthly survey.

What seems to have occurred with the September household survey is that one time in 100. The household survey, a relatively small sample, reported an astounding 418,000 jobs jump in the labor force at a time when it has been steadily shrinking, and the survey reported an 873,000 jobs jump in employment at a time when the economy is stumbling.

To be sure, there are instances when the household survey’s result can jump, as when the government gears up for the decennial census or when the Labor Department resets the entire series. Aside from these understood anomalies, none of which apply today, the last time the household survey showed such a huge jump in employment was in 1983 during the Reagan-era economic boom. Today’s economy does not look much like the Reagan boom.

The September household survey is one to set aside to wait for a more reliable report next month, which will almost certainly reverse the odd results from September. If it does, then we have both confirmation of the power of statistics and of the weakness in the economy. If the next household survey is like the September survey, however, then we will know the Obama Administration was playing games with the numbers as alleged. The odds against two such anomalous reports back-to-back are not 100 to 1, but 10,000 to one, or about once in 800 years.