No U.S. economic statistic is more important than the unemployment rate, which was officially reported as 6.2 percent in July. Unlike gross domestic product, it’s not subject to revision, adding to its aura. But a new academic paper, highlighted in the New York Times, finds that a long-known source of potential error in the unemployment rate “has worsened considerably over time.” The actual rate may be higher than the reported one, though it’s hard to know for sure, the paper says.
The unemployment rate is calculated by averaging the results from eight separate samples of the population, known as rotation groups. In theory, the people in all eight rotation groups should have the same likelihood of being unemployed, but they don’t. In the first half of 2014, the newest rotation groups had an unemployment rate of 7.5 percent. The oldest rotation groups had an unemployment rate of just 6.1 percent.
That disparity of 1.4 percentage points is huge, considering that economists pay close attention to changes in the jobless rate down to the tenth of a percentage point. The Bureau of Labor Statistics puts a heavier weighting on the older rotation groups, so the official unemployment rate for the first half of 2014 was 6.5 percent.
The new paper (PDF) is important not just because of its finding but also because one of its authors is Alan Krueger, a Princeton University economist who was chairman of President Obama’s Council of Economic Advisers from 2011 to 2013. His co-authors are Alexandre Mas of Princeton and Xiaotong Niu of the Congressional Budget Office.
Gary Steinberg, a spokesman for the Bureau of Labor Statistics, says the agency has done its own research on the topic of what’s known as “rotation group bias” but would have no immediate comment on the paper by Krueger et al., whose title is “The Evolution of Rotation Group Bias: Will the Real Unemployment Rate Please Stand Up?”
The BLS creates a new rotation group each month by drawing a random sample of people from the U.S. population for the Current Population Survey (CPS). A rotation group is included in the calculation of the unemployment rate for four months, then left out of it for eight months, and then brought back in for four months before being dropped for good. The first researcher to notice discrepancies in the unemployment rates of the various rotation groups, way back in 1975, was a statistician named Barbara Bailar, who later became executive director of the American Statistical Association. At the time, the discrepancy between rotation groups was on the order of half a percentage point.
Krueger and his co-authors write that while other economists have delved into the topic, “the magnitude and evolution of rotation group bias in the CPS has not been documented since Bailar’s (1975) study.”
The jobless people in the rotation group that’s being interviewed for the first time are more likely to say they are actively looking for work, so they’re counted as unemployed. In the rotation groups that have been interviewed repeatedly and are about to be dropped from the survey, jobless people are more likely to say that they are not actively searching for a job, so they aren’t counted as unemployed. The increase in the bias over time, the authors say, seems to be from changes in the way people answer the survey questions, and an increase in people not responding at all.
The obvious question is what the real unemployment rate is. The authors say that the groups reporting high unemployment rates correlate more closely with independent measures of labor slack, namely total capacity utilization and the insured unemployment rate. But tweaking the official unemployment rate to give heavier weight to the new groups doesn’t seem to improve the accuracy of predictions of prices, wages, and GDP growth.
