Thread by @bencasselman

Mismeasurement in the Current Population Survey is worse than we realized -- and it's skewing our understanding of the labor market.
Remarkable paper from Hie Joo Ahn and James D. Hamilton.
papers.nber.org/conf_papers/f1…

I'm going to leave it to the actual economists to walk through the technical details (I'm looking at you, @Claudia_Sahm), but this paper is important for all of us who follow the labor market and work with CPS data. So less technical thread to follow.

First, the blockbuster: Properly measured, the "true" unemployment rate could be 1.9 percentage points higher on average than the official estimate. And in the Great Recession, the gap was >2.5pp.

The participation rate, meanwhile, is 2.2pp higher (on avg) than the official measure shows, and has fallen by less since the recession.

BLS also overstates the average duration of unemployment, and suggests it remains elevated, when the adjusted measure in this paper shows it returning pretty much to normal.

OK, what are the sources of this bias? First, it's critical to understand that the authors are talking about "bias" in a statistical sense -- they are *not* suggesting political interference (and the problems predate both Trump and Obama).

Rather, this is about the inner workings of the Current Population Survey. The details are too long for Twitter, and the paper does a nice job explaining it in (relatively) plain language. But I'll give one example:

The way the CPS works, a household is in the survey for eight total months. But not every household is successfully interviewed in every month. And it turns out that unemployed people are more likely to fall out of the survey -- biasing the unemp. rate down.

That problem, and others they identify, were known, but they're the first (I believe) to add them together and come up with a comprehensive alternative estimate.

Crucially, this mismeasurement is counter-cyclical -- it's worse during recessions. Some of our most important measures of the labor market are failing when we need them most.

The authors note that the employment-population ratio is less affected by these problems: "For some purposes this ratio might therefore serve as a more robust measure of economic slack than the unemployment rate."
cc @ModeledBehavior @ernietedeschi and the rest of #epoptwitter

One final note: This paper and others like it do *not* mean that we can't trust BLS data. They mean we need to be aware of the ways it is incomplete and imperfect (and work toward making it better).

I'm also not sure they change our understanding of the current state of the labor market in any fundamental way: The labor market is very strong, but not quite as strong as the official unemployment rate makes it look. I'd argue we already knew that.

Anyway, fascinating paper. Slides (with the charts in this thread) are here:
papers.nber.org/conf_papers/f1…