Bad statistics are worse than useless, they’re harmful. Bad statistics mislead. And bad statistics undermine the credibility of good statistics. Unfortunately, the U.S. Department of Labor has been releasing bad statistics.
This morning, it released the latest data on initial unemployment claims. This is potentially an extremely important indicator: At its best, it’s a complete count of all people newly claiming unemployment benefits. Because it’s a complete count, there’s no measurement error. The data are available quickly, so they potentially are a leading indicator. And because it’s a weekly series, it allows policy makers to keep their fingers on the pulse of the economy. At least in theory.
In practice, the Labor Department often releases badly flawed data. This morning, for instance, it reported that initial unemployment claims were 292,000, a decrease of 31,000, and a number fully 35,000 below expectations. It’s the sort of number that might lead you to become more optimistic about the recovery.
But you shouldn’t.
Apparently, two states are in the midst of retooling their computer systems, and so they reported smaller than usual tallies. How do I know this? You could try reading the full data release, but you won’t find it. I’m only aware of it because Bloomberg News's Jeanna Smialek was in the press lockup, and the Labor Department spokesman there read a “technical note” out loud to the assembled reporters, while declining to name the states affected. The department made no such details available to those of us who weren’t in the lockup. Moreover, we have no idea if the same problem also afflicts the count of continuing unemployment claims.
Labor Department officials are publishing bad data, they know where it’s bad and by how much, and they’re not telling you.
In fact, related problems arise reasonably often. The problems arise because sometimes changes in initial claims simply reflect the ability of bureaucrats to process claims, rather than the number of newly unemployed people trying to file claims. Such difficulties are to be expected. But officials should be completely transparent when they know that bureaucratic issues are distorting their data. Such transparency would allow economists to provide simple statistical fixes, making the data more reliable and useful.
There are other problems, too. For instance, these same data have a habit of being systematically revised upward after every release. The reason is that some data always come in late every week. Rather than making a useful estimate of this, they simply keep assuming -- contrary to all experience -- that there will be zero late data. The result is that we’re often told that initial claims are falling, even when later revisions show them rising.
The deeper problem is that the Labor Departments bumbling leads a skeptical public to become mistrustful of all official statistics, and it yields more fodder for conspiracy theorists. The result is that the outstanding work done by the statisticians at the Census Bureau or the Bureau of Labor Statistics in building trust in their numbers is being systematically undermined by the bad data published by the Labor Department.
Fortunately, there’s a solution. The root of the problem here is that the mandarins currently publishing the unemployment statistics lack statistical expertise. It’s time for these potentially important data to be handled by the independent statistical agency inside the Labor Department, the BLS, which has the expertise to get it right. It’s standard operating procedure at the BLS to make the corrections that are needed, to appropriately disclosure data quality issues, and to not favor those few analysts in the lockup.
This simple shift in bureaucratic responsibilities will yield transparency for the public, and a timely and useful data series that can be relied on by both market participants, and policy makers.
Trustworthy federal statistics are critical to good economic policy. Ask the Greeks.
(Justin Wolfers is a Bloomberg View columnist. Follow him on Twitter.)