The big economic news to start the week: The U.S. factory sector slumped unexpectedly in May.
The bigger news: No, it didn’t.
The Institute for Supply Management on Monday reported that its manufacturing index, a measure of U.S. manufacturing activity, fell to a three-month low of 53.2 in May. The drop was unexpected — most economists had expected a modest increase — and markets immediately fell on the news.
At about 11:25 a.m., nearly 90 minutes after the original report, the ISM issued a correction: The index had actually risen to 56, a five-month high. Hooray! Markets quickly rebounded, more than regaining the ground they’d lost.
Then, 45 minutes later, another correction: Never mind, the real figure was 55.4, a modest improvement over April and pretty much what economists had expected in the first place.
A bit of background: The ISM index is little known to most members of the public but is closely watched by economists and investors, in part because it comes out in the first week of the month, before most government data. The privately produced index measures manufacturing activity by polling purchasing managers, the executives responsible for ordering supplies, who in theory should have an up-close look at businesses’ activity. (The institute also releases an index of service-sector activity.) Readings above 50 represent expansion; the higher the number, the faster the pace of growth.
The source of the error appears to have been in the ISM’s seasonal-adjustment calculations. Seasonal adjustment is tricky and often leads to odd results, but this wasn’t a case of a quirk in the formula. It was just an old-fashioned screw up. The institute somehow applied April’s adjustment formula to May’s data. Oops.
The institute hasn’t returned my phone calls, but in an interview with Bloomberg, the chairman of the group’s survey committee, Bradley Holcomb, blamed a “software error” for the mistake. It’s unclear why the group got its first correction wrong.
Credit for catching the error appears to go to Kenneth Kim, an economist at Stone McCarthy Research Associates. The investment firm highlighted the mixup on Twitter within half an hour of the release:
I talked to Kim, and he said his first reaction to the ISM report was the same as everyone else’s: “Wow, that’s pretty weak.” But then he dug deeper: “As I started plugging in their figures, something wasn’t adding up.”
Kim said he assumed the mistake was on his end, not the institute’s. But even after double- and triple-checking his formulas, he couldn’t make his numbers match the ones in the report. Almost on a whim, he tried applying April’s seasonal formula, and everything fell into place.
There’s a lesson here for data watchers: When economic numbers depart from expectations, don’t assume it’s the expectations that were wrong. The mistake could be in the numbers themselves. Most of the time, of course, the errors won’t be as clear cut as the ISM’s, and even when they are, casual readers might not be able to catch them. But you don’t have to be an expert to view surprising data with appropriate skepticism — a theme we’ve hit on before. It’s a central tenet of the Bayesian approach to thinking about statistics.