Data defects

When the Labor Department announced on Oct. 5 that unemployment had fallen to 7.8 percent, bringing it below 8 percent for the first time in 44 months, some Republicans suggested the administration manipulated the numbers to favor Obama’s reelection effort. Former General Electric CEO Jack Welch led the charge with a tweet: “Unbelievable jobs numbers … these Chicago guys will do anything … can’t debate so change numbers.” Welch later admitted he had no evidence of manipulation but maintained the government’s numbers “defy logic.”

At first glance, they do. The report said the economy added 114,000 new jobs in September, but it also said 873,000 more people had jobs. How can that be? The answer is complicated though hardly satisfying, for it highlights major flaws in the way the government gathers data.

First, it’s important to know that the Bureau of Labor Statistics’ (BLS) monthly report comes from two surveys, not one: of households and of employers. These surveys measure different things with significantly different methodologies. The payroll survey is like counting the seats on a bus, while the household survey counts the people. The likelihood that they will agree exactly is exceedingly remote, so they almost never do.

Another key factor is the very different margin of error of the surveys. The survey saying the United States created 873,000 new jobs—the household survey—has a margin of error of 436,000. The survey of employers is larger and more accurate. The margin of error in that survey is about 90,000, and the adjustments in recent months have been upward. The bottom line: We might discover in the months ahead that adjustments will wipe out most of the difference between these numbers.

Other factors play minor but statistically significant roles. Every December the BLS uses Census Bureau data to make population adjustments to its data set. That could mean—for example—in a year when Hispanic growth is strong, they could be slightly undercounted in monthly surveys until the annual adjustment is made. And if Hispanics have a higher-than-average unemployment rate (and they do), undercounting them could depress the unemployment rate, especially late in the year.

So could administration technocrats create a conspiracy of tweaks to give Obama good news just a month before the election? That’s highly unlikely. Conservatives understand that large bureaucracies are difficult to manage. That makes them just as difficult to manipulate. But these dramatic discrepancies point to a need for an update in methodologies that have been around for too long.

People who follow these numbers closely agree with Robert Oak, the blogger behind “The Populist Economist.” He said, “The BLS doesn’t cook their books and this isn’t an October surprise. That doesn’t mean we shouldn’t get our juices going over [the] poor monthly labor force statistics.” He said the Department of Labor needs “much better raw data collection methods,” including larger survey sizes and better correlation to other labor data sources, such as W-2s and IRS tax data. The real problem, Oak said, is not conspiracy but complacency: “Data collection needs to be brought into the modern era.”

Romney bounce

The day after the Oct. 3 presidential debate, in which Romney performed well, the major stock markets rose. Analysts, including Bob Pisani of financial channel CNBC, said the rise came because of a “long-held belief that a Romney win might be better for stocks.” Those reading stock market tea leaves also pointed to the fact that healthcare stocks, which have risen with Obama’s rise in the polls, moved counter to the overall market and fell on the day after the debate. Conventional wisdom is that Obamacare will keep money flowing into healthcare. The markets have decided Romney’s promise to “repeal and replace” Obamacare likely means more competition among healthcare companies.