Still Don't Trust the Jobs Report? Here's How to Build a Better One

Here's how to Welch-proof the jobs number in two easy steps


Jack Welch needs our help. The former CEO of GE has had a hard time understanding the September jobs report. Torn between the competing hypotheses that falling unemployment was a case of mean reversion or a shadowy conspiracy on the part of career civil servants, Welch has chosen the latter. Because ... Occam's Razor? Of course, this isn't to say we can't make the jobs number more accurate. We can. Here's how to Welch-proof it in two easy steps.

First, a big caveat. There's no need to Welch-proof the jobs number. House Oversight Chair Darrell Issa is concerned that the "constant revisions -- significant revisions -- tell us that it's not as exact a science as it needs to be," but the reality is the Bureau of Labor Statistics (BLS) already does a pretty magnificent job. As University of Michigan professor and former Labor Department chief economist Betsey Stevenson told me, it's important to think in terms of changes to the level and not just revisions to the change. In other words, the revisions sound big because we compare them to the initial change, but they aren't big compared to the 143 million people currently working. They're a rounding error.

But they're a rounding error we could eliminate. It's just a question of how much we're willing to pay. Right now the BLS gets its job numbers from a pair of polls, one of establishment businesses and the other of households. The first, and easiest, way to increase accuracy would be to increase the survey size and increase the number of questions -- that is, reduce the margin of error.

This is straightforward enough, but we can do better still. Justin Wolfers, a professor at the University of Michigan and Betsey Stevenson's husband, points out that the establishment data actually comes from two sources. There's the monthly poll, and the annual benchmark revision -- the latter of which looks at so-called "universe counts" of who has paid into unemployment insurance over the previous quarter. In other words, it's a very precise measurement of employment rather than a very precise poll of employment. What kind of difference are we talking about here? Well, in the year ending in March 2012, the benchmark revision found 386,000 more jobs than the establishment poll had found. That's not nothing.

Obvious question time: If the benchmark data is better than the establishment survey data, why don't we just use the benchmark data? The answer is one part infrastructure and another part convenience. The crazy thing is every state has a record of how many jobs there are, but we can't get to the records fast enough. The data are stuck on different, sometimes antiquated, computer systems, and only come in every three months. We could revolutionize how we calculate the jobs number if we harmonized these computer systems across all 50 states and required businesses to file unemployment insurance payments every month instead of every quarter. It would become an administrative, rather than a statistical, task, as Justin Wolfers put it to me.

There is, of course, a rather large irony here. The Republicans are effectively attacking the BLS for not being invasive enough when last year they attacked the American Community Survey (ACS) for being too invasive. The ACS is one of those boring but essential government functions that policymakers and business leaders alike can't do without -- it's an annual survey of three million households that covers everything from demographics, internet usage to how often they flush the toilet. Despite its 200-year history, House Republicans decided the ACS was unconstitutional, if not worthless since the surveys are "random" and "not scientific". (Pro tip: The randomness makes it scientific).

So does this brouhaha over the BLS mean conservatives care more about increasing jobs accuracy than about increasing the regulatory burden on businesses? I'm just asking questions.