Timothy Ni, a research assistant at the San Francisco Federal Reserve, was kind enough to share with me this data he assembled from ALFRED showing the difference over the years between the Bureau of Labor Statistics' initial reporting of monthly jobs gains and its final estimate.
As you can see, though the average error is small, the scale of the revisions on a month-to-month basis is extremely large. What's more is that it's been large for a long time. There's no new problem with the data but little evidence of progress being made on estimating job growth more precisely. Which means, basically, that the monthly initial estimates that make so much news are actually an incredibly poor guide to understanding what's going on in the world. For a lot of data series, the agencies responsible for tracking the indicator in question (be it the BLS, the Commerce Department, the Census Bureau or what have you) don't even try to give these quasi-real-time numbers. If the BLS simply said, "We can't do the data accurately enough that quickly" and did everything on a one-month delay, we'd probably get better initial estimates. But people want to know what's going on as soon as possible. That's a very understandable impulse, and it's nice for the world that the BLS hustles and tries to indulge that impulse. But people ought to understand that the initial estimate is quite rough and ready. In any given month it's the revision to past data that provides the most information.