Mysteries of the Seasonal Adjustment Process

When we talk about jobs, we’re normally talking about the seasonally adjusted figures. But when we’re talking about the weather, I think it makes sense to remember the difference between the actual count and the seasonal adjustment algorithm. Brad Plumer talks about construction and the cold and says, “Construction jobs actually dropped by 7,000 in March after growing the previous two months.”

But look at the unadjusted data. They show construction adding a bunch of jobs in March. The seasonally adjusted series says construction employment dropped because the February-to-March increase wasn’t as large as the seasonal adjustment algorithm would have predicted, perhaps because the unseasonably warm February through off the normal seasonal cycle of construction work. Or perhaps for some other reason. You could drive yourself insane thinking too hard about this stuff. The fundamental point is that while you hope for a better number than a worse one, the closer you peer at the details the more you’re drowning in a swamp of possible sampling error and murky adjustment considerations.