I keep hearing that the US economy grew by only 1,000 jobs in December, and I am forced to wonder: how accurate are these employment statistics? Are 1,000 jobs even within the eror margin of these measurements? There are what, maybe 200 million jobs in the US, so 1,000 is 1/200 of one percent. Are the statistics and sampling methods even remotely this accurate?
This ought to get you started:
Bureau of Labor Statistics Technical Notes
According to Table 2-E, relative standard error for total nonfarm employment estimates is estimated to be 0.2%. Total nonfarm employment is hovering around 130 million workers.
However, you’re particularly interested in this “only 1,000 new jobs from November to December” business. There’s at least two things you need to know: this figure is seasonally adjusted, and it is preliminary. Root mean squared error for preliminary estimates (as compared with final estimates) is 50,000 jobs, according to Table 2-D of the technical notes.
What does all this mean? The total change in jobs was assuredly not 1,000. It is an estimate, and it is seasonally adjusted (the non-adjusted change was negative 202,000 jobs, according to the BLS).
So why do it this way, and why is a modest, seasonally adjusted preliminary estimate of any use in the first place? Because if we measure it the same way each month and compare it across time, it can give us an estimate of how the economy is doing relatively speaking. Maybe the seasonally adjusted estimate is wrong, but it’s pretty likely that we’re worse off than we would be if the same estimate was up 250,000 jobs, no?
why would employment be seasonally adjusted upwards in December? You’d think that any seasonal adjustment would be downwards to get the temp holiday jobs out of the statistics.
You’d think that, wouldn’t you? In fact, non-adjusted employment was negative 9 Decembers since 1990. If I were at the office I could boot up my copy of E-Views and run you a quick estimate of what the coefficient would be (or, possibly more accurately, feed the data into the Census Bureau’s X-12 program). As it stands, I can leave you with this bit of advice: all theory is secondary to cold, hard data. For whatever reason, the BLS is adjusting it upward, and has since 1980 at least from a cursory glance at the data (though the magnitude of the adjustment appears to have increased significantly). The adjustment was negative in 1939, which also seems counter-intuitive. Obviously, Christmas related employment is currently completely offset, and then some, by decreases in some other sector, but which one that is I’d have to think about (it shouldn’t be agriculture, this data comprises only nonfarm employment).
You will be pleased to know that there is still a massive upward adjustment from December to January, though (lately, the non-adjusted decrease has been about 2.6 million jobs).