In an earlier post on the September employment report, I suggested that, on balance, the data was looking pretty good. In a comment, The Nattering Naybob took a pin to my bubble by reminding us that this data is, let's say, subject to revision. It's a fair point. Let's go to the record:
The picture shows the difference between what the Bureau of Labor Statistics indicates was the actual change in employment (at establishments) in each month from 1997 through 2004, and the estimated value of that change in real time. For example, the height of the bar for January 1997 is derived by taking what we now know was employment growth for that month and substracting the value that was reported in February 1997. (The vintage data comes courtesy of the St. Louis Fed's ALFRED archive.)
Yes, the discrepancies can be quite sizable. But the surprises are just as likely to be positive as negative -- the accumulated revisions for each year turn out to be (in thousands) 438 in 1997, 347 in 1998, 808 in 1999, 145 in 2000, -516 in 2001, -555 in 2002, 358 in 2003, and 227in 2004. Except for the recession year 2001 and it's immediate aftermath (otherwise known as 2002), the misses over this period have on average been in the direction of underestimating job growth.
So, I will consider myself fairly warned, and keep my bubble warily afloat for now.
For a description of the methodology used in constructing and revising the establishment employment figures, follow this link.