Statisical Aspects of Fool Employment

by the Sandwichman

Dean Baker points out that among the discrepancies in the "better than expected" April employment number from BLS is the fact that the "birth/death" adjustment for April 2009 is 50,000 jobs higher than it was for April 2008 -- an unlikely eventuality.

The explanation for this is mind boggling. For the most part, the BLS simply assumes that the more firms go out of business, the more new firms start up...

To account for this net birth/death portion of total employment, BLS uses an estimation procedure with two components: the first component excludes employment losses from business deaths from sample-based estimation in order to offset the missing employment gains from business births. This is incorporated into the sample-based estimate procedure by simply not reflecting sample units going out of business, but imputing to them the same trend as the other firms in the sample. This step accounts for most of the net birth/death employment.

So, for example, if 227,000 jobs were lost at firms that went out of business in April but other firms shed only about .5% of their employees that month, then the BLS pretends (roughly) that new firms created 226,000 new jobs (with some minor adjustment to take into account the historical trend over the last five years). This actually might make some kind of sense in normal times, because the BLS survey systematically misses employment at firms that are starting-up. But there's a recession on. And a credit crunch.

More egregious than the questionable BLS imputation is the tendency in media commentary to interpret the result as some kind of turning point. 539,000 job losses is huge, but it's an improvement over the 699,000 jobs lost in March or the 681,000 in February... right? No. It's a further deterioration of the job market at a decelerating rate.

But wait, minus the birth/death imputation the trend is: February 815,000, March 813,000, April 765,000 . And, net of the extraordinary, one-time-only hiring of 62,000 people for the 2010 census, the April figure would be 827,000. It's also worth mentioning that these losses are from a progressively shrinking base, so a given number of job losses represents a larger percentage loss.

[Correction: the birth/death model imputation is not seasonally adjusted, so, strictly speaking, it's not cricket to subtract it from the monthly unemployment numbers, which are seasonally adjusted. But since my point is mainly that the 539,000 figure is no cause for celebration, these numbers will do as 'ballpark estimates'. The BDM is, after all, itself a sort of ballpark estimate of an unknown number.]

Discounting for statistical anomalies, the trend appears to be one of accelerating job losses. The best that can be said for the April situation is that the acceleration of job loss is within the margin of error. Whoop-de-doo!