## Thursday, May 30, 2013

### Statistical noise and labor force math

Take a second and actually look at these numbers.

The state reported yesterday that the unemployment rate in the Pittsburgh MSA dropped by a tenth of a percent from 7.2% in March to 7.1% in April.  Note that the Pittsburgh MSA is in its entirety made up of 7 component counties.  The unemployment rate changes in each of those counties is as follows:

Allegheny - increased 2/10ths of a percent

Armstrong - stayed same

Beaver -  increased 2/10ths of a percent

Butler - increased 3/10ths of a percent

Fayette - increased 1/10ths of a percent

Washington - increased 2/10ths of a percent

Westmoreland - increased 2/10ths of a percent

Hmmm.....  I don't think I could devise a weighting scheme that makes all that consistent without violating a fixed point theorem of some kind somewhere. Yes, there is an explanation of sorts for it all, but why try to rationalize?

OK, ok....   not to get into it too much, but it all is saying that the seasonal adjustment factors used by the federal wonks are so different from the seasonal adjustments made by the state wonks that the region's unemployment rate could differ by as much as 3/10ths of a percent between them.  Compare that to the punditry and parsing based on on monthly changes of just 1/10th or 2/10ths of a percent.

Unknown said...

Isn't this just an example of Simpson's paradox?