Friday, January 04, 2013

Apples and oranges and counting workers

OK… this is for labor force wonks only.

So if you read the official press release on the monthly dump of labor force statistics, a headline the state points out is that the count of total unemployed in the region dropped by 2,000 between October and November, and that was the largest monthly drop since May of 1999! A meme some of the media picked up on.

Well… sort of.
If you look back in the news, there were plenty of months were unemployment drops of 2K or much more were reported.  But that data has all been revised and virtually all larger month over month changes were dampened down (which begs a question, what would the 2K unemployment drop have been under the old data?) So it all depends how you look at it.

Some may recall that the state recently switched the method of seasonal adjustment for this data.  I went into that in some detail earlier.  Basically, the state stopped applying their own seasonal adjustment, and instead standardized on the US Bureau of Labor Statistics data.  OK. No problem.  They also did what is a good analytical thing and switched their historical data to reflect the new adjustment, even though it was different from what was reported at the time.  OK as well.   They did that ‘backcasting’ all the way back to January of 2000 which is what the BLS was providing. 

Soo…  is the current unemployment drop the biggest since 1999?  Basically you have apples and oranges.  The new seasonal adjustment clearly smooths out a lot of month over month variability than in the past.  So ove the last decade there were plenty of months where unemployment dropped by 2,000 or more in the region.  But with the seasonal adjustment they went away.   No surprise that the last big jump  was in data from the earlier decade, which reflects the older seasonal adjustment that allowed for bigger monthly jumps in the data.  
How different are the new vs. old seasonal adjustments?  Just compare what the time series looks like before and after January 2000.  Lot's of variation just gone per this graphic of whatis nominally supposed to be consistent data looking backwards.    
Note the whole time series is for seasonally adjusted data.  But there is no need for my highlighting to show where the seasonaly adjustment algorithms differ.  Two pretty different realities. 

And this is not a story of a decade ago vs. now. The data that was coming out earlier last year was really the older data. Lots of contemporeanous month by month analysis of that data over the decade would actually be very different if the data now being reported was used. Basically a lot of apparent 'news' at the time just got wiped away by the new data.

So the punchline?   Know your data.  Goes beyond repeating a number.

For some simpler punditry.  Employment and Labor Force for Pittsburgh are again hitting new all-time highs in November.   I will always argue to look at trends more than the monthly numbers for the reasons above and more.  So it turns out this is the 7th straight month in a row the region has hit a new all-time labor force peak.  I think we can begin to talk about it well beyond any monthly variation. Not that there has been a single mention of the factoid (all time peak labor force in Pittsburgh) by anyone.  Odd. 


Post a Comment

<< Home