Here's an interesting finding from Jed Kolko, the stronger the labor market recovery in any given metropolitan area the worse Obama did in 2012 relative to 2008.
I think we should see this first and foremost as a case study in the perils of univariate analysis. Kolko also finds that racial and ethnic composition issues impacted not only the level of support for Obama in a given place but also the rate of change. The whiter the place the worse Obama did not just in absolute terms but relative to his 2012 performance. It seems to me that for basically coincidental reasons, some of the biggest falls in unemployment have happened in very white places. It'd be interesting to rig up a much richer collection of data and see how the change in unemployment impacted the change in Obama support after controlling for various other demographic variables that we know to be politically relevant.