The Economist has an interesting article, “No City for Old Men,” about the inexplicably high death rates in Glasgow, Scotland. After punching holes in just about every explanation for these high death rates, ranging from deindustrialization to poverty to income inequality, it concludes:
The “Glasgow effect” may well be a problem without a solution. It is as if a malign vapour rises from the Clyde at night and settles in the lungs of sleeping Glaswegians. Improvements are likely to emerge from the cumulative effect of many small changes, rather than from a single big thing. Just do not expect anyone to be able to explain it when it happens.
There is another possible reason, which is that there is no reason. This simply could be stochastic variation. In these types of analyses, there is never a perfect correlation among variables (“R” doesn’t equal one), nor could a model ever fully account for all of the variation among different cities (“R2”). In other words, somewhere has to do better than expected, and somewhere has to do worse. Poor Glasgow just happened to get the shit end of the stick.
Just something to keep in mind when you read about these kinds of analyses. They have their utility, but it’s foolish to think one can account for every last drop of variation.