Deirdre McCloskey is in fine
fighting form, at least by the standards of statisticians:
Good fit is not the same thing as importance. In fact, usually it has nothing
to do with importance.
I think she’s absolutely right. People who use a lot of statistical analysis,
economists among them, tend to gravitate towards the measurable. Often, important
things aren’t easy to measure, and things which are easy to measure aren’t important.
While statistical analysis should be used to examine some a priori hypothesis,
too often hypotheses are constructed around whatever data might be lying around.
And if the data is crap, the results of any statistical analysis on that data
will also be crap, no matter how good your fit.
(Earlier: The
Limits of Empiricism)