The Hanover Collaboration

Statistics is changing. All fields of knowledge change, sometimes quickly, sometimes slowly. Statistics has been changing quicker than usual over the past two decades or so.

I think that it is becoming much harder to justify including statistics as a discipline within the field of mathematics. (Others, much more learned than I, have made similar observations .)

Lately, statisticians have become too wrapped up in their identity as mathematicians. This has meant that much of the really exciting work with data being done today is being done by people with CS backgrounds. It’s pretty depressing to read a lot of statistics journals these days. It seems like much of the work being done is filling in theoretical gaps that have relatively little impact on real world data analysis. Many of the details being investigated mathematically assume that a stochastic model is right, and then try to make the estimators better. But I seriously doubt that the biggest limitation of multiple regression analysis in social sciences is really the relative efficiency of one estimator over another. (I’m generalizing; not all research is like this, obviously.)

What makes this worse is that the actual math underlying statistics (beyond the very basics) is just plain boring. There’s a reason that statistics papers put their proofs in the appendix: they aren’t needed to understand the result!