Over the weekend a NYT op-ed on “physics envy” in the social sciences made the rounds.* At the same time, a blog post (also at NYT) from January about the rapid expansion of statistical techniques made the rounds (via @brendannyhan). I find this interesting because the first is criticizing empirical models, while the latter is singing their praises.
These two pieces do not necessarily present a fundamental contradiction, since the second (earlier) piece does not have social science exclusively or primarily in view. However, it is worthwhile to spend a bit of time trying to reconcile these two perspectives–the critique of statistical social science on the one hand, and the surging popularity of statistics more generally.
The simplest possibility is that physics envy is driving the popularity of statistical social science. This seems improbable for two reasons, though. First, social scientists have been “teching up” for quite a while, employing more and more advanced models over the past twenty years.** Second, the current hype (not intended derisively) around statistics is much broader than the social sciences, including such fields as biology and finance in addition to economics, political science, and the like.
A second possibility is that the hype is driven by results: businesses are investing in analytics because understanding customer/market data is useful. There is some risk that the current popularity of statistics may drive a bubble, resulting in the application of sophisticated techniques even when the juice is not worth the squeeze. But right now there is probably much low-hanging fruit to be had simply by analyzing data that companies already have or could easily obtain. One objection that could be raised to this is that we have not seen dramatic, conclusive discoveries in empirical social science–for every published statistical finding, there is approximately one opposite finding, sometimes using the same statistical method or the same data. I personally do not endorse this criticism wholeheartedly, but it is sufficient to reject the second explanation as a reconciliation of the two viewpoints with which we began.
The third alternative is that social scientists (and others) see promise in the results of contemporary empirical work, not just satisfactory answers in extant results. The “physics envy” piece fails to recognize the exciting developments at the frontiers of statistical science and computational social science. Methods such as ensemble Bayesian model averaging or computationally-aided mechanism design are serious attempts to account for the complexity of social behavior. Ignoring these developments in order to rehash an old, tired argument is counterproductive.
*Erik Voeten comments on it at TMC, but I’m waiting until after this post to read his thoughts.
**As should become clear later in the post, this refers to the frontier of the discipline(s), not the general level of statistical acuity.