This article is a truly fantastic critique of how modern social sciences (besides economics) are carried out and publicized in modern society. Any critique one dislikes - typically "your theory predicts X which is false" - can be dismissed by "nuance", "things are more complicated", "that's different in unspecified way", etc. In statistical terms, these dismissals are little more than a call for overfitting - if I ca…
> In statistical terms, these dismissals are little more than a call for overfitting. can arbitrarily add more dimensions to a theory, I can make it fit any data set.
In order to fit any social data set, you may need dozens if not hundreds of variables. If your data set contains many millions of data points, telling you that there may be four variables instead than two is certainly not "a call for overfitting"; it is no more excessive nuance than the entirety of special+general relativity. Or, don't use this as an excuse to stop questioning your theory once it predicts behavior in one setting. You may have completely missed the mark by misjudging the order of the equation. Let alone, when your theory doesn't even fit the data we have -- just the data you have -- it's nothing more than amateurism; don't confuse amateur smart-assedness with any sort of profundity; it is often wrong in the most boring way: uninformed wrong.
Intuition that comes from years of study of human behavior may help a lot, but some believe they can come up with theories without spending the time to earn their intuition first. Whether overly nuanced or over-simplified, I think the chances that any of their theories are correct -- especially if they put little effort in forming them -- are low.
And one thing is certain: the less you know, the far more likely it is that you'll err on the side of over-simplification than over-nuance. This paper is a warning to experts, who know enough to err with over-care. Amateurs (of the non-humble kind) are indeed completely protected from this particular kind of error by their ignorance, especially if they don't have the basic tools to even assess the system's level of complexity.
This paper is a call against obsessive, exaggerated nuance. Not a call for oversimplification and sloppiness. And while one can err on both sides, historically, errors of over-simplification (when it comes to social theories) have been much more detrimental to the human race than errors of nuance. The latter kind may cause boring science; the former has led to atrocities. I think that many of the calls for more nuance are just warnings against careless errors of the first kind, simply because they are far more dangerous than errors of the second kind.