Two sides, no vig: The problem with generative and inferential reasoning in social science

Above is the title, and here’s the abstract:

The logic of social science can work in two directions: generative modeling predicts behavior given assumed preferences, and inferential reasoning deduces preferences given observed behavior. Both these modes of reasoning can be valuable, but in choosing which mode to use, social scientists have the freedom to come to essentially opposite conclusions for any problem that comes in. We give examples from economics, political science and public health. On the plus side, recognition of this danger has the potential to improve social science for those researchers who come to terms with this disjunction and move to a more holistic integration of data and theory.

Now I just have to write the paper. The idea is to have a cleaned-up version of this post from a few years ago and various related material. And here’s a good example from Tyler Cowen. Just to be clear: by calling that “a good example,” I’m not speaking ironically and saying that Cowen’s post is a good example of mistaken thinking; rather, I’m saying it’s an example of a good discussion of the problem, and it’s an example I’d like to use in this as-yet-unwritten paper.