Thread by @xftrebbi

Many students ask me why I work with structural methods, that is theoretical models mapped directly to data, parametrically or nonparametrically. [1/n]

To many, used to think about applied work in economics as mainly answering causal questions – i.e. focusing on one parameter of interest or a combination of parameters capturing some causal effect of interest – the assumptions of the models I run appear heavy. [2/n]

Even to some researchers in IO, who mostly focus on policy counterfactuals, structural Political Economy structural papers appear “weird”. Too much focus on estimation & fit. [3/n]

The reason I do is that in most situations I am focused on model selection. We have this enormous amount of theory in Economics for almost any aspect of social behavior, and certainly for political behavior. [4/n]

Yet most of us dedicate minimal time to pruning this space of knowledge of models that are false – not in the sense of the proofs being wrong, but in the sense of not being able to match empirical moments or doing so without parsimony. [5/n]

Structural approaches are great at disciplining model selection. If your model doesn't have forces pushing towards minimum winning coalitions & in the data you mostly observe those, this approach will make that immediately clear with poor fit or nonsensical estimates. [6/n]

In this, structural approaches fulfill the important scientific role of rejecting theories that are not supported by data. [7/n]

We are seeing Theory in Economics journals taking less and less relevance. I think structural methods are the scientifically sound way of going back to theory and giving it back its relevance. [8/n]

We do so by discriminating good theoretical frameworks from others not in terms of elegance, but ability to match the data. Falsifying theory is key to the scientific method. This is what model selection does. [n/n]