Here’s a paragraph I wrote about fifteen years ago.
The culture of economic expertise places inadequate weight on integrating insights from multiple perspectives, that it frequently places an unreasonably high ‘burden of proof’ on heterodox views, and that it has a penchant for spelling out the normative implications of its analysis by way of ‘peremptory rules’ which narrow the scope for dialogue.
This is just a sketch of a post, because I don’t have time for more, but for some time I’ve wanted to put down a marker about one of the ways in which results from Kaggle are instructive. The winners of Kaggle competitions often win not by trying to build the One True Model of the phenomena they’re trying to model, but rather by building a large number of models all of which have some explanatory power and which are independent of each other and then aggregating their insights. Random forests approaches often perform very well. One might call it a ‘wisdom of algorithmic crowds’ approach.
In any event, this is the antithesis of much economic modelling which involves reaching for the One True Model and then parameterising it.
I was reminded of this by David Colander’s latest piece urging humility on economists (pdf). His proposal? That instead of thinking of themselves as analogous to dentists (which Keynes suggested) economists should think of themselves as engineers:
An engineer’s approach to modeling such a complex system as the macro economy would likely focus much more on statistical models and methods of pulling patterns out of the data. It would explore a wide variety of formal models to gain analytic insight, and then would integrate the many variety of models with the statistical models to interpret the patterns. That would involve a fundamentally different way of doing macro and of thinking about macro problems. Similarly, with micro. Economists focus much of their applied micro policy discussion on Pareto optimal solutions even though we know that all actual policies will violate Pareto optimality. We can contort our micro policy models designed to provide Pareto optimal solutions to provide insight into non-Pareto optimal solutions, but, generally, that contortion comes at a cost. It means that we spend less time discussing other models that betting fit not Pareto optimal solutions, but “reasonable person solutions” that more closely reflect society’s value judgments. An engineering applied microeconomics would likely have an entire branch devoted to measuring society’s value judgments and integrating those judgments into applied policy instruments. Our scientific applied micro leaves the topic almost totally undiscussed.
Other required reading on the point is this paper entitled “Identification Problems in the Social Sciences and Life”.
I made the latter point in a discussion of discursive collapse a while back.