Models, Modeling, and Social Science

This is on the first page of
Google Images results for
"models."

Kaiser Fung writes about the difficulties of modeling social behavior, particularly in economics; Andrew Gelman picks up the thread here.

Well, it's always encouraging to hear people much smarter than me say they find modeling behavior to be hard, since I do too. It's much more discouraging to hear people much smarter than me say that modeling is going to be well-nigh impossible. But I think that this is an overly pessimistic conversation. Yes, detecting causal inference is hard; yes, it is probably epistemologically impossible for us to uncover the "real" drivers of human behavior; and, yes, the measurements of human behavior and of the motivating forces of those actions that we routinely deploy are pretty bad.

Nevertheless, the question is not whether we should ask political science to perform as well as physics. (It won't.) The question, rather, is whether we can reasonably expect social science to outperform our intuitions and our folk wisdom, and to become more sophisticated and more certain about at least some propositions over time.

That, I think, is likely. Think of the implications just of (say) the Arrow theorem or the Hibbs bread-and-peace model for understanding American politics. Arrow takes a lot of high-school civics course pabulum off the table, while Hibbs should remind us that much of the churn of polling in the general election is irrelevant even as campaigns are determined, at least in a sense, by the fundamentals that we wish that they would be determined by.

It's odd to see Rodrik cast as the defender of economic orthodoxy, and in particular of quant-led orthodoxy. (Really? That Dani Rodrik?) Yet it's equally strange to see Fung and Gelman so glum about the chances that knowledge can lead to better estimates of both the current social scene and also of the broader regularities of human behavior. Without taking too much of a swipe at social science before, say, 1975 (and I mean from ca. 10,000 BC to AD 1975), I think it's pretty clear that we've learned a lot, and that there are good grounds to think we'll learn more.