Where Economists Dare to Tread

Over at The Frontal Cortex, Jonah has a blog referring to a WSJ article impugning economic jurisdiction in questions outside the traditional bounds of economics. Specifically, the article cites a paper recently publicized by Cornell University claiming to establish a causal link between early childhood television viewing and autism. The thrust of the article is that the statistical tools used by economists are ill-equipped to address such questions and should be treated as suspect by the natural sciences.

The answer, as with all things, is a bit of yes and a bit of no.

I would like to begin by stressing that I do not intend to support the particular TV-autism paper cited in the WSJ. I have read the paper and find it both bad science and bad economics. First among my concerns is that the author purports to use an instrumental variable (IV) analysis in establishing a link between early television watching and autism, and then does not use that technique. He uses precipitation as a proxy variable instead of an instrumental variable, which changes both the meaning and the significance of his results. Whereas an IV regression may have indeed provided some information about the link between TV watching and autism, this proxy regression tells us nothing about it. There's your bad science. The bad economics comes in when he makes recommendations to parents based on the results of his study. A pharmaceutical researcher wouldn't immediately prescribe a new drug to every cancer patient in the country on the basis of one experiment. Economics is at its best when it is descriptive rather than normative until the best prescription is overwhelmingly demonstrated.

That said, I fear that people who readily dismiss potential economic contributions are throwing the proverbial baby out with the bath-water. Properly applied, econometric techniques can give us a very good idea of correlation and perhaps even causation between statistical events. Even the much-maligned instrumental variable approach, if carefully constructed and convincing, can offer some information about relationships among observed data relatively cheaply and quickly.

The statistics used by economists were developed over decades as a way to get around the fact that economists can't perform controlled experiments. You can't just call up a central bank somewhere and say, "Hey, would you mind raising the interest rate .25% suddenly? I'd like to see what happens." So we have to make use of data born of so-called "natural" experiments--those that occur either by natural events or government policy--to tease out causal relationships and explanations. We've gotten pretty good at it, with a few caveats. Firstly, none of these techniques is 100% foolproof. There can always be something else that explains the observed outcomes lurking in the background to foil our neat little regressions. This is where peer review and repeated analysis can be a saving grace. For every economist who publishes a model to explain the causal effect of A on B, there's another who looks at the results and says, "Hey, but what about THIS?". Over time, with enough heads together, we can hopefully converge on an explanatory model that satisfies just about everyone, and then start thinking about what we can do with the results.

Secondly, as the saying goes, "there are lies, damned lies, and statistics". It is very easy to use numbers to obfuscate complexity, as in the TV-autism paper. This is where we need qualified people to evaluate research methods and results to separate the good science from the bad before it gets too far. It is very easy for the media to seize hold of a soundbite from a particular paper and make it seem as if the result in question were sacrosanct, when in fact the opposite is true. I often feel that the misrepresentation of what economics can and can't show is what encourages such widespread scepticism of the discipline amongst researchers in other fields.

So, having said all that, what is it that I claim economics has to offer the natural sciences? Jonah hinted at this in his post, as have some of his commenters, but the quick and relatively easy preliminary examination of hypotheses strikes me as one of the biggest contributions econometric techniques can make. No regression, however subtle, will ever substitute for a controlled experiment. But in an time when scientific research is getting more and more expensive, it makes sense to try to find ways to eliminate spurious hypotheses before embarking on huge, controlled studies. And with modern statistics collection practices, we have mountains of data ripe for analysis. In the case of the TV-autism link, it would probably be quite time-intensive and difficult to experimentally establish (or not) a causal link. If someone were to do the actual IV regression on the idea and find no significant relationship, perhaps the time and funds could be better spent pursuing another avenue of autism research. This applies to other areas as well. Indeed, it applies to all the areas mentioned by the article as questionable venues of economic pursuit: education, politics, history, epidemiology. How useful economics is in these areas is more a question of proper application than intellectual turf.

****************************

Finally, I hope Jonah will forgive me for jumping all over what was probably just a throwaway comment, but I have to take this opportunity to clear up one of the fundamental misunderstandings lay people often have about economics. Namely, this whole business of "rationality". No economist anywhere will ever, EVER tell you that human beings are rational. That's just silly, and most of us consider ourselves quite reasonable people. The concept of "ideal rationality" is a tool. It is a simplifying assumption that we ascribe to the agents in our theoretical models because it makes analysis easier. After looking at our analytical results, we consider what would happen if we relaxed these assumptions, and whether or not the outcomes change significantly (and what it means if they do). It's perhaps a facile comparison, but think of it as keeping animals in cages in a laboratory. Maybe the lab environment changes things and maybe not (depending on what you're studying), but it sure makes it a hell of a lot easier to work with them.