…is often to find clever instrumental variables in order to take advantage of what the social sciences like to call “natural experiments”. The basic rationale is that an economist (or other social scientist) doesn’t always have the luxury of running a controlled experiment in the real world as, say, a physicist would in the laboratory. However, in order to reach a valid conclusion, the economist must be able to identify a control and a treatment group, or have some form of exogenous (meaning coming from outside the system in question) variation in order to be able to identify cause and effect rather than just correlation.
The best general interest description of this topic (in my opinion) has come on February 27 in the Wall Street Journal in an article titled “Is an Economist Qualified To Solve Puzzle of Autism?”. (Apologies that the link will only be functional for 7 days.) The article mentions several examples of natural experiments, and I will go through the autism one here. Researchers have found that children that watch more television have higher rates of autism. Many people would jump to the conclusion that television watching in fact causes autism, which is not only incorrect logically but also runs counter to biological knowledge. The general problem in evaluating this cause and effect is that television watching is generally a choice variable, and it could be the case that either autistic children just like watching TV more or there is an outside force that results in both autism and desire for TV watching. A scientist could tease out cause and effect here by doing a large-scale study where children in the control group got no television and children in the experimental group had to watch TV all day. In addition to being potentially prohibitively expensive, this plan also suffers from various ethical issues. This is where economists come in. The economist basically says “gee, it would be really nice if there was an outside factor that randomly placed children into low TV and high TV groups so that I could measure the difference in outcomes that result.” This is exactly what an instrumental variable is. In the autism case, the instrument used is rainfall. The amount of rainfall within a region is fairly difficult to predict ahead of time, so families are not likely to be making decisions of where to live based on this information. That said, children in areas that turn out to be rainer are likely to be inside more, and thus likely to be watching more television. If the economist can show this link, and he can also show that kids in rainer areas have higher rates of autism, then there is fairly convincing evidence that there is a causal link between TV watching and autism. (Full disclosure: biologists are very up in arms over this conclusion, since autism is considered to be a hereditary rather than an environmental condition. The burden is also on the economist to rule out other potential explanations- for example, as pointed out in the article, one could also reach the probably erroneous conclusion that carrying an umbrella causes autism.)
So why am I bringing this up now? An article in today’s NYT talks about variation in medical treatment by ZIP code. Since people likely don’t choose the specific ZIP code to live in based on the potential medical care available (especially since an earlier article argued that high cost did not correlate with high quality), it seems like there is potential for ZIP code to be used as an instrument for…something. (It is also important to note here that the authors are careful to point out that cost is not correlated with rates of illness or cost of living.) For what? I’m not sure yet, but it’s interesting to think about. Suggestions welcome.
I will follow up with another post relating to the economics of specific points in the article.