Monday, January 12, 2009

A lesson in correlation and causation

One of the foundational texts in voting behavior is the American Voter, published in 1960. This book kick-started the behavioral revolution in the discipline, moving political science away from the study of institutions and legal systems to the examination of individual political behavior.

Campbell et al discuss in the early chapters of the book the funnel of causality. To use their words directly:

"Events are conceived to follow each other in a converging sequence of causal chains, moving from the mouth to the stem of the funnel. The funnel shape is a logical product of the explanatory task chosen. Most of the complex events in the funnel occur as a result of multiple prior causes. Each such event is, in its turn, responsible for multiple effects as well, but our focus of interest narrows as we approach the dependent behavior. We progressively eliminate those effects that do not continue to have relevance for the political act" (p. 24).

What causes someone to vote in a particular way? Lots of things. But the best predictor--the most causally proximate--would be partisanship. It is the best single predictor of political behavior.

I mention all of this in the context of a news story I saw floating about on the web the other day. Apparently, traffic fatalities in Montana were down in 2008, compared to 2007. The reasons offered included increased use of seat belts and more police patrols. And yes, these probably do relate to the decrease in traffic deaths. But are these factors the most causally proximate predictors of the observed behavior (i.e. a decrease in traffic fatalities)? Probably not.

I was surprised that the article did not mention the most likely reason for the decline: fewer highway miles travelled. This, from a press release generated by the US Department of Transportation:

"Americans drove more than 100 billion fewer miles between November 2007 and October 2008 than the same period a year earlier, said U.S. Transportation Secretary Mary E. Peters, making it the largest continuous decline in American driving in history." Read the whole release here.

Now, sure, one might say Montana probably didn't experience as big of a decline. Wide open spaces, big distances, there's only so much travel you can reduce in a state as big and as sparsely populated as Montana. Well, wrong. Montana's August 2008 travel numbers suggest that we had between a 5 and 6 percent drop in travel compared to August of 2007. That's among the biggest drops in the country (18 other states were in the top category). Check out these maps here.

The moral of the story: the simplest explanation is probably the best, and correlation does not equal causation.

No comments: