Thursday, May 08, 2008

Correlation is not Causation

In this election season, commentators keep telling us about polling data. They tell us which candidates are preferred by different ethnic groups, different economic groups, different age groups, groups characterized by their different educational achievements. The elections themselves tell us something about which candidates are preferred by self declared Democrats and self declared Republicans in different states.

The commentators, in the way that they frame their comments suggest causality. They seem to suggest that older women who vote Democrat prefer Clinton because they are older women who vote Democrat, or that blacks in North Carolina predominantly voted for Obama because they are black.

Recall, however, that these variables are not independent. Blacks and Hispanics tend often to be poorer than whites; Asians tend to get more education than blacks; people who reach retirement age often move to states with better climates. So maybe people are voting their economic interests, but the correlation shows up also with race, or people are voting their interests are retirees but the correlation shows up with geographic variables.

Competent social science analysts spend a lot of time trying to figure out which variables show the best predictive ability, but even that variable found to best predict election outcomes is limited by the sets of measurable indicators and the ability to obtain data. Good social scientists are much more reluctant to attribute causality to correlations than are the talking heads on TV!

The problem with the facile framing of the election by the pundits may be not so much that it is substantively erroneous as that it may be morally wrong. They do not suggest that people as individuals make reasoned decisions as to which candidates will do a best job for the country, based on their own information and their own points of view, but rather that they act in groups through prejudice. The pundits may well be promoting divisiveness in their audiences through the way they incorrectly explain voting behavior.

No comments: