Predicting election results
November 20, 2005
Posted by on
Interesting article Why are American presidential election campaigns so variable when votes are so predictable? (pdf) by Andrew Gelman and Gary King. First, I didn’t know that presidential elections were so predictable, but apparently they are (close races are predictably close, not-so-close races are predictably not-so-close) based on certain relatively well-known independent variables. Second, the authors make a good case about why voter polls are so variable. Basically, people eventually gather enough information to make a decision based on their real preferences; and at the point (which is usually near the election). Until then, they poorly estimate how they are likely to vote. Thus, the polls really are just snapshots of voters’ identification of a candidate’s match to the voters’ preferences. This means the campaigns and news coverage seem to actually provide the information voters need to judge candidiates; without them, the voters remain relatively clueless about how well candidates match their preferences.
A couple of notes: this doesn’t work for primaries apparently, and it predicts only the likelihood of a candidate winning. In the last election, I presume the fundamental predictions would have forcast about a 50/50 chance for Bush or Kerry to win; if so, the fundamental prediction was correct, even it if show a slight preference for Kerry. I wonder if this was behind the ‘get out the vote’ efforts by both sides in the last election–Bush probably won because he was better able to mobilize voters to actually vote.
(I was woefully wrong about the 2004 election, and this article helps me to understand why.)