« CAPS Seminar--Larry Bartels, "What's the Matter with What's the Matter with Kansas?" | Main | Tolbert on Same-Sex Marriage Ballot Initiatives in 2004 »

9 February 2006

Predicting election outcomes

Predicting the future as well as explaining the past is often a major goal of political science. Every four years, we're treated to various statistical models going head-to-head to predict the presidential election outcome using various quantitative measures. The recent elections in Canada, however, saw a different kind of test, as different methods, not just different models, of predicting the outcome were put to the test.

Although several polling outfits put out predictions of the seats each party would win based on their final days of polling, of greater interest are two other, more unusual methods. First, the UBC Election Stock Market acts as a financial market, where traders buy and sell contracts representing the political parties competing in the election. The ultimate value of each party's stocks are then used to predict the number of seats each party is likely to win in the election.

Second, the Election Prediction Project solicits information and opinions from the public, which is considered by a panel of political pundits, before making a somewhat informed guess about the likely winner of each riding.

So how did they do?

The Election Stock Market helpfully shows their results compared to other methods. Polling firms differed largely in their results, with the four polls shown having an average error in party vote shares of around six percentage points. In contrast, the Stock Market incorrectly predicted just 15 of the 308 seats - an impressive record, especially given that they extrapolate from national party stocks to individual riding outcomes.

But what surprised me was that the Election Prediction project actually did better than either of these methods, incorrectly predicting only 12 of the district outcomes. For a supposedly less scientific way of predicting the future, the guesses of pundits and the public seemed to outdo the informed opinion of stock market traders.

Posted by Phil Jones at February 9, 2006 10:22 PM

Comments

Several researchers have argued that the Iowa Electronic Market data are more informative that polling of the general electorate. See Shaw and Roberts (2000 BJPS).

Posted by: Barry Burden at February 10, 2006 11:26 AM