Study: When election polls fail, models may prevail

Nathalie Tadena

According to a new study conducted partially by Northwestern professors, economic models are more accurate than public opinion polls at predicting election outcomes.

The research team analyzed prediction markets, a model that is used to forecast future events. According to data from the Iowa “winner-takes-all” market that the researchers studied, traders can buy contracts that signify which candidate they want to win.

In futures markets during the 2008 election, traders could buy shares in what candidates they thought would win. Those markets judged President Barack Obama as more likely to win throughout the campaign.

This year, the media portrayed the race between Democrat Barack Obama and Republican John McCain as very competitive, said Kellogg and political science Prof. Daniel Diermeier. But the markets showed Obama as a much more favorable candidate than McCain.

“When people try to understand events in a campaign, there is a lot of discussion about the public’s perception of candidates and the way in which the population is going to vote,” said one of the researchers, McCormick Prof. Luis Amaral. “Our proposed model identifies some events that signal changes in a campaign.”

The research team was composed of Amaral, Diermeier, NU postdoctoral researcher Saikat Majumder and University of Iowa finance Prof. Thomas Rietz.

“Prediction markets behave like typical financial markets,” Diermeier said. “We can take this as a real-time measure of how respective elections’ fortunes and candidates are affected by certain events.”

The team’s model does not predict which candidate will win or lose, but instead examines how voters’ views on a candidate change as campaign events occur. These movements can indicate the effect of certain events on an election’s outcome.

“If something weird happens, you can use the model to identify events or information arriving up to that marker,” Amaral said.

Though economic models draw from several mathematical assumptions, they can be applied to many fields besides political science.

“No model is perfect, especially political ones,” said NU economics Prof. Jannet Chang. “But the field is growing, and there are very strong and interesting models to explain what we see in the U.S.”

The researchers found that some traders make decisions based on partisan beliefs instead of information about the candidates. One of these traders will still buy a contract for a particular candidate even if the candidate is not likely to win, the researchers said.

The team plans to examine more data from the most recent presidential election in the future.

[email protected]