New ‘prediction survey’ forecasts Georgia runoff elections

• Link to: Northwestern Now Story

• Georgia residents participated in surveys to predict the Senate outcomes
• New method combines opinion polling with prediction market
• Prediction surveys reward participants for correct predictions, without risking their own cash

EVANSTON, Ill. — A new type of political forecast, called prediction surveys, predicts that both runoff elections for Georgia’s two U.S. Senate seats are likely to turn blue.

Developed by Northwestern University’s Thomas Miller, prediction surveys combine preference polling with the prediction market. Miller partnered with a consulting company to run daily polls among a sample of adults who have lived in Georgia for at least one year.

In the survey, participants identified who they want to win the elections and then predicted who they think will win the elections. For each correct prediction, participants will receive a small cash reward. This incentivizes unbiased predictions without risking cash buy-ins to participate.

As of Jan. 5,  about 1,200 Georgia residents have participated in the survey — conducted with the help of Isometric Solutions, Panel Consulting Group and Prodege LLC. Miller’s forecast indicates that both runoff elections are essentially tied, but numbers have been trending toward blue in recent days. Drawing on preference surveys, prediction surveys, markets and survey participation data, the forecast’s final predictions are:

• Democrat Jon Ossoff will beat Republican David Perdue by 1 point.
• Democrat Rev. Raphael Warnock will beat Republican Kelly Loeffler by 2 points. 

Viewers can follow the updates on Data Science Quarterly and on Twitter. 

Miller was inspired to develop his prediction survey method after running a successful forecast using data from PredictIt, a prediction market in which users bet real money on political races. He noticed that the data appeared biased toward Republican voters and people with higher incomes compared to the general population. Although preference polls have more representative samples of the population, those samples are often small and have a slight democratic bias. 

“Prediction surveys combine the best aspects of polls and the best aspects of the prediction market,” Miller said. “I put them together and split the difference to get the best of both worlds.”

Miller is the faculty director of the master’s in data science program in Northwestern’s School of Professional Studies. He runs The Virtual Tout, which continues to make forecasts from prediction market data. 

Miller said that by running his own surveys, he gains more data. When using data from other polls, forecasters only receive the polling numbers. With all the survey data, Miller can analyze relationships among income, demographics, voting preferences and voting predictions.

“By having all the data, you can do so much more as a modeler,” Miller said.

Earlier in the Georgia runoff campaigns, Miller’s forecast predicted Republicans would snag both seats. In recent days, however, Miller has watched his forecast shift from strongly red to somewhat red to light blue.

“I think voters are continuing to have issues with Trump’s inability to accept the results of the presidential election,” he said. “People, including Republicans, aren’t happy with his behavior. That appears to have an effect on Georgia.”