Election Analytics: Predicting the Next President

4/18/2013

Prof. Jacobson's Election Analytics website employs Bayesian estimators using poll results to predict election outcomes.

Written by

Want to know who's going to win the presidential election this year? So did Sheldon H. Jacobson, a professor of computer science and director of the Simulation & Optimization Lab. To answer the big question, Jacobson, along with a group of computer science and industrial engineering students, created a sophisticated computer model that takes an advanced computing approach to predict the outcome.  Their results are presented online at electionanalytics.cs.illinois.edu.

Illinois computer science professor Sheldon Jacobson
Illinois computer science professor Sheldon Jacobson
Illinois computer science professor Sheldon Jacobson

While most opinion polls touted in the national news media only provide estimates of the nationwide popular vote, Jacobson's mathematical model uses state polling data, running the results through a dynamic programming algorithm before making a prediction. "The results from the 2000 and 2004 presidential elections suggest that it can be difficult to predict the winner of the presidential election based on popular vote," says Jacobson. "In fact, it is possible that the popular vote and the Electoral College vote can lead to significantly different results."

Working with St. Louis University’s Steven Rigdon, Jacobson’s model employs Bayesian estimators, using the available state poll results to determine the probability that each candidate will win each of the states. Presently, he's using state polling data available from realclearpolitics.com, including polls from Rasmussen, Quinnipiac, Gallup, and Survey USA, among others. These state-by-state probabilities are then used in a dynamic programming algorithm, developed with Edward C. Sewell, Southern Illinois University Edwardsville, to determine a probability distribution for the number of Electoral College votes that each candidate will win in the 2012 presidential election.

He has reason to believe that his model provides a more realistic method of predicting the results. In 2004, when most other polls showed Senator Kerry with a clear edge, his model consistently showed a Bush victory.  In 2008, he was able to forecast all but one state correctly (Indiana was the lone outlier, by a very small margin). 

One reason for the accuracy could be Jacobson's method of assigning "safe" states. "We take into account 'safe' states—the states that each candidate is certain to win," says Jacobson. "In 2004, once you took into account Bush's 'safe' states, he had a much narrower gap to close to get to 270 electoral votes than Kerry.  In 2008, Obama’s safe states guaranteed his victory.” In Jacobson's model, a safe state is one in which the candidate is computed to have an 85% chance or greater of winning the state.

Another interesting feature of Jacobson's model is its treatment of undecided voters. "Undecided voters can have a significant role on the outcome of the election. In fact, they are likely to be the ultimate deciders of who will win this election," says Jacobson. The model accounts for five different voting scenarios involving undecided voters, each considered individually. A "Neutral" scenario provides an unbiased handling of undecided voters. "Strong Republican" and "Strong Democratic" scenarios provide two extreme envelopes around which results can be judged and evaluated, while "Mild Republican" and "Mild Democratic" provide more realistic possibilities if late-breaking information surfaces that shift or sway voter preferences.

University of Illinois Computer Science Ph.D. candidate Jason Sauppe and undergraduate students (Calvin Shipplett, Dimitriy Zavelevich, Taylor Fairbank, and Bhargava Manja) update the latest predictions at electionanalytics.cs.illinois.edu. Industrial Engineering student Angela Ding also participates in the project.

So what outcome is the model predicting today? In all five voting scenarios (Neutral, Mild Republican, Strong Republican, Mild Democrat, and Strong Democrat), President Obama is predicted as the winner. However, in mid-September 2008, Senator McCain was predicted as the winner, but by the end of September, President Obama was the new predicted (and eventually) winner.  Therefore, Jacobson warns, opinions can change quickly. The web site forecasts will be updated daily as we approach election day.


Additional Coverage:


Share this story

This story was published April 18, 2013.