Election Analytics 2016: Forecasting Presidential & Senate Races
9/7/2016 Update: Now that the primaries are over and the election season is in full swing, Election Analytics has added its analysis of the U.S. Senate races. This model correctly predicted 31 of the 33 Senate races in 2012, including all 19 of the so-called non-battleground states. Currently, there is a 27% chance of a tie for control of the Senate.
Original story, posted 8/8/2016:
Students at the University of Illinois at Urbana-Champaign—under the tutelage of CS Professor Sheldon Jacobson—have fired up the Election Analytics website, which has been one of the country’s most accurate reflections of both the presidential and U.S. Senate races for the past six years. The analytics and algorithms that undergird the site are what Jacobson sees as setting this election projection site apart from other prognosticators.
“Our stance is that with our methodology, in close races we will be able to provide better insights than others,” explained Jacobson, who is director of the Simulation and Optimization Laboratory at Illinois. “This is the year to test that, because it’s going to be close.
Although the site makes political forecasts, it is non-partisan.
“Our real mission is to advance STEM—Science, Technology, and Engineering, and Mathematics—providing students the opportunity to see how what they learn in the classroom can be used to provide insights in a real-world event, in this case, the national election,” Jacobson said.
“The results from the 2000 and 2004 United States Presidential Election suggested that it can be difficult to predict the winner of the presidential election based on popular vote. In fact, it is possible that the popular vote and the Electoral College vote can lead to significantly different results. To address this, we created a new prediction model based on the Electoral College vote to determine the winner of the next presidential election and Senate races across all 50 states (plus Washington, D.C.).”
The mathematical model employs Bayesian estimators that use available state poll results (at present, this is being taken from PPP, Survey USA, and Quinnipiac, among others) to determine the probability that each presidential candidate will win each of the states (or the probability that each political party will win the Senate race in each state). These state-by-state probabilities are then used in a dynamic programming algorithm to determine a probability distribution for the number of Electoral College votes each candidate will receive (or Senate seats that each party will secure).
Since 2008, Jacobson and his students have gathered polling data from a number of different sources and developed the Election Analytics project to analyze that data and predict election outcomes. The site weights polling data based on a variety of factors, for example, when the poll took place; more recent polls have more weight than earlier polls. Also, polls are weighted more heavily if they have a larger sample. Polls with 1,000 respondents would be weighted more than those with 500 respondents.
In 2008—Obama vs. McCain, the Election Analytics team forecast the Electoral College votes as 359.52 for Obama and 178.48 for McCain. The actual numbers: 365 and 173 respectively. In 2012, the group expected 304 Electoral College votes for Obama and 234 for Romney. The actual numbers, 332 vs. 206 showed how much difference there can between the polls and the final Electoral College tallies.
“We also got 10 of 11 battleground states correct 2008 and 2012,” Jacobson added. “Our model correctly predicted 35 of the 36 Senate races in 2014, missing only North Carolina. The election-day forecast incorrectly predicted that the race in Georgia would be decided by a runoff, but it did predict the overall winner correctly.”
So why does the site focus solely on this year’s Presidential and Senate races? Data quality.
“The strength of what we do depends on the strength of the polls and the quality of the polls,” said Wenda Zhang, a graduate student studying industrial and enterprise systems engineering and senior advisor for Election Analytics. “There’s not enough quality data with the House, so we eliminated those races in 2014.” Alumnus Jason Sauppe (CS PhD '15) is updating and further developing the website along with four computer science undergraduates—Siddhartha Duri, Rishi Jain, Niraj Pant (web developers), and Victor Jarosiewicz (mobile app developer).
The Election Analytics website is updated daily as new polls are published. Website users can select an election and forecast, choose poll filters and swing scenario based on undecided voters (to the left or right), or select candidate combinations—two-party, with or without third-party races.
The ideas and methods used in the Election Analytics website originated in a research paper written by Sheldon H. Jacobson and his collaborators, published in 2009.