Election prediction models simulate and predict the outcome of the upcoming presidential election. As the election draws near, these models draw increased attention as they try to accurately demonstrate the breakdown of votes in the Electoral College.
Different from traditional polls — which attempt to show the outcomes of nationwide or statewide popular votes if the election were to happen that day — election prediction models compile several polls, betting markets and economic data, and taking into consideration trends over time, simulate the Nov. 8 election millions of times.
This process is similar to the simple probability experiments, conducted by many elementary and middle school students. The safest way to get the most accurate results is to conduct the experiment many times.
Using the results from these simulations, the models predict the likely distribution of Electoral College votes. The prediction for which these models are most popular, however, is the percent chance that each candidate will win the election.
A widely used prediction model, FiveThirtyEight, gives Hillary Clinton a 71.2 percent chance of winning, as of Nov. 1.
Another popular model, the New York Times’ Upshot, indicates Clinton has an 88 percent chance of winning the election. The Princeton Election Consortium shows Clinton with a 97 percent chance of winning, while PredictWise — which compiles data from betting markets — gives the Democratic candidate an 84 percent chance of winning.
No major prediction model forecasts a Trump victory, according to the New York Times’ forecast comparison.
The new prevalence of election prediction models likely results from their strengths when compared to singular polls, such as those conducted by major news agencies and universities. The results among singular polls often vary, while models that incorporate the results of many polls are more accurate in their predictions.
The compilation and use of multiple polls is called poll aggregation, and it can produce more reliable results, according to political science professor John Frendreis.
When taking prediction models into consideration, it can be helpful to consult more than one, according to Gregory Matthews, a statistics professor at Loyola.
“[Although the different models] are reaching slightly different answers, they’re all pointing in the same direction. It’s not like we have Trump at 90 percent in one, and Clinton at 90 percent in the other. There’s wide agreement on where this election is [going],” Matthews said, referring to the predicted Clinton victory.
It’s this agreement that draws some students to election prediction models as they read election coverage.
“I read [prediction models] because it makes me less nervous about the possible outcome,” said Amy Al-Salaita, 19. The first-year political science student expressed curiosity in the prediction models as a new form of electoral prediction.
However, Trump voters may be less inclined to trust the predictions.
“I feel that those predictions haven’t accounted for all the recent events,” said Christian Geoppo, a Trump supporter and the Vice President of Loyola’s Republican Club, which does not officially endorse a candidate. By “recent events,” Geoppo refers to the FBI discovery of emails possibly related to Clinton’s use of a private email server, which has been a major weakness for the Clinton campaign.
The 19-year-old sophomore economics major also expressed doubt about predicted voter turnout.
“A lot of people might say they’re supporting Clinton, but after these recent events, they may just stay home from the polls,” Geoppo said.
Ultimately, one poll will matter more than all the others in the end, according to Frendreis.
“The most important poll is the one that results from everybody voting. So, [students] should vote,” Frendreis said.