Behind the scenes of the Blair ORACLE

On Saturday, Nov. 7, Joe Biden became the 46th President-Elect after a nail-biting few days. In the wake of the 2020 Presidential Election results, political observers and professional election predictors alike are now comparing which election forecast model most accurately predicted Biden’s win percentage. Blair’s Political Statistics class, led by teacher David Stein, is no different. 

ORACLE of Blair

Photo: ORACLE of Blair forecast model predicted Biden win at 93% (courtesy of polistat.mbhs.edu)

ORACLE of Blair forecast model predicted Biden win at 93% (courtesy of polistat.mbhs.edu)

Starting in September, seniors in Political Statistics created a Presidential election model, named the Overall Results of an Analytical Consideration of the Looming Elections (ORACLE) of Blair.  The ORACLE predicts the statewide and national win percentage of both candidates, and the model is complete with interactive maps, timeseries’, electoral vote distributions and blog posts written by students analyzing specific outcomes and trends. With the conclusion of the 2020 Presidential Election, students are now learning statistics metrics to compare how well the ORACLE did in its predictions compared to professional models created by 538 and TheEconomist.

In the midst of a COVID-19 pandemic, the 2020 Election has been a historical election, with many notable election moments like Texas surpassing 2016 vote numbers despite the pandemic. Despite accurately predicting Joe Biden’s national win, students who worked on the ORACLE model note many major flaws with their model, the biggest one being polling data.

According to the New York Times, election polls in 2016 infamously hugely under predicted Trump’s win in key states as well as nationwide. And with most votes counted and accounted for, it seems as if the polls missed the mark again this 2020 election cycle, albeit not as blatantly as in 2016. The ORACLE incorrectly predicted that Biden would turn sunbelt states such as Florida and North Carolina, and in the historically red states that Biden did win, like Michigan and Wisconsin, the vote percentage for Trump was significantly higher than predicted. 

Senior Carlos Hernandez, who worked on poll weighting and theoretical aspects of the ORACLE model, states that the polls were the deciding factor in such an inaccurate picture of the election. “The polls were really, really, really off, and it's hard to make a good model with garbage data," Hernandez comments. "Polling inaccuracy is something that our model did account for, but we concluded that a massive polling error would be needed to give Trump a shot at winning, and that was what we ended up getting." Hernandez explains that most good models do keep inaccurate polls in mind, and that the ORACLE model handled this polling error as reasonably as could be expected. 

Polling error hinges on a multitude of factors including incorrectly predicting that all Latinx voters would lean Democratic, when in reality, Cuban and Venezulan voters vote differently than immigrants from Mexico and Central America. “I think it's definitely related to the country of origin. I was born in Venezuela, and I wasn't surprised that many bought into the anti-socialist narrative Trump was pushing. Georgia was a major surprise for me. When I saw the situation in NC and FL on election night I thought for sure it was going red,” Hernandez continues.

Polls also didn't account for the Trump voters who are uncomfortable publicly supporting him, and misrepresented demographics that swing heavily for Trump.

Despite polling inaccuracies underpredicting Trump’s vote count, the biggest win for seniors working on the ORACLE was the opportunity to put together a professional-like forecast model in less than one month. Senior Kevin Higgs says that for a high school team, he is very proud of the work that the Political Statistics class was able to put together. “I wanted to join Political Statistics because I'm always intrigued by these real world sorts of projects. I was excited by the prospect and it lived up to my expectations,” Higgs says. 

Creating an election forecast model required tremendous work on many fronts to not only create an election model but also develop a website and design the graphics. “I was most involved in developing and hosting the website. I pulled an all nighter the night before it went live because I had to implement a ton of features and incorporate everyone's feedback,” Higgs concludes.

The ORACLE of Blair model saw major wins and losses this election cycle, but the seniors hope their model will increase awareness for the Political Statistics class as well as encourage other high school students to take on similar projects. The ORACLE has been retweeted by Rep. Jamie Raskin and Founder of Center for Politics, Larry Sabato.

To see how the ORACLE of Blair did in 2018 races, read our article here.