So We Called An Election
By: Ally Bernstein-Naples
On November 3rd The Conard Courant formed its first Decision Desk in order to keep readers informed throughout a chaotic week. We provided live updates and projections as well as comprehensive swing state biographies. After intense deliberation and excruciating calculation The Conard Courant accurately projected the results of the swing states of Florida, North Carolina, Nevada and Arizona before most other news outlets. When election day spilled into election week, myself, Ethan Mathieu, Ryan Lafferty, and Jayanth Karuturi split our attention between our televisions, computers, and phones scrutinizing every exit poll, every campaign update and every Secretary of State, telling constituents results would be delayed yet again, in order to provide you with complete and comprehensive election information. Our first indication that Biden may underperform the polls was when reporting came from Fayette County, Kentucky, showing noticeable disparities between Biden’s performance in the polls, and Biden’s performance in exit polls and state reports . Half-way through the night it was clear to those of us on the Decision Desk, that the blue wave anticipated by some would not be crashing on the shore tonight. Additional exit polling data from the Rio Grande region of Texas clued us in to Biden’s vast underperformance with Latinx populations, and by midnight we were all left wondering how the polls could have been so wrong.
Well, first we have to acknowledge that the polls were not that wrong. In most polls, Biden underperformed by 3-4 points, slightly, but not significantly higher error than usual. In some swing state polls, however, such as Wisconsin’s, Biden underperformed by around 6-7 points on average. There are many theories as to why Biden underperformed in the polls; below are the most common. First, there is the “shy Trump voter” theory from 2016. This argument hypothesizes that many Trump voters feel that they will be judged and ridiculed if they express their support for the president in a poll so when they are surveyed they either refuse to answer, or state that they are undecided. The most glaring hole in this theory is that in red states and counties, where it is much more likely to be socially unacceptable to be a Democrat supporter than a Trump supporter, Trump still underperformed in the polls and overperformed on election day. Most pollsters agree that there is very little credible evidence supporting this hypothesis. That brings us to the “social trust” theory. Supporters of this theory argue that since Trump supporters are known to exhibit lower levels of social trust (faith in the media, government systems, and society as a whole), they are less likely to answer their phones or agree to be polled when asked, leading to an underinclusion of Trump supporters in polling data (there is no conclusive data tying low social trust with not answering the phone, however reasonable inference may lead us to believe this to be true). Another common one is the “weighting by education” theory. Weighting by education is when pollsters extrapolate from data based on the education level of participants; for example, if a pollster gets a hold of 10 non-college educated white men, and 8 out of 10 say they plan to vote for Donald Trump, a pollster uses some algorithm to transfer that to the population of the state; if there are 80 non college educated white men in a state with only 100 people, the pollster would give Donald Trump a very high polling average (in reality the sample size would be much larger, I am reducing it for simplification). The problem with educational weighting is it assumes that the sample of people fitting into a certain educational demographic will be representative of that demographic. While no definite conclusion can be drawn until more exit polling is released, the Associated Press and Edison exit polls suggest that weighting by education overestimated college-educated white women’s support for Joe Biden. The next theory is the “enthusiastic voter” theory. Research finds that the majority of people who answer polls are more politically active than the average American; they are more likely to donate money, phone bank, and attend rallies in support of a candidate. The enthusiastic voter theory hypothesizes that people who are particularly excited to vote for a candidate are the ones most likely to answer polls, leading to an overestimation of support for whichever candidate receives the most enthusiasm and an underestimation of support for the other candidate. Many pollsters speculate that in 2012, Barack Obama outperformed the polls because Mitt Romney had succeeded in eliciting more enthusiasm than him. The final common theory is one born from the Coronavirus pandemic; the “stereotypical democrat” theory. Supporters of this theory hypothesize that democrats were overcounted in polls for two reasons 1. White Democrats are more likely to have white collar jobs where they can work from home than Republicans and 2. Democrats are more likely to be at home social distancing, ready to answer the phone, while Republicans are more likely to be out socializing. Interestingly, the states with the highest rates of Coronavirus also experienced the most inaccurate polling, suggesting that there may be some correlation between Coronavirus and polling inaccuracy.
In reality, it will be hard to determine which, if any, of these theories account for the 2020 polling disparities, however you can trust that if there are any developments, we here at the Conard Courant will keep you updated. Thank you to everyone who tuned in to our election night coverage, and please, stick with us as we continue bringing you comprehensive political analysis.