How the FiveThirtyEight model affected my experience of the 2020 Presidential election.
*Note: written week of November 9, 2020, assuming Joe Biden is President-elect.
Election polling has been a significant aspect of American presidential elections since roughly the mid 19th century, when representatives for candidates conducted “straw polls”, going door to door to get a sense of sentiment. Roughly 60 years later, statistician George Gallup created the first election surveying model based in statistics, county demographics, and sampling. The “Gallup Poll” is still used today.
Today, statistician Nate Silver’s FiveThirtyEight model is one of the more interesting recent developments in political polling. Silver’s model is unique in that it measures the reliability of all national polls, including Gallup, and then rates them according to their accuracy and methods of research. Better polls are given higher value ratings (“A” and “B”) and more weight once the averages of polls in each state are accumulated. Gallup, for example, was given a “B” rating for its model, which FiveThirtyEight determined has a slight “mean-reverted bias” towards Republicans, which is a fancy way of saying their model has historically bent to favor Republicans.
Silver gained fame for almost perfectly predicting all state outcomes in both the 2008 and 2012 state elections. However, in 2016, Silver’s model clearly predicted the favorite Hilary Clinton to defeat underdog President Donald Trump. Although their model did give Trump a roughly a one in three chance of winning, much higher than many other established polls, the site was reported to have gotten the result wrong. FiveThirtyEight, as well as all other polls and election predictors, were grouped into the narrative of the “Fake News Media” who live in a bubble and do not take half of America seriously.
Silver and his team faced a problem entering the 2020 election: they needed to demonstrate how their model is merely a prediction based on an accumulation of national data, not a clear pick for who the winner would be. To accomplish this and give the user a clearer understanding of the process of election data, the team turned to user-based web design.
The 2020 FiveThirtyEight forecast site stands out from other prediction models through its emphasis of narrative and art. First, instead of having a simple percentage line, showing the odds of each candidate winning, the model visualizes the percentages in a unique and aesthetic ways. Instead of a single bar with numbers, the model pulls 100 possible sample electoral map outcomes from the simulations they run to show possible ways the race could go. Thus, instead of giving President Trump a 33% chance of winning, they show 33 ways in which Trump could, based on the data, win the election. This visual shows how poll predicting is still often unpredictable, more like rolling a dice than definitively picking the winner.
The coupled narratives reflect the relative unpredictability of the election. In fact, on election day, the site even reminded the reader that the model shows Trump’s odds were “about one in six, the same odds as rolling a one on a six-sided dice.” Most of the site’s narrative is conveyed through a cartoon fox character, FiveyFox, on the margins of each design. The same way Steve Jobs wanted a handle on the Macintosh to make computers less intimidating, the cartoon character makes the model feel more approachable, like something even a child could comprehend. FiveyFox explains each map and what it signifies. He also constantly reminds the reader that “upsets can happen” and not to count the underdog out.
Why is this significant? Despite what many say, history has proven that predictions and opinions from experts do, in fact, affect the result of the elections. In October 1864, President Lincoln wanted to appear to not be involved in campaigning or predications. He wanted the decision to be left to the people, saying, “The will of the people is the ultimate law for all…it is their own business, and they must do as they please with their own.”
Even back then this principle was a bit idealistic and naïve. Straw polls were being conducted, and from every election since then, people have been influenced to act or not based on what they hear about the election from others. Those who hear results will be close in their district may do more to rally their neighbors. Those who hear their candidate will lose in a landslide may not bother voting at all.
In the days following the election, Republicans were, somewhat legitimately, outraged at Fox News (No relation to FiveyFox) for calling the state of Arizona “prematurely” for Joe Biden. The race was close, and it was reported that people waiting in the long voting lines saw stories of the race being called and many chose to step out of line and not bother.
Our culture is being flooded by misinformation at the moment, and it may be our largest threat to the healthy continuation of our democracy. Fair elections are the cornerstone, the most fundamental aspect of who we are as a nation. I am grateful for the FiveThirtyEight team’s thoughtfulness in presenting users with a responsible explanation for how election polling works. Without them, the process would have felt mysterious and unclear to me, but they showed what is really behind the curtain while using facts, numbers, and statistics. The site’s design was transparent and effective, and assured me that, at the end of the day, no scary greater powers are at work. Elections are still our own business.
Comments
Post a Comment