The Polls Weren't Missing Anything, They Were Ignoring Everything

FiveThirtyEight's Excuse for Their Error is Worse Than the Flub Itself

So, it turns out FiveThirtyEight was wrong about the election in ways we could’ve easily predicted. Nothing new there, nor about the standard commentary on the nature of polling that came out following the flub. For an example outside the site itself, take the explanation given by Kristen Soltis Anderson of Echelon Insights. In an amiable back-and-forth with Matt Lewis of the Daily Beast, she gently reminds us that polls aren’t really about election predictions at all, but are actually tools for researching opinions on political messaging and policy. 

That’s fine, I guess, except that sites like FiveThirtyEight don’t seem to mind the traffic boost they get once every four Octobers. Which in and of itself seems unobjectionable enough were it not for the fact that they, too, seem to be attempting to mislead consumers. Consider this piece published on Tuesday in which they attempt to cover up their mishandling with the following diversion: 

“In our pre-election survey on the strength of Americans’ social networks, we found that nearly one in five Americans (17 percent) reported having no one they were close with, marking a 9 percentage point increase from 2013. What’s more, we found that these socially disconnected voters were far more likely to view Trump positively and support his reelection than those with more robust personal networks. Biden was heavily favored by registered voters with larger social networks (53 percent to 37 percent), but it was Trump who had the edge among voters without any close social contacts (45 percent to 39 percent).

Now, undoubtedly, the 4Chan crowd with its over-ripe adolescents had a big role in electing Trump in the first place. (The documentary Feels Good Man, focused on the evolution of the Pepe meme, does an excellent job of explicating this very phenomenon.) As with everything FiveThirtyEight presents, the math is good — and a narrative always fits handily into the mold that the data prepares. 

But remember what Soltis Anderson said? The polls are about measuring opinion, not crafting predictive equations. And from that vantage, what we’re seeing here in this cover-up is a failure to address what is really going on: an entire industry is attempting to gather meaningful data using archaic technology. 

When was the last time you were questioned by a pollster? Soltis Anderson reminds us that this question is statistically insignificant (as pollsters are always reminding us all of our opinions are statistically insignificant). But the question remains, albeit colored in a different shade than the one preferred by data mongers. What the question really indicates is how we go about our daily routines of expressing political opinions — and it ain’t by answering questions. 

Much relevant data is suggested indirectly via a host of expressions that appear not in answering a question directly, but instead by communicating culturally in different manners and methods across an array of platforms. How people represent themselves and their (political) personality can be seen all over Instagram, Facebook, Reddit, Craiglist, 4Chan, Pinterest, Tik Tok, Tumblr and even OkCupid. This data is available and is as plain as day for anyone who wants to tune in to hear how the proletariat is convinced that Liberals support cop-killing Antifa, are swaying the election with mail-in voter fraud, enjoy decorating their front lawn with handmade Trump mailboxes, or whatever else you want to know. Similarly, The Federalist turned up plenty of vocal Trump supporters just by visiting bars in the Midwest, even in populated cities like Minneapolis and Milwaukee. They also were able to find surprising amounts of Trump support among Unionized laborers.

Victor Davis Hanson of the Hoover Institute, vocal in his support of Trump, has suggested that he knew Trump was going to win in 2016 because of a dinner he had with Peter Thiel. (He doesn’t name Thiel directly, but hints. It may have been Mark Zuckerberg.) But one needn’t have access to the graphs numericizing the keywords to catch a whiff of what could easily be detected with a good nose to the ground. This talk was published on YouTube, after all.

A polling institution, one might think, should know that. What causes the oversight? I know enough about the motivations of anyone serious about their career that laziness is not the answer. My hunch is that the answer resides in the same realm as the one that makes us uncomfortable during struggle sessions. It’s easier to craft a question that detects a certain ideology than it is to go a-scrollin’ the social media of the very co-workers and family members you typically try to ignore when politics comes up. I mean, who wants to listen to them? They hardly know how to speak properly, let alone espouse a political idea. 

But, you know, they vote. And as Davis Hanson has many times articulated, Trump had a broad palette to paint with when appealing to these sorts of voters. But beyond this, here’s Hanson in May making an even stronger point:

“You know in your own experience, in your circle of friends and associates, maybe even neutrals and enemies, that all of the people that said that they were not going to vote for him (although they were conservative) in 2016, most of them are going to vote for him in 2020. And of the people who *did* vote for him in 2016, you don't find very many who said 'I just can't do it this time around. He tweets too much.'

I understand I’m wrong to expect too much of polling. It is what it is, and it quantifies what it does to the best of its ability. But the inconvenient reality portrayed in the above quote points to a problem that I’ve been wrestling with for months. As poll after poll showed great leads for Biden, I tried to argue over and over that Trump still had a good chance of winning. Over and over I was dismissed. When I considered writing a column on the matter, a confidant suggested to me I had “nothing really to go on but ‘a vibe’.” 

Well, the election is over now, and we have the map to look at as we consider the polls that wrongly contradicted my vibe. It may well be that the data we needed was there, it was simply hidden in plain sight.