The first thing we do, let’s kill all the pollsters. The smart money may still be on Joe Biden to eventually prevail as the 2020 presidential election winner, but I’ve seen enough to declare the polling industry, and its cousin, “probabilistic election forecasting,” as the biggest losers.
I’m not alone in thinking this. Republican pollster Frank Luntz told Axios “the political polling profession is done” after Tuesday. “Political polling is a fraud,” writes conservative John Podhoretz in the New York Post. It turns out that “Trump PTSD” from 2016 wasn’t a paranoid or out-of-touch response; it was a smart defense mechanism. At one point, the night felt eerily reminiscent of 2016, which is to say, the so-called experts blew it. Again. Even if Biden goes on to win, it’s a far cry from the landslide we were led to expect.
Probabilistic forecasting, which is tightly tied to the polling industry, has harmed our understanding of elections and the electorate. That’s because it creates an inherent veneer of certainty, no matter how hard you insist otherwise. We were all better off when folks used to just say “polls are snapshots and things can always change.” The whole allure of data journalism was that it would bring objective clarity, and it just doesn’t.
The problem is that the data geeks and their sophisticated methods are only as good as the polls. Which is to say, they suck. A week ago, there was an ABC/Washington Post poll showing Biden winning by 17 points in Wisconsin. Dismiss it as an “outlier” if you want, but this either signaled that (a) Biden was about to win BIG, or (b) there are some serious flaws in some serious polls. “This is the third race in a row (presidency 2016, governor and Senate 2018, and this) in which Florida polling was almost comically wrong,” complains Podhoretz. Heading into Tuesday night, poll averages had Ohio as a virtual tie. Trump ended up winning pretty comfortably.
But it wasn’t just the horserace numbers in specific states they got wrong. For months, we’ve been told that a major shift had occurred, whereby seniors were abandoning Trump (ostensibly because of COVID-19) in favor of Biden. According to exit polls, this was either wrong—or suddenly changed.
Of course, the data nerds always give themselves an out. There’s always the CYA reminder about how you have to understand probability, which belies their smug confidence—caveats like, “Trump might only have a one-in-100 chance of winning, but you have to understand that one-in-ten chances happen all the time.”
Fine, but when you give Biden a 90 percent chance of winning, as FiveThirtyEight did, don’t act surprised when people expect Biden to win bigly and early. The tweets and the headlines are what count. You can caveat it all you want in the text, but most people are going to think this means the election is a lock for Biden. It doesn’t. As the data nerds will tell you, “If I said you had a one-in-ten chance of dying in a plane crash, you wouldn’t feel comfortable taking that flight.” Well, maybe that’s true, but the last plane crashed and this one has hit some major turbulence. It’s almost like they never learn their lesson.
In the last decade, an entire cottage industry has developed around political oddsmakers. They’ve got computers and data and algorithms and modeling. And you know what? It’s all bullshit. They can’t predict with any more specificity than I can using the 270towin website. This specific industry has proven to be a net negative in terms of its contribution to understanding elections and setting expectations. What they have accomplished is to further erode any trust that Americans have in the news and the experts.
Now, I’m not a pollster, and neither are most analysts, pundits, and commentators I follow. One thing we all have in common is that our political commentary is informed by polling, which is obviously viewed as more scientific and valid than our feelings or anecdotal observations. But it’s garbage in, garbage out.
At the very least, you would hope we would learn from our past (embarrassing) mistakes. We haven’t. They haven’t. We keep making the same mistakes over and over. And not only do the polls drive our analysis, they often drive the news, which is to say the polls are the news. This becomes even more problematic when you consider their track record for accuracy. Having been chastened by 2016, I asked whether the pollsters had corrected their problems and was assured by countless friends in the industry that they had. What happened with Trump, after all, was a once-in-a-generation fluke. The pollsters, I was told, had an obvious incentive to make sure they got it right this time. Besides, there are more high-quality, district-level polls this time around. This seemed to be the consensus of the experts—the Charlie Cooks, the Sean Trendes, the Dave Wassermans, and the Henry Olsens of the world—which is to say, professional political analysts I generally respect. My friend Bill Scher may wish he had this tweet back, as well. Now, each of these people could surely find something I got wrong, as well, because we were all making some of the same flawed assumptions.
Throughout the campaign, there were occasions where my gut told me something (e.g., the Republican National Convention was better than people think, and the urban unrest has to be hurting Biden), but the polls never changed. Indeed, one of the reasons people were so confident in Biden’s victory was that his lead was so consistent. Maybe my gut was right, though. Who knows?
Again, it seems likely Biden will go on to win this election. But that doesn’t exonerate the pollsters and the press. We got many things wrong. To the extent that the media has some blame, apart from believing the polls (our biggest mistake), I think media elites assumed COVID-19 redounded to Biden’s benefit, when the lockdown backlash (especially as it pertains to school closings) might have mattered more. The media also failed to recognize Trump’s success in wooing minorities, particularly Cuban-Americans living in South Florida.
Regardless of the outcome, we in the media need to take a close look in the mirror and do some deep introspection. Both political parties need to do the same. But even if they dodge this bullet and avoid two consecutive presidential races with egg on their faces, the so-called experts simply must confront the reality that there are serious problems. Maybe it’s the fact that people simply aren’t responding to pollsters. Maybe it’s due to the proliferation of cell phones and the decline of landlines?
Maybe it’s the fact that the liberal media, fixated on polls, largely missed a whole conversation going on among Trumpers on Facebook. To put it simply, conservatives dominate the platform, but (unlike cable news hits or paid TV ads, for example), this mostly flies under the media radar. Or maybe it’s simply that our progressive intellectual betters are out of touch with regular Americans—acting as if half the country doesn’t even exist—and are therefore making flawed assumptions about modeling and weighting. These are the same people who scoffed at the notion there might be any “shy” Trump voters out there. Maybe there’s another explanation for what happened, but the lack of humility didn’t allow for even entertaining the possibility. Again, garbage-in-garbage-out.
This is going to require an autopsy, even if the patient lives.