Biases in political journalism and election forecasting
An aggressive pursuit of centrism will blind us to truths otherwise evident in the data
A brief aside: I assure you that this is a post about political journalism and forecasting, but I must begin with the ideological. Forgive me.
I am not a political moderate. That much should be evident from many of my tweets and other newsletter dispatches. Fundamentally, I think that center-left policies and mixed-market economics usually solve our collective problems more than individualists and proponents of center-right small government like to admit. And more recently I have also taken up the belief that the political right in America is more predisposed to fascist tendencies (and specifically here I use “fascism” to mean the oftentimes violent politics of “us-versus-them”) than those on the left. I think ethnonationalism and militarism are bad—seeing that these are more common on the right, I am opposed to most of the candidates who associate themselves with America’s current Republican Party.
I am lucky to have a job where I can be open about these beliefs. Especially in the “mainstream” media, it is hard to find employers who will tolerate an employee publicly sharing them. But The Economist is built on the idea that attitudes and beliefs that are empirically justifiable are valuable. In the production of the weekly print magazine, the company operates under the premise that aggregating up all of our individual empirically-sound attitudes—about politics, economics, liberalism and what have you—is a useful and productive way to produce journalism. (Not having bylines is a somewhat-unfortunate consequence of this theory.)
These attitudes are related to what I consider to be one of my strengths as a political journalist: the ability to write about the dark sides of our psychology, political parties, institutions etc even when it reflects poorly on one side of the aisle and might bring accusations of partisanship to my critics’ tongues. And that’s what I want to write about today.
Editor’s Note: I will continue writing election dispatches at a higher-than-usual frequency throughout the fall election campaign. If you have friends that you think might benefit from this newsletter, you should feel open to forward it to them regardless of their status as a free or paid subscriber—though I would love if you urged them to sign up by following this link or hitting the button below:
A friend of mine recently called to talk about this very subject. He wanted to say that my coverage of polls, election forecasts and political data offered him something “different” than the other big names in the field. The Nate Silvers, Nate Cohns, Harry Entens and David Wassermans of the election journalism world, he said, operated with this sort of “hyperaggressive impartiality” that can often tend to “both-sides” political issues. On race and policing, for example, they could write only about the horse race impacts of candidates’ speeches and statements, whereas I also had the freedom to cover the psychology that pushes a lot of Donald Trump’s supporters to violence or liberals to tolerate looting. (I also have the freedom to say which of these two is worse: it’s the first one, obviously. Shooting compatriots be it with bullets or paintballs is plainly worse than burning down a storefront.)
My friend also offered up another anecdotal criticism of their work. The New York Times in October of 2019 released polls of the Midwest that lead them to imply overconfidently that Elizabeth Warren would not be able to triumph over Trump in the 2016 general election. According to the reporters, she polled too poorly with non-college whites. But I operated in a more mild-mannered space. Instead of writing off her candidacy, I published an article in February of 2020 that said Joe Biden’s relative ideological moderation would best prepare him to beat Trump, but that more extreme candidates such as Warren and Bernie Sanders could still have a shot because they benefit the constraining effects of partisan polarization. But the other analysts ignored this nuance, preferring instead to write off left-leaning candidates almost entirely.
This raises an important question about how our biases shape our political coverage, and also our election forecasts. I think it’s fair to say that my peers (or, in this context, competitors) in the election forecasting (and forecasting-adjacent) space have all carefully cultivated a different form of the hyper-aggressively objective, centrist political analyst who has no biases and no predispositions toward either candidate. It is my theory that they also tailor their coverage—be it actively or subconsciously—to buttress their persona. If Cohn and his Times colleagues were not so caught up in pursuing the talking points of a political moderate, would they have paid more attention to the caveats of their analysis?
I think this “hyperaggressive impartiality” makes us worth off, in two ways. First is in improving our understanding of politics. That’s because the constant search for middle ground can distort some ground truths of political behavior. You will never hear Wasserman or Cohn or Silver write about how racism and sexism helped convert white Obama voters to Trump supporters. They prefer other narratives about elitism and income—despite the very thin evidence to back up those conclusions. They can’t admit those conclusions because it makes them seemed biased; and if they’re biased, they can’t be impartial; and if they’re not impartial, they can’t be trusted to divine the future. At least, that’s how the story goes.
And that brings us to the second way that such aggressive impartiality, such blindness to our underlying biases hurt our ability to forecast elections.
Here, it’s all about incentives. When a forecaster is incentivised to suppress their biases in order to seem “non-partisan” and therefore “trustworthy,” they might be more willing to accept models that make them appear more grounded in the political middle than elsewhere. For example, I have argued that Nate Silver’s incentive toward seeming non-partisan — and maybe his incentive to provide cover in case his model misfires again — has led him to believe that a forecast that gives Donald Trump about 2x the chance to win the election as the other high-quality statistical models inherently has more face validity than one that joins the consensus in rather confidently (but not certainly) projecting a Biden victory. Maybe if he didn’t have so many external incentives to want to believe that lower Biden forecast, he would have put more thought into what I believe to be a pretty flawed model. That’s the thing about biases: only if you’re aware of them can you check your work against them.
But this isn’t just about Silver. Take Wasserman now. He has been on a tear recently about how Joe Biden is losing ground with white voters who do not have college degrees—a group that is core to Trump’s strategy in the Midwest. I believe he’s incentivized toward producing analyses that the election is closer to 50-50 because of his pressure to appear non-partisan. Why else would he ignore the obviously more important (and good news for Biden!) that Trump’s margin among non-college whites is about 10 percentage point smaller than it was in 2016. That shift alone will break the president’s ability to win re-election—if it holds.
…
I could provide other examples, but this is—in a nutshell—why I don’t shy away from writing pieces that might cause others to accuse me of being biased. Not only do I typically only adopt attitudes are offered a robust defense by hard evidence, but I also believe that openly and forcefully adopting them both makes us better forecasters and is crucial to improving our shared understanding of politics and political behavior.
Isn’t that what this is all about, anyway?
I too would rather see the butcher's thumb. But what I think you might add is that having a PhD in political science should-- and usually does-- teach you to challenge your own assumptions. If anything this tends to make us too cautious (regardless of ideology). My one-time colleague and friend, the late Stanley Kelley, liked to say that the best answer when asked to predict the impact of events on the future was "nothing much will come of it." Stan was admittedly wrong when applying it to the early years of Viet Nam, but I think most of us tend, too often perhaps, to invoke Kelley's law.
The fiction that we can act (or write) without bias is costly -- to everyone. Better to expose, reflect and discuss our biases (e.g., priors) as thoroughly as we can.