Popularism

So this means that every time you open your mouth, you have this complex optimization problem where what you say gains you some voters and loses you other voters. But this is actually cool because campaigns have a lot of control over what issues they talk about.

David Shor’s Unified Theory of American Politics

I can’t claim to know exactly what the electoral effects would be of doing these things. But all of the polling I’ve seen suggests that things like HR 1 and adding states are above water. They’re not as popular as a lot of economic issues, but they’re above 50 percent. Electoral backlash doesn’t typically come from doing things that poll at 53 or 54 percent. It comes from doing things that poll at 30 or 40 percent. And so I think that the downside of this stuff is low. I think the level of voter interest in procedural issues is low. If we lived in a world where voters punished politicians for playing procedural hardball, we would have a lot fewer Republicans in office.

David Shor on Why Trump Was Good for the GOP and How Dems Can Win in 2022

David Shor is a pollster and, perhaps more significantly, an interpreter of polls. He calls himself a data scientist. Shor worked for Obama during his 2012 re-election campaign and was responsible for building a statistical model, combining polling and demographic data, to forecast the election result. (He did remarkably well.) He later worked at Civis Analytics, a political consulting firm founded by an Obama campaign alumnus. As I understand it, Civis is paid by large corporations to understand how their messaging affects public opinion. (In what follows, I am applying my own judgment and interpretation to the corporate jargon-laden gloss in Civis’s description of itself.) Suppose, for example, we are Facebook, and we want to improve our public standing following the revelation that we ignored internal research showing that Instagram was bad for teen girls’ self-esteem. We would pay Civis, and work with them, crafting many different messages designed to burnish our reputation. Civis would then survey the population, ask the respondents to assess these messages, apply some fancy data science techniques to more robustly infer their impact on various demographic subgroups and the population as a whole, and, in the end, use this information to select the “winning” message. Shor describes himself as a socialist, and I have no doubt that he and I share many political values (although clearly not all), but what Civis does, to a large extent, is data-driven corporate propaganda.

Shor obviously realized that Civis’s techniques apply well beyond corporate messaging and public relations. In fact, he has married his earlier work in politics with his more recent work at Civis, and has become, according to Politico, “one of the most in-demand data analysts in the country”. Shor tries to figure out how the public responds to various Democratic Party messages, and is paid by various party actors (whom he has not disclosed) to optimize their campaign messages. He is now doing for a swing state congressional candidate what he used to do for Verizon.

But that isn’t all. Shor doubles as a political pundit. Ezra Klein writes that Shor has achieved public stature as “the Democratic data guru who refused to soften an analysis the left often didn’t want to hear. He became ubiquitous on podcasts and Twitter, where Obama posts his analyses and pundits half-jokingly refer to themselves as being “Shor-pilled.”

Shor is so “ubiquitous” that much of my Twitter feed has been consumed by debates over his “theory of politics”. He conveys supreme self-confidence, and bathes his ideas and statements in the language of data science and statistics. To dispute him would be to argue with the numbers: a clearly futile task.

I am more amenable to futility than most people, so I will press on. What disturbs me about Shor is that so much of his theory is either questionably grounded or not really a theory at all. Before I get to the criticism, I will try to summarize what I am criticizing. Here’s Ezra Klein:

Atop this analysis, Shor has built an increasingly influential theory of what the Democrats must do to avoid congressional calamity. The chain of logic is this: Democrats are on the edge of an electoral abyss. To avoid it, they need to win states that lean Republican. To do that, they need to internalize that they are not like and do not understand the voters they need to win over. Swing voters in these states are not liberals, are not woke and do not see the world in the way that the people who staff and donate to Democratic campaigns do.

All this comes down to a simple prescription: Democrats should do a lot of polling to figure out which of their views are popular and which are not popular, and then they should talk about the popular stuff and shut up about the unpopular stuff. “Traditional diversity and inclusion is super important, but polling is one of the only tools we have to step outside of ourselves and see what the median voter actually thinks,” Shor said. This theory is often short-handed as “popularism.” It doesn’t sound as if it would be particularly controversial.

It is.

Now, as Klein later explains, “There’s no comprehensive paper or experiment in which he has constructed and footnoted a full theory, in which his data can be rerun and his footnotes picked through.” To the extend Shor’s theory exists, it is by piecing together Twitter replies, off-the-cuff interviews, screenshots of slides he makes for clients and posts without context, and so on. (This is obviously not the proper way to conduct science and make an argument, but who am I to cast aspersions on the “Democratic data guru”?)

So if I am guilty of misrepresenting his work, it is largely because he has not properly represented it in the first place. That caveat aside, let’s dive in. Here are the assumptions that underpin Shor’s “popularism”:

1. A properly conducted poll can be used to infer public opinion. We can use polling and other data, combined with machine learning, to determine, with high accuracy, how various demographic subgroups view Democratic party candidates, policies, and messaging.
2. Democratic candidates can choose between various policies and messages. However, they should not do so based on ideology, but instead based on the polling data. As Shor puts it, “you have this complex optimization problem where what you say gains you some voters and loses you other voters”. So, Democrats should indeed “optimize” in this way: they should choose policies and messages that gain them more votes than they lose.

Now, even at this point, there is plenty to object to, and many have. But what I find particularly odd is that Shor himself does not seem to follow this logic in practice.

One example is this tweet of Shor’s: “politicians really are correct to mostly ignore academic MRP estimates saying that all their constituents want them to pass liberal policies”. To cut through the jargon slightly, “MRP” is a statistical technique used to generate more accurate estimates of public opinion by correcting for “non-response bias”: the fact that a lot of people simply don’t respond to pollsters anymore. So, Shor is saying that even if polls show broad support for liberal policies, politicians should ignore these polls and not talk about or support these policies. The example he cites involves polling about a carbon tax. Academic pollsters found that, even in red states, the public supports a carbon tax coupled with “equally reducing other taxes”. But, when Washington state (a fairly liberal state) had a ballot initiative to implement such a measure, it failed, with 43% in favor and 57% opposing. Shor later wrote, “The reality that the public usually trust big business more than liberal/left-wing interest groups on taxes is too depressing for folks to seriously engage with.”

Again, what we have here is not a theory, let alone a data-driven one, but let me again try to be charitable. If I understand Shor’s point, it is that the public may support an issue in the abstract, but, in practice, when that same public hears anti-tax sentiment from actors they trust, like big business, their opinions may change. If that’s true, what does it mean for popularism? We have drifted rather far away from the prescription of “talk[ing] about the popular stuff and shut[ting] up about the unpopular stuff”. Issues may be popular but that popularity may erode after propaganda takes its toll; conversely, issues may be unpopular but that popularity might increase if the Democrats make a concerted effort to bolster support (unlikely, but we can hope). What we started caring about was the popularity of issues and messages; what we’ve ended up caring about is that same popularity, filtered through the debates of a political campaign. Shor completely elides that distinction in most of his tweets and public statements, and, if I’m being cynical, the reason is that it is much harder to explain and analyze using the techniques he’s comfortable with. You can, perhaps, do polling where a respondent is told about arguments from “both sides”, and Shor argues that he does this at his new gig, Blue Rose Research, but it is highly questionable whether that would represent political debate in the real world.

To take another example: Shor argues that comprehensive immigration reform may be popular, but it is ultimately bad for Democrats to talk about it. Raising the “salience” of immigration angers a bloc of electorally important non-college educated white voters and harms Democratic candidates. Now, this is a possible story, perhaps even a plausible one, but it seems tough to square with naive popularism (like Shor’s statement that “Electoral backlash doesn’t typically come from doing things that poll at 53 or 54 percent”). It seems as if talking about popular stuff is good, unless it’s actually bad…or something.

I empathize to some extent with Shor’s idea that we can reduce politics to some kind of reinforcement learning system, where we pick topics that generate reward and avoid those that don’t. It would certainly make decision-making easier! But, besides the obvious philosophical and ethical issues, the main problem is that a static and decontextualized sample of public opinion doesn’t get us there, not even close. We would have to be able to figure out what the public would really think about a particular issue in a real political situation, with special interest groups and mainstream media and Youtube conspiracists and Facebook algorithms, not in the sterile environment of a Tuesday evening phone call.

Shor himself admits that he often doesn’t understand why political opinion changes. He was asked about his prescription that Democrats should stop talking about defunding the police and Black Lives Matter:

But we have seen a surge in support for the Black Lives Matter movement and police reform. We’ve seen Biden boasting a bigger advantage over Trump on the question of which candidate can best handle race relations — and all while progressive activists have been associating the left with the exceptionally unpopular concept of defunding the police.

and he replied

Yeah. I’m not going to pretend that I would have predicted that this is how it was going to shake out. But I do think it’s actually consistent with what we’ve been discussing.

If you are able to contort your theory into giving the right answer even when a straightforward interpretation would appear to give the wrong one, the problem might be that there is no theory at all: that the theory is so elastic that it is capable of absorbing any possible future that might arise. I think what is called for is some epistemic humility. Polls suffer from severe biases, are often internally or externally inconsistent, provide little indication about how opinion might evolve in the future, cannot be used to make causal, let alone normative, claims, and so on. Shor’s tools are far too weak for the conclusions he wants to draw, and thus he engages in storytelling wrapped in the veneer of science. (It is also, of course, difficult to get a man to understand something, when his salary depends on his not understanding it.) The chain of deductive links that separate Shor’s finding that 2.5% of voters apparently change their minds on the child tax credit (CTC) if it is not means-tested, and his claim that Democrats should support means-testing, is so long and tenuous that I wonder if he’s actually being serious.

Polling is but a small slice of politics, despite the tendency of “data-driven journalists” and pollsters to talk up its importance. And, to end on a somewhat romantic note, the remainder is what I find far more exciting, and what Shor finds puzzling and unexpected: not when public opinion drives politics, but when politics and organizing and activism change minds.

Advertisement