There has been a lot of attention dedicated to increased political polarization. And a lot of it has focused on the factors causing voters to be more polarized. I want to discuss a different, but related question. Given that voters have become more polarized, why are politicians more polarized? At first blush, you’d think this follows inevitably from the preferences of voters. Politicians try to position themselves around the central view of their voters, so if voters are more polarized, so will be the politicians’ views, right? Not necessarily. Let’s see why.

The puzzle: politicians’ polarization

Let’s describe this in a schematic (modeled on the US general elections), where we assume there are only two political candidates and that political views can be described as a single continuum[1] going from left to right, with each voter voting for the candidate which is closer to their position, and the candidate who gets more votes wins. You can imagine it might look like this:

This seems reasonable, until you consider what would happen if one of the candidates (or parties) decided to make their views more moderate. This could take the form of e.g. adopting one of the policies favored by the other party and less favored by the candidate’s party, to appeal to moderates (e.g. a Democratic candidate declaring that they do not intend to pursue any further gun control measures). They would move closer to the center, and grab the votes of the more moderate voters from the other side, and always win by a wide margin.

The other candidate would have to respond by moving closer to the middle as well, and then the original candidate would become more moderate still, until they converge to the median voter. This is how the Median Voter Theorem works. So even in the case of a highly polarized populace, it still predicts that candidates will be very moderate, with almost identical views.

This is decidedly not what we see in the US (or in any country) - the two parties’ views differ substantially. They might have even become more polarized recently. So something else must be going on. What could it be?

Hypothesis 1: Irrational Politicians, or “the echo chamber”

In our simplified model, we’ve been assuming that politicians position themselves optimally to garner the most votes, but maybe they are not rational, or don’t know which position will get them the most votes. The echo chamber hypothesis is that party elites, politicians and donors (who have much more radical views than the median voter of their party) hang out with one another in an echo chamber, and mistake their preferred policies for everyone's favorites, and so adopt and support them, not noticing that it takes away votes. This is the point Matt Yglesias and others constantly make on Slow Boring about Democrats.

If this were true, though, you might expect another candidate to pop up who is not affected by the echo chamber, or is naturally more moderate in his or her views, adopt more moderate views, be extremely popular and be overwhelmingly likely to win the elections against the other party. Wouldn’t people realize this and select this candidate to represent their party?

Hypothesis 2: Radical voters not voting for a moderate-but-preferred candidate

So far we’ve been assuming that all voters vote for the candidate who is closer to their views, irrespective of how close they are in absolute terms (this would make sense - after all, that candidate is still preferable to the other candidate). However, suppose that at some point, when the two main candidates are too far from your radical position, you vote for neither (instead of voting for the lesser evil, you vote for a third candidate or abstain entirely). Maybe you’re angry at both candidates and want to punish them. Maybe from such a great ideological distance they seem the same to you, and you can’t get emotional enough about voting to spend the time. So in the end you don’t vote. Here’s what this would look like schematically.

As you can see, by moving towards the center, the left candidate might lose more radical votes than win moderate votes. In that case, it stops being always optimal for the candidates to move closer to the median voter, and instead they also have to get closer to the radicals to get their votes. This is known as “appealing to the base”. It doesn’t even have to really be the case that radicals won’t vote - it’s enough for the politicians to think this is the case in order for them to adapt to it. Why would they think that? Maybe because of the “echo chamber”?

Interestingly, this means that from a game-theoretic perspective threatening not to vote could be a better rational strategy for radicals themselves, if they want to affect candidates' policy and not just leave all the influence to the median voter. But is it realistic? I’m sure there are some Bernie Sanders supporters who did not vote for Hillary Clinton in the general elections (or at least said they would, and maybe even did it if nothing was at stake). And there are some hard-core communists for whom both candidates are capitalist and nationalist and are barely distinguishable. But are they enough to sway the elections? I don’t have data on this, but I’m skeptical. If they’re not very numerous, then they’re not a good reason for candidates to change their positions, certainly not so drastically.

Hypothesis 3: Radical voters not donating or becoming activists for a moderate-but-preferred candidate

This is a variation on the previous hypothesis. I’m not convinced that radical voters will actually not vote in large numbers, but I find it much more credible that they won’t donate and will not become activists for candidates which are too far from their views. Imagine you’re a Democrat and the general elections are between Mitt Romney and Donald Trump. You have a preference, and if you could vote you probably would (because what else will you do with your vote). But would you donate to your preferred candidate? I think it’s far less likely (because there is a lot else you can do with your donation). Will you become an activist for your preferred candidate? Maybe even less likely.

In fact, we have a test for this. Democrats are free to donate to their preferred candidate in the Republican primaries, but I think almost none do. And I’m pretty sure none become activists for their preferred candidate in the other party’s primaries. True, there’s a possibility that your donation will be wasted because your party will win the general elections. But there’s also a possibility that the same amount of donation/activism will sway the chances of the candidate much more in the primaries than in the generals, so not sure how this balances out.

To the extent that donations and activism are important resources in winning the elections, and that they are really reduced when the candidate’s views are too moderate, they could make it rational for candidates to become more polarized in response to the electorate. Or rather in response to the donating / activist group, who could be even more polarized than the general public.

Hypothesis 4: Primaries

We glossed over how candidates are selected within each party, and assumed there is a single candidate from each party with the liberty to select their position optimally. But that’s not how it works, at least in the US. You have to become the party’s candidate. This happens through primaries. Suppose that in the primaries there are also only two candidates, that only voters of that party vote, and each one votes for the candidate closest to their position. In this case, the Median Voter Theorem applies in the primaries and forces the candidates towards the Median Voter of their party, who is far from the overall median voter. If voters in the primaries are more extreme in their views than the general party voters, their median will be even more radical.

Importantly, this requires voters in the primaries to vote for the candidate closest to their positions, without taking into account “electability” - the chances of the candidate winning the general elections. Electability is a crucial consideration if the goal of voters is to get their preferred policies implemented, for which they have to win the general elections.

But then why doesn’t the chosen candidate change their positions before the general elections, in order to get more votes by moderates and win? To some extent candidates do, but mostly they don’t abandon all the policies they advocated in the primaries. That would be viewed as “flip flopping”, reducing their credibility, and punished by voters. This yields a slightly different model for voters’ selection criteria, not based only on policy positions, but incorporating other factors such as desirable character traits - preferring a candidate who is brave, honest, competent and not corrupt.

OK, then it could be primaries at fault. But that still leaves a mystery - why are the general elections so close? How come there is drama on election day, and a percentage point or two in each direction could sway the result? Primaries don’t provide a natural mechanism to cause this. It could have been the case that one party is much less radical than the other, so their median voter is much closer to the overall median voter, and they would win the general elections every time by a wide margin. Schematically, it would look like this.

It could be that the natural balance this causes is 60% of voters prefer the candidate of the more moderate party, and that party always wins by a similar very wide margin. But this is not what we see - general elections are in fact close, sometimes very close. What’s more, they are often closer in the metric that matters (in the US - electoral votes) than by metrics that don’t decide the winner (in the US - the popular vote). It could be pure coincidence - we don’t have that many elections with high polarization to make statistics, and electoral votes are more noisy than the popular vote. But it could also imply some version of the Median Voter Theorem effect going on. What form would it take to achieve this situation?

I mentioned before that we were making the assumption that voters in the primaries do not consider electability, but only their policy positions and those of the candidate. But that is a bit naive - we know voters also consider electability because it is openly discussed in the primaries. If voters in the primaries only decided according to electability, they would select candidates which are very close to the overall median voter (out of all voters in the general elections), and we would have the Median Voter Theorem applying again in the national elections, with both candidates having very moderate positions. If they only decided on the primary candidate closest to their preferred positions, we would not be able to explain the mystery of the general elections being close. That leaves us with a middle possibility - that voters in the primaries weight electability just enough so that they get a candidate with views close to theirs, but who has a chance to win with the general electorate. But they don’t make them too electable, otherwise the general elections would not be as close as they are. I still find it surprising that voters would be able to aim so precisely. And of course if one party’s voters in the primaries decided to weight electability a little higher, they could reach an equilibrium where they get 55% of the vote and win easily. The other party might respond by nominating a candidate which is also more electable, and we’d have the equilibrium of the Median Voter Theorem again, so it’s an unstable equilibrium and without resorting to some of the other explanations it’s hard to explain how it happens.

Hypothesis 5: the model is wrong

It could be that voters don’t base their votes on policy positions almost entirely. They are entrenched partisans, or have distorted views of the other party, vote to signal various things about themselves and so on. It could also be that politicians are polarized not in response to voters’ preferred positions, but due to their personal preferences instead of positioning themselves according to the optimal election strategy. I don’t want to go into all the ways this could go wrong, but I do think we have evidence that politicians react to the views of the electorate (they certainly change their views more often than personal preferences would imply), and this could also take the form not of a single politician changing their positions, but of other politicians with more favorable positions being elected.

Conclusion

I’m still a little confused by this. I don’t have a good model which explains both politicians’ polarization, and elections being very close. But I think the factors I pointed out play a role, and a combination of some of them could even be a full explanation. Or it could be something completely different. Maybe examples from other countries could shed light on this?

—------------

[1] If we model voters on more than one dimension, the median is not defined. But it is still the case that if a candidate has positions closer to their opponent, they would get a larger share of the vote, so the same dynamic described here holds.

* Thanks to ZBH and ALB for reading an early draft of this post.