Primaries, cooler?

Center squeezed. (Photo: Mario Tama/Getty Images)

This is a guest post by Dr. Jameson Quinn, a long-time activist and theorist of voting methods who’s a friend of the blog. You may recognize him as a frequent commenter here but please don’t mention his pseudonym in explicitly googleable form on this post.

Right now, the best-case scenario is that Joe Biden will be the next president of the USA; the worst-case is that Trump is the last one. That is to say, we will have a choice between a guy whose primary campaigns twice flamed out from self-inflicted errors and who, the day he takes office, will be the oldest president the country has ever had;¹ and a narcissistic, mobbed-up reality television star whose platform is focused on his core base of racists, trolls, and racist trolls.

If you were hoping for something different, you have been robbed, and you have a right to be furious. You have been robbed, not by a “smoke-filled room” conspiracy of powerful party insiders, but by a voting system with predictable and recurring pathologies. And if we want things to work out better in the future, now is the time to be planning the reforms that will make that possible.

Our voting method was especially unfair to Elizabeth Warren. But it’s not just Warren fans who should be furious. Other people who never got a fair chance in this primary: Pete, Amy, Kamala, Cory, Julián, Jay, Kirsten. It’s debatable, but maybe even Bernie belongs on this list. And frankly, the 2016 GOP primary was even worse. If you hoped for any other options but the ones we have, you should want to fix things.

To explain this, I’m going to do 4 things:

• Define what I mean by “a fair chance”.
• Discuss, using examples, the multiple ways that our current system predictably fails to give candidates a fair chance.
• Discuss the best-known reform proposal, “Ranked Choice Voting” (RCV), and why it would fix only some of those problems.
• Discuss other reform proposals that could at-least-mostly fix all the different problems.

What I won’t do in this post, though we could discuss in comments:

• Discuss pure multi-winner voting methods; the kind of thing that, applied to US House or state legislative elections, could fix gerrymandering and break the zero-sum polarization “doom loop” that makes modern US legislative politics so dysfunctional. I have plenty of ideas and opinions on that, and it’s arguably more-important than presidential politics, but it’s a separate set of issues from the broken primary that we saw today.
• Talk about whether the US system — with partisan primaries and a strong president — is a good idea as compared to other models, or discuss how it could be reformed.

Goals and status quo

So, what does it mean to have “a fair chance”? Arguably, a candidate should win if they are an honest pairwise winner; that is, if they would be preferred by a majority over any possible rival in a one-on-one race. (Another term for this is “honest majority Condorcet winner”.) In theory, it’s possible that there is no such candidate, if there’s a cycle where A would beat B, B would beat C, but C would beat A; but such cycles seem to be rare in practice.³

Aside from finding the right winner, a good primary voting method should be reasonably easy to understand and use, both when voting and when looking at results. That’s why, even though I said above that a method should be designed to usually elect a Condorcet winner, I don’t think it should necessarily be a strictly Condorcet-compliant method. That’s because, in order to determine who the pairwise winner is, you would need something like a table which compares every possible pair of candidates. Since it’s tough to read such a table at a glance, it may in practice be better to use a voting method whose outcomes are easier to show graphically, even if it doesn’t quite guarantee electing a pairwise winner when one exists.

But our current Democratic primary process predictably fails to elect honest pairwise winners. To review the steps of that process:

• It starts out with a staggered state-by-state series of choose-one, semiproportional primaries, in which votes for candidates under the 15% threshold are wasted.
• Interspersed with the primaries are a few low-turnout, high-nuisance caucuses which at least allow supporters of “nonviable” sub-15% candidates to realign to viable ones. But since this viability threshold is applied locally, your voting options depend crazily on where you happen to live.
• The results from those primaries and caucuses are converted to pledged delegates in a “proportional” process that almost seems designed to magnify rounding error and inter-regional inequities.
• The delegates use iterated choose-one voting at the convention until some candidate gets a majority. If that process goes for more than one round, “superdelegates” (mostly elected in some other choose-one process for some other original purpose) get to vote as well.

In other words, it’s a hodge-podge of different voting methods, but the one thing they all have in common is a choose-one ballot format. (In the case of caucuses, the “ballot” consists of standing in a certain part of the room, but it’s still a choose-one format.)

Problems with the status quo

Among the pathologies of this Frankenstein’s monster of different choose-one methods are:

1. Unless and until the race narrows to two candidates, the “winner” is usually decided by a minority of voters. This makes all the other pathologies that follow more serious, both individually and in combination.
2. The primary predictably functions as a “Keynesian beauty pageant”, giving more power to voters who correctly guess which candidate(s) will be the frontrunner(s) among the electorate as a whole (or at least, who will pass the 15% threshold). This pushes voters to over-weight “objective” criteria like the amount of money a candidate has, the highest office they’ve held previously, their fame/name recognition, or the amount of media coverage they’re getting, and under-weight “subjective” criteria like the quality of their proposals or their qualifications for the job. That’s especially true when there are blocs of voters who want the eventual winner to “owe them one”, and so who are motivated to vote for the winner above and beyond how favorably they feel to that winner. But, since only votes for one of the two frontrunners can help those candidates get to a pledged delegate majority, even voters who only care who wins, and not who votes for whom, are pushed to vote strategically.
3. The primary predictably has problems with “center squeeze”. That is, “center” candidates whose support overlaps with those of multiple other candidates will be fighting a battle for voters on several fronts, and will probably end up having to drop out first. Meanwhile, more “extremist” candidates whose support groups only overlap with other candidates’ on one side, will be able to concentrate on just one opponent, and thus are more likely to survive. This is true even though it’s more likely that a “center” candidate is the honest pairwise winner.
4. The sequencing of primaries effectively gives more power to voters in early states. In particular, the first two states are two of the whitest.
5. If there were a contested convention, delegates would have no good way of knowing the true priorities of the voters who elected them, aside from the one top preference. This makes it less likely for the convention to find a good compromise winner in that situation, and even if they do, it would damage the democratic legitimacy of that winner.

Note that all of the above problems except #4 are related to the choose-one ballot format. So even though the primary uses different voting methods — choose-one semi-proportional in primaries, two-round choose-one semi-proportional in caucuses, and iterated choose-one majority threshold in the convention — the same choose-one pathologies run through the entire process.

Let’s run through a selective list of non-Biden candidates and see which of these pathologies they suffered from:

Cory Booker: Hurt by 1, 2, & 4

Pete Buttigieg: Hurt by 2. Helped by 1, 3, & 4; without that leg up, he probably would not have been a factor.

Julián Castro: Hurt by 1, 2, 3, & 4

Kirsten Gillibrand: Hurt by 1 & 2; dropped out before 3 & 4 could help her.

Jay Inslee: Hurt by 1, 2, & 3; dropped out before 4 could help him.

Kamala Harris: Hurt by 1, 2, 3, & 4

Amy Klobuchar: Hurt by 2. Helped by 1, 3, & 4.

Elizabeth Warren: Hurt by 1 & 2. Hurt by 3 perhaps more than any other candidate; I’d say that her campaign could function as a textbook example of center squeeze. Helped by 4 (and maybe would have been helped by 5), but not enough to cancel out the others.

That leaves Bernie Sanders, who was arguably “helped” by all five of the above pathologies. However, I think that even he was ill-served by this process. The advantages he got were all more-or-less fleeting and illusory. They all pushed him to campaign as more of an extremist, but in the end, that was a dead-end strategy. In a healthy primary process where a crowded field didn’t make it so hard to see voters’ true preferences, he would have been pushed instead to appeal more broadly, and it’s entirely plausible that in that environment he could have found a message to beat Biden.

So, can these problems be fixed?

The best-known proposal to reform choose-one voting is RCV: ranked-choice voting. Note that like “choose-one”, “ranked choice” is really just a ballot format, not a full voting method. In practice, it’s used to refer to two related voting methods: IRV (instant runoff voting) for single-winner contests, and STV (single transferable vote) for multi-winner ones. In both cases, after the voters rank the candidates, the candidates are sequentially eliminated in a bottom-up process, transferring their tallied votes to voters’ next preferences, until all remaining candidates are above some threshold (50% in the case of IRV; lower in the case of STV).

Saying “the primaries should use RCV” sounds like a simple proposal, but in practice, it’s inevitably just as complex as the system it would replace. You’d have to use STV with a 15% threshold in each individual state to choose delegates, then IRV at the convention itself;⁴ in other words, it would still be an ungainly hybrid of different voting methods; difficult to explain, audit, or analyze.

Furthermore, of the pathologies I listed above, RCV would only ameliorate 1 (minority winners) and perhaps 2 (establishment advantage), but would do nothing to address 3 (center squeeze). In IRV, if the final 3 candidates are a “centrist” and two opposite “extremists”, it’s likely that the centrist will be the honest pairwise winner, but will nevertheless be prematurely eliminated, leaving a runoff between two extremists, neither of whom could have beaten the centrist one-on-one. Furthermore, because of the way RCV results are reported — inflating the apparent vote totals of frontrunners while minimizing those of other candidates — some aspects of problem 2, establishment advantage, could actually be worsened.

Are there other voting reforms that would do a better job at resolving the pathologies above? Yes. We’re looking for a voting method that’s easy for voters; and where the results can be counted separately in multiple states (either simultaneously or sequentially) and then added up to find a clear national winner; and which tends to find Condorcet winners. A few methods that would do that are:

• Approval voting: the simplest of all voting methods. Each voter approves of as many candidates as they want to; the candidate with the most approvals, wins. The main organization promoting this method is the Center for Election Science (disclaimer: I used to be on their Board of Directors.) This method is easy to understand, and easy to add results across states. Unlike the current situation, voters in later states would have a strategic advantage over voters in earlier states. This isn’t my favorite option — I’d rather one that had fewer strategic incentives — but it certainly would be a HUGE step up from the existing system. It would resolve problem 5 (principal-agent issues in the convention); certainly ameliorate problems 1 and 2 to at least some degree, and, depending on voter behavior, might also help ameliorate problem 3.
• STAR voting:⁵ voters rate each candidate 0-5 stars. The two candidates with the most stars are the finalists. Of the two finalists, the one rated higher on the majority⁶ of ballots is the winner. This method is promoted by the organization Equal.Vote, which is campaigning to get it adopted for local elections in several Oregon cities & counties. This would resolve problem 5, almost eliminate problems 1 and 2, and substantially ameliorate problem 3 (center squeeze).
• Other voting methods such as ranked-pairs or 3-2-1. These have some minor theoretical advantages over the above suggestions, but probably not enough to justify their additional complexity.

You may notice that I didn’t mention problem 4, the White-voter-frontloaded schedule, in talking about the above reforms. That’s not to say they couldn’t resolve it; any of the above proposals would work just as well if all states voted simultaneously. But there are arguments that the current system has some advantages. Certainly, not enough to justify always letting Iowa and New Hampshire go first; but I can sympathize with the idea that “retail campaigning” might give a better chance to a high-quality candidate who didn’t start out with an established fundraising machine. 

This is less a matter of voting theory than the above, so in this case I don’t pretend to have any special expertise that you don’t have; but, if it were up to me, I’d replace the staggered primaries with a staggered series of regional Citizens’ Juries. That is, I’d give randomly-selected groups of citizens — a couple hundred or so in each group — a chance to meet the candidates, analyze their proposals, and take non-binding but highly-publicized votes. The results of those votes would then help inform the actual binding vote, which would be simultaneous across all states, with early voting only starting after the last of the Citizens’ Juries.

Conclusion

Our primary system failed us. Ultimately, it was unfair to almost all the candidates except Joe Biden. We can fix it. RCV would help a little, though it would be more complicated than it sounds; approval voting, a bit more; and STAR voting would be an excellent system in all regards.

Footnotes

¹ I don’t want to exaggerate Biden’s faults or minimize his strengths. His friends say that he’s a good, loyal man, and while I don’t know him, I think they’re probably right. As a presidential candidate, he has arguably the most-progressive platform any nominee in US history, with the possible exception of Hillary Clinton’s. And overall he has pretty good relationships with the people he will need in order to get stuff done: Democratic incumbents, domestic bureaucrats and wonks, and foreign leaders. It may be hard to get excited based on these factors, but they could end up being very good for the country, especially if they end up minimizing the backlash in 2022 and 2024.

² There’s no question that in 2016, Bernie lost the primary fair and square; that is, if you believe in small-d democratic primaries at all, Hillary was the normatively “correct” candidate. But on a purely practical level, there is a real argument that with a reasonable voting method, Bernie could have won any race among the top 3-5 candidates. You can certainly argue that he would have lost such a race due to the trove of opposition research that was never really used; but that’s a speculative argument that, reasonable as it is, goes against a naive reading of the polls just before election day.

³ Some argue that, because of the possibility of a majority preference cycle, a better goal would be to elect the candidate that maximizes “utility”; the one that makes the average voter “happiest” by winning. However, because self-reported utility measures are subject to strategic and other distortions, it seems that voting methods designed purely on this basis may have more-common strategic pathologies than pairwise-majority-based voting methods.

⁴ Why couldn’t the primaries use IRV on a nationwide basis, instead of the state-by-state-STV-followed-by-IRV-at-the-convention hybrid I suggested above? Because RCV methods (both IRV and STV) require at least some steps of the ballot-counting process be centralized. Doing that on a nationwide basis would be difficult to execute and nearly impossible to audit.

⁵ The name “STAR voting” is nominally an acronym, standing for “Score Then Automatic Runoff”.

⁶ In STAR voting it is in principle possible for neither of the finalists to be rated higher on a majority of ballots, if they are tied with the same number of stars on the median ballot. Usually, STAR voting would then choose the one rated higher on a plurality of ballots. In a Democratic primary, this might be the case where the superdelegates would help decide. In any case, it would be rare.