The primordial election that is never held
Consider the following phenomena:
- Before entering a race, many prospective candidates ask whether they have a good chance of winning. If they think their odds are poor they won’t run.
- Candidates who are perceived as frontrunners have a much easier time fundraising. The invisible primary is an important phase of a presidential race, and my research on Australian elections also suggests that a link between viability and spending shapes elections.
- The media focuses on perceived frontrunners, and they determine who the frontrunners are partially based on who has money. Dark horse candidates tend to remain in the dark.
- A major reason elected representatives avoid compromising is the fear of losing a primary election to an extremist challenger when they expect voters in their party’s primary to be anti-compromise. This affects their behavior even if a primary challenger hasn’t declared their candidacy, and being sufficiently extreme can prevent primary challengers from appearing at all.
The common thread: Expectations shape reality, and expectations of losing a race prevent candidates from being serious contenders. And these expectations are based on counterfactuals. A would-be candidate wants to know: If I run for office, can I win the race? A donor wants to know: If this candidate gets as much money as the frontrunners, can they win? An incumbent wants to know: If I get primaried, will primary voters support me? These counterfactuals form what I call the primordial election: A hypothetical contest in which every would-be candidate can run, receive ample campaign contributions, appear in the debates, and receive enough media attention to be competitive.
Candidates who do well in the primordial election can run serious campaigns. Candidates who lose the primordial election are unlikely to run in the actual election at all; if they do, their campaigns will be ill-funded and mostly ignored.
An exception is when a race is a foregone conclusion. If one candidate is bound to win then the media wants to look for some drama, someone will want the attention of being the challenger, and a doomed candidate can mount a relatively serious campaign. But this exception isn’t particularly important; having well-funded doomed candidates isn’t going to affect anything.
It should be noted that talking about “the” primordial election is somewhat inaccurate because different people are considering different counterfactuals; a prospective candidate is considering what would happen if she runs, not if every other potential candidate throws their hat in the ring with her. But it’s a lot easier to think about a single primordial election, and it’s a relatively benign simplification when we’re not focused on voting methods that are vulnerable to vote-splitting.
Voting methods and the primordial election
Consider two pathologies:
- The center squeeze, which occurs in Instant Runoff Voting (single-winner Ranked Choice Voting). Here, the candidate closest to the center of the electorate would be preferred over any other candidate head-to-head, but is eliminated before the final round because he doesn’t have enough first-choice support.
- A Condorcet cycle where Candidate A beats Candidate B head-to-head, B beats C head-to-head, and C beats A head-to-head.
Which of these should we expect to be more common? Intuitively, center squeezes seem like they should be common in districts where both parties are competitive. Suppose the candidates are a Democrat, a Republican, and a centrist Independent in a district with comparable numbers of Democratic and Republican voters. In this case, it seems like the centrist should defeat either party’s candidate head-to-head, but have the least first-choice support unless the parties nominate extremists. By contrast, Condorcet cycles seem weird and contrived. Unlike the center squeeze, there isn’t an intuitive story for the electorate should have the preferences A>B>C>A.
The empirical data tell a different tale. Out of hundreds of American IRV elections for which we have sufficient data, there have been just two center squeezes — and also two Condorcet cycles.
Why are these numbers so different from what we’d naively expect? The primordial election provides an answer. Practically speaking, Condorcet cycles only occur in elections that are extremely close between three or more candidates. On the other hand, center squeezes can be blowouts; it doesn’t feel contrived at all to imagine a Democrat and a Republican each being the first choice of 40% of voters and a centrist being the first choice of the remaining 20%. If there’s a Condorcet cycle in the primordial election, every candidate in that cycle will have come close to winning it. If there’s a center squeeze in the primordial election, and a losing candidate loses by a mile, that candidate won’t be able to mount a serious campaign in the actual election, if she decides to run at all. Even if she runs, her diminished campaign resources and media attention will prevent her from attracting as much support as she received in the primordial election — likely averting the center squeeze. A center squeeze in the primordial election can prevent a center squeeze from appearing in the actual election.
A center squeeze in the primordial election can forestall one in the actual election if a losing “extremist” candidate declines to run a serious campaign. Here’s how Lee Drutman describes the 2022 Alaska Senate election between Democrat Pat Chesbro and Republicans Lisa Murkowski and Kelly Tshibaka, in which Chesbro got 10.4% of votes in the first round:
Murkowski survived to be re-elected only because Democrats effectively stood down and supported her. Murkowski raised $11.2 million in campaign contributions and spent $10.4 million. Chesbro raised $188,164 and spent $178,681 — which, in current campaign spending, is the equivalent of sneezing into the wind.
Murkowski’s reelection strategy relied on Democratic cooperation as much as the voting system.
If there was a center squeeze in the primordial election, Chesbro got crushed in the final round. She never had any real hope of winning, so her campaign wasn’t a serious one. In this race, a center squeeze in the primordial election was a self-correcting problem; the worst that can be said of IRV is that it prevented a doomed candidate from campaigning more.
If the extremists are about evenly matched, however, the centrist who loses the primordial election may be unable to run an effective campaign. In this case, the center squeeze will have still occurred in the sense that an extremist will have won instead of a candidate at the center of public opinion — but this center squeeze won’t be visible in any election data. A centrist who loses the primordial election either won’t appear in the election results at all — because he decided not to run — or he will appear to have only narrow support, perhaps losing head-to-head against either extremist due to lack of voter outreach.
A primordial center squeeze in 2016
Detecting a phenomenon that does not appear in your data set by definition is problematic. Nonetheless, we can identify probable cases in which a primordial center squeeze prevented a well-positioned centrist from gaining widespread support.
In the 2016 U.S. presidential election, both Trump and Clinton were less popular than any previous nominee for President in recent history. Intuitively, it seems like the Libertarian candidate, Gary Johnson, should have crushed either Trump or Clinton head-to-head given how divisive the major-party nominees were. However, Igersheim et al. did a study asking voters to vote in that election using different voting methods and found that either Trump or Clinton would have defeated Johnson by more than a 2 to 1 margin.
One possible explanation for this is that Gary Johnson had poor name recognition. The authors write,
For instance, a 2016 Gallup Poll indicated that two-thirds of Americans didn’t know who Gary Johnson and Jill Stein were. This voter ignorance is likely due to a number of causes, including the fact that the Commission on Presidential Debates excluded these two candidates from every presidential debate.
A study by Potthoff and Munger looked at ANES feelings thermometer data to predict how voters would behave in various pairwise matchups. Potthoff and Munger mitigated the issue of name recognition by only considering the opinions of voters who had an opinion on both candidates in a given pairwise comparison. They found that Johnson would have most likely beaten either Trump or Clinton head-to-head.
We can view this through the lens of the primordial election. In the primordial election (which used Plurality + Electoral College) Gary Johnson lost by a mile despite not being burdened by poor name recognition and spending a comparable amount to Trump and Clinton. So he had far less money to spend in the actual election and couldn’t muster enough name recognition to be the Condorcet winner in Igersheim et al.’s survey. But it certainly seems that he was the Condorcet winner over Trump and Clinton in the primordial election, and probably by much more than the statistically insignificant margin found by Potthoff and Munger.
Had the actual election taken place using a Condorcet method (and without the Electoral College), Johnson, or perhaps a candidate we haven’t even thought of, would have won the primordial election. And he would have made it into the debates and received at least as much in the way of donations as Trump and Clinton, who would have received substantially less than they actually did. If the disparity in campaign spending was large enough it’s possible that Johnson also would have received the most first-choice votes — making him the Plurality winner in the (hypothetical) actual election, despite being in third place (or worse) in first-choice support in the primordial election. Had Igersheim et al. performed their study in this alternate reality, they might have again found that the Plurality winner and the Condorcet winner were one and the same — only in this case, it would have been Gary Johnson.
I also think it’s probable that there were primordial center squeezes in just about every swing electorate in the 2022 Australian elections. But I can’t point to any more likely cases. Primordial center squeezes don’t appear in the data. It’s possible to see the outlines of one if you’re knowledgeable enough about the perceptions of candidates in a particular election, but broadly speaking, you should only expect to identify a small fraction of them in such a manner.
Computer simulations and the primordial elections
In Jameson Quinn’s original Voter Satisfaction Efficiency simulations, he found that the Plurality winner differed from the Condorcet winner roughly 50% of the time in the voter model he used. By contrast, the Plurality winner has differed from the Condorcet winner in 26 out of 407 US IRV elections (6.4%). Given that Jameson’s voter model is designed to reflect a lot of real-world complexities, this seems like a very large discrepancy.
My main explanation for this is that models like Jameson’s are modeling the primordial election rather than the actual election. To model the actual election you need to include the possibility of candidates deciding not to run, money in politics, decisions to drop out of the race, limited media attention, and more. This is difficult, and if you try to do it all you end up with a whole lot of guesswork in your model. The primordial election is much cleaner.
Furthermore, if you’re not going to somehow create an elaborate and well-justified model that takes into account campaign spending and differences in media coverage under every voting method, modeling the primordial election seems like the best option. When there’s a primordial center squeeze that prevents a centrist from winning the actual election, we want our models to capture this. Modeling the kinds of candidates and their levels of support that you get under Plurality or IRV means that you’re effectively asking, “What if every candidate and donor thinks there will be an IRV or Plurality election, but instead, BAM! You switch it to being an Approval/STAR/Condorcet contest on election day.”
That said, only modeling the primordial election is not entirely satisfactory since many problems get mitigated, and not merely concealed, between the primordial and actual elections. When a candidate opts not to run because her presence in the primordial election caused the vote to get split or when an extremist doesn’t campaign seriously because of a primordial center squeeze, we would ideally like to capture this when we’re simulating election accuracy. It’s a difficult problem, and I don’t have a good solution.
Conclusions
I’ve sometimes heard people say that better voting methods are a “solution in search of a problem”. It’s easy to look at some recent municipal election results and see that vote-splitting has only occasionally been a factor, or look through IRV election results, and conclude that all voting methods will elect the same winners the vast majority of the time. But our data are incomplete. Scrolling through election results doesn’t tell you when a potentially great candidate declined to run because she thought she’d split the vote. IRV results don’t tell you when a well-positioned centrist seemed unlikely to win and couldn’t mount a sufficiently serious campaign to become the Condorcet winner.
These problems appear in the primordial election, but the primordial election doesn’t get tabulated. It’s both influential and invisible.
I’ve found it frustrating to write this blog post because I’m positing something that is extremely difficult to study. How, exactly, are we supposed to test the theory that there are many primordial center squeezes that don’t show up in actual elections? Do we search for centrists on the street, give them thousands of dollars to launch well-funded campaigns, and see if become victims of the center squeeze? And, if we accept that primordial center squeezes that don’t appear in election results are real and important, how common are they? Do they show up equally as often as actual center squeezes? 20 times as often? It could be anything!
When operating under uncertainty, the right thing to do is to acknowledge that uncertainty and try to do whatever is best in expectation. If we could find a clever way of getting a half-decent estimate on the frequency of primordial center squeezes that would be excellent, but, in lieu of that, we should expect that the numbers we see are seriously understating the extent of the problems with our elections.