Rarely-occurring pathologies can often be relevant
How we elect our president is badly flawed:
- The Electoral College means that a candidate can get the most votes and still lose because of how those votes are geographically distributed.
- The use of choose-one Plurality Voting means that similar candidates can easily split the vote, third-party candidates can “spoil” the election, and voters’ preferences beyond their favorite candidate are irrelevant.
- Partisan primaries favor candidates who can appeal strongly to either the Democratic or Republican base; a candidate who appeals to Democrats and Republicans equally would have a difficult time winning either party’s primary since his support among Democrats would be useless in the Republican primary and his support among Republicans would be useless in the Democratic primary.
All of these problems were on display in the 2016 election. Clinton won the popular vote, but Trump won the presidency. The spoiler effect may have been decisive in swing states; Trump won Pennsylvania by 0.7%, but Johnson and Stein got a combined 3.2% of the vote. And Trump was extremely unpopular among Democrats compared to other Republicans, but he won the nomination handily since this didn’t matter. Some polling even suggests that Gary Johnson would have defeated either Trump or Clinton head-to-head, yet the use of Plurality meant that this wasn’t remotely apparent from the results.
But the failure of the American electoral system in 2016 isn’t the point I want to make here. Instead let’s ask: How relevant were these problems in the 2008 Presidential election?
By modern standards, 2008 was a blowout, and Obama probably would have won handily under any voting method. Obama didn’t win because of the quirks of the Electoral College; he won the popular vote by 7.3%. So can we say the Electoral College didn’t matter? Here’s a map of ad spending:
There was a similar pattern in campaign stops; FairVote counted 17 visits by the presidential and vice-presidential candidates to Colorado without a single visit to California.
One might argue that this is irrelevant. What I want from elected officials isn’t for them to bombard me with campaign ads, visit my neighborhood, and make me feel important; I want them to govern effectively in accordance with my values. But ad spending and campaign stops are also indicators of how presidents and other elected officials prioritize various constituents when making important decisions. The competitiveness of a state is a strong predictor of how likely a president is to declare a disaster in a state (for natural disasters like hurricanes) — and such disaster declarations come with billions of dollars for recovery efforts. Of course, presidents prioritizing some Americans above others isn’t limited to disaster declarations; they’re just the easiest to study since they’re amenable to statistical analysis.
Partisan primaries probably played an analogous role. Obama faced fierce competition from Hillary Clinton in the Democratic primary, but Republican voters couldn’t weigh in on that. My best guess is that Republicans would have preferred Obama to Clinton, so I doubt this altered who won. Nonetheless, it distorted Obama’s incentives. In the Democratic primary, Obama had a strong incentive to make campaign promises that would win over the Democratic base. In the general election, Obama had a strong incentive to win over undecided voters. But at no point did he have an incentive to appeal to the Republican base.
As with the Electoral College, the exclusion of Republicans (and many unaffiliated voters) from the Democratic primary wasn’t the reason Obama won. Neither problem changed the outcome of the election, but both the Electoral College and partisan primaries incentivized Obama to care less about some voters than others.
The Center Squeeze
Let’s turn from how we elect the President to Instant Runoff Voting (IRV, or single-winner Ranked Choice Voting). Here, the pathology in question is the center squeeze: a situation in which there’s a centrist candidate who would defeat any other candidate head-to-head, but doesn’t have enough first-choice support to make it to the final round of tabulation.
Just as the winner of the popular vote for US President doesn’t often lose the Electoral College, there have only been two IRV elections in the US where we have the data to prove a center squeeze occurred: the 2009 mayoral race in Burlington, VT, and the 2022 special election for US House in Alaska. When advocates of IRV discuss the center squeeze, their main argument is that center squeezes are rare enough that any issues are negligible. (Their secondary argument is a claim that center squeezes are actually good.) But the center squeeze is relevant in many elections without a centrist who gets squeezed for the same reason the Electoral College was relevant in 2008: it distorts incentives.
I studied this in my recent paper on Candidate Incentive Distributions:
Under IRV, the incentive for candidates to appeal to voters who are more supportive of them than the median voter is about three times as strong as the incentive to appeal to voters who are less supportive of them than the median voter. This is because of the center squeeze; Minimax, a method that is not vulnerable to the center squeeze and yields different outcomes from IRV mainly in center-squeeze scenarios, gives candidates far more balanced incentives.
More concretely, we can consider the 2022 race in Alaska State Senate District E as an analog to the 2008 Presidential race.
There was no center squeeze in this election; to the best of my knowledge, it was not pathological at all. Giessel would have won under any reasonable voting method, and there was nothing strategic voters who preferred a different candidate could have done to elect someone they liked more. I’ve criticized Alaska’s system for replicating the problematic structure of partisan primaries previously, but that didn’t happen here. The final round saw both dominant-party candidates face off, and Democratic voters’ preferences for Giessel over Holland were reflected in the tabulation.
A different picture emerges when we look at the incentives. Giessel won in the final round with a 1,932-vote margin. On the other hand, Giessel avoided elimination in Round 1 with a 134-vote margin. Had 135 of her first-choice supporters stayed home there would have been a center squeeze, so her incentive to get first-choice support was dramatically stronger than her incentive to be ranked over Holland by Democratic voters. If she could have won over a Holland>Giessel>Cacy voter at the expense of causing three Cacy>Giessel>Holland voters to vote Cacy>Holland>Giessel instead, it would have been worth it. Giessel may still have had an electoral incentive toward moderation to convince center-left voters to rank her first instead of Cacy, but she also would have had this incentive under Plurality. There wasn’t a center squeeze in this election, but it came close enough to happening that the possibility of it dominated the incentives faced by Giessel.
Interestingly, Giessel didn’t follow the incentive to prioritize voters who might rank her first. She knocked on doors regardless of partisan affiliation. Her campaign paid for a mailer urging Democrats to vote Cacy>Giessel>Holland. She ran a campaign that would have made perfect sense if Alaska used a Condorcet method instead of IRV, while Holland focused on first-choice support. It’s like one presidential campaign pursuing a 50-state strategy while the other targetted swing states: foolish for the purpose of winning, but there’s a lot to admire in valuing all voters equally. We just need to elect people — both for president and state senate — in a way that makes the admirable strategy the smart strategy.