RCV and Core Support

When Instant Runoff Voting (IRV, more commonly known as single-winner Ranked Choice Voting) fails to elect the Condorcet winner, people who prefer other non-Plurality voting methods interpret it as a failure of IRV and say that IRV elected the wrong winner. This happened in the 2022 special election for the House of Representatives in Alaska, in which Republican Nick Begich lost the election despite being preferred over both the other candidates, Democrat Mary Pelotola (by a 7-point margin) and Republican Sarah Palin (by a 22-point margin). Do the supporters of IRV agree that this is a problem? Here’s what Steven Hill has to say concerning advocates of non-RCV voting methods:

It never occurs to these advocates that, given the fact that Begich is the “Condorcet winner” in the Alaskan election, this is clear and compelling proof that the Condorcet winner standard is as obsolete as its 18th century origins.

This isn’t especially surprising; if he thought that the failure to elect the Condorcet winner was a problem, he and other RCV advocates would probably just recommend switching to using a Condorcet method.

Why does Steven think IRV yielded a better victory than Condorcet here? Because IRV takes into account something that he calls core support.

Here’s what I mean. With RCV, winning candidates in a single-winner contest (like a governor, mayor or a representative of a legislative district) must have BOTH a broad base of support AND a strong core of support. RCV rewards both deep support as evidenced by a high number of first rankings, and broad support, as evidenced by many backup rankings. But with Condorcet and these other methods, winners only need to have a broad base of support, and do not need much core support.

Let’s say in an RCV election there is a candidate who is ranked second on every single ballot in the election. That would be a clear indicator of broad support. But let’s say that same candidate is ranked first on only a single ballot. That candidate will be the first one eliminated from the race, because that candidate did not have strong enough core support — enough first rankings — to prevent being rejected early in the round by round vote tally.

So core support is at least correlated with being ranked first by a lot of voters, and broad support is correlated with receiving a somewhat high ranking by even more voters. Steven argues that having both strong core support and broad support is essential:

RCV rewards candidates who show real leadership and can attract a nucleus of voters who strongly believe in that candidate. But not exclusively, because the truly defective plurality voting method also rewards candidates with a strong core of support. So RCV also rewards candidates that have a broad base of support. The values and principles of RCV are boldly declaring: “A candidate who has virtually no core support, even though the candidate did have broad support, that candidate should not win. You need BOTH core and broad support to win an election under RCV.”

But with Condorcet voting, that same candidate who is ranked second on every ballot and first on hardly anyone’s ballot will be the winner. Being no one’s first choice but everyone’s second choice is not a losing strategy in Condorcet voting, quite the contrary. It encourages candidates to be bland and take few strong stands on anything. They are incentivized to water down their true beliefs and hide them from voters so as to not offend or be opposed by anyone.

As someone who strongly prefers Condorcet to IRV (I prefer the term “IRV” since “RCV” is an imprecise umbrella term), I agree with Steven that something along the lines of broad support is important; the crux of our disagreement is over core support. But this begs the question: What, exactly, is core support?

In this post, we will start by considering a definition of core support as the amount of first-choice support a candidate has and why this definition is problematic. We will move toward a definition based on cardinal preferences and design some cardinal voting methods that implement it more concretely. We will then consider the question of whether IRV does a good job of ensuring the winner has a large amount of core support, as defined based on cardinal preferences. Finally, we will answer Steven’s most important question: “What are the values in your voting system?”

The limitations of ordinal preferences

Steven seems to be roughly conflating core support with first-place rankings, writing that “RCV rewards both deep support as evidenced by a high number of first rankings” and claiming that a candidate who was ranked first on only a single ballot but second on all other ballots wouldn’t have enough “strong core support”. This probably isn’t an exact equivalence — “as evidenced by” implies that there’s more to it, and maybe strong core support is a separate concept. But for now, let’s suppose that core support = first-place rankings. Where does that lead us?

By this standard, in the actual election, Nick Begich had less core support than either Peltola or Palin. Okay, this seems like it could be a good reason for him to have lost. But let's step back from the actual election and imagine a god’s-eye view: Suppose you can read the minds of every registered voter in Alaska, and know every nuance of their opinions towards each candidate and all of their views on all the issues, ranging from tax policy to natural resource extraction to healthcare. With this knowledge you are asked to answer a question: How well would each of the candidates represent the people of Alaska?

Steven almost certainly thinks that, were he given this knowledge, he would most likely conclude that Pelota would represent the Alaskan people better than Begich. But there’s an important subtlety at play here: How well Petola represents Alaskans depends only on Peltola (along with the Alaskan voters we’re trying to have represented). How well Begich represents Alaskans depends only on Begich. Whether Petola or Begich better represents the Alaskan people can only depend on Petola and Begich. It cannot depend on whether or not Palin is also under consideration.

From the god’s-eye perspective, we may still care a great deal about something along the lines of core support, and I expect Steven would indeed deem it important. But this cannot simply be first-choice support since how much first-choice support Begich gets depends on whether Palin is in the race. To equivalate first-choice support with core support is to give the spoiler effect a place within our values. Imagine the following dialogue with me:

You: Who do you think would better represent you, Peltola or Begich?

Me: Hmm, I’ll go with Begich.

You: I forgot to mention, you can also choose Sarah Palin.

Me: Oh, I didn’t know Palin was an option! In that case, I think that Petola would represent me better than either Palin or Begich.

From this exchange, you should conclude that I am stark raving mad. It would be reasonable for me to switch from Begich to Palin or to stick with Begich, but switching from Begich to Peltola, when I had just decided that Begich represented me better than Peltola did, makes no sense whatsoever. And it doesn’t matter if I’m being asked to decide who would represent me better or who would represent all Alaskans better. But this is what happens when core support is considered to mean first-choice support. With Palin out of the race, Begich has more first-choice support than Peltola. With Palin in the race, he has less first-choice support than Peltola.

(Readers who are familiar with Arrow’s theorem may feel a bit confused. Isn’t it provably impossible to create a reasonable voting method in which the addition of one candidate (Palin) in the race can never shift the winner from one candidate (Begich) to another candidate (Peltola)? This is, indeed, impossible — but I’m not making a claim about voting methods. Rather, I’m considering an idealized preference aggregation system that doesn’t make use of ballots and doesn’t even involve comparisons between candidates. While this isn’t useful for actually selecting members of Congress, it is useful for formalizing desiderata for a voting method. No voting method will be able to satisfy our desideratum perfectly on account of us being limited to the information that’s submitted on ballots, but some voting methods will still outperform others.)

More generally, ordinal preferences of any kind cannot be what we care about intrinsically. (The word “intrinsically” is doing a lot of work here — we’ll get to that later.) To claim otherwise is to say that the addition or removal of non-winning candidates to or from a race can affect who should win; it is to make our values themselves subject to the spoiler effect.

Core support with cardinal preferences

While we can’t define an intrinsically important version of core support based on ordinal preferences, we can define one based on cardinal preferences. Ordinal preferences say who you think is the best candidate, who you think is second-best, who you think is worst, etc. An example of ordinal preferences is, “I prefer Palin to Begich and Begich to Peltola”. Cardinal preferences say how good you think every candidate is, or at least how good the candidates are relative to one another. So “I think Palin is great, Begich is just okay, and Peltola is terrible” is one example of cardinal preferences. “Palin is my favorite candidate and I greatly prefer her to the other candidates, but I slightly prefer Begich to Peltola” is another example of cardinal preferences. Ordinal preferences say nothing about how big the difference between two candidates is; cardinal preferences do tell you this.

Which cardinal preferences correspond to being a core supporter of a candidate? Let’s consider a few hypothetical cardinal preferences, and see if these voters should be considered core supporters of Nick Begich:

1. “I like Begich a lot and he’s my favorite, but Palin is nearly as good. I’d never support a Democrat.”
2. “I’m a lifelong Democrat and I think Peltola is amazing, but among the Republicans, Begich is the more acceptable.”
3. “I’m a committed Republican, and it’s hard to say! If you forced me to choose then I guess I prefer Palin to Begich by a hair, but they’re both great candidates. Peltola is unacceptable.”
4. “I think all three of the candidates are terrible, but I guess Begich is the least awful of them. Even though all the candidates are basically the same.”

(1) is clearly an example of core support, and (2) is clearly not an example of core support. Less obviously, (3) must be an example of core support. This voter really likes Begich and would enthusiastically rank him #1 if Palin wasn’t in the race. Having (3) not be an example of core support would mean that our definition of core support was subject to the spoiler effect. Finally, (4) seems like it really shouldn’t be an example of core support; if there was a fourth candidate in the race who this voter actually liked there is no way we’d say he was a core supporter of Begich. However, pretty much any voting method we come up with would wrongly consider him to be a core supporter of Begich (as IRV does) since he has no reason not to give Begich the maximum amount of support. (We could try something unorthodox like quadratic voting or giving voters the option of standing in line an additional three hours in order to cast “core support ballots”, but these options have their own problems.)

It’s worth noting that embracing a cardinal notion of core support that says that Peltola represents Alaskans better than Begich entails biting a bullet in admitting that IRV (along with every other voting method) would have elected the wrong winner in a two-way race between Peltola and Begich. But this is a pretty reasonable bullet to bite. If we can about the intensity of voters’ preferences (as I think we should), then sometimes the candidate with fewer supporters in a two-way race is still the one who should win.

Cardinal voting methods that ensure core support

Let’s try designing a cardinal voting method that, in Steven’s words, elects “winners with a broad base AND core support”. We’ll make sure that (1) and (3) are counted as core support and that (2) isn’t, and we’ll give up on classifying (4) correctly.

For a first attempt, let’s (a) make a numerical definition of how much core support each candidate has; (b) make a numerical definition of how much broad support each candidate has; (c) multiply the two numbers together to create an overall support score that can only be high if a candidate has both core support and broad support; and (d) elect the candidate with the highest overall support score. We’ll use 5-star ballots, where voters score the candidates from 0 to 5 and can give multiple candidates the same score; this will allow for easy comparisons with STAR Voting. (We could also have a STAR-style runoff between the two candidates with the highest overall support score to encourage voters to use a wide range of scores and vote more honestly, but let’s not focus on such optimizations.)

For (a), if we’re using 5-star ballots it’s clear that every voter who gives a candidate 5 stars should be counted as a core supporter. Should 4-star supporters also be counted? Hmm, maybe — or maybe they should be counted somewhat, but substantially less so than 5-star supporters. We can try measuring core support by taking the sum of the cubes of all the scores given to a candidate, so a ballot that gives the candidate 5 stars contributes 125 points, a ballot with 4 stars contributes 64 points, and a ballot with 3 stars contributes a mere 27 points — about 22% of the support provided by a score of 5.

For (b), we can consider the square root of the candidate’s scores — a score of 1 star contributes one point, and a score of 4 contributes 2 points. So for maximizing broad support, it’s important to turn 0s into 1s and 1s into 2s, but turning a 4 into a 5 doesn’t matter much at all — the opposite of the incentives for maximizing core support.

The fact that we’re multiplying core support and broad support together means that, for a candidate with a lot of core support and little broad support, additional broad support is more valuable than additional core support. Similarly, for a candidate with a lot of broad support and little core support, additional core support is more valuable.

How does this voting method compare to STAR? Suppose candidate A receives 5 stars on 40% of ballots and 1 star on 20% of ballots, and 0 stars on all other ballots. Candidate B receives 3 stars on every single ballot. (There can also be other candidates in the race, but we’ll assume they aren’t very popular.) Under STAR, Candidate B absolutely crushes Candidate A, receiving both a much higher average score and easily winning a runoff between the two of them — and I’m pretty sure Steven would consider this a travesty since Candidate B is absolutely nobody’s first choice.

How does this go under the (core*broad) method? Let’s suppose there are 5 voters. Candidate A has a core support of 5³*2 + 1³*1 = 251 points, and B has a core support of 3³*5 = 135 points. Candidate A has a broad support of 2*sqrt(5) + sqrt(1) ≈ 5.47 points, and B has a broad support of 5*sqrt(3) ≈ 8.66 points. So candidate A wins, with an overall support score of about 1373 compared to B’s 1169.

However, I’m not convinced that this voting method is properly in line with Steven Hill’s values. Suppose that every single voter who ranked Begich first and didn’t mark a second choice had instead ranked Palin #2. In this case, Palin would have beaten Peltola in the final round — and I’m virtually certain that Steven would say that this would be the correct outcome since Palin, unlike Begich, had enough core support, and this added support from Begich voters would have given Palin enough broad support as well. But Palin was only the first choice of 31.3% of voters, compared to 28.5% of voters for Begich. Numerically, this is not a large gap. For instance, it is roughly half the size of the gap between Begich and Peltola in a head-to-head contest (which Begich would win with the support of 52.6% of voters who expressed a preference). But Steven considers it to be a very big deal:

It never occurs to these advocates that, given the fact that Begich is the “Condorcet winner” in the Alaskan election, this is clear and compelling proof that the Condorcet winner standard is as obsolete as its 18th century origins.

Think about it: a candidate who finished in third place, who was nearly 12 points and 22,000 votes behind the eventual winner, and 2.7% and 5000 votes behind the second-place finisher, and who would not have won this election in Alaska’s old closed primary system either, because he would have lost to Sarah Palin in the Republican primary; and who has lost to Palin three times now, including in 2 primary elections and 1 general election…yes, THAT candidate…under Condorcet rules…is the “real winner.”

In fact, using Condorcet rules Begich would have stomped Palin by a landslide of 22 points and Peltola by six points. So according to the Condorcet standard, a candidate who finished in a distant third place, who was in fact dead last, is actually the legitimate winner by a wide margin.

Huh?

No wonder the Condorcet voting method is not used for any public elections anywhere in the world. In fact, I don’t know of any country or sub-country that uses an election method in which a candidate can win after finishing dead last. The idea that Begich is the Condorcet winner is kind of an “alternative universe” argument to make. As we saw in that Alaska election, the Condorcet winner criteria is saying that a more popular candidate (either Peltola or Palin) spoiled an election for a less popular candidate (Begich).

(Steven is wrong about public uses of Condorcet; the Schulze method is used by several European cities and political parties.)

Okay, so apparently we need a voting method in which the 2.7% gap between Palin and Begich is much, much more important than the fact that Begich would defeat Peltola head-to-head by over 5 points. But, in order to keep the spoiler effect out of our values, we have to express this in cardinal terms and not focus on the differences between particular candidates.

Since Steven clearly thinks Begich didn’t have enough core support, perhaps we can implement this requirement as a threshold? Let’s make another voting method that uses 5-star ballots:

1. First, for each candidate, determine the fraction of ballots on which they are given a full 5 stars. This is their core support. If no candidate’s core support exceeds 30%, elect the candidate who received 5 stars on the greatest number of ballots. Otherwise, eliminate all candidates with less than 30% core support.
2. Elect the Condorcet winner from among the remaining candidates. (Condorcet cycles can be dealt with using Smith//Score.)

This seems to fulfill Steven’s requirements; a candidate with little core support can’t win, and Steven considers the Condorcet winner to be a candidate with a great deal of broad support. But might it go too far? There are plenty of IRV elections in which no candidate received over 30% first-choice support. However, all of these elections involved more than three candidates. As shown earlier, we cannot conflate core support with being a voter’s #1 support. When there are more candidates, voters will be more likely to be core supporters of multiple candidates. It may seem extreme for a candidate who is given 5 stars by 31% of the electorate to defeat a candidate who is given 5 stars by 29% without any regard for the preferences of other voters, but I don’t think it’s more extreme than what happened in Alaska.

What have we found here? First, we can meaningfully formulate the idea of core support in terms of cardinal preferences, such that core support can be something one can care about intrinsically without making their values subject to the spoiler effect. Second, we can construct cardinal voting methods that elect winners who have both strong core support and strong broad support, but don’t elect candidates with only one of the two. Granted, these voting methods aren’t necessarily any good — they’re significantly more complicated than popularly-advocated voting methods and I haven’t even tried analyzing them from the perspective of strategic voting. But they work as a proof of concept.

IRV and (cardinal) core/broad support

A formulation of core support based on ordinal preferences, unlike a formulation based on cardinal preferences, is something we cannot care about intrinsically. IRV only considers ordinal preferences. Based on this, can we conclude that Steven and everyone else who talks about “core support” as the reason for preferring IRV to Condorcet have screwed up and would have their values better satisfied by some cardinal voting method?

Not even close. There are tradeoffs to consider; IRV is probably better than these cardinal methods in terms of simplicity and strategic straightforwardness. And while IRV cannot directly embody a cardinal formulation of core support the way a cardinal voting method can, perhaps it can still do very well at “usually electing winners who have the best combination of core support and broad support in the cardinal sense”. While we cannot care about an ordinal formulation of core support intrinsically, we can still care about it as a proxy for some cardinal formulation that does have intrinsic significance.

(The cardinal formulation of core support lets us answer another question: If core support is so important, why is IRV better than “Top Two IRV”, in which the two candidates with the most support advance to the instant runoff without any additional rounds of elimination and transfers? Answer: Because some second-choice support is core support, and IRV captures this whereas Top Two IRV does not.)

Let’s imagine an “alternate history” version of the Alaskan election in which both Palin and Begich had 2.7% more first-choice support, and all of these additional supporters ranked the other Republican second; Peltola would have 5.4% less first-choice support. In this alternative history, Palin would have won the election, easily defeating Peltola in the final round. But Begich would still crush Palin head-to-head by at least a 16-point margin. Most significantly, Begich would have had as much first-choice support as Palin did in the actual election. Unless we deny that Palin had enough core support in the actual election, we must conclude that, in this alternate election, something went wrong. (And, of course, saying that whether or not Begich has enough core support to deserve to beat Peltola depends on how much support Palin has results in incoherent values that are susceptible to the spoiler effect.) Begich would have enough core support to be eligible to win and vastly more broad support than Palin, but Palin would still beat him. The problem is that IRV completely ignores the fact that Peltola voters overwhelmingly prefer Begich to Palin — this broad support doesn’t factor into the IRV algorithm in the slightest.

Broadly speaking, ignoring voter preferences is a really inefficient thing to do no matter what kinds of winners you’re trying to elect. IRV does ensure that a candidate with only a tiny amount of first-choice support can’t win, but it does this by outright ignoring many candidates’ second-choice support — including some candidates that do have sufficient core support. But this inefficiency can be a price worth paying — if core support is so overwhelmingly crucial.

Values

Steven says that his disagreement with people who think Begich should have won instead of Peltola ultimately boils down to a disagreement over what values our voting methods should express:

[I]t really comes down to what values and principles are embedded into an electoral system. The voting method used is fundamental to any political system because when you select a particular method for your elections you are selecting a set of values and philosophy of government, as well as a range of accompanying effects and consequences.

For the purposes of this discussion, it’s really not a matter of which system is better, either ranked choice voting or Condorcet or approval voting or range voting — it’s a matter of what are the incentives for winning? What kinds of candidates do you want to reward with an electoral victory?

I agree. I strongly agree. My experience is that RCV advocates tend to talk about incentives a lot more than the advocates of other voting methods do, and incentives matter more than picking the best winner. But nonetheless, I much prefer the incentives provided by Condorcet and STAR Voting.

Here are the values that make me prefer STAR and Condorcet to IRV:

• I think the opinions of all voters should matter equally.
• I think candidates should be equally incentivized to appeal to all voters.

How do these three options compare in terms of these values? Consider two voters. One is a diehard Republican who greatly prefers both Palin and Begich to Peltola. Another is a diehard Democrat who greatly prefers Peltola to both Republicans. However, both are deliberating over which Republican is the better choice. What incentives do Palin and Begich have under each voting method?

• Under (most) Condoret methods, if the Republican votes Palin>Begich>Peltola and the Democrat votes Peltola>Begich>Palin, their ballots will have equal and opposite effects, canceling one another out. Palin and Begich have equal incentives to appeal to the Democrat and the Republican.
• Under STAR Voting, if the Republican votes Palin: 5, Begich: 4, Peltola: 0, and the Democrat votes Peltola: 5, Begich: 1, Palin: 0, these ballots again will have equal and opposite effects on the outcome. Again, Palin and Begich have equal incentives to appeal to the Democrat and the Republican.
• Under IRV, if the Republican votes Palin>Begich>Peltola and the Democrat votes Peltola>Begich>Palin, the Democrat’s preference for Begich over Palin will not affect the outcome (barring a massive swing that results in Peltola having the least first-choice support). Palin and Begich have no incentive to appeal to the Democrat.

Part of this extreme discrepancy is due to the inefficiency of IRV in ignoring certain preferences altogether, but another part is an inevitable consequence of valuing core support. If core support is part of a voting method’s values, then candidates must care about potential core supports more than about voters who cannot be persuaded to become core supporters. This inequality of incentives isn’t some side effect, it’s the goal.

Steven implies that this difference in value is irreconcilable:

So the next time you are at a dinner party with your politico acquaintances and wonky confrères, and you hear someone suddenly hiss with conviction, “But Begich was the Condorcet winner!”, just smile and nod and then make a dash for the rest room. Because this is an argument that you cannot win, since the different electoral systems are by definition trying to elect different kinds of candidates. Instead, you have to first decide what values and goals you are trying to fulfill, and then design your electoral system to fit that mission.

I agree with him on a high level, but I disagree on a lower level that I consider to be more relevant. A fundamental moral disagreement is usually intractable. But I don’t work on voting methods because I think that certain democratic processes are intrinsically morally superior. I work on voting methods because I think certain democratic processes will lead to better outcomes — and I don’t think Steven and I differ all that greatly in how we would define “better” here.

I see the division in American society and I am terrified. A civil war following a contested election, of Democrats versus Republicans, seems like a real possibility. Democratic backsliding, in which one or both sides decide it’s okay to bring an end to American democracy in order to prevent the other side from gaining power, is even more likely (if not already happening). And on top of that, we have congressional dysfunction stemming from politicians being incentivized to appeal more to primary election voters who hate compromise than to general election voters on the other side of the aisle who want it.

I believe that incentivizing politicians to care about all voters equally would help with these problems immensely. If Democratic voters can meaningfully support Republican candidates who say the 2020 election was legitimate over those who say it was fraudulent (while still offering greater support to Democratic candidates), Republican candidates will be a lot less inclined to claim fraud. More generally, if alienation is bad politics then politicians will stop making our politics so divisive. Equitable incentives can save our democracy.

Steven does offer an argument for why a requirement for core support will lead to better outcomes:

But with Condorcet voting, that same candidate who is ranked second on every ballot and first on hardly anyone’s ballot will be the winner. Being no one’s first choice but everyone’s second choice is not a losing strategy in Condorcet voting, quite the contrary. It encourages candidates to be bland and take few strong stands on anything. They are incentivized to water down their true beliefs and hide them from voters so as to not offend or be opposed by anyone.

I have several disagreements with this.

First, this argument subtly misrepresents Condorcet. Steven writes,

The values and principles of RCV are boldly declaring: “A candidate who has virtually no core support, even though the candidate did have broad support, that candidate should not win. You need BOTH core and broad support to win an election under RCV.”

But the key word here is not “BOTH”, the key word is “need”. Steven is right that it’s mathematically possible to have a Condorcet winner who has no core support whatsoever, but Condorcet still incentivizes core support; candidates still prefer to be ranked #1 to being ranked #2. It’s just that, under Condorcet, core support and broad support are more fungible than they are under IRV. Condorcet lets candidates make up for having relatively little core support by having a lot of extra broad support; IRV says that if a candidate doesn’t have enough first-choice support, whatever broad support that candidate has is irrelevant.

Second, I think the factual claim (“It encourages candidates to be bland and take few strong stands on anything”), insofar as it’s actually correct, refers to a mild tendency. If Condorcet were adopted nationwide, I agree that it’s at least possible that we’d see more candidates dodging questions on controversial issues. But there wouldn’t be a tidal wave of word salad coming to replace substantive debate. As I’ve written previously about the center squeeze:

The 2016 presidential election seems a lot like a center squeeze — Gary Johnson seems far more acceptable to Democrats than Trump and far more acceptable to Republicans than Clinton. Actual research shows that there might have been a center squeeze, but the “centrist” who got squeezed out was Bernie Sanders. Head-to-head, Johnson would have lost in a landslide to either Trump or Clinton.

I don’t think it’s unreasonable to say that Gary Johnson seemed a lot blander than Bernie Sanders. But, for the purpose of becoming a Condorcet winner, blandness didn’t pay. I don’t know of any empirical evidence that being bland is an effective strategy under Condorcet methods. Granted, this isn’t saying a great deal since we don’t have much empirical evidence — but the evidence I do know of points in the opposite direction.

Another question that Steven doesn’t address: How can a candidate be “everyone’s second choice” and win while being bland? This seems basically impossible in a two-party system; if all the Republican voters prefer the bland candidate to all the Democratic candidates and the Democratic voters prefer the bland candidate to all the Republican candidates, either a Democrat or a Republican will still beat the bland candidate head-to-head unless the bland candidate has an appreciable amount of first-choice support. For being “everyone's second choice” via blandness to be a good campaign strategy, it seems there would have to be at least three factions that all hated one another enough that they’d rank Blandy McBlandface over every single candidate in the other factions. While mathematically possible, it seems like a stretch sociologically.

Third, this argument seems a whole lot weaker than the one for equal incentives. I mean, I prefer for candidates not to be bland, but this problem seems a whole lot smaller than that of a possible civil war. Steven makes it clear that he’d consider it an utter travesty if a candidate with less than 30% first-place support won a three-way election, and that Condorcet advocates are bizarre mutants for thinking that such a thing could possibly be reasonable. I find it hard to understand how the possibility of a few more bland-seeming candidates can justify this vehemence.

There’s a sense one gets from reading Steven’s post that IRV values both broad support and core support, whereas Condorcet only values broad support. But he’s careful never to say this explicitly — he knows it’s false. The real difference is that IRV is absolutely uncompromising in its demand for core support — and to have this demand satisfied, IRV goes to the extreme of ignoring some voter preferences altogether.

Under Condorcet methods, every preference for Begich over Peltola, or for Peltola over Begich, matters equally. It doesn’t matter how you feel about Palin — your preference will be counted equally. But the values of IRV — “A candidate who has virtually no core support, even though the candidate did have broad support, that candidate should not win. You need BOTH core and broad support to win an election” — contradict this. When core support is what’s necessary, candidates must put the needs of voters who may give them core support over the needs of voters who almost certainly won’t. Under IRV — or Alaska’s old system of partisan primaries — Palin and Begich had a strong incentive to be the first choice of reliably Republican voters who were always going to rank one of them #1. And they had had a strong incentive to appeal to “swing” voters who could vote for someone of either party. But they did not have an incentive to appeal to voters who were always going to rank Peltola first.

In deciding between the values of IRV and the values of Condorcet, a single question suffices: Should politicians (and elected officials) primarily care about their core supporters, or should they care about all voters equally? If you think politicians should care about their core supporters more than other constituents, and that a moderate like Begich is unfit to lead because he failed to properly prioritize potential core supporters over other voters (e.g. committed Democrats), then IRV fulfills your values. But if you think everyone’s preferences should matter equally, and that politicians should value all voters equally, then Condorcet (or STAR) is what you want.