Hacker News Reader

RCV != IRV.

Everyone is reporting this as a failure of Ranked Choice Voting, but it's actually a failure of Instant Runoff Voting. There are other, better, more reliable Ranked Choice alternatives, and any serious expert in voting systems would tell you that IRV has as many weird failures as FPTP.

http://zesty.ca/voting/sim/ is a good overview of different methods.

In US political vernacular, RCV == IRV. This is just a case of terminology getting warped, but it is what it is.

Yeah, but we should push back against this, lest it poison every attempt at getting better non-IRV RCV implemented.

that's the idea behind falsely mislabeling irv as rcv...

I think it's far too late for that.

I don't know if I have ever seen any movement to push back against "wrong" vernacular succeed. Consider the more recent "coronavirus" which is equivalent to COVID-19 in the vernacular, even though it's really just a family of viruses and COVID-19 is an illness caused by a particular coronavirus instance (SARS CoV-2).

Agreed - you can't fight back against labeling by telling people they are wrong. Instead to you must fight-forward with a new proposal going in a different direction.

99% of people who call it "RCV" have no idea how it works. Just call it "Hare RCV" in the same context as "Schulze RCV" or "Minimax RCV" and they'll be forced to confront their ignorance.

Honestly I have the term RCV. It is a specific type of ranked choice systems and now we have to refer to that class as Ordinal to not be confused with the specific variant. And it leads to a lot of misunderstandings like this.

This whole conversation is a good demonstration of why FPTP succeeds and is popular -- it is easy to understand, and most of the other systems are not. When a voting system is hard to understand, people will tend to see conspiracies and fraud even though there isn't any, and they will lose trust in the system.

With FPTP, every voter casts one vote for one candidate, and the candidate with the most votes wins. Simple.

If you need a website with diagrams to explain the abstruse differences between different voting mechanisms, I think you've already failed. This stuff is certainly of academic interest but the average voter cannot handle it.

This is why I endorse STAR voting: https://www.equal.vote/star. The explanation behind it is simple enough for most people to understand, because it's a combination of 5 star ratings, which are everywhere in society, and FPTP.

At worst, you've made single vote FPTP with extra steps. Give your guy 5, everyone else 0. The goal is to get in the runoff and to keep others out. If everyone adopts this strategy, it's the same as FPTP.

At best, you're made RCV/IRV with extra steps. Because still you give your guy 5 stars, but now there's people you do not want at all. They're 0s. Then there are people you're ok with, but would rather have your guy win. They're 1s. That way, if the runoff is between one of your 0s and one of your 1s, you're making sure a 0 doesn't win. Which is exactly RCV/IRV where you've ranked them 1, 2, and 3.

Past that, candidates towards the bottom of the ballot are going to get shafted. There is incredible decision fatigue in having to rank everyone 0 - 5 stars. After a while, you will either give up or just start picking more arbitrarily.

RCV/IRV is fine. Condorcet is fine. Star voting is just bad.

https://www.equal.vote/accuracy

Star voting outperforms IRV on both 100% honest and 100% strategic election strategies.

Decision fatigue is easier for Star than for IRV—just rate everyone you don't care about as a 2, whereas for IRV, if you want to distinguish between "8 people I don't know" and "person I hate", you have to decide on a ranking order for the 8 people.

> At best, you're made RCV/IRV with extra steps

I'd argue that every single word in this small claim is wrong.

Scoring isn't equivalent to ranking unless the conditions are that you cannot score two candidates the same and that you must fill in every possible score vale (i.e. rank). Neither of these conditions are true for any cardinal system. You can score candidates the same and you can score candidates with variable distances (e.g. 5,3,2,1. Note there is no 4). This has different embedding properties. You actually can express more information in cardinal this way.

Second, run-off voting is much more complicated. Cardinal systems are determined by argmax of a reduce-sum. Which is trivially parallelizable and only requires a singular election. Run-offs require multi-elections. Even in simple cases like this one, there were 2 elections (and that's with 3 candidates!). This doesn't scale well either.

But as we've seen in the UK, and the US it leaves a great deal of people, especially those in safe (or gerrymandered) districts feeling unrepresented.

UKIP getting 13% of the vote in a general election, showing a large amount of support, but not winning a single seat is, in my opinion, a travesty (and I vehemently disagree with their platform and what they stand for). It left 13% of the electorate feeling that their views were not represented in any way whatsoever .

That seems like more of a problem with proportional representation than ranked-choice voting.

If only 13% of voters support UKIP, UKIP would not win elections no matter what kind of voting system you have: ranked choice, instant runoff, etc.

On the other hand, with proportional representation, if a party gets 13% of the vote it would end up with 13% of the seats in Parliament. That's a different type of system, though, and it is hard to reconcile with a per-constituency election system.

> That's a different type of system, though, and it is hard to reconcile with a per-constituency election system.

One way to reconcile it is with different legislative bodies. Exchange the house of lords for a proportional body, and keep the commons local to constituencies.

FPTP has more failures than IRV in general, they're just less visible because you don't have the ranking data to work out the hypotheticals. That may be good for FPTP's popularity, but doesn't make it a decent voting system.

FWIW I also disagree with the suggestion that people can't handle it. Plenty of countries use ranked or proportional voting systems and their citizens understand how to use them just fine. They may not be able to perfectly optimise their preferences, but then they can't do it in FPTP either (and it's provably impossible to do so in every case whatever the voting system).

ELI5: why do we care about Condorcet failures?

From https://fivethirtyeight.com/features/what-democrats-win-in-a...

> 40 percent of voters had chosen Peltola as their first choice, 31 percent had chosen Palin and 29 percent had chosen Republican businessman Nick Begich III. Under the rules of ranked choice voting, Begich — as the candidate with the fewest first-choice votes — was then eliminated, and his votes were redistributed to whomever his voters ranked second.

> Unsurprisingly, most of Begich’s votes (50 percent) went to his fellow Republican, Palin. But an impressive 29 percent went to Peltola, and 21 percent were “exhausted,” meaning there was no second-choice pick, and the votes were essentially thrown out. That combination was enough for Peltola to win. While Palin gained more votes from the redistribution than Peltola did, Peltola was starting from a higher total, and receiving 29 percent of Begich’s votes was enough to keep her ahead of Palin. In the end, Peltola received 51 percent of the votes counted in the final round, while Palin received 49 percent.

Sounds fair to me. Some of Begich’s voters expressed a “Begich or nobody” preference. Many Begich voters expressed a “Peltola over Palin” preference. It’s not clear to me why pairwise preferences are meaningful, they don’t encode nuances like the above. Though maybe that’s a feature and not a bug?

I’m aware this gets pretty technical with Arrow’s theorem, maybe there isn’t an ELI5.

The problem with IRV is that it asks for a full list of preferences but doesn’t actually respect the voters full preference, and specifically it violates monotonicity.

This short video gives an example that’s easy to understand:

https://youtu.be/Xekj6gycFpU?list=PLz0n_SjOttTcmP-7y4beuYv49...

Monotonicity intuitively seems like it would be desirable, but I'm struggling to see how the effect described in the video actually matters in practice. It seems vanishingly unlikely that candidates 2&3 would be eliminated in a tie (and you can just change the rules to force a re-vote/runoff in that case if you really think it's a bad outcome, similar to how you'd handle a tie in FPTP). And even then, the outcome of picking the first candidate when 2&3 are tied doesn't seem explicitly unfair. I'm sure there are other cases where this can show up too though, and maybe some of those do seem unfair.

I can see how ex post, one could say "if I had ranked my choice lower, they would have done better", but it's not clear to me that this property actually informs any real-world voting strategy ex ante.

Basically, this seems like a somewhat academic property that most voters probably wouldn't care about, so it's not clear that a lack of monotonicity would actually impede IRV from popular acceptance.

You’re totally right, but I don’t think I’ve ever heard RCV being called more consensus-oriented. Just that it makes it less harmful for third party candidates to participate in elections since they won’t play spoiler in the same way (Ralph Nader). I think the success of RCV just has to do with its relative simplicity compared to other alternative voting methods.

Even leading up to this election, I saw some mainstream coverage of the fact that the election would come down to who finished 3rd. i.e. if Palin got eliminated in the first round, most of her votes would’ve gone to Begich and he probably would’ve won. Doesn’t speak well for RCV that it can come down to such a slim margin, but I still think RCV > FPTP for the third party reason above.

I agree that RCV is better than FPTP, but only marginally. It's not clear to me if the marginally better system is worth the complexity (in the eyes of some voters, at least).

As you allude to, RCV is fine when the minor parties essentially have no chance at winning. They get knocked out in the early rounds and then don't matter. In cases where two candidates have similar vote shares, the result is fairly chaotic.

I've taken to looking at "RCV" as the UI/interface, which can then have multiple implementations; IRV being only one of them. Condorcet would be another implementation. (Of course, many IRV adherents would object to this interpretation.) I think it's fundamentally dishonest to take a term like "ranked choice voting" and insist that it means IRV and IRV only.

If you can get a city/county/state to adopt Condorcet, by all means have at it. IRV is much better than FPTP because FPTP encourages candidates to drop out of a race so that their supporters can unite around one ideologically aligned candidates - meaning we lose the "signal" from those other candidates.

IRV, on the other hand, does not encourage candidates to drop out, allowing the voters to express their views more clearly.

The Alaska data shows that this is not the case. For instance, if Palin had dropped out, Begich would have won. Palin pulled enough votes from Begich in the first round that Begich was eliminated, even though he would have beaten Peltola in the final.

Almost all Palin voters would have preferred Begich to Peltola, so they could have gotten a better result by lying about their preferences and strategically voting for Begich as their first choice. Expressing their views honestly resulted in a worse outcome than if they had expressed their views dishonestly.

Among voting nerds this failure mode is referred to as "favorite betrayal" -- https://electowiki.org/wiki/Favorite_betrayal_criterion

> IRV, on the other hand, does not encourage candidates to drop out, allowing the voters to express their views more clearly.

That's literally the opposite of what it does. Palin should have dropped out, or her voters should have dishonestly ranked her second, so that the win would have gone to a Republican instead of a Democrat. IRV did not prevent Palin from acting as a spoiler and did not make it safe for voters to honestly rank the candidates.

The winner here isn't the Condorcet winner, true. But in an FPTP election it seems likely the same result would occur so it's at least not worse than the 'default' US system in this case.

(Palin beats Begich in a one-winner Republican primary, given that she had more of the votes cast for the two R candidates which is ~= the votes cast in a primary with only Republican voters, then loses to Peltola in the general by 51.4 to 48.6.)

It might be that non-transferable votes as a result of preference failures would change this a bit, of course.

I don't agree- Begich wins with 61.4% against Palin, as mentioned above? It seems clear that he would've prevailed in a primary

The argument is that he wins against Palin due to Democrats, who wouldn't have voted in the primary, preferring Begich over Palin.

Well Alaska had open primaries before (only 9 states have truly closed primaries). So it's not necessarily the case that Dems wouldn't be voting in the Republican primary- or more likely, independent voters who lean left on some issues. That independent point is key, because you don't know that everyone who voted for the Dem candidate is necessarily a Dem

"It seems clear" -> "it's not necessarily the case" is the exact shift in certainty I was trying to argue for. I agree with this comment.

Alaska did not have open primaries before. Except if you mean way before. In which case, if they ran this election like the elections back then, then Pelota and Palin would be running in the general as they were the top two vote recipients.

Now I'm not sure if Alaska was semi-closed, but that's not moving the needle as hard as you want it to. Because if you're already registered for a party, you cannot vote in the other party's primary. Only unregistered voters may, and only in one.

And typically, independent voters don't do that. As reflected in the low vote totals in the various primaries in previous elections.

I was simply responding to OP's assertion that anyone who voted for Peltola as their 1st choice was necessarily a Democrat, and so would not have voted in a theoretical Republican primary. This is not the case. I don't agree that 'typically, independent voters don't' vote in party primaries

The suggestion was that they weren't likely to be Republicans, given that they've chosen a D over two different Rs, and presumptively wouldn't be voting in a Republican primary. I agree than with (semi-)open primaries that's not guaranteed, but it seems valid as a first approximation at least.

Extremely important observation.

The two-tier primary/general election system produces more extreme-position winners, exactly by this mechanism.

And consider another common outcome - shift the odds very slightly here and the Palin-esque candidate wins the primary, and the general, even though the eliminated primary candidate (Begich in this case) would have been more acceptable to the electorate at large.

Palin got more 1st votes than Begich. That tells me republican voters generally prefered Palin and she would've won the primary. Dem voters preferring begich over palin tells us nothing about how a primary would go.

Whether this is a "failure" of course is a matter of opinion.

Observation: contingent preference is a valid concept; the "problem" for Condorcet is that it is reductionist.

Among other things this means that it is not a given that you can compare percentages as shown.

Now... how to vote for our voting system of preference?!

Range/Score voting in general would just be much better. Its much easier to aggregate, better expresses relative preference and actually leads to a more fair more understandable outcome. Its also easier as its the same method that people rank almost everything in the real world.

This is interesting because it's probably due to exactly how the candidates stack against each other to the respective political groups.

I think Condorcet kind of fails here because people are voting "not Palin" more than they are voting for anyone with their second choice.

If there were primaries, Palin beats Begich, not by much, but she does. In which case, the general is Palin and Peltola. Palin is a particularly odious candidate to Democrats. Very few Democrats are going to vote for her. So Peltola will win any general election that's just them two.

So Republicans are split among Begich > Peltola > Palin; Palin > Begich > Peltola; and possibly some degree of others. Probably some degree of Begich > Palin > Peltola as well.

Democrats are mostly going to be Peltola > Begich > Palin. Because while they would prefer Peltola, they absolutely do not want Palin.

Independents are going to independent.

So I think that's what we're seeing. Condorcet is showing us results for elections we would never actually see. Palin against just Begich in an election where Democrats can vote just doesn't happen.

To address your point about why IRV is seen as more consensus based than FPTP is because FPTP doesn't care about 50%+1. If Alaska was FPTP, then Peltola still wins the election here, because she had the highest vote tally in the initial round. But then she wins with only about 40% of the vote. With IRV, you don't stop until someone gets to 50%+1.

I think the biggest problem with Condorcet is simply math. Already, Condorcet is very "set-theory" based. And once you hit an edge case, you all of a sudden get dropped into a whole world of combinatorics.

Another issue is that the expectation is that your down ranked choice can't cause your first rank choice to lose. But that's what happens in Condorcet. The people who voted Peltola > Begich > Palin are the ones causing Begich to win even though they prefer Peltola.

Every voting method is imperfect when things are split near the middle. And hilariously enough, they're all perfectly fine when most people want something in particular.

I don't understand the focus on Condorcet- In a 3 party election run via IRV, the winner may represent the most conforming combination of positions of the people, but may not be the Condorcet winner. Therefore, Condorcet is not as aligned with best possible outcomes as IRV is. We've already learned that FPTP (which IS a Condorcet vote, quite directly) is suboptimal.

How is FPTP with > 2 candidates a Condorcet vote? A Condorcet method seeks to pick the candidate who would beat all other candidates in a 1-on-1 election. An N-way FPTP election is by definition not a 1-on-1 election.

FPTP never has > 2 candidates because everyone realizes that creates a spoiler.

The 2000 US presidential election where Ralph Nader spoiled for Al Gore seems like a notable counterexample.

And that election is why everyone now knows to not vote third party.

FPTP in non-US contexts very regularly has >2 candidates, despite the risk of a spoiler. England has a 2.5 party system where the Lib Dems compete in every seat and win some. Scotland and Wales have nationalist parties in addition to this (and the Scottish National Party is currently overwhelmingly dominant in seats won, as a result of one of FPTP's other failure modes). Northern Ireland has a different party system with sectional voting and ~5 effective parties. Canada has the NDP and (in Quebec) the Bloc.

One of the classic results in political science is Duverger's Law, sometimes formulated as 'under first past the post the number of effective parties tends towards two', but 'tends towards' is doing an awful lot of work. One could unkindly suggest that, like a lot of political science results, it works better in the US than in the rest of the world. And I say that as a political scientist...

It's not meant to be more concensus oriented. It's meant to give third parties/candidates a chance while preventing the spoiler effect.

With the current US political climate, a third party would likely win a lot of the time if we were simply doing pair preferences like you laid out, because people are essentially voting by party for the lesser evil in most elections these days.

RCV isn't actually that great at avoiding spoiler effects. You would do better with approval or range/score/star voting.

In the IRV version, what spoiler effects are left? Since they're eliminating the lowest voted candidates in each round, you're essentially left with a two candidate election where it's possible to have 100% of voters voting for one or the other.

Palin is the spoiler in this election, taking away first-preference votes from Begich until Begich is eliminated, and then losing to Peltola. If Palin had strategically dropped out, or if Palin voters had dishonestly voted for the "lesser of two evils", then the outcome would have been better from their perspective.

Some ranked choice people might tell you that's a feature not a bug, they usually cite increased ability for third parties to win which implies less mainstream appeal to everyone, i.e. not the "least of evils" candidate who beats others.

> I've never understood why it's somehow considered more 'consensus'-oriented than say FPTP. Biden won his election with 51% of the vote...

Odd example here. That election wasn't FPTP, it just happened to work out that way. Overall people don't think FPTP is consensus because you have to vote pragmatically for people you don't necessarily like. Therefore the winner is not the consensus "favorite" choice of everyone who voted.