Item — HN Reader

STAR and Approval have different pros and cons, they are not objectively superior. I think RCV is better, in my opinion.

STAR and Approval both have their own tactical voting problems. If you really like A, are meh about B, and dislike C, it’s clear you should rate A high / give A your approval and rate C low / do not give C your approval. But what about B? It has nothing to do with what you actually think about them, it has to do with what you think their odds of beating C and A are. You want B to do well enough to beat C, but not well enough to beat A. This adds too much distortion to people wanting to express their sincere preferences, which they can easily do in RCV: A > B > C

And, more importantly, RCV is a stepping stone to multi-member district STV proportional voting (plug for https://www.fixourhouse.org/), which is the golden goose; way more important to implement proportional representation than squabble about STAR vs RCV, and RCV gets us closer to proportional representation.

5 comments

This adds too much distortion to people wanting to express their sincere preferences, which they can easily do in RCV: A > B > C

RCV actually literally has the exact same problem you're describing with voting strategically, it's just subtler and harder to understand. The problem is, if you rank A > B > C, then your subsequent preferences after A only matter if A gets eliminated. If A comes in second place, then your "B > C" preference has no impact, and C may be elected whereas if you "dishonestly" voted for B first you might have allowed B to win.

This is a feature, not a bug.

The point of ranking is not to have your vote redistributed many times, but rather for your #1 preference to go to your preferred candidate.

If your preferred candidate does not win, your vote gets a second chance (or more!) to get counted.

In the case your describing here, the only way that C wins, is that he gets more votes than A or B.

If your preferred candidate does not win, your vote gets a second chance (or more!) to get counted.

But that's not always the case; this is exactly the point I'm trying to make. The whole problem is that your vote only gets a second chance to be counted if your preferred candidate gets eliminated. If your preferred candidate loses because somebody else wins before your preferred candidate gets eliminated, then none of your later rankings affect the outcome whatsoever. This is why you can't actually vote honestly in IRV if you want your vote to count.

Let me try to illustrate this with an example:

Let’s imagine the US switched to RCV before the 2020 election, and Mitt Romney decides to run as an independent. Now let’s look at, say, West Virgina, which is a pretty red state, and went about 70% Trump 30% Biden in the 2020 election in real life

WV Democrats are still going to vote for Biden, and they’d probably prefer Romney over Trump, so let’s say they all vote in that order, but imagine roughly a third of Republican voters in West Virgina like Romney better than Trump, so we end up with this:

45% vote Trump, Romney, Biden; 25% vote Romney, Trump, Biden; 30% vote Biden, Romney, Trump

Romney gets eliminated first, his voters fall back to Trump, and Trump wins. The Democrats all wanted Romney over Trump, but Biden was never eliminated, so their second choice of Romney had no bearing on the outcome. But, what if they’d strategically voted for the “lesser of two evils”?

now we have: 45% vote Trump, Romney, Biden; 25% vote Romney, Trump, Biden; 30% vote Romney, Biden, Trump

Now Romney is a clear winner, he has the majority of first choice votes, so nobody’s second or third choice even matters. So basically in this example, Biden was a spoiler candidate for Romney, and all those Democrats are going to be super disillusioned when they realize that the “it fixes the spoiler effect" and "you can vote honestly” lines are both wrong, and they could’ve had a better outcome if they’d voted strategically, just like in the old system.

That's not quite right. In ranked systems you're saying that your preference of A over B is the same as your preference of B over C. This is often not the case. In fact, this is why ranked systems almost always fail the favorite betrayer and monotonicity criteria. I explain more with examples in a longer post below.

Whether later-no-harm is a feature or a bug is the biggest difference, I think, between people who like STV/IRV and people who dislike it.

You have to consider factors beyond just the decision algorithm — elections must be inclusive, fair, and accurate as well. RCV makes auditing a lot harder due to centralization, the complexity means there is more room for software problems and process problems, and the spoiled ballot rate increases, especially among people who are less privileged. When you look at the whole picture, RCV is probably a net negative.

Approval doesn't have these problems; it's a clear win.

> RCV makes auditing a lot harder due to centralization,

I'd argue that the algorithm is far more complex. It is `while no winner: pop(min(candidates)); reallocate(min(candidates))`. This while loop is quite a problem. Cardinal systems (except STAR) are `argmax(sum(candidates))`, which is actually identical to Plurality. The difference is that Plurality requires a one-hot vector. This algorithm is trivially parallelizable too. STAR doesn't require a while loop but is more complex because it does have 2 phases (which RCV is almost always going to be more than 2 phases given that while loop).

The trivializes of cardinal systems tallying algorithm makes statistical audits also trivial. I just can't imagine the mess we'd be in if what happened in Arizona happened with an ordinal method like RCV. It would be much easier to create a narrative of fraud.

Yea what we need is to get rid of districts. Proportional representation is the only great solution.

Districts are good to make sure there are some local ties between citizens and representatives and not only represantitives from one region remain.

Take the German system: there are district level elections, so that there is one person elected in each electoral district. These get half the nominal seats. Then the other half of parliament is filled up with candidates to match the proportional vote. The fill up is a bit complicated, especially if a party has a lot more directly elected folks than proportionally mandated, which leads to growth of parliament and discussions how that growth can be avoided. (Larger districts and only 1/3 directly elected or not every directly elected candidate really making it or not holding up the exact proportionality or whatever) But it combines the factors mostly fine and avoids many of the problems of the pure systems.

That is one solution, another is multi-member districts

Depends on your tradeoffs.

If you elect, oh, 5 people in each district then your proportionality is still granular to about 17% of the vote in the district (for single member districts it's granular to about 50%, of course). And your districts are 5x as large as single member districts are for the same number of representatives. What it does give you though is better electability for local independents.

With mixed-member proportional you can easily get down to couple-percent granularity while maintaining medium-size districts (typically twice as large as single-member).

This right here is the problem, these things get way too damn complicated for normal people really fast.

You don't think normal people can understand: "Rate this candidate from 0 to 5 stars?" If so, we really need to rethink a lot of how we ask for people's opinions, because this is probably the most common way.

The problem with RCV is there's a big difference between giving an honest opinion and voting tactically.

In RCV it's safe to put your first choice first if either they're an overwhelming favorite to win, or they're so weak they're sure to get eliminated right away. If it's a close three-way election, you might cause your favorite candidate to lose by ranking them first. It's hard to develop any kind of intuition around how your vote will impact the race in these situations -- I think in the long run voters tend to learn that voting for third parties isn't safe, so two parties continue to dominate.

> If it's a close three-way election, you might cause your favorite candidate to lose by ranking them first.

Could you explain that? I can't think of a situation where putting the candidate I want to win as #1 would cause them to lose. Ranking other close candidates as #2/#3 could cause my #1 to lose, but I don't think voting for someone as #1 alone would possibly harm them.

So what I think the parent is referring to is called the Favorite Betrayer Criteria[0]. The short is that there are two forms of the spoiler effect: strong and weak. RCV handles the weak case but does not handle the strong case. I explain part of why cardinal methods handle the strong case in this comment[1]. The truth is that we're far more concerned with the strong case than weak case because from our current position this is the more frequent type of spoiler (and arguably is more frequent in general).

[0] https://electowiki.org/wiki/Favorite_betrayal_criterion

[1] https://news.ycombinator.com/item?id=33111242

I wonder if fixing the "strong case" spoiler actually is NOT what some "spoiler voters" would want, simply BECAUSE they want to spoil for the other party to try to move that party toward where they want it to be.

Probably. But I don't think the public really wants their voting systems to be highly vulnerable to strategic voting. That does just give elites an edge and ends up with less equitable representation.

(strategies exist for ALL voting systems. But not all strategies are equally as bad)

Let's say you have three candidates A, B, C.

A is left, B is center, C is right leaning.

Polling shows that voters are leaning:

- 15% BAC

- 15% BCA

- 25% ABC

- 45% CBA

If everyone voted as they desired, then the instant runoff would be between B and C, with B winning with 55% of votes.

If a subset of C voters instead strategically vote ACB (6%), this would change the runoff to be between A & C, with C winning 60%.

First, a run off between A and C wouldn't have C with 60%. C would have 54%, the 6% of C voters would have to come from CBA. And if A is in the runoff, that 6% stays with A.

These are post facto strategies though. You can only implement that strategy if you can guarantee all of that information. Which you can only do after the election.

The first problem is that polling is an estimate. If you are wrong by a couple of percentage points, the strategy fails. If the results come back 16 (BAC), 16 (BCA), 23 (ABC), 39 (CBA), 6 (ACB), B now has 32 first votes and A only has 29.

Next is getting the right number of people to comply. You could easily just mess it up and have either not enough or too many people cast the strategic ballot. If only 4% or less vote ACB, the plan doesn't work. If 11% or more vote ACB, A wins.

In the end, strategic voting is a bit like voter fraud, talked about way more than is actually done. And implied to be a much bigger problem than it ever could be.

Oh I don't disagree, and am in the strong support of anything that moves away from FPP. The only addition I will make is that these (potentially unfounded) fears is what leads people to continue to vote strategically, such as voting BAC instead of ABC to avoid the potential of C using the strategy.

And just to be clear, this makes sense.

Because the relevant fact here isn't that some CBA voters put A at the front (ACB). It's that they moved B from 2nd to 3rd. So of course, that lessens support for B, so B no longer wins.

People seem to present this as a flaw, when it's reflecting actual changes in voting. And this is basically impossible to do so strategically, because nobody has this level of accurate detail prior to the election.

The only issue is that the (potentially unfounded) fear of strategic voting on the other side leads people to vote strategically in the same FPP manner by voting ABC instead of BAC to avoid C. But I greatly long for anything that moves us from the anti-democratic FPP.

> you might cause your favorite candidate to lose by ranking them first

That's a flawed characterization that keeps getting repeated.

As I explain in another comment below, in theory you could get your favorite candidate to win by further downranking the candidate who would otherwise win. One way of accomplishing that is to move other candidates up higher above your actual preferred candidate.

But in reality, there's no way to know what the exact precise vote breakdowns will be, so voting strategically (lying about your preferences) is impossible in a practical sense. There's no intuition to develop -- just vote your preferences.

It isn't a flawed characterization. It is called the Favorite Betrayer[0] (also known as Strong Spoiler Effect) and has been seen in practice. We should also talk about rank's failure of the Monotonicity Criterion[1][2] because that's relevant here too.

To claim it is impossible in practice is to deny history.

[0] https://electowiki.org/wiki/Favorite_betrayal_criterion

[1] https://electionscience.org/library/monotonicity/

[2] https://electowiki.org/wiki/Monotonicity

> I think in the long run voters tend to learn that voting for third parties isn't safe, so two parties continue to dominate.

The empiricval evidence in countries with long term ranked voting (Australia, say) is that bicameral party systems (only Democrats or Republicans have any realistic chance) wither and die, and smaller blocs have a better chance of survival and impact | influence in the House and Senate ..

Two parties "dominate" in Australia (or rather three parties, which are effectively two - the Nat-Liberals being an Rural-Urban conservative coalition of long standing) but nowhere near the degree to which two parties dominate in the USofA; more importantly minority opinions have gravitas and can sway the major vote - none of this bipolar US Dem-Rep deadlock.

Also it’s not just about the nominal parties that win, it’s the composition of politicians in those parties, and RCV selects for more broadly agreeable candidates.

I think comparing the US to others is often a bad comparison. Most others have coalitions and you'll find that two coalitions dominate. In the US parties act as coalitions. We can see the wide range in opinions from people like Warren/Sanders/AOC compared to those like Pelosi/Biden/Harris. Even in the Republican side you have this massive division, though they've been more effective lately at achieving a party over principle strategy. So two coalitions isn't really any better than what's here in the US because it is in essence the same thing.

They can do that - but the game theory around "don't rate the candidate you don't want ..." stuff is way too complicated.

And if it is not as simple as "Bob got the most votes, Bob won" it's going to get harder and harder to explain exactly what is going on, which is going to get people annoyed and angry.

As long as the result is the same as would have come from first past the post, nobody will care.

> They can do that - but the game theory around "don't rate the candidate you don't want ..." stuff is way too complicated.

I'm not understanding your point. In cardinal systems a non rating is equivalent to a rating of zero. Just like in ordinal systems a non ranking is equivalent to ranking last place.

> And if it is not as simple as "Bob got the most votes, Bob won" it's going to get harder and harder to explain exactly what is going on

This makes it seem like you're in favor of cardinal methods actually. Tallying their votes is pretty trivial. You sum the scores of all candidates and then argmax the totals. I don't think people have a hard time understanding this. It is essentially the same as what happens with FPTP! I mean this is objectively a simpler algorithm than those used by even the simplest ordinal methods.

The problem is not understanding how to rate a candidate but understanding what the ratings will accomplish and what voting strategy should be used to get your intended outcome.

If you're looking to game the system, the the point is to make this hard. The entire point is that we're trying to understand your preference in candidates. That's all that needs to be understood. Moving your preferences around slightly has little effect on purpose. If it had large effects, like RCV has, it would violate the monotonicity criteria and make it likely for the favorite betrayer criteria to fail. Both criteria that RCV fails btw and I think most people care about these criteria a lot, even if they don't fully understand them.

> "And, more importantly, RCV is a stepping stone to multi-member district STV proportional voting"

I agree that proportional representation is a huge deal, but I suspect we can do better than STV by changing what the election results mean. Currently in every system I know of we treat it as a binary choice; a candidate is either a member of the set of candidates who won, or they aren't.

A different way to do it is to allow every candidate who receives some minimum threshold level of support to be elected, but when voting on issues each elected person's influence is weighted by the level of support they received in the election.

This sort of system is compatible with FTPT, RCV, or approval-voting style ballots, but the differences matter less as long as you can say of the vast majority of voters (assuming an omniscient view -- privacy of votes would obscure the actual data) that that voter clearly has a representative in congress, and that voter delegated the same amount of influence to that representative as any other voter. (People who only voted for fringe candidates that don't meet the minimum threshold lose out, but that would probably be rare among people who actually pay attention to politics and aren't casting a protest vote.)

If this were enacted for the U.S. House of Representatives for example, the likely outcome would be that each state would have at least one Republican and at least one Democrat who would wield different amounts of power depending on how popular Republicans and Democrats are in that state. You'd also have Libertarians and Greens and Independents and whatever else. A state might have two representatives after one election and ten after another, and it wouldn't be a problem because influence is the same.