STAR is just score with a single automatic runoff. It gets everything you are talking about in terms of intuition, explanation… everything you like about IRV is present in STAR and basically everything wrong with IRV is fixed by STAR.
Intuiting what "giving one star more or less to a candidate" means is really difficult to understand as a voter. "If more voters had given Kodos four stars instead of three, he could have won" is a headache inducing explanation.
Approval voting is the only alternative to IRV or plurality voting that seems to match the level of intuitiveness of both the voting process and the tabulation process.
If IRV complexity and explanations were not a problem, we wouldn't see today's reality where IRV proponents themselves constantly make incorrect claims about it.
Not always (and not in the linked article here), but often people get the intuition that their 2nd choice will get counted when their 1st choice is eliminated, and then try to get them to understand why their 2nd choice was not counted (because their 2nd choice had already been eliminated) while someone else's 2nd choice was counted…
What happened in Burlington before they repealled IRV shows not only people getting intuitions about IRV wrong but people often don't even get what happened in Burlington because it seems so counter-intuitive. https://www.equal.vote/Burlington
And yet, that scenario is common enough. Burlington is just the case where all the ballot stats were released so we can study it.
The problem, though, is one of correctness, not complexity -- I would maintain that almost anyone could look at that election summary and understand why the conclusion that was reached was reached. And if they had had a traditional non-instant runoff election, the result would likely have been the same. It's hard to even explain why the outcome was "wrong" because the tabulation of the results is so intuitive.
> I seriously doubt that anyone would get a headache over "if people had given that losing candidate higher scores, they would have won". I don't even see how you can honestly claim that to be anything but completely intuitive to everyone.
The question for the intuition is "how many stars to I give this candidate". What does it mean to give them one more star? One person's four-star vote carries exactly as much weight as four people's one-star vote, which seems incredibly odd. An Amazon product with a hundred one star reviews is intuitively worse than one with 10 five star reviews.
Certainly not! In Burlington, people know the Republican can't win. So, like everyone in the current system, people would (and have both before and since IRV) vote strategically. So, the less-hardcore Republicans would vote Democrat in order to stop the Progressive.
In fact, if IRV had not been repealed, a good portion of the Republicans would betray their favorite and vote Dem as 1st choice (dishonestly) in the future in order to stop the Progressive from ever winning again.
Furthermore, when vote splitting happens in a current system, everyone readily gets why it was "wrong" and they adapt their strategies accordingly.
> What does it mean to give them one more star?
It means they are one more point ahead of the others.
> One person's four-star vote carries exactly as much weight as four people's one-star vote, which seems incredibly odd
The weight is relative, that's all that matters. As long as people understand the system, they can realize that they are throwing away part of their vote if they only put a 1-star and nothing else. Four people who put 1-star for candidate A might have put 5 stars for candidate B. They are influencing the race just as much as anyone else.
Here's what it means to be equal: Any voter can score the opposite of another voter and they will cancel out. Our votes can have equal weight if that's possible. That's not always possible in our current system or with IRV.
> An Amazon product with a hundred one star reviews is intuitively worse than one with 10 five star reviews.
If there were 110 voters, then every item is relative to that vote. There's 100 1-star votes for candidate A and 10 5-star votes for candidate B? That means candidate B got zeros (blanks are no points) from 100 candidates. A candidate with 100 zero-star votes and only 10 5-star votes is a worse option.
With FPTP voting, C would have won of the three candidates, and A would have placed last. If you removed A from the race, B would have won - and did!
Alternate voting systems are supposed to produce different results from plurality voting, and that's often unsettling the first time it happens.
That page makes the claim that IRV produced a bad result and that's why it was replaced. But at least the candidate with the 2nd-most 1st place votes won. If instead the Condorcet winner, A, had won, do they really think it would have been better? A got the _least_ 1st place votes of the three, who's to say there wouldn't have been an even larger backlash?
FPTP is even worse than IRV, except that people know how it is bad so they adapt accordingly.
Of course, it's speculation about the specific backlash, but the core issue in IRV is this (and it happens in the Burlington case):
ALL the voters whose 1st choice loses in the final round never get any of their other preferences counted and they lose their 1st choice. They get NOTHING, no say, totally screwed. They could be as much as 49% of voters. Now, WITHIN IRV, they can later learn (as people have learned about strategy in FPTP) that if they betray their favorite and vote for their lesser-evil choice as 1st, then they WILL swing the election to their lesser-evil instead of the greater evil.
By ignoring the preferences of some voters and counting others, the weighting is unequal and people will feel disenfranchised.
Maybe this older more thorough discussion of the Burlington case will help you understand: https://www.rangevoting.org/Burlington.html
The core point is that voters in IRV can get a preferable outcome via favorite betrayal (a dishonest strategy). Either they do use that strategy and we're back to the lesser-evil problems we have now, or they don't use that strategy and we're back to vote-splitting like we have now, where a candidate choosing to run can both lose and cause a worse outcome for their supporters.
Another way of seeing this could be that such voters had their 1st choice considered and given value all the way until the last round, whereas many other voters had their lesser choices considered in their previous rounds. The 1rst-choicers-all-the-way-till-the-end could be considered more favored than the latter voters I mentioned, who lost their favorite candidates earlier on in the process.
I'm not saying that this is a better interpretation of the situation you posit. I'm just using it to illustrate that it's hard, maybe impossible, to get away from subjective criteria when we consider the virtues of various voting systems.
by the way, Fargo North Dakota just adopted approval voting for their City elections tonight.
Within a strict ranked-ballot approach, the overall best tabulation is probably Ranked Pairs.
STAR has some similarities with Bucklin but does not have its problems. STAR's advantages are several and aren't directly just a forgone conclusion merely by showing what's wrong with IRV. STAR is much simpler to tabulate than Ranked Pairs and doesn't need to be tabulated at the highest level, it can be counted at precincts like FPTP.
If there were an objective way to measure actual utilities, score-based methods would be great (and economics and policy assessment would be much simpler, and...)
In STAR voting, you have up to 5 points to give to each candidate. You effectively get to push each candidate ahead in the race relative to the others by as many as 5 steps forward. There is NOTHING subjective about the meaning of points. They are just points. More points wins. They have no cultural signficance, no opinion, nothing subjective.
Yes, there's a mapping to a (typically 0-5) numerical rating like those popular in online review systems, which is a motivation for the forced acronym, and exactly the similarity that is viewed as making it particularly intuitive and accessible.
Cultural differences in how such rating systems are used for similar preferences have been studied considerably.
It's true that you can also view STAR as a limited bucket (and so forced-tie for any but the smallest candidate pool) ranked-ballot method, in which you force voters to arbitrarily select which intercandidate ranking preferences to suppress, and then do a completely wacky (from a ranked preference perspective) method of tallying the ballots.
It is indeed a form of ranking that allows ties and has a limited resolution. But the limited resolution does help with simplicity.
From a UI perspective, presenting a ranked ballot that allows ties is unfortunately complicated. That's what STAR actually could/should be though.
And the tallying method is fine, not wacky. In a sense, it's like Borda Count to get to two finalists and then IRV for the finalists. That avoids all the worst aspects of IRV (which are all symptoms of the way its multi-round elimination ignores the preferences of many voters while counting the preferences of others).
I disagree. Generally, unforced rankings (allowing full expression of preferences with optional ties) are simpler than forced rankings (with no ties) which in turn are simpler than limited bucket arrangements that force ties; “Between X and Y, do you prefer X, Y, or neither?” is the simplest question about voting preferences, and unforced preferences can be answered by answering that question alone; forced rankings may, as the name suggests, force the invention of an artificial preference in some cases, but limited buckets methods force deciding which preferences are most important and which to suppress. And limited buckets methods which apply something more than ordinal meanings to buckets,as STAR does, are most complicated.
> From a UI perspective, presenting a ranked ballot that allows ties is unfortunately complicated
Tallying unforced preference ballots is complicated, “number your preferences starting with ‘1’ for most preferred, ’2’ for next most, and so on, with equal preferences getting the same number” with a ballot with appropriate entry blanks isn't hard UI either in paper or machines.
Even if you want paper optical scan ballots, the main problem is the size of the ballot (which is a logistical problem when you've got multiple simultaneous races) more than the UI.
> And the tallying method is fine, not wacky.
Viewed as a preference ballot, giving non-ordinal meaning to rankings is wacky, since it is inventing information. The wacky description was in that specific “viewed as” context.
> In a sense, it's like Borda Count
I don't disagree with this characterization, though I view it as more of a condemnation than a defense.
> That avoids all the worst aspects of IRV (which are all symptoms of the way its multi-round elimination ignores the preferences of many voters while counting the preferences of others).
If you want to eliminate all the worst aspects of IRV, which you view as all due to ignoring preferences via loser elimination, the simple solution is just to drop loser elimination (giving Bucklin or a form of Majority Judgement, depending on where you use forced preferences or allow ties; classic Majority Judgement even uses limited buckets, though you can use a full unforced preference ballot, so if you really think the preference compression is justified by UI benefits, you can use exactly the same ballot as STAR, with a much more straightforward resolution procedure.)
This is certainly far simpler than STAR, whether or not you use the same ballot.
You are right in noticing that it's not exclusively an unforced ranking (which I agree is vastly superior to forced rankings!). So, in STAR, there's a different between 3 candidates scored as 5,1,0 versus 5,4,0. They count identically in the runoff stage (which is effectively the rank part of the system). But they count different in the score.
I'm not convinced that difference is necessary, valuable, or intuitive to people, but I'm not convinced it isn't. I think these things need more real-world case studies. I definitely think the distinction isn't fatal or anything. I see how it could have a positive impact on outcomes.
I don't agree that multi-round eliminations are simpler than STAR per se.
I'm certainly inclined to see 3-2-1 voting as a competitor with STAR for best overall system.
What it's missing compared to IRV is that you can't go, it's just like what people already do (runoffs), just in a single round.
In STAR, 5 points is simply "most points" i.e. most support. It doesn't mean something semantically. You can give up to 5 points, like if there's a race, you can push each candidate forward as much as 5 steps. The only thing it means is that you pushed them forward more than another candidate.
That's not that hard to get intuitively.
IRV is not what people already do but in a single round, it's much more complex than that. And whoever's 1st choice in IRV loses in the final round, they (and that could be as much as 49% of the voters!) never get any of their other preferences counted at all, unlike either STAR or our current two-round system.
Yes, so you import the well-known cultural variation in numeric rating systems without concrete semantics for rating levels into your voting system. That's a great idea.
More data on ballots is good only as long as it has a consistent meaning and is used in a way that is consistent with that meaning.
STAR (and other score voting systems) gather noise and pretend it is signal.
STAR Voting allows voters to rate all the candidates similar to how we rate books on Amazon
Ah yes, Amazon, that site with famously reliable and unbiased user reviews.
This is a terrible idea. It’s a bad voting system made worse by appeal to similarity with popular apps and websites, which themselves use a really bad voting system.
STAR voting is NOT like rating books on Amazon, it's just superficially similar.
At Amazon, if you don't like any of a bunch of books, none of them get 5 stars. With STAR voting, you award points not to mean anything, but only to decide how much ahead in the race you want to push each candidate versus the others. Whoever you want to win most, you push them ahead with all 5 available points. That's it.
Your reaction and misunderstanding does further my annoyance at the people explaining STAR using the Amazon analogy, and I think that bad comparison is harming support for what is truly an excellent voting system.
My issue with all of these point- and range-based voting systems is that when you see the results, you might want to tweak your scores.
Let’s say I’m range voting out of 10 and I give four candidates 1, 3, 8 and 10 points respectively. But the candidate I rated 3 wins, with the candidate I rated 8 a close second. Damn it! If I’d known that was going to happen, I would have voted 1, 1, 10, 10.
So with extra information, range voting boils down to approval voting, and I think it all ultimately becomes ranked voting. It’s hard to imagine any scenario where I would want to change my rankings, even if I have information about the results. If I like both candidates A and B, but A slightly more, I’ll always put them at the top of my list and in that order.
Now, it’s true that in IRV there are scenarios where I can get a better result for myself by voting for candidate ranks in a different order; but those are rare (I think) and definitely unintuitive. In the example above, if B were a real long shot candidate, it might be worth ranking them above A in case B gets eliminated early. But it’s unlikely to make a big difference because, you know, B is a long shot. And that’s just IRV; other ranked voting systems are much more robust.
This is the EXACT reason STAR voting was invented. Unlike plain score, STAR has a runoff where it compares your preference of the top two and puts your full vote to your preference.
So, this is like an exact case study of why STAR is worthwhile. STAR proposes 0-5, but we can stick with 1-10 for this case. You vote 1, 3, 8, 10 and the top two candidates are the ones you voted 3 and 8. With STAR, it doesn't matter that your 3 candidate got a total score higher than your 8 candidate, your vote in the runoff goes to your 8 because that's rated higher. If more voters rated that candidate higher, they win even if their plain score total was lower.
Thus, you don't have to exaggerate all your scores. In fact, you have an incentive not to, because if you just give ties, you risk abstaining in the runoff instead of having a say there.
I mean literally you just spelled out the most common critique of plain score voting which is itself THE argument for why we should use STAR instead.
I’m reading https://www.equal.vote/starvoting and I still don’t entirely follow. It’s lacking some detail I’d like to understand. For example, it looks like it is not a Condorcet method. That seems bad to me -- if you’re not electing the Condorcet winner, you need a very good explanation why; and I don’t trust an appeal to numerical ratings that each voter will interpret and use in different ways.
Aha, yes, Wikipedia has a clearer explanation and confirms that it is not Condorcet: https://en.m.wikipedia.org/wiki/STAR_voting
I do agree that it looks probably better than IRV, but I’m not yet convinced that it’s a really good system and is worth pushing as a good approach for real elections.
One short point: the primary problem with not electing Condorcet is that coordinated strategic majority could, in hindsight and for future elections, force the Condorcet by refusing to support later preferences. In STAR, that strategy is risky and unpredictable. If the Condorcet is a strong case, it will win STAR anyway. If it's an edge case, nobody knows in advance if it will be the true majority or just a bit less, so strategically zeroing all by the favorite is so risky in that edge case, most people won't do it (it could backfire for them to refuse to support their 2nd favorites if their hail-mary hope that they just might have a Condorcet turns out to be wrong). No strategies with STAR are reliable. The best outcomes come from just being honest.
Note: STAR doesn't use rank as a tie-breaker only. It can flat-out overrule a score difference with the 2nd-highest scoring candidate winning. It's not just for ties.
> I do agree that it looks probably better than IRV, but I’m not yet convinced that it’s a really good system
Well, since IRV gets pushed for real elections and STAR is better on a large number of factors, and both are better than FPTP…
You otherwise want to promote 3-2-1 https://wiki.electorama.com/wiki/3-2-1_voting or approval or maybe some good PR approach like PLACE voting https://medium.com/@jameson.quinn/place-voting-the-elevator-...
If you want to discuss this stuff more with other folks, https://forum.electionscience.org is the place to go, btw
You’re combining ranked voting with range voting.
Ranked voting is objective and clear (the Condorcet criterion is very compelling) but unfortunately incomplete (Arrow’s theorem, which in this case boils down to unresolvable three-way cycles.)
Range voting is more tractable (counting is easy, no awkward cycles) but subjective (no-one knows what the scores actually mean) and prey to tactical voting.
STAR starts with range voting (so you lose Condorcet) and adds cut-down ranked voting as a final step (only after some rank information has been discarded; and ranking isn’t a sound way to resolve ties).
It would be better to use ranked voting first, then range voting to resolve tie-breaks. That would make better use of the unique advantages of each voting style.
I still don't buy your assertion that no one knows what scores mean. That's no stronger of an argument than the opposing critique of ranking: nobody knows in a rank ballot whether it's a slight preference between two supported candidates or a huge difference between a liked and a hated (but maybe some other candidate is even worse) situation.
In both ranking and scoring, the ballot does not give a transparent view of the voters' beliefs. They only mean what they mean on the ballot, that there are relative scores or a ranked order. We can't and shouldn't assert that we can know more deeply what's behind those marks.
Condorcet is not a bad principle per se, but it's not the absolutely most desirable. If it were the only factor, nobody would accept IRV. And Condorcet means that a candidate who is 2nd choice of 100% of the voters, maybe even a strong 2nd choice (rankings don't carry that info) will always lose even in a case where all the other candidates are grossly polarizing and will lead to civil war. There are at least cases where a consensus is superior to Condorcet.
> It would be better to use ranked voting first, then range voting to resolve tie-breaks.
How does this proposed idea work? What actual system are you suggesting that would do this? EDIT: I see this referenced in your other post.
That's not correct, if there are 4 or more candidates.
If there were exactly 3 candidates the split would have to be something like 51-49, and then the candidate with 100% second place would come second. But it's arguable whether that's a bad result, and it seems like a very unlikely scenario anyway. Even 10% of people voting that candidate in first place would be enough for them to win it.
What I meant to express was that the 100% 2nd choice candidate will lose to Condorcet winner if there's 51% support for the Condorcet winner. I.e. majority support. That's true also for non-Condorcet systems. My point was that Condorcet systems are always also strict majoritarian, and there are reasonable arguments that majoritarianism can be undesireable (such as cases of polarizing slim-majority winner.
If, in a two-candidate election, one candidate gets more votes, should that election be the winner, even if the losing candidate has more passionate supporters?
In democratic elections that respect one-person/one-vote, the answer to that question is simply yes, because that is the definition of one-person/one-vote democracy. Even if there are 100 voters, and it's 51-49, and the 51 voters are lukewarm while the 49 voters are passionate, the candidate with 51 votes wins. No matter what.
The reason that matters is because it is actually fairly slippery to measure passion. If one person submits a 10 rating, and another submits a 7 rating, what does that mean? It could mean that one person is passionate and the other person is lukewarm. But it could also mean that one person is emphatic while the other person is meek. Or that one person is authoritarian and the other person is abused. The entire reason we have one-person/one-vote is to make sure that voters are treated equally, even if society is otherwise telling certain people that they are "less than".
And for those sorts of elections, Condorcet voting is basically perfect. All the flaws people attribute to "Condorcet methods" are only attributable to when there are multi-member Smith Sets.
Now there are other kinds of elections; elections where we don't follow the one-person/one-vote principle. Smaller electorates, non-governmental, self-governing organizations. In those cases, "consensus" is more possible without risking systemic abuse. And in those cases, it might be entirely appropriate to award the winner to the 49% candidate with more passionate support. In which case you would perhaps choose something other than Condorcet.
You should come to terms with the fact that whatever you mean is completely unintuitive to many/most of us, and that almost everyone will just this like they use Amazon. Whether or not the page says "it's like Amazon" or "it's not like Amazon".
The ballots are proposed as labeled "no support" to "most support".
It isn't entirely unlike Amazon, it's like Amazon if your standard for what's a 5, like "the best book" is "the best of THESE candidates". All people need to do to get the system is care about their impact on the outcome of the election in order to realize that 5 doesn't mean "love" or some other semantic thing, it just means the top of the options.
And if people just use it like Amazon and some give no candidate a 5, the system doesn't fall apart, it works pretty well anyway.
http://star.vote is not a controlled scientific survey or anything, but people overall seem to have no trouble using the full range.
In fact, in my anecdotal experience, even over years in the past when I told people about plain score voting and tried to get them not to use the full range (because I misunderstood that issue myself), many people insisted that they would use the full range anyway. A lot of people find it intuitive to give their favorite all the points they can.
But the real answer is that, yes, we need more studies and trials of all these things.