Successive Rank Voting



  • @Jack-Waugh As far as I know, there is no case law establishing a precedent for what is acceptable and what isn't. But I wouldn't want to jeopardize an open primary initiative with any method that looks like "voting for more than one candidate". Since ranked choices are specifically allowed, the safe thing to do is to have voters rank choices and treat those choices as discrete ranks.



  • @Keith
    If all you care about is that your favorite wins and you do not care about any compromise then you should bullet vote.

    I am glad we agree on that point. The reason I am interested in open primaries is that there is not enough compromise in politics these days.

    Majority support is that the candidate who is preferred by the majority of voters wins. It is implied that this is the majority of voters who have a preference.

    This is where we disagree. I interpret “majority” as meaning over 50%. The “majority of voters who have a preference” would be a plurality. I worry about the worst-case scenario of the election winner being the most opposed candidate of a majority (>50%) of voters. STAR does not prevent that.

    That is not just a bad idea because ranking is bad but because you are forcing people to show FALSE preference.

    Unless the candidates are clones, I expect all voters to have some preference, whether they express it on the ballot or not. If they really don't have a preference, then they could randomly rank one candidate above the other without any regret. I see no harm in forcing voters to express a preference between near-equally-supported candidates. Requiring unequal ranks or scores also prevents voters from “voting for more than one candidate”, which is unconstitutional in my state.



  • @robertpdx said in Successive Rank Voting:

    Since ranked choices are specifically allowed, the safe thing to do is to have voters rank choices and treat those choices as discrete ranks.

    Given that constraint, I wonder whether one of the answers that would work best against spoiler effect might be Coombs.



  • Or how about a hybrid of Borda and Coombs and Baldwin?

    Voters rank the candidates.

    For each ballot, the tallying algorithm arrives at a rating of the candidates by assigning numbers linearly to the ranks (or with a logistic function instead??). If a voter leaves candidates unranked, those receive the score that would be given to the bottom rank.

    Each round eliminates the candidate with the lowest total score. When one sole candidate remains, that one wins.

    When going from one round to the next, for any ballots that don't use the full range, they are expanded linearly to fill it.



  • This is where we disagree. I interpret “majority” as meaning over 50%. The “majority of voters who have a preference” would be a plurality. I worry about the worst-case scenario of the election winner being the most opposed candidate of a majority (>50%) of voters. STAR does not prevent that.

    Majority does mean over 50% so you are correct. However, what if there are two candidates where 90% of the public view as identical? Deciding based on the preference of the 10% makes sense. STAR voting does prevent the case of the winner being most opposed by a majority. The only way this could happen is if they score the most opposed the same as everybody but their winner. This is why people do not bullet vote. Bullet voting will hurt you. There is no incentive to do this.

    Unless the candidates are clones, I expect all voters to have some preference, whether they express it on the ballot or not. If they really don't have a preference, then they could randomly rank one candidate above the other without any regret. I see no harm in forcing voters to express a preference between near-equally-supported candidates. Requiring unequal ranks or scores also prevents voters from “voting for more than one candidate”, which is unconstitutional in my state.

    This conflicts with your other point. You are basically arguing for what I said above about clones but in a roundabout way.



  • @Jack-Waugh I considered a similar runoff scheme using Borda count as the elimination criterion. But it comes back to the question of whether voters approve or disapprove their lower ranks. Voters shouldn't be expected (or required) to rank candidates they don't like unless there is a "later no harm" option. And if ranks are left blank, the winner may not have a majority of votes in the final round.



  • @Keith said in Successive Rank Voting:

    However, what if there are two candidates where 90% of the public view as identical? Deciding based on the preference of the 10% makes sense. STAR voting does prevent the case of the winner being most opposed by a majority. The only way this could happen is if they score the most opposed the same as everybody but their winner. This is why people do not bullet vote. Bullet voting will hurt you. There is no incentive to do this.

    You are assuming that if voters oppose two candidates enough to score both of them zero, then they have no preference between them. I don’t think so. I think the incentive to give both candidates a score of zero is that they don’t like either of them. But they still may hate one more than the other. Also, candidates have an incentive to encourage their supporters to bullet vote. The only way to know for sure if there is a preference is to allow a “later-no-harm” option (which voters can leave blank if they truly don’t care). You expressed concern that this would lead to “tyranny of the masses”. That is always a possibility. But the only thing worse than" tyranny of the majority" is "tyranny of a minority". I think a good election method should at least prevent that.



  • @robertpdx said in Successive Rank Voting:

    Voters shouldn't be expected (or required) to rank candidates they don't like unless there is a "later no harm" option.

    Why not just assume that if a voter leaves a candidate unranked, that candidate has earned that voter's maximum opposition?

    And if ranks are left blank, the winner may not have a majority of votes in the final round.

    With an elimination system, the winner is guaranteed all the votes in the final round, because the other candidates will have been eliminated. It is not only a "majority" (of the non-exhausted votes), but is unanimous.



  • @robertpdx said in Successive Rank Voting:

    You are assuming that if voters oppose two candidates enough to score both of them zero, then they have no preference between them. I don’t think so. I think the incentive to give both candidates a score of zero is that they don’t like either of them. But they still may hate one more than the other.

    This is a fair point. Some voters may hate both the STAR finalist and then should have given them both a 0. This is guaranteed to be a minority of voters since the two finalists are the top two utilitarian candidates. STAR is designed exactly to balance these two issues.

    @robertpdx said in Successive Rank Voting:

    candidates have an incentive to encourage their supporters to bullet vote.

    Maybe but people will know about this. It might make the candidate look bad to do so.

    @robertpdx said in Successive Rank Voting:

    The only way to know for sure if there is a preference is to allow a “later-no-harm” option (which voters can leave blank if they truly don’t care). You expressed concern that this would lead to “tyranny of the masses”. That is always a possibility. But the only thing worse than" tyranny of the majority" is "tyranny of a minority". I think a good election method should at least prevent that.

    “later-no-harm” is one extreme end of the balance. STAR is a nice middle ground and it outperforms the other methods when strategy is simulated.



  • @Jack-Waugh The problem with any method without a "later no harm" option is that voters have an incentive to avoid showing any level of support for opposed candidate, even if they don't oppose them all equally.



  • @Keith said in Successive Rank Voting:

    This is a fair point. Some voters may hate both the STAR finalist and then should have given them both a 0. This is guaranteed to be a minority of voters since the two finalists are the top two utilitarian candidates.

    How is that guaranteed to be a minority of voters? If voters bullet vote among four popular candidates you could have a winner with 26% scores of 5 and 74% scores of zero. This is an extreme example, but you could relax the bullet voting quite a bit and still not have a majority winner.
    I take your claim seriously that bullet voting might be unlikely. But as I mentioned earlier, I worry about worst-possible scenarios. If an initiative gets on the ballot, it will be attacked by pointing out potential bad outcomes.



  • @robertpdx Hi Robert, I see you found the forum. I'm glad to see this thread active.

    Your Bucklin voting type proposal is interesting. Bucklin is the system PDX and then repealed around 100 years ago. It tests as better than IRV and Approval but not as good as STAR on a few metrics. It's also harder to explain.

    As to STAR and some of the concerns you raise, I agree that a few are misunderstandings or points that can be cleared up.

    • No voting method can guarantee a majority. Even with multiple elections and a top 2 runoff the last PDX mayoral race failed to elect a majority preferred winner because one simply didn't exist. Portland had 3 polarized factions. IRV and STAR and other methods can find a majority winner if one exists, but can never guarantee one. IRV finds a majority of remaining ballots only so it's a manufactured majority. STAR looks at every ballot.

    • Bullet voting can and does happen with any voting method. It wasn't incentivised in Buckin voting either but voters do it anyways because they are polarized, because they don't have a more nuanced opinion. Some people just want to vote for their favorite and get on with their day. Voting theorists love to dive into the game theory and strategic incentives to explain this behavior, but at the end of the day most people do it if they haven't taken the time to learn about the full field of candidates. STAR does a good job of incentivizing and allowing expressive voting, but that's all we can do. Laws that would force voters to rank all candidates would just result in lower turnout or voided ballots anyways.

    • On "real" runoffs. A "real" runoff is a second election with less candidates than the first. For elections that means a primary and general.

    • Any time you have a preference voting ballot you can do an instant or automatic runoff. These terms are synonymous. The only reason STAR isn't STIR voting is that the acronym didn't work. A STAR voting runoff is a more real runoff than an IRV runoff because every ballot is counted in the runoff, which is not the case in IRV.

    • On bullet voting. STAR voting never incentivises strategically dishonestly bullet voting. It's more strategic and smart to show at least some support for other candidates besides your favorite, just in case your favorite can't win. Failing to show that you prefer your 2nd choice over your 3rd choice means that if your favorite can't win you don't care what happens. The only time a dishonest bullet vote would be no-risk is if you were sure your favorite would win no matter what. In that case you might as well vote more expressively because it can't hurt.

    • Overall, STAR Voting does a better job at incentivising honest voting than IRV or Approval or Bucklin. Bucklin and strategy hasn't been studied a ton in modern voting theory because there are other proposals that are simpler and perform better.

    • One thing that is worth noting is that no voting method can be perfect incentivising voters to vote for their honest favorite as best (not doing lesser evil voting) and also be perfect at incentivising voters to show support for others. (Incentivising voters to be willing to compromise to some degree.) STAR does a pretty great job at both, and that's as good as we can ask for without going for much more complex options.

    • Your example shows a dead tie between candidates A and B. Any of the voters could have been more expressive, and so the incentive would be for all three to do so in the next election. Your example does a good job of illustrating why it's safe to assume that voting behavior and more expressive voting would likely improve over time as voters learn that unexpressive and stingy voting isn't a good tactic in STAR.
      In your example if we had the same election again a reasonable vote would be:
      Voter 1: (5,0,1)
      Voter 2: (1,5,0)
      Voter 3: (0,1,5)
      That's still a tie, but at least now each voter is doing the most they can to ensure their favorite wins while doing as much as possible to ensure their least favorite doesn't win.

    In a dead tie there will always be prisoners dilemmas, but a strategic dishonest vote doesn't give anyone an edge, it's just as likely to backfire in STAR Voting. And voters want to be honest. The reason strategic voting is so prevalent now is because voters need to be strategic to get a fair shake. In STAR an honest vote is a good smart vote.



  • @robertpdx All voting methods which pass Later No Harm fail Favorite Betrayal Criteria, meaning that it's not always safe to vote for your honest favorite and you should consider voting lesser-evil instead.

    Voting methods which pass this criteria fail to eliminate vote-splitting.

    It's not a good pass/fail criteria. It's an important consideration to balance to find a well rounded system that does a good job overall.
    This article explains: https://www.starvoting.us/farewell_to_pass_fail



  • Again, what about Coombs?



  • Evidently "PDX" has come to mean Portland, Oregon, by virtue of being the code both of its airport and its railroad station.



  • @SaraWolk said in Successive Rank Voting:

    All voting methods which pass Later No Harm fail Favorite Betrayal Criteria,

    Also worth pointing out that you must have "later No Harm" (non-compromising) or the chicken dilemma. Later no harm can cause you to get the wrong winner when everybody is honest but Chicken Dilemma stratigies only have pay off if the strategy is used effectively and can also cost you. In STAR the cost of the alternative vs the favorite is 1/5. ie if you want to minimally score a lesser evil you would give them a 1. This makes the chicken dilemma 5 times less incentivized than in Approval.

    If you want to take this to the extreme you could argue that giving the lesser evil 1/5 of our favorite is too much. We could do STAR on a [0,100] scale or even more. Clearly this would allow more expression at the cost of complexity. If it was a zero to 1 million then I am sure your complaint about STAR would vanish. Fundamentally I do not think your issue is with STAR itself but in this one scenario where 5-STAR is not expressive enough. Where is the line where you would make a cut off on the ganularity?



  • @Keith It's worth noting that we considered STAR 0-9 and other scales, but 0-5 is the sweet spot; best for cognitive load, great outcomes, and better strategic resilience. The bigger the scale the bigger the impacts of tactical voting would be.

    Practical considerations like ballot design have to be taken into account too. In order to prevent spoiled voided ballots due to bad handwriting and people whose 1s look like 7s and so on you really need a row of bubbles. To many bubbles is not good. Too few choices is also not good.



  • @robertpdx
    In STAR Voting your vote goes to the finalist you prefer. Its constitutional, fully, and dovetails well with the Oregon constitution's wording and history. We have never had any problems or concerns raised on this by our lawyers, by the constitutional review that was done for our ballot initiatives, and there were no concerns on this raised when the Oregon Legislative Committee drafted HB 3250 STAR Voting for Oregon.

    We are not sure if Approval would have problems, but as mandated, in STAR your vote only even goes to one candidate and that only happens once, in the runoff.

    Here's the relevant text from their own initiative, Section 3-5:
    (5) Each ballot cast shall be tallied to determine, for each office, which of the two candidates identified in subsection (4)(b) of this section receives the vote of each elector that
    cast a ballot in the election. For the purpose of vote tallying under this subsection:
    (a) The candidate who received the higher score from an elector, as calculated under
    subsection (4) of this section, shall receive the vote from that elector.
    (b) If the two candidates received the same score from an elector, neither candidate may
    receive the vote of that elector.
    (c) The individual who receives the most votes, as calculated under this subsection, is
    nominated or elected to the office.

    https://d3n8a8pro7vhmx.cloudfront.net/unifiedprimary/pages/547/attachments/original/1614330748/HB_3250_Establishes_STAR_as_the_default_voting_method_for_the_State_of_Oregon.pdf?1614330748



  • What State are you in?



  • @SaraWolk Sorry I haven't had time to formulate an answer to your replies, and even this reply will be minimal.
    One comment I have is that it IS possible to ensure a majority if all voters express a preference when there is no harm (to more favored candidates) in doing so. The Portland election used plurality voting and allowed write-in candidates, so it was not a true runoff between two candidates. I think RCV/IRV would have produced a majority. In what sense do you mean that IRV does not count every ballot in the runoff?
    I would like to know your criticism of the last method I proposed:
    http://www.classicalmatter.org/Election Science/BAIR Voting.pdf
    It is Bucklin voting for three rounds, and if no candidate has a majority then there is a "true" runoff between the two most approved candidates. By "true" runoff I mean that every voter may express a preference without risk to more favored candidates.
    I think your arguments against later-no-harm do not apply if the voter can choose between later-no-harm and no friend betrayal.


Log in to reply