• mindbleach
    link
    fedilink
    arrow-up
    2
    ·
    1 year ago

    A simpler demonstration:

    45% vote A > B > C.

    35% vote C > B > A.

    20% vote B > C > A.

    FPTP says A wins despite 55% of people preferring B or C instead. This is classic vote-splitting.

    RCV eliminates B for having the fewest top votes, making the election 45% A>C vs 55% C>A, so C wins. But 65% of people preferred B instead.

    Ranked Pairs does 1v1 comparisons: A-B is 45-55, A-C is 45-55, B-C is 65-35. B wins every possible runoff and therefore the election.

    … and yeah, Approval would look something like 45% A, 65% C, 100% B. If everybody checked their top two candidates, everybody would vote B.

    • sugar_in_your_tea
      link
      fedilink
      arrow-up
      2
      ·
      1 year ago

      Thanks for the simple example!

      And this is why I tend to pitch approval voting. You usually get the same results as more complex models, it’s simple to understand, and there’s no scoring model that people can bicker about. You mark the candidates that are acceptable, and don’t mark the candidates that are unacceptable.

      I think STAR is technically better, but it’s a harder sell because people need to understand RCV and STAR well enough to understand the differences.

      Approval is the same as FPTP conceptually, but with multiple votes thrown in. We already do something similar when there are multiple seats available (e.g. city council) where the top X candidates get the position, but we just set X to 1 for single seats like Senator or President. It’s intuitive, and high numbers for the result is a nice side effect.

      • mindbleach
        link
        fedilink
        arrow-up
        2
        ·
        edit-2
        1 year ago

        Approval should be the default. There’s no good reason not to do it everywhere. “Check anyone you like, most votes wins.” Done. Good results, low regret, no invitations to self-defeating strategery.

        Incidentally it’s how presidential elections used to work: every voter wrote two names, and the runner-up became vice president. That ended with the twelfth amendment after a comically inept tie-breaking session needed thirty-five attempts.

        Ranked ballots are better but I’m always going to endorse Condorcet specifically because ‘more people wanted this guy’ are what elections are for. RCV finds the first candidate who can scrounge together a simple majority. Condorcet finds anyone who’d beat them. The possibility of someone who’d beat them is all people need to know about why RCV sucks.

        Score systems… I don’t see the draw. Intensity of support is individually crucial and collectively impactful, but if everybody counts equally, then the math can’t care. You don’t get to vote harder based on strong feelings. You also shouldn’t be able to throw away half your vote by scoring your favored frontrunner low. STAR does paper over that honesty-punishing shortcoming, but only by expressing what Ranked systems do naturally.

        As an off-the-cuff illustration, if C>B>A and B>C>A voters score everyone 3-2-1, but A>B>C voters score everyone 5-3-1, C can be eliminated despite supermajority preference over A. If it’s split about evenly, A averages 1+1+5, B averages 3+2+3, and C averages 3+2+1. The runoff comes down to A and B. B wins, in this case… even if C>B>A ballots outnumber B>C>A ballots.