3 ms·

IRV is a Schelling Point, and I think it's important to consider it as such. The biggest critique of alternatives to FPTP is "they are too complicated". When yo

IRV is a Schelling Point, and I think it's important to consider it as such. The biggest critique of alternatives to FPTP is "they are too complicated". When you look at an individual system like IRV it seems hard to sustain that objection, but then a bunch of voting system nerds enter the room, and now we're comparing Condorcet to Borda, and critiquing a bunch of mathematical properties of each system, and drawing graphs of preference-space; at which point the average voter's eyes gloss over and they say "yup, too complicated". I think this is a perfect case for the maxim "Let not the perfect be the enemy of the good". It seems IRV is sticking, and nobody has made Condorcet or Borda stick; let's go with IRV for now. Once we've abolished FPTP, and everyone is familiar with ranked systems, we can come back and argue for a better option. As far as I'm concerned, IRV captures 99% of the utility of changing voting systems. (All that said, I do like the work that https://electionscience.org https://electionscience.org is doing with Approval Voting, and if we can run experiments in local elections across the country with two new systems, I think that's also positive.)

ajmurmann 4y ago
IRV is hard to evaluate though. How long would it take to teach a member of the general public to evaluate IRV results? You also cannot have meaningful estimates until all results from every district are in. How long it took in New York to get any results might leave people disillusioned and opposed to anything but FPTP. Using Borda and Condorcet is also clearly a straw man, as AFAIK nobody is actually pushing for those in practice. The real runners seem to be IRV, Approval and STAR. Of those IRV is clearly the most compared to evaluate. IMI they are all pretty simple and in an ideal world it wouldn't matter that much. However, we need to be highly aware that as soon as a major election gets won with anything but FPTP, we'll see the losing party "flood the zone with shit" to undermine the election results and the method. We've already seen the beginning of this after the election in Alaska. We need to keep it as simple as possible, otherwise the shit will drown everything else out.
- godelski 4y ago
  > IRV is hard to evaluate though. How long would it take to teach a member of the general public to evaluate IRV results? This is always an interesting point to me. I often hear arguments that STAR and Approval are too difficult for the voters and far more complex (there's plenty of comments here suggesting that). While I disagree, that's a matter of opinion (I think both are trivial fwiw). But on the other hand, how votes are tallied is much harder to argue. Ordinal systems use a while loop where you pop an element each iteration and reallocate votes. Cardinal methods are sum then argmax. It is a joke to think that voters would understand Ordinal methods better than Cardinal methods with respect to how votes are tallied.
- theptip 4y ago
  > How long would it take to teach a member of the general public to evaluate IRV results? I'm puzzled by this. You can do it with pencil and paper in a few minutes? It's simple addition. You can easily build a website that visualizes each step if you think addition is difficult for people. It's trivial to compute in a spreadsheet. > might leave people disillusioned This is speculative; if it is true, you can find evidence by talking to, e.g. Alaskan voters. (IOW, Citation Needed.) > clearly a straw man I'm simply painting a picture by recounting my experience of what happened in the last thread on HN where voting systems came up, not making a strong claim about these specific systems being the runners up. My point still stands if you substitute any other systems, in this thread for example https://news.ycombinator.com/item?id=33111242 https://news.ycombinator.com/item?id=33111242. The point I was making had nothing to do about the merits of those specific systems; the point was that a big technical can of worms gets opened, and the average voter has no interest in digging in to compare N solutions on M technical axes, Arrow's theorem, etc. This is what people mean when they say "alternate voting systems are complicated".
  - godelski 4y ago
    > It's simple addition. Not quite. The tallying algorithm is `while (no candidate > 50% of votes): pop min(candidates), reallocate_votes(min(candidates))`. On the other hand cardinal methods are `argmax(sum(candidates))`. For comparison, FPTP is `argmax(sum(candidates))` too. The only difference is that FPTP is a one-hot vector. I'd argue that the average person intuitively understands cardinal methods without any training.
    - ajmurmann 4y ago
      To add to the simplicity case: STAR and Approval voting are on essence how many online rating systems like IMDB, Yelp or Rotten Tomatoes work. Those seem very intuitive to users. (Edit: ignoring cases where those platforms hide votes etc)
    - theptip 4y ago
      You're right, my bad. Computing `reallocate_votes()` is not as trivial as I suggested. I do wonder if many people actually care to work through that step; in all of the public discourse around the Alaska vote, it's simply presented as a table with the reallocations already summed. For example, https://fivethirtyeight.com/features/what-democrats-win-in-alaska-tells-us-about-november/ https://fivethirtyeight.com/features/what-democrats-win-in-a.... The general algorithm of "reallocate the votes from the candidate that is being eliminated" is conceptually simple, even if it's annoying to tabulate. I do stand by my claim that it would be very easy to build a website that can visualize this process without getting bogged down in the tabulation. See every site that's giving analysis on the Alaska results for examples. I want to see actual UX studies on this; for example has anyone polled in Alaska to see whether people understand the result?
gamegoblin 4y ago
A group in Seattle actually got an Approval voting ballot initiative enough signatures to get on the ballot this year (where RCV has been trying and failing for decades), but the City Council, sensing their re-election chances under such a system in danger, intervened and modified the initiative to add RCV as a 3rd option, because they know they have a better election odds with RCV, and RCV will probably beat Approval on name recognition alone. A cynical take would be that IRV has gotten traction in the US precisely because it does not seriously threaten the existing power structure. It might nibble away on a few edges, but if you look at all the countries and cities with IRV, they are not obviously better than those still stuck on FPTP.
tunesmith 4y ago
Condorcet is actually quite simple, and the voting community does it a huge disservice by insisting on talking about it in very complicated ways. Condorcet is just "if a candidate would beat all others head to head, then that candidate is the winner". But the voting community insists on immediately jumping to all the crazy tie-breaking algorithms you need to implement if there is a "loop" (Smith Set). My opinion has always been that those two concepts should be severed. All the "flaws" of Condorcet actually only apply to when tiebreakers are used. The flaws do not apply when there is a Condorcet Winner. (The Smith Set isn't hard to understand either. It's just "these candidates beat all others head to head.") And when there isn't a Condorcet Winner, that is actually valuable information. It means that the population was not ready to decide. It may warrant a later runoff, giving the population more opportunity to research the remaining Smith-Set candidates, and decide through additional debates or whatever else. As an alternative, people can always choose to apply any old method for the tiebreaker - even IRV! Condorcet-IRV (where IRV would only be applied to the Smith Set) strictly dominates IRV-only, in every way.

ajmurmann 4y ago

IRV is hard to evaluate though. How long would it take to teach a member of the general public to evaluate IRV results? You also cannot have meaningful estimates until all results from every district are in. How long it took in New York to get any results might leave people disillusioned and opposed to anything but FPTP. Using Borda and Condorcet is also clearly a straw man, as AFAIK nobody is actually pushing for those in practice. The real runners seem to be IRV, Approval and STAR. Of those IRV is clearly the most compared to evaluate. IMI they are all pretty simple and in an ideal world it wouldn't matter that much. However, we need to be highly aware that as soon as a major election gets won with anything but FPTP, we'll see the losing party "flood the zone with shit" to undermine the election results and the method. We've already seen the beginning of this after the election in Alaska. We need to keep it as simple as possible, otherwise the shit will drown everything else out.

godelski 4y ago
> IRV is hard to evaluate though. How long would it take to teach a member of the general public to evaluate IRV results? This is always an interesting point to me. I often hear arguments that STAR and Approval are too difficult for the voters and far more complex (there's plenty of comments here suggesting that). While I disagree, that's a matter of opinion (I think both are trivial fwiw). But on the other hand, how votes are tallied is much harder to argue. Ordinal systems use a while loop where you pop an element each iteration and reallocate votes. Cardinal methods are sum then argmax. It is a joke to think that voters would understand Ordinal methods better than Cardinal methods with respect to how votes are tallied.
theptip 4y ago
> How long would it take to teach a member of the general public to evaluate IRV results? I'm puzzled by this. You can do it with pencil and paper in a few minutes? It's simple addition. You can easily build a website that visualizes each step if you think addition is difficult for people. It's trivial to compute in a spreadsheet. > might leave people disillusioned This is speculative; if it is true, you can find evidence by talking to, e.g. Alaskan voters. (IOW, Citation Needed.) > clearly a straw man I'm simply painting a picture by recounting my experience of what happened in the last thread on HN where voting systems came up, not making a strong claim about these specific systems being the runners up. My point still stands if you substitute any other systems, in this thread for example https://news.ycombinator.com/item?id=33111242 https://news.ycombinator.com/item?id=33111242. The point I was making had nothing to do about the merits of those specific systems; the point was that a big technical can of worms gets opened, and the average voter has no interest in digging in to compare N solutions on M technical axes, Arrow's theorem, etc. This is what people mean when they say "alternate voting systems are complicated".
- godelski 4y ago
  > It's simple addition. Not quite. The tallying algorithm is `while (no candidate > 50% of votes): pop min(candidates), reallocate_votes(min(candidates))`. On the other hand cardinal methods are `argmax(sum(candidates))`. For comparison, FPTP is `argmax(sum(candidates))` too. The only difference is that FPTP is a one-hot vector. I'd argue that the average person intuitively understands cardinal methods without any training.
  - ajmurmann 4y ago
    To add to the simplicity case: STAR and Approval voting are on essence how many online rating systems like IMDB, Yelp or Rotten Tomatoes work. Those seem very intuitive to users. (Edit: ignoring cases where those platforms hide votes etc)
  - theptip 4y ago
    You're right, my bad. Computing `reallocate_votes()` is not as trivial as I suggested. I do wonder if many people actually care to work through that step; in all of the public discourse around the Alaska vote, it's simply presented as a table with the reallocations already summed. For example, https://fivethirtyeight.com/features/what-democrats-win-in-alaska-tells-us-about-november/ https://fivethirtyeight.com/features/what-democrats-win-in-a.... The general algorithm of "reallocate the votes from the candidate that is being eliminated" is conceptually simple, even if it's annoying to tabulate. I do stand by my claim that it would be very easy to build a website that can visualize this process without getting bogged down in the tabulation. See every site that's giving analysis on the Alaska results for examples. I want to see actual UX studies on this; for example has anyone polled in Alaska to see whether people understand the result?

gamegoblin 4y ago

A group in Seattle actually got an Approval voting ballot initiative enough signatures to get on the ballot this year (where RCV has been trying and failing for decades), but the City Council, sensing their re-election chances under such a system in danger, intervened and modified the initiative to add RCV as a 3rd option, because they know they have a better election odds with RCV, and RCV will probably beat Approval on name recognition alone. A cynical take would be that IRV has gotten traction in the US precisely because it does not seriously threaten the existing power structure. It might nibble away on a few edges, but if you look at all the countries and cities with IRV, they are not obviously better than those still stuck on FPTP.

tunesmith 4y ago

Condorcet is actually quite simple, and the voting community does it a huge disservice by insisting on talking about it in very complicated ways. Condorcet is just "if a candidate would beat all others head to head, then that candidate is the winner". But the voting community insists on immediately jumping to all the crazy tie-breaking algorithms you need to implement if there is a "loop" (Smith Set). My opinion has always been that those two concepts should be severed. All the "flaws" of Condorcet actually only apply to when tiebreakers are used. The flaws do not apply when there is a Condorcet Winner. (The Smith Set isn't hard to understand either. It's just "these candidates beat all others head to head.") And when there isn't a Condorcet Winner, that is actually valuable information. It means that the population was not ready to decide. It may warrant a later runoff, giving the population more opportunity to research the remaining Smith-Set candidates, and decide through additional debates or whatever else. As an alternative, people can always choose to apply any old method for the tiebreaker - even IRV! Condorcet-IRV (where IRV would only be applied to the Smith Set) strictly dominates IRV-only, in every way.