3 ms·

> While this is easy to write down it isn't easy to tabulate. Ranked pairs etc. have no problem tabulating it. If the change to allow equal ranking breaks some

> While this is easy to write down it isn't easy to tabulate. Ranked pairs etc. have no problem tabulating it. If the change to allow equal ranking breaks some obscure metric, that's a sign the metric isn't very important. After all people can just rank everyone if they worry about it. I've read a lot about this, and I argue with the usual suspects range people every time the topic shows up. I don't understand how you don't understand the argument. You can come into situations where you'll regret you didn't min-max your choice (rank everyone 0 or 100). You can't come into situations where you say, "gee, I'm glad I ranked that candidate somewhere in between". Consequently, everyone will do that, whether it's good for the collective or not. Sure, I want what's best for society, but I'm not volunteering to have my vote diluted for the common good - especially since I know that the people I disagree with most, certainly won't do it.

godelski 4y ago
RP has a resolve time of O(n4) so I'm not sure that's a method you want to argue for under a thread where (an even lower) complexity led to a critical mistake. It would be naive to maximize VSE over all other criteria. I cannot stress enough that there are no optimal systems and that we cannot look at criteria independently. > I don't understand how you don't understand the argument. Because pure honesty is never an optimal strategy? I'm not aware of a system where that's the optimal strategy. But as we showed here in range, the best strategies are some form of honesty. The dishonest part is that you rescale, which I don't think most people would be upset about. What is frustrating to me is that you appear (I'm willing to admit a misread because these conversations are difficult and it is clear you're primed on my position given your usual suspects statement) to be ignoring identical strategic issues in ordinal systems. But I've also stated pretty clearly why I have a preference of cardinal systems. It is because I care a lot about the resolve complexity (the thread topic is a good example), monotonicity (something IRV/RCV fail (RP and Schultz pass but fail the prior criteria)), favorite betrayer/strong spoiler effect (again IRV/RCV fails), and an important factor is the embedding aspect. That the distance between all preference is not identical. We just can't have as richly expressive latent embedding as we can in a rated system. This has a lot of benefits that I think we need to discuss, including a smoother VSE solution space[0]. That dictates how resistant a system is to dishonest strategies and how representative we are. [0] https://www.youtube.com/watch?v=-4FXLQoLDBA https://www.youtube.com/watch?v=-4FXLQoLDBA
- vintermann 4y ago
  > Because pure honesty is never an optimal strategy? I'm not aware of a system where that's the optimal strategy It's not in a system alone. Strategies are something you have when you have information about the other players and their payoffs. As it happens, honesty is an excellent strategy in Condorcet methods (even the "bad" ones, like those who aren't monotonic) when you have a realistic amount of information about other people's preferences/strategies. People who are honest because they can't be bothered to be strategic, do great with those methods. In approval, even the information about who the front runners are, causes trouble. And in range, even with NO information, you'll know it's better to just use the max and min values. Those systems don't encourage honesty, and the "extra information" range sneaks in from self-reported strength of preferences on an absolute scale, is junk information. Nothing good can come of using it. Certainly it's no good to use it to reject a candidate that is preferred by a majority over all the other candidates.
  - godelski 4y ago
    https://electionscience.github.io/vse-sim/ https://electionscience.github.io/vse-sim/
    - vintermann 4y ago
      We've been through this so many times. The original version of those simulation-based utility arguments was pushed by Warren D. Smith on RangeVoting.org. There are so many ways to lie with simulation. When you took away even the most obvious lie in the simulations painting range in a good light - namely when you took away the simulated people's magical ability to honestly compare interpersonal utility on an absolute scale, so they always ranked their most preferred 100 and their least preferred 0, no matter where they were in "opinion space" - then most of the good properties of score vanished. Smith for a while conceded that, and advocated range with top-2 runoff if you were concerned with that (as if this was some obscure technicality). Second, those simulations had lousy ideas of "strategy", as if strategy was just a percentage slide. Strategies are inherently dynamic, and require coordination - if I see you instructing your supporters how to vote, I can take that into account in how instruct my supporters to vote.

godelski 4y ago

RP has a resolve time of O(n4) so I'm not sure that's a method you want to argue for under a thread where (an even lower) complexity led to a critical mistake. It would be naive to maximize VSE over all other criteria. I cannot stress enough that there are no optimal systems and that we cannot look at criteria independently. > I don't understand how you don't understand the argument. Because pure honesty is never an optimal strategy? I'm not aware of a system where that's the optimal strategy. But as we showed here in range, the best strategies are some form of honesty. The dishonest part is that you rescale, which I don't think most people would be upset about. What is frustrating to me is that you appear (I'm willing to admit a misread because these conversations are difficult and it is clear you're primed on my position given your usual suspects statement) to be ignoring identical strategic issues in ordinal systems. But I've also stated pretty clearly why I have a preference of cardinal systems. It is because I care a lot about the resolve complexity (the thread topic is a good example), monotonicity (something IRV/RCV fail (RP and Schultz pass but fail the prior criteria)), favorite betrayer/strong spoiler effect (again IRV/RCV fails), and an important factor is the embedding aspect. That the distance between all preference is not identical. We just can't have as richly expressive latent embedding as we can in a rated system. This has a lot of benefits that I think we need to discuss, including a smoother VSE solution space[0]. That dictates how resistant a system is to dishonest strategies and how representative we are. [0] https://www.youtube.com/watch?v=-4FXLQoLDBA https://www.youtube.com/watch?v=-4FXLQoLDBA

vintermann 4y ago
> Because pure honesty is never an optimal strategy? I'm not aware of a system where that's the optimal strategy It's not in a system alone. Strategies are something you have when you have information about the other players and their payoffs. As it happens, honesty is an excellent strategy in Condorcet methods (even the "bad" ones, like those who aren't monotonic) when you have a realistic amount of information about other people's preferences/strategies. People who are honest because they can't be bothered to be strategic, do great with those methods. In approval, even the information about who the front runners are, causes trouble. And in range, even with NO information, you'll know it's better to just use the max and min values. Those systems don't encourage honesty, and the "extra information" range sneaks in from self-reported strength of preferences on an absolute scale, is junk information. Nothing good can come of using it. Certainly it's no good to use it to reject a candidate that is preferred by a majority over all the other candidates.
- godelski 4y ago
  https://electionscience.github.io/vse-sim/ https://electionscience.github.io/vse-sim/
  - vintermann 4y ago
    We've been through this so many times. The original version of those simulation-based utility arguments was pushed by Warren D. Smith on RangeVoting.org. There are so many ways to lie with simulation. When you took away even the most obvious lie in the simulations painting range in a good light - namely when you took away the simulated people's magical ability to honestly compare interpersonal utility on an absolute scale, so they always ranked their most preferred 100 and their least preferred 0, no matter where they were in "opinion space" - then most of the good properties of score vanished. Smith for a while conceded that, and advocated range with top-2 runoff if you were concerned with that (as if this was some obscure technicality). Second, those simulations had lousy ideas of "strategy", as if strategy was just a percentage slide. Strategies are inherently dynamic, and require coordination - if I see you instructing your supporters how to vote, I can take that into account in how instruct my supporters to vote.