have you ever worked with pivotal Labs? One of my biggest faults I guess is that I worked with femme and they have such an incredibly high bar for understanding, design patterns and cyclomatic complexity and solid and test driven development principles and so forth that once you've worked with them, anywhere else you go is going to feel like a dumpster fire. so then you'll just pull your hair out thinking everyone's an idiot, even if it's really just a lot of bad incentives.
oh man, as a veteran of cruise this hits hard. thinking back to all the astonishingly dumb code I tried to refactor to do the right thing, and just get no recognition for it. politics > proficiency
i had a chance to visit arrow at his palo alto condo circa 2014. his theorem is nice and all, but it only makes sense to apply it to social welfare functions, not voting methods. yes, the correct social welfare function is just the utilitarian sum of all voters' individual utilities.
once you know that, that's the function you use in your VSE metrics. then the performance of the voting method is measurable without having to think about any specific criteria.
> it seems like Approval voting just moves the spoiler effect into how people vote
that's orthogonal. ranked voting methods already have (arguably more severe) response to strategic voting AND ALSO can fail IIA even with no strategy applied, just by changing an irrelevant alternative.
> Personally I think the possibilities of circular ties under Ranked Pairs is oversold.
what does that even mean? we have VSE figures that measure the combined effect of all failures, including when the Condorcet winner isn't the favorite candidate of the electorate (not the social utility maximizer).
https://electionscience.github.io/vse-sim/vse-graph.html
that's not under or oversold, it's just measured performance.
Restating my disclaimer of "It's been a while since I've studied the details of voting systems"...
> ranked voting methods ... can fail IIA even with no strategy applied, just by changing an irrelevant alternative.
Can you clarify whether you're referring to some ranked methods (eg IRV), or all ranked methods (ie including ranked pairs) ?
> that's not under or oversold, it's just measured performance.
Isn't this due to defining "performance" in a way that is congruent with Approval (/ Score) ? A quick skim of that VSE page has it talking about "utility", which I would imagine is a scalar per candidate representing "happiness" ?
The problem I have with Approval is that coming from our two-terrible-party system - do I Approve my latent terrible party or not? That choice seems purely down to strategy, compared to being able to rank them to say I completely prefer the new party/candidate over my latent terrible party, and my latent terrible party over the other latent terrible party. The dynamic also seems exacerbated knowing there will be a lot of people who continue to vote exactly as they did under plurality.
it's mathematically proven that all ranked methods can fail IIA. see arrow's theorem.
> Isn't this due to defining "performance" in a way that is congruent with Approval (/ Score)
1. i did not define performance in a way that is congruent with approval/score. scores are not utilities. they are the modification of utilities via ignorance, normalization, and strategy.
2. that the correct social welfare function is just the sum of all voter utilities (the definition of "performance") is mathematically proven.
https://www.rangevoting.org/UtilFoundns
approval voting obliterates IRV ("RCV") with any mixture of strategic or honest voters, so i'm not sure why you're bringing up strategy.
https://www.rangevoting.org/StratHonMix
of course they are. market forces are what you want for everything. granted, you may want to internalize a negative externality like the climate change induced by greenhouse gas emissions, but that's literally correcting market forces.
