4 ms·

Looked it up just now - Approval voting is basically when you can pick more than one candidate off the list, all with equal weights. The candidate with most vot

by eps 6y ago

Looked it up just now - Approval voting is basically when you can pick more than one candidate off the list, all with equal weights. The candidate with most votes wins. It is a simple scheme, but it's too simple. In reality a voter will prefer one candidate over others and may have a secondary choice so not to see their vote lost. The approval voting means that if your candidate has a decent chance of winning, you will need to vote for this candidate only and no one else. That is, your voting will depend on how well your candidate is doing. An alternative - if I can't have A, then I want B - is the ranked voting. It maps directly on what a voter wants and it doesn't require adjusting their voting strategy based on the standing of their preferred candidate.

godelski 6y ago
So go range or STAR. The appeal of approval over these is the simplicity while maintaining most properties. I prefer STAR then range then approval. But these are all very close. There's a bigger gap to Condorcet (a class of ranked systems), then big gap to IRV (RCV), then a small gap to FPTP/plurality. There's a lot more to this than than just these simple options. I suggest you look more into the subject before forming strong opinions. The subject is highly complicated and there is no global optima. There's a lot more to be concerned about besides VSE and solving the spoiler effect (which IRV/RCV doesn't do btw). Ranked doesn't actually perfectly map. Consider the case where you like two candidates equally? You can't express this with ordinal systems, but you can with cardinal. There are plenty of non-binary selection systems in cardinal votings that allow for more expressiveness. The thing is that this expressiveness doesn't get you much while it adds complexity (granted, not much). But cardinal systems are also fairly resistant to strategic voting (one of those factors that matters).
- eps 6y ago
  > Consider the case where you like two candidates equally This is an exceptionally improbable case though. Between two choices there's always one that's better. Unlike when you need to select the worst, when all options may be equally bad.
  - jabl 6y ago
    With an ordinal voting method if you vote A > B > C you're stating that the difference between A and B is the same as the difference between B and C. Which is a poor match to reality.
    - eps 6y ago
      Still way closer to the reality than A = B = C that you get with the approval voting.
      - godelski 6y ago
        You could just approve A, or A & B. People will set different criteria. The reason equality is a feature is because you can't be infinitely expressive. Let's say that we're using a 5 point scale. If A is 4.9 and B is 4.8 and C is 2, some voters will express both A and B as 5 while some will split. They will do this because they are strategizing (one thing STAR handles well). But the key here, like said before, is that you can express how much more you like A and B over C. This can have major results on the election (also, this ability makes practicality simpler. Less prone to errors because you are scoring candidates and not ranking. So you can have an infinite number of candidates with no added complexity. You can make "mistakes" like scoring or "ranking" the candidates the same value. There are a lot of advantages to the reduced complexity. And you're not making any trades in features, only upgrades).
jabl 6y ago
> In reality a voter will prefer one candidate over others and may have a secondary choice so not to see their vote lost. The approval voting means that if your candidate has a decent chance of winning, you will need to vote for this candidate only and no one else. That is, your voting will depend on how well your candidate is doing. All voting methods are affected by strategic voting. Approval voting, however, is relatively little affected by it compared to many other voting methods, such as ranked voting methods. > An alternative - if I can't have A, then I want B - is the ranked voting. It maps directly on what a voter wants and it doesn't require adjusting their voting strategy based on the standing of their preferred candidate. Ranked voting methods, while better than the first-past-the-post most of the world uses today, tend to suffer from many deficiencies. And while not a technical objection per se, excessive complexity eroding trust in democracy is a pretty big one as well. For a nice overview, see https://ncase.me/ballot/ https://ncase.me/ballot/ . In particular, see the chart showing the Bayesian regret simulations for different voting methods. In the presence of strategic voters, approval and score voting are the best methods, whereas with 100% honest voters score voting wins.
derbOac 6y ago
I think the idea is that across voters, the average difference in preference for A vs B is revealed by choice of voting strategy along the lines you're mentioning. If your preference for A is strong enough that you want to secure a vote solely for A then that choice of strategy does so. But if you want to preserve votes for A and B both, you can do that. Presumably someone else would feel differently, and it would be reflected in expected voting patterns. No system is perfect, but at the same time, at some point "strategic voting" becomes revealed preference. The issue is how well a voting system encodes these things.

by eps 6y ago

godelski 6y ago

So go range or STAR. The appeal of approval over these is the simplicity while maintaining most properties. I prefer STAR then range then approval. But these are all very close. There's a bigger gap to Condorcet (a class of ranked systems), then big gap to IRV (RCV), then a small gap to FPTP/plurality. There's a lot more to this than than just these simple options. I suggest you look more into the subject before forming strong opinions. The subject is highly complicated and there is no global optima. There's a lot more to be concerned about besides VSE and solving the spoiler effect (which IRV/RCV doesn't do btw). Ranked doesn't actually perfectly map. Consider the case where you like two candidates equally? You can't express this with ordinal systems, but you can with cardinal. There are plenty of non-binary selection systems in cardinal votings that allow for more expressiveness. The thing is that this expressiveness doesn't get you much while it adds complexity (granted, not much). But cardinal systems are also fairly resistant to strategic voting (one of those factors that matters).

eps 6y ago
> Consider the case where you like two candidates equally This is an exceptionally improbable case though. Between two choices there's always one that's better. Unlike when you need to select the worst, when all options may be equally bad.
- jabl 6y ago
  With an ordinal voting method if you vote A > B > C you're stating that the difference between A and B is the same as the difference between B and C. Which is a poor match to reality.
  - eps 6y ago
    Still way closer to the reality than A = B = C that you get with the approval voting.
    - godelski 6y ago
      You could just approve A, or A & B. People will set different criteria. The reason equality is a feature is because you can't be infinitely expressive. Let's say that we're using a 5 point scale. If A is 4.9 and B is 4.8 and C is 2, some voters will express both A and B as 5 while some will split. They will do this because they are strategizing (one thing STAR handles well). But the key here, like said before, is that you can express how much more you like A and B over C. This can have major results on the election (also, this ability makes practicality simpler. Less prone to errors because you are scoring candidates and not ranking. So you can have an infinite number of candidates with no added complexity. You can make "mistakes" like scoring or "ranking" the candidates the same value. There are a lot of advantages to the reduced complexity. And you're not making any trades in features, only upgrades).

jabl 6y ago

> In reality a voter will prefer one candidate over others and may have a secondary choice so not to see their vote lost. The approval voting means that if your candidate has a decent chance of winning, you will need to vote for this candidate only and no one else. That is, your voting will depend on how well your candidate is doing. All voting methods are affected by strategic voting. Approval voting, however, is relatively little affected by it compared to many other voting methods, such as ranked voting methods. > An alternative - if I can't have A, then I want B - is the ranked voting. It maps directly on what a voter wants and it doesn't require adjusting their voting strategy based on the standing of their preferred candidate. Ranked voting methods, while better than the first-past-the-post most of the world uses today, tend to suffer from many deficiencies. And while not a technical objection per se, excessive complexity eroding trust in democracy is a pretty big one as well. For a nice overview, see https://ncase.me/ballot/ https://ncase.me/ballot/ . In particular, see the chart showing the Bayesian regret simulations for different voting methods. In the presence of strategic voters, approval and score voting are the best methods, whereas with 100% honest voters score voting wins.

derbOac 6y ago

I think the idea is that across voters, the average difference in preference for A vs B is revealed by choice of voting strategy along the lines you're mentioning. If your preference for A is strong enough that you want to secure a vote solely for A then that choice of strategy does so. But if you want to preserve votes for A and B both, you can do that. Presumably someone else would feel differently, and it would be reflected in expected voting patterns. No system is perfect, but at the same time, at some point "strategic voting" becomes revealed preference. The issue is how well a voting system encodes these things.