Hacker News new | past | comments | ask | show | jobs | submit login

I'm not sure it can be explained. To borrow a phrase from economics, moving to instant runoff has always seemed Pareto efficient to me. Which would mean that there are no possible counterpoints.



It's definitely not Pareto efficient.

If you're a member of the party establishment in one of the big two parties of a system, you stand to lose a lot by moving to an instant runoff and nothing to gain.

Unfortunately, most of the power to change the voting system is in the hands of these kinds of people, so it's going to be pretty hard to change it. Kind of a Catch-22.


Why are we considering the preferences of the parties? We're talking about the most efficient way to allow the electorate to express its representational preferences.

I understand that things get a bit muddled because the individual representatives are also members of the electorate, and thus they may have warped incentives relative to the vast majority. But they are such a small fraction of the overall electorate that we should be able to ignore them in any kind of efficiency discussion.


> Why are we considering the preferences of the parties?

I mentioned it because you said something about Pareto efficiency, and moving to instant runoff would not be Pareto efficient because some people (the parties are made up of people, remember) would be worse off.

Declaring something Pareto efficient has a specific meaning, and it applies to all people, not all the people that stanleydrew thinks are important. Though it's a comparatively small number of people, it's not some weird edge case that can be discounted. Rather, these are precisely the best people to convince to get the law changed. Note well that even this story in Maine, the measure was passed by ballot initiative, not by a legislature vote.


I wrote this in a sibling reply, but I'll post it here too. Perhaps we are talking about two different things?

I am talking about the efficiency of the voting system itself, and its ability to express the preferences of the electorate.

Note that I'm not talking about changes related to the election outcomes that such a shift in voting systems would potentially cause.

Put another way, it would seem to be efficient to move to any voting system that would allow a voter to express more preferences than just one, since then we're getting strictly more information.


"Pareto efficiency" is a very strong claim, and it's invalidated by any single individual doing worse.


You seem to be missing the point. Or perhaps we are talking about two different things?

I am talking about the efficiency of the voting system itself, and its ability to express the preferences of the electorate.

Note that I'm not talking about changes related to the election outcomes that such a shift in voting systems would potentially cause.


The same reason we care about the rights of multinational corporations when it comes to voting on things that concern them - they have lots of entrenched power and/or money so ignoring them is unwise and unrealistic.


Arrow's theorem[1] demonstrates that no voting system can be "efficient" in the normal sense of the word. That is, no voting system can guarantee:

1. If every voter prefers A over B, the electorate prefers A over B

2. If no voter changes preference between A and B, the electorate's preference of A and B is unchanged

3. No single voter can always change the electorate's preference

[1]: https://en.wikipedia.org/wiki/Arrow%27s_impossibility_theore...


That's a different issue from what the parent comment is talking about. He's not saying that IRV will be fully efficient, but that the switch itself will be Pareto efficient (which means that each entity is better off, or the same).

This actually isn't true, because major party candidates will be worse off, but I think he meant something like Hicks-Kaldor efficient, or positive sum.


Yes, I wasn't claiming that there is a perfect voting system. Just that IRV seems strictly better for pretty much all voters than FPTP.

I am also seeing several comments that try to take into account the preferences of the candidates themselves, or their parties. Why? Aren't we talking about efficiency from the perspective of the electorate? To the extent that the candidates are themselves members of the electorate, and can vote for themselves, shouldn't we ignore them in this analysis because their preferences are inherently biased?


> I am also seeing several comments that try to take into account the preferences of the candidates themselves, or their parties. Why? Aren't we talking about efficiency from the perspective of the electorate? To the extent that the candidates are themselves members of the electorate, and can vote for themselves, shouldn't we ignore them in this analysis because their preferences are inherently biased?

I fully agree with you. I was just being precise in my usage of Pareto efficiency, which includes all agents in a situation. Especially because it's used so often when Hicks-Kaldor improvement is meant (which means that there's the possibility for Pareto improvement after redistribution of gains).


You missed a criterion: All voters' preferences are expressed through a ranking.

As an impractical example, if you could get every voter to tell you exactly how many utils a particular candidate is worth to them (maybe through complex incentives), as a quantity and not as a ranking, you would end up with a perfect voting system.

As a more practical example, approval and range voting don't fall victim to Arrow's theorem.


> As an impractical example, if you could get every voter to tell you exactly how many utils a particular candidate is worth to them (maybe through complex incentives), as a quantity and not as a ranking, you would end up with a perfect voting system.

Would end up with a perfect system, or could end up with a perfect system? I assume that latter, because I can easily design a system that uses that extra data and is far from perfect.

Assuming could, not would, my question then is that known? Or is it just known that Arrow's theorem does not apply to such systems, and so we cannot rule out that such a system could be perfect?


As mentioned in the article, Arrow's theorem only applies to rank-order voting systems (including FPTP and IRV). There are other types of voting systems (e.g. approval voting) that do not require the voter to be strategic in expressing her preference.


That's a bad argument to make, though, because nothing about Arrow's theorem says how much the criteria will be broken, only that they can't be fully satisfied. Moreover, it doesn't apply to "range voting" methods (cardinal information-based voting systems).

See here:

https://www.reddit.com/r/badeconomics/comments/5a9a1b/arrows...




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: