Shortly after the 2016 election I did some research into election and social choice theory. First-past-the-post is clearly the worst system possible. But ranked-choice also has strange artifacts where voting for your actual favorite as your first-rank choice means they might have a lower chance of winning.
A much simpler method in my opinion is approval voting: the ballot instructs, "Who do you think would perform the duties of the office to your approval? Check all names that apply." Then whoever gets the most checkmarks in aggregate is the winner. No spoiler effect, easily-understood vote tallying system, and reflects true approval in the population of voters. It can be adapted to multi-seat races just by taking the top n winners.
Ka-Ping Yee simulated elections using a number of recognized methods for counting votes[1]. I believe that the best system will be one which is easily-understood at the ballot box and behaves as similarly to the Condorcet method as possible. According to these simulations, it appears that approval voting is the best.
No election method will meet all the requirements for an ideal one[2] so it comes down to which conditions you want to enforce and which are less important. IMO, monotonicity is very important while the later-no-harm criterion can be interpreted as a compromise which is necessary for strategic voting.
This algorithmic instability isn't merely of academic interest, it was encountered in the 2009 election for Burlington's mayor [0]. Note that IRV advocates don't see this sort of discontinuity to be a problem [1]. Perhaps we need something like Star Voting [2], which seems to be an improvement over range [3] or approval [4] voting. Star Voting is rather new and could use analysis, like Ka-Ping Yee's and Nicky Case's efforts.
I always recommend this pair of videos when discussing STAR voting vs other alternative systems. The first video is an easy-to-follow explanation of how spoiler effects can happen in both plurality voting and IRV [1]. In the second [2], the guy who created Star voting applies Ka-Ping Yee's methods to Star voting and animates it side-by-side with alternative systems to show its performance.
Interesting, this must be relatively new. They introduce a slight modification to IRV that mitigates its artifacts. But you can see even in the video that some remain.
I'm not sure if or why people might feel more satisfied casting their votes as a partial ranking versus as a binary approval. But it seems like STAR works great for the former group, approval for the latter.
Star voting (along with Borda) introduces intensity of preference, which introduces extra concern when we're talking about democratic elections.
For instance - in a 99-voter election, say that 50 voters prefer A, and 49 prefer B. But the A supporters are lukewarm in their support, while B supporters are really passionate. Who should win? For democratic elections, that answer always needs to be A.
STAR has a 2-phase mechanism. The 1st phase, by intensity, is used to determine the finalists. In the 2nd phase -- the actual vote -- it is one person one vote, regardless of intensity. By your example, STAR poses no problem -- both A & B would be the finalists by intensity. Then, A would be chosen by majority. Anyway, it's examples like this that could use some analysis & help visualization.
addendum: if a candidate has broad, but shallow support, they'd not make it in FPTP nor in RCV. Even so, it'd be great if we could have visualization tools /w STAR voting included so that we could point to specific examples and have fun arguments about them. Perhaps there are some very weird discontinuities in STAR voting as well.
Yeah, I know, but is it possible for a candidate with broad shallow support to be boxed out of the top two? There's no pure reason to have a cutoff point there, it's arbitrary.
Was going to say exactly this. STAR is basically approval with the added metric of approval amount, which I personally think is important, and has the benefit of being similar to something people already understand (star ratings for products/services).
I feel logically the amount would just be embedded in people's vote of approval. Since score is the average anyways. Most people would probably approve of all 3+ amount, and won't approve of who they would have scored 3 or less.
It gets weird making the 0s and 5s, and the 1s and 4s fight I feel. Since those are just ultra polarized candidates.
It depends what we want to optimize for, but I think most people would agree that we want to avoid having a very large number of absolutely discontent voters. So I think a system which either makes the least hated candidate win (even if no ones favorite), or makes an overwhelming favorite win, is what we'd want. That way we minimize the number of people who are absolutely disgruntled.
> It depends what we want to optimize for, but I think most people would agree that we want to avoid having a very large number of absolutely discontent voters. So I think a system which either makes the least hated candidate win (even if no ones favorite), or makes an overwhelming favorite win, is what we'd want. That way we minimize the number of people who are absolutely disgruntled.
If the least hated candidate wins, I worry that they would lack the mandate to get anything done. I'm not sure if that's a good thing or a bad thing.
I'm not sure why that would be the case. Why would they face any more opposition than any other candidate being elected would to put in place their electoral promesses?
In fact, I'd expect less opposition, since more people approved of their proposed platform.
As someone who did some academic research on the subject, I second this. Approval voting is the one which makes most sense, as it is a method which lies on the pareto curve of "optimal" voting methods. On top of that, it is also easy to be understood by the laypeople and easy to perform the counting (so we do not need to resort to "SPOF" machinal counting).
Oh fantastic, I hadn't realized Nicky Case did a post on this too!
The chart at the bottom demonstrating the range of "Bayesian regret" caught my eye when I first started looking into all of this 3 years ago. It roughly measures aggregate societal satisfaction with respect to the result of the election versus their own preferences. Strategic ballots in approval voting reduce to FPTP -- there's not really much room for a concerted effort to sway an election one way or another, since this strategy will be shared by large swaths of each candidate's voter base and so is roughly negligible. And people will feel that "their voice is heard" which I strongly encourage as political engagement is contingent on feeling like it's worth it.
As a casual observer, I wonder why ranked-choice captured the mind share when people who are passionate about voting systems seem to universally prefer approval. Maybe there's something about voting for multiple candidates that just doesn't sit right with people?
In my experience, when you ask people to do approval voting, they seem to rank the candidates then try to figure out where to put the cutoff between the approvable candidates and non. The system doesn't give any guidelines on where the cutoff goes, so people recoil because they feel like they're doing it wrong. The fact that people can set arbitrary cutoffs and it'll work out in aggregate is pretty surprising.
This may be too cynical, but I have to wonder if it's because of its flaws that the two parties in power are allowing RCV to gain ground. RCV eliminates the spoiler effect for non-competitive third-party candidates, but conveniently fails to do so when third parties start to become threatening.
Approval voting, on the other hand, always allows people to show support to a third-party candidate without risking throwing away their vote....
In my mind it's something a little more subtle but to the same effect. I think that ordinal and cardinal voting systems attract/repel people with different psychological tendencies, probably not unlike the tendencies described here, related to tolerance for uncertainty: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3092984/.
Ordinal systems impose a strict rank on peoples' choices, whereas cardinal systems allow people to settle on superpositions of choices, to explicitly give up control.
To me, being forced to order my choices is an unfair imposition. I want to be able to say "I'd be okay with fish or vegan food, someone else who cares more should make the final call". I don't want Rob Richie telling me that I should vote for fish over vegan food because I happen to really like fish. In doing that, I'd be alienating my vegan friend, and because I'm totally up for eating vegan food, I find it unacceptable to be pressured into ranking one over the other. This is the very dynamic that creates splitting, and I'm offended by it.
Being forced to rank my choices is unnatural to me, but I've heard ordinal voting advocates claim that ranking choices is natural and intuitive to everyone/most people, even after sharing my position that it feels funny. Go figure.
I'd expect ordinal voting advocates to be more likely to agree with statements like "everyone has a favorite color".
A good ranked choice system like Schulze, avoids that issue.
You can give multiple people the same rank, to express no preference between them. So you could have two or more choices you don't want to express an opinion between both marked 1, then perhaps the lesser but still acceptable candidates marked 2, the unacceptable candidates as 3, and extremely unacceptable candidates as 4.
Meanwhile, somebody else can come along and rank all the candidates with a unique number if they prefer.
In Schulze, all the numbering is doing is expressing who you would pick in a one-on-one race any pair of candidates.
Ranking candidates the same (tied), simply means you would have left that question blank on a traditional FPTP ballot because you don't care which of the two wins, either because you find them equally good/bad or you feel they sufficiently close that you would prefer to let those who care more decide.
For example, I don't own a home, or have kids, so i will usually leave school property tax levy questions blank. I may have some super slight preference one way or another, but I'm mostly unaffected, so I will let those who actually care about this decide.
Maybe I'm just projecting, but I think the expressiveness of being able to rank candidates appeals to the same type of informed voters who would advocate for electoral reform.
One reason is constitutionality. The "One man, one vote" principal likely has to be upheld for a voting system used in the US to be ruled as constitutional.
I don't think it's obvious that FPTP is the worst. It has the advantage that it forces coalitions of voters to figure out their bargain before the election rather than after. The party that can eliminate factional infighting will stand a better chance at winning elections than two sets of people with substantial overlap but who are still trying to beat on each other.
That is part of what was at stake in the 2016 US Presidential election. Both sides were deeply divided, but one side decided to pull behind its candidate despite that, and the other one was still at odds with itself. The third candidate wasn't on the ballot, but his supporters were still carping from the sidelines. That would have been even louder had he actually been on the ballot. It's possible that some alternative system would have brought out enough additional voters to put one or the other over the top*, but it seems at least as likely that they'd develop even more bad blood in public.
You could explain to them that an approval or second choice for their semi-aligned opponent would leave them better off, but the longer they continued to fight, the more likely they'd just consider each other enemies rather than friends. Either way, their opponents who came behind their candidate would have an easier time attacking both at once.
My point isn't that the alternative systems are bad, but that in the end every election has exactly one winner, and that winner will always represent a coalition since people never see exactly eye to eye. They have to work out the rules of that compromise at some point, and the advantage of FPTP is that they get to do so among themselves rather than mixing it in with the broader contest. Those who do so will find themselves at an advantage regardless of the system.
[0] And yeah, that's made even more complex by the weird double-aggregation system they have going on, but that's a somewhat separate issue.
A much simpler method in my opinion is approval voting: the ballot instructs, "Who do you think would perform the duties of the office to your approval? Check all names that apply." Then whoever gets the most checkmarks in aggregate is the winner. No spoiler effect, easily-understood vote tallying system, and reflects true approval in the population of voters. It can be adapted to multi-seat races just by taking the top n winners.
Ka-Ping Yee simulated elections using a number of recognized methods for counting votes[1]. I believe that the best system will be one which is easily-understood at the ballot box and behaves as similarly to the Condorcet method as possible. According to these simulations, it appears that approval voting is the best.
No election method will meet all the requirements for an ideal one[2] so it comes down to which conditions you want to enforce and which are less important. IMO, monotonicity is very important while the later-no-harm criterion can be interpreted as a compromise which is necessary for strategic voting.
[1] http://zesty.ca/voting/sim/
[2] https://en.wikipedia.org/wiki/Arrow%27s_impossibility_theore...