At market close, I'll run an election to determine which voting system is the best, lasting 3 days. Any manifold user can, and should, vote.
The election will be run using the electoral system with the highest average probability for the entire market duration. The winner(s) of that election determine the final outcome of this market.
Related questions
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ343 | |
| 2 | Ṁ125 | |
| 3 | Ṁ83 | |
| 4 | Ṁ48 | |
| 5 | Ṁ43 |
@traders Below is the Election! We proceed by approval voting, so you can like as many of the options as you wish. The poll will close around this time on the 2nd of april.
RCV is undemocratic junk; don't fall for it!
Voters express their rankings of all candidates on the ballot, but RCV doesn't actually count all of them. It gives some voters more power than others, and it suffers from vote-splitting, spoiler effect, center-squeeze effect, etc. causing it to devolve into a polarized two-party system.
@Snarflak that's approval voting which does not count in all your candidates, only the highest scoring ones. RCV collapses to comparing the highest scorers only after it shows, that even with maximal support some candidates would not win, and then does no throw the votes away but uses second level preferences.
Let's say in approval voting I do not want to nominate any of the A and B (who are expected to be leaders of the run), but want to vote C and D. If I do the honest vote, my opinion will be thrown away. The approval system forces me to use the knowledge/guess about who would be the two toprunners, and vote for the lesser evil between them. And that only increases the dominance of the two.
In RCV I would make honest prioritisation [C,D,A,B] and every member of that array is tested whether they have enough support. If not, the same "safety against bigger evil" is built in at the end, but without hurting my own candidates.
that's approval voting which does not count in all your candidates, only the highest scoring ones.
I don't think you understand my comment. Approval voting counts every mark on every ballot. Hare RCV does not.
and then does no throw the votes away but uses second level preferences.
That's exactly the problem. It doesn't count second level preferences until after eliminating a candidate. It puts the cart before the horse and eliminates candidates based on incomplete data.
Voters express preferences such as "I prefer Abe over Betty" and "I prefer Abe over Claude" and "I prefer Betty over Claude", but Hare RCV only counts some of those preferences and discards others. A majority of voters can express "I prefer Abe over Claude" on their ballots, while a minority expresses "I prefer Claude over Abe", yet RCV can still elect Claude, because it counts more of the minority's preferences than the majority's.
The approval system forces me to use the knowledge/guess about who would be the two toprunners, and vote for the lesser evil between them.
Yes, the optimal strategy in Approval is to look at polls and see who the frontrunners are, and then approve of the frontrunner you prefer, as well as honestly approving of every candidate that you like more than that frontrunner, and disapprove of every candidate you like less. The only "strategy" involved is where to set your approval threshold. Without polling information, the optimal strategy is to approve of everyone you like more than the average candidate.
Every vote is counted, it is always safe to vote honestly for your favorite, and there is never any incentive to dishonestly approve of a candidate you dislike while not approving of a candidate you like. This results in the most-preferred consensus candidate winning the election, reduces polarization, and allows third parties to become viable.
In RCV I would make honest prioritisation [C,D,A,B] and every member of that array is tested whether they have enough support.
No, every member of that array is not counted. If you vote C>D>A>B, then you have expressed C>D, C>A, C>B, D>A, D>B, and A>B. But your complete preferences will only be counted if the candidates happen to be eliminated in the same order that you listed them. Otherwise, only some of your preferences are counted, and others are ignored, which can result in undemocratic outcomes. For that reason, this scenario is always used in RCV marketing, but it's not the only scenario that can happen.
If not, the same "safety against bigger evil" is built in at the end, but without hurting my own candidates.
No, it's not safe to vote honestly under Hare RCV. You must still rank the "lesser evil" as your first choice, or you risk wasting your vote. Since RCV only counts 1st-choice rankings when determining who to eliminate, if you give your 1st-choice to your true favorite, then you risk taking enough votes away from the lesser evil that they are eliminated first, and then your preference for lesser evil > greater evil is discarded, and greater evil can win, even if they are not the preference of the majority.
if you give your 1st-choice to your true favorite, then you risk taking enough votes away from the lesser evil that they are eliminated first, and then your preference for lesser evil > greater evil is discarded
Then it is discarded, and what is left on my ballot is a first level preference versus greater evil, and in this "second round standoff" my favourite gets a strong preference over it, because it is still number 1.
(Just in case, my favourite system is neither of the two, I just want to discuss).
Then it is discarded, and what is left on my ballot is a first level preference versus greater evil,
Yes, but your preference for lesser evil > greater evil is not counted, and has no effect on the outcome of the election. Even if a majority said they preferred the lesser evil, the greater evil wins. If you had tactically ranked lesser evil first, then lesser evil would win and you would have a better outcome, just like under FPTP.
and in this "second round standoff" my favourite gets a strong preference over it, because it is still number 1.
RCV doesn't count strength of preference. Your preference for favorite > greater evil is counted with the same weight regardless of which round it is counted in.
Cardinal systems like STAR Voting allow you to express strong or weak preferences, though.
Yes, but your preference for lesser evil > greater evil is not counted
It should not count if some of evils have already dropped of. And In approval system even less of such pairs is accounted.
You have to mark lesser evil and not mark great evil (strategic but not honest), just like you suggested placing lesser evil on top of your ranked list. We have not established whether a good system should have such obvious strategic methods like in Approval. In RCA it is much less obvious who will be the final pair due to redistribution.
approve of the frontrunner you prefer, as well as honestly approving of every candidate that you like more than that frontrunner,
I don't agree that all your marks are counted. There will almöst always be no effect whether I vote for only the lesser evil frontrunner or for him AND everybody I like in Approval. In RCA at least some of your secondary marks will be counted after eliminations, and your opinion will is guaranteed to be counted at the very last round. In approval it can happen, that two actual frontrunners both recieved from you the same mark, and here your vote is not counted at all.
It should not count if some of evils have already dropped of.
Yes, it should. All votes should be counted. A supermajority of voters can rank Lesser Evil over Greater Evil, yet RCV can still elect Greater Evil because it pays attention to some of those preferences while discarding others. That's undemocratic and harmful to society.
You have to mark lesser evil and not mark great evil (strategic but not honest)
That is honest; what do you mean?
We have not established whether a good system should have such obvious strategic methods like in Approval.
All voting methods have strategic incentives, as proven by Arrow and Gibbard. It is impossible to avoid them except by adding randomness.
In RCA it is much less obvious who will be the final pair due to redistribution.
You mean RCV, right? RCV's not being able to predict whether your vote will help or hurt a candidate is a bug (non-monotonicity), not a feature.
In RCA at least some of your secondary marks will be counted after eliminations
They should all be counted. Throwing away voter preference information in order to make the counting easier is unfair.
In approval it can happen, that two actual frontrunners both recieved from you the same mark, and here your vote is not counted at all.
Your vote is still counted, just not between those two. There are better systems like Approval+Runoff and STAR that have more expressive ballots.
That is honest; what do you mean?
Honest vote is when i only place the marks in front of all those candidates, who in my opinion would have a total positive impact (no matter how small the impact is). When marking lesser evil I am being dishonest, i do not actually want him to win, I just strategically reduce risk of the worst case.
(I don't know, maybe game theory has some other definition to being honest, i am using common meaning)
All voting methods have strategic incentives
Not all of them have obvious strategic choices. More complicated a system is, möre people would not even think about calculating strategy, thus voting "from heart".
That is why I ask, whether we consider having obvious strategy a good feature or not. If we have different parameters to score systems then the dialog would not have common ground to even understand each others priorities.
As far as I heard from people, systems that increase that type of honesty and reduce obvious strategies (thus absolute majority does not mess with the system), are considered good. Do you think those are good features?
Could you write, please, your personal preference list of the voting systems?
Your vote is still counted, just not between those two.
I see absolute zero difference. One system "throws away" your vote when one of the candidates was eliminated (which will happen if you did not predict well), and the other one does not give you to even set the preference between the two unexpected frontrunners. Both systems do not count your preference when something unexpected happens. It looks absolutely the same, but you blame only rcv.
Honest vote is when i only place the marks in front of all those candidates, who in my opinion would have a total positive impact (no matter how small the impact is).
OK, but even if you vote for a frontrunner that you don't love, you're still expressing an honest preference for them over everyone you didn't approve of. There is no incentive to approve of a candidate you don't like while disapproving of a candidate you like more, and it is always 100% safe to vote for your favorite candidate.
Not all of them have obvious strategic choices.
Yes, but making strategic voting more difficult is not the only goal of a voting system.
That is why I ask, whether we consider having obvious strategy a good feature or not.
Yes, it's a good feature, but it doesn't trump representativeness.
Picking a candidate at random has perfect resistance to strategy, for instance, but is not good at electing a good representative, so is worthless overall. RCV likewise behaves randomly in multi-candidate competitive elections, which makes strategic voting difficult but also makes it unrepresentative.
Could you write, please, your personal preference list of the voting systems?
Of the ones I know decently well:
STAR
Condorcet
Approval + Runoff
Score
Balanced Approval
Approval
Coombs
RCV (Hare / Instant Runoff)
Final Five / Top Four
Contingent Vote
Runoff / Two-Round System
Supplementary Vote
FPTP / Plurality
Roughly similar to this Kialo debate.
Both systems do not count your preference when something unexpected happens.
No, under Approval, every mark you put on the ballot affects the outcome of the election. Under RCV, some are completely discarded and have no effect. If B is eliminated first, A vote of A>B>C has the same effect as A>C>B. The 2nd preference is not counted at all.
@traders Below is the Election! We proceed by approval voting, so you can like as many of the options as you wish. The poll will close around this time on the 2nd of april.
@Bayesian Post a top-level comment that represents the election. Post each voting method as a comment to that one. "Likes" are votes.
@Bayesian another way to do it on manifold is to make a poll where each option is an element of the powerset of the set of candidates 😈
@Bayesian one thing that might be fun is to run the election in such a way that many different methods can be run against the set of ballots.
If you require each vote to assign a score within a predetermined range to each candidate and to clearly state what point within the range is their approval cutoff, you can run plurality, runoff, any cardinal method (approval, score, star...) and any ranked method (IRV, Schulze, Borda...) on the resulting ballots. As an example, for a range between 0 and 10, a ballot could be:
PR 10, approval 8, quadratic 7, [approval cutoff here] IRV 6, Vickrey 4, lottery 0, FPTP 0, two-round 0.
@BrunoParga right, that's just asking the agents to display their preference list, and then creating the social choice function out of that
@BrunoParga right yeah i noticed that. ig approval is preference list + expected strategic cutoff? idk
@Bayesian oh, if you want a secret vote I don't think you can hold it on Manifold, but then I don't think any external services allow for the kind of complete ballot I described.
Maybe you could act as the election authority and people would message their votes to you? You'd then publish the anonymized set of ballots and we'd run the various functions on that input.
@BrunoParga yeah tbc right now the election will run approval voting only. asking for ppl to dm their vote is an option, yeah.
@Bayesian I would like to run different methods on the ballot set myself, and for that I would like it to be public (anonymized is fine). Ofc, since approval won this market, it's the approval winner that counts; but I'm curious if this complete ballot experiment would result in different winners from different systems.
@BrunoParga I think that is a separate experiment. It has been run before in other contexts, eg political science discussion forums.
@MartinRandall sure, but not with this electorate ;) I suspect people would react differently to "did you know that theoretically different voting methods can give different results on the same set of ballots" than they would "my favorite method lost but if we had run the election using it, it would have won!"