mccJul 11, 2024

What is your favorite alternative voting system?

[Notes: • If you vote in one poll in this thread please vote in all of them • For a real world example of what I mean by "Multiround runoff" consider the French system]

Condorcet
15.8%
Instant Runoff
48.6%
Approval
23.3%
Multiround runoff
12.3%
Poll ended at .
Show threadmccJul 11, 2024

What is your favorite alternative voting system? (Poll #2)

[Notes: • If you vote in one poll in this thread please vote in all of them • For a real world example of what I mean by "Multiround runoff" consider the French system]

Condorcet
21.1%
Instant Runoff
39.2%
Approval
20.6%
Multiround runoff
19.1%
Poll ended at .
Show threadmcc

What is your favorite alternative voting system? (Poll #3)

("Instant Runoff Voting" is when you rank your choices and a multiround runoff is simulated by bumping off candidates one by one and redistributing their votes. "Approval voting" is when you get checkboxes by each candidate and can select more than one; whoever gets the most "approvals" wins.)

Instant Runoff
73.3%
Approval
26.7%
Poll ended at .
Jul 12, 2024 at 9:20pmWeb
Show threadNeia masksJul 12, 2024
@mcc I'm just hoping we get to implement Dodgson's Method some day.
Show threadmccJul 12, 2024
@neia Do you know how this differs from the "Cloneproof Schwartz" method?
Show threadNeia masksJul 13, 2024
@mcc Do you mean Schulze? It's been a hot minute since I looked into it in detail, but IIRC Schulze is basically looking at how much Candidate A could beat someone who could beat Candidate C and vice versa, and it uses that to resolve cycles where A beats B beats C beats A.

Dodgson's Method alters individual ballots, switching the minimum number of individual preferences to produce a Condorcet winner.

In terms of how well they perform, performance is the biggest one. Schulze is O(n³) in the number of candidates. Dodgson is exponential in the number of ballots. Worse, Dodgson is not in NP. So if a large country like Liechtenstein decided to use Dodgson, they might have to wait a few centuries to get the result of a national election. And if you need independent verification, it will take centuries longer.

For pretty much all other standard voting criteria, Dodgson fails. Like, it fails the anti-Condorcet criterion: if you held a bunch of 1v1 elections and one person lost against every other candidate, Dodgson sometimes elects that one person.
Show threadConor NashJul 12, 2024
@mcc brb implementing PR-STV as an ActivityPub voting mechanism
Show threadmccJul 12, 2024
@Conornash I was thinking about making an app you log into with oauth, but that sounds sick too
Show threadJames WidmanJul 12, 2024

@mcc which one is better at expressing my desire to see one of the candidates eliminated?

i.e. if we have an election where there are 3 candidates, X, Y, and Z, and to me, the most important thing is to keep candidate X out of office, which voting system does a better job of supporting that priority? (If i *also* get to express my preference for Y vs Z, that's an added bonus.)

Show threadmccJul 12, 2024
@JamesWidman This is an excellent question. In my opinion it is IRV. Condorcet advocates insist that Condorcet does eliminate the need for strategic voting, which to me would imply the property you ask for, but when I do the math that's not what I see. I do not know if this means the Condorcet advocates have missed something, or if it means I did the math wrong. I'm kinda inclined to the second theory, but my point is I'm not sure.
Show threadjmeleskyJul 12, 2024
@mcc @JamesWidman I think they’re largely equivalent from an expression perspective, right? I thought the difference between IRV and Condorcet was in the counting, not the ballots, but I’ll admit it’s been a while since I was evaluating them closely.
Show threadmccJul 12, 2024
@jmelesky @JamesWidman Yes, I previously ran a ranked voting poll (a yearly poll about the best video game that year) and I was able to produce borda, condorcet and IRV results from the same ballots. Actually, the IRV and condorcet results only rarely differed. (I was never able to find a situation where borda and condorcet differed at all. Did I implement it wrong? Still not sure.)
Show threadjmeleskyJul 12, 2024
@mcc @JamesWidman Thanks for verifying. I had completely forgotten about Borda, but now that you mention it, it always seemed like a strange amalgam of IRV and approval. It may well be mathematically equivalent to Condorcet.
Show threadmccJul 12, 2024
@jmelesky @JamesWidman Yeah, I actually didn't know about Borda when I started implementing the script. I actually implemented Borda *by accident* while trying to implement Condorcet. Then later someone who looked at my method said "that's not Condorcet, that's Borda". So I wrote a much more complicated script implementing Condorcet proper, and it worked worse (didn't provide "scores") and seemed to give the same results. Like I said, I'm still a bit confused what happened there.
Show threadjmeleskyJul 13, 2024
@mcc I suspect it would require a very niche scenario for Borda and Condorcet to differ (if it’s possible at all). On the day that actually happens, I’ll happily tolerate the very few people in the two camps loudly yelling “I told you so!” because either one is such an improvement over what we have now that I will be calm and serene in my dealings with the world.
Show threadBerkubernetusJul 14, 2024
@mcc @JamesWidman have you published anything with that math? As someone who works on voting systems, I'd be very interested
Show threadmccJul 14, 2024

@fuzzychef @JamesWidman Uhh, that'll be here https://msm.runhello.com/p/category/video-games/goty You'll notice each post in this category, over time, got less and less certain about the accuracy of the Condorcet.

It looks like I didn't publish the outcome of my "real condorcet" test. I remember it being really slow (my implementation was PHP+SQL and there were over 150 options in the poll)

I might have also published the poll software somewhere, I don't remember. This was a while ago, the poll ran yearly from 2003-2013.

Run Hello » Game of the Year poll

Show threadmccJul 14, 2024
@fuzzychef @JamesWidman Also hm the 2013 post may be missing from the blog? Whatever.
Show threadBerkubernetusJul 15, 2024

@mcc @JamesWidman I would suggest that if you have a poll with 150 options, Condorcet is not a good voting algo to use. It's really designed for elections with 3-16 options, where a large number of voters might conceivably rank all options.

I wouldn't be surprised at all if a case of 150 options, with 95% of voters ranking only 10 of them each, causes condorcet to produce unpredictable results.

Which gives us the general mantra: choose the vote tally method that suits your poll.

Show threadmccJul 15, 2024
@fuzzychef @JamesWidman Yeah, that sounds about right to me. As I remember, the point where Condorcet started to seem weird was when it had to produce a relative ranking between two candidates which never appeared ranked on a single ballot simultaneously. Borda, of course, could do this without any conceptual difficulty.
Show threadBerkubernetusJul 15, 2024
@mcc @JamesWidman That's why a lot of Condorcet voting tools mandate ranking all candidates.
Show threadmccJul 15, 2024
@fuzzychef @JamesWidman 🇦🇺
Show threadbswolfJul 13, 2024
@mcc Approval is the only one I didn't vote for in approval.
Show threadanyMouseJul 13, 2024
@mcc Fondly remembering the time when my Linear Algebra professor made one entire lesson about "Mathematically proving that a fair voting system doesn't exist"
Show threadMarcoJul 13, 2024
@anymouse_404 @mcc Physics For Birds just did a video on the mathematical issue with “fair” voting from a topological approach: https://youtu.be/v5ev-RAg7Xs
The Topological Problem with Voting

YouTube
Show threaddroqenJul 13, 2024
@mcc ever since i heard of approval voting ive never approved of any other voting system
Show threaddroqenJul 13, 2024
@mcc why is approval for approval voting so low across the board?? am i missing something??