The DAO's approach to multiple-choice voting

From memory, the changes to the WG rules were quite rushed. The intention was to make a few amendments, but many of the rules had to be completely reworked. I don’t know that every single change was communicated in an itemised way in the forum but the reworked rules were available for everyone to read before it was put to a vote.

I know that each time we used ranked choice for steward elections, the process left a lot to be desired. In removing the requirement to use ranked choice voting, I wanted to leave it to the metagov stewards to choose the voting method. Perfectly happy to specify the voting method and tighten up the WG rules if we land on a method that works well in the upcoming elections.

Candidates are ranked following the outcome of a vote, based on the number of votes they receive. This is true irrespective of the voting method used. The person with the highest number of votes is ranked number one, the person with the second highest votes is ranked number two etc.

I’d like to add here a screenshot I made from the feedback here and on twitter, showing how it would display a more complex vote, where there are cycles and winners have ties.

I still believe that this is an almost self explanatory system, but will be running a few tests on Usability hub with a more general population to better data. Of course, this requires building such system which I would be happy to contribute. Meanwhile I am in favor of using approval voting on our next important election while we wait for Snapshot to support Ranked Pairs.

So my vote for votings systems currently is:

  1. Ranked Pairs
  2. Minimax
  3. Shultz
    2.5) Any other Condorcet
    :white_check_mark: Approval voting
    :+1: Approve-oppose
    0.3) Weighted Voting
  4. Ranked Choice with IRV
    :x: First past the post

I’d also like to add an ncase graph since @carlosdp posted it. I made a Sandbox example to illustrate my main point:

Good example of the third place spoiler failure of IRV. The two main candidates are more extreme far right and far left than anyone in the population. There’s a third, less popular candidate in the middle who is closer to most users. In FPTP, two round systems and in IRV, the third candidate is eliminated, and the next round is done with the two extremist candidates.

But by any other method, including Borda, Approval and Condorcet, the candidate in the middle wins. In all rounds he beats his opponents by over a 70% margin. I’d argue that this is not an example I picked exclusively to make it work, but rather a typical situation in the world: candidates are getting more extreme, pushing their main opponents to also be more extreme in the other direction. Most voters are not voting on their favorite but rather voting in their least hated to prevent “the other guy” to win. This is precisely what Ranked Choice voting was supposed to fix, but IRV is bad at it. (It’s not what occurred on our endowment vote however)

You can play with it here.

Yea they were in the proposal too, I just totally missed that part was removed is all.

That’s understandable, but yea I don’t think it’s wise to let the group being elected get to decide the election method. At minimum that seems like an easy point of contention that can turn into a flash point in any controversial election.

So, it seems like in this thread, we have two competing proposals: Approval Voting, and Ranked Pairs.

Perhaps we run a temp check on a proposal that puts those two up for a simple one-choice vote for the steward elections? How does that sound to everyone?

We don’t have an implementation of tanked pairs on snapshot anyway so there’s no sense doing a vote. I guess I’m looking mostly for a rough consensus that the ranked pair implementation makes sense so I can try to find it and use it on the future

That’s true, but we also technically don’t have a working one for IRV too, which is why we relied on @nick.eth 's Sankey graphs.

If the DAO really preferred ranked pairs over approval, I imagine it wouldn’t be too difficult to replicate the visualization you’re suggesting in the same manner, outside of Snapshot, and just use snapshot RCV, just like in the last vote.

That said, I’m also down to propose we do approval voting, try that out, and go from there!

The IRV implementation on Snapshot works fine. The only problem is that it doesn’t provide clear insight into how the results were arrived at, which is where the Sankey charts come in.

I’ve wanted to share that I’ve written two longer articles on my experience on the vote. They breach basically two topics:

It’s my opinion that on the next votes we should try to:

  • On the short term: Use approval voting as it’s the system that Snapshot has
  • On the longer term: work with snapshot to implement Copeland implementation of Condorcet with a Minimax tiebreak. In short words: whoever wins the most races is the winner and ties are broken by whoever had the least rejection. It’s self explanatory by just presenting a table of results.