Someone just shared this website with me. I have not read it over fully yet but it looks like a ballot initiative for RCV which is some IRV/Condorcet hybrid with 1/2 votes for people who gave the same ranking.
Can anyone make sense of this proposal?
This website is not to be confused with the other new group also proposing an IRV initiative in Oregon. Oregon-RCV. https://www.oregonrcv.org/
Seems like the proposal (I think from http://www.rankedchoiceoregon.org/) is for https://electowiki.org/wiki/Instant_Pairwise_Elimination. It’s only Smith-efficient when there are no Condorcet cycles among candidates not in the Smith set; otherwise it can fail even the majority criterion. The inventor, who I’m guessing is behind this website, often claims that something like IPE can cash in on IRV’s “eliminate least popular candidate” formula while being much closer to a Condorcet method.
I checked it out a bit more thoroughly, and it’s essentially IRV with fractional equal-rankings and a step which eliminates the Condorcet loser repeatedly in each round before defaulting to eliminating the FPTP (IRV) loser.
I have to say that handling equal-rankings fractionally seems fundamentally wrong, though. It’s a literal way of retaining vote-splitting. Here was a discussion held over the method:
@AssetVotingAdvocacy, I’d like to hear more of your thoughts on what you say here “handling equal-rankings fractionally seems fundamentally wrong, though. It’s a literal way of retaining vote-splitting.”
I met with the two organizers today, along with Alan Zundel (who founded the original group Ranked Choice Voting Oregon, which officially endorsed STAR Voting in 2017 and coalitioned with Equal Vote later that year.)
Reading their site it’s still not 100% clear what the details of the method are though I believe your assessment is correct after talking with them. If there is a concise explanation from them which can be quoted directly please share it here.
It looks like the leaders are Richard Forbes (votefair.org) and Joseph Hoffman (who is new to voting theory.) They started on this a month ago.
Regarding that, here’s what their ballot measure says on handling equal-ranking:
(8) If a ballot ranks two or more candidates at the same ranking level and two or three or four or five of these candidates are continuing candidates who have risen to become the highest-ranked candidates on this ballot then temporarily while looking for the candidate with the fewest continuing votes these continuing candidates shall split this ballot’s single vote into equal decimal or fractional counts that add up to no more than one count per ballot. The Secretary of State is allowed to determine how to handle ballots on which there are more than five continuing candidates at the same highest ranking level. […]
(9) “Equal decimal or fractional counts” means a decimal value such as 0.50 or 0.33 or 0.25 or 0.20 for two, three, four, or five (respectively) continuing candidates […]
So when I say that’s like literally retaining vote-splitting, I mean that a voter can not always fully support more than one candidate at a time, since the voting method treats their vote as splitting equally between all their supported candidates when it defaults to RCV. I wrote a critique on this in that discussion, and to take the most salient points from that:
The good thing is that your voting method reduces to Approval voting when all voters use only two ranks to evaluate all candidates, which regular RCV doesn’t.1 But your method might retain some unnecessary vote-splitting during the RCV stage because of its FPTP-like behavior during that stage, which could be reduced if voters were allowed to give one vote to each and every candidate they ranked equal-top.2 A related example of how this is problematic is that under section 4, part 11 of your measure, “For this purpose if a ballot ranks two or more candidates at the same highest ranking level then this ballot does not contribute support to any political party.”; according to that logic, a voter who equal-top ranks 1 Democrat supports the Dem party, but not if they top-rank 2 Democrats.
1 This is because you’re guaranteed to have a Condorcet ranking when voters rank the candidates Approval-style (ignoring ties in vote counts), so there’s always a Condorcet loser (i.e. the Approval loser) at each stage of elimination.
2 This seems like the most sensible behavior for the IRV stage, since after all that’s exactly what the method is doing during the pairwise counting stage. Consider that if a voter ranks two candidates equal-top and does not rank any other candidates, he gives those two candidates one vote in every pairwise matchup, which is essentially like he approved those two candidates and disapproved all other candidates.
@AssetVotingAdvocacy, Can anyone break down how this compares to Condorcet methods like Minimax, Ranked Pairs, or Schultz? Specifically, I know that some of these include variations where voters can give equal rankings. How does that work in those methods? When that’s the case, do any of them pass the Equal Vote Criterion, and thus eliminate vote splitting and spoilers in the vote itself?
They seem to be fairly antagonistic towards those already:
STAR voting combines two flawed vote-counting methods and depends on the confusing interaction between the two methods to overcome the weaknessesof each method.
To clarify what this means, let’s start by saying that STAR voting combines Score voting with Approval voting, and both of these methods are vulnerable to voting tactics that can increase a voter’s influence over other voters.
Score voting is vulnerable to ballot-marking tactics that exploit the fact that gaps between score levels can increase a ballot’s influence over other ballots. For example, a single voter who marks one candidate at score level 1 and another candidate at score level 5 cancels out the influence of 4 ballots on which those two candidates are separated by just one score level, in the opposite direction.
Approval voting is vulnerable to the tactic of not approving an OK-but-not-great candidate if the voter knows that not a lot of other voters will be approving that candidate. This increases the chances that the voter’s “great” candidate will win. But this tactic requires knowing how other voters will vote.
An important part of STAR voting is to confuse voters about how these different voting tactics will interact. Supposedly this confusion will cause voters to vote more sincerely.
Yet STAR voting is so new that it has been used in only a few governmental elections. Therefore claims about how voters will mark STAR ballots amount to speculation.
Like I said above, the proposed method is not always Smith or even Condorcet-efficient when there are Condorcet cycles among candidates not in the Smith set; so one possible strategic vulnerability would be voters trying to start cycles.
Most Condorcet methods treat equal-rankings in a pairwise comparison the same way STAR does equal-ratings in its runoff. There are subtle variations on how to handle the strength of a victory in a pairwise comparison, but IIRC there is at least one mainstream option (I think it’s the margins option, as opposed to winning votes, meaning that a candidate with 70 votes is said to pairwise beat a candidate with 50 votes by a margin-strength of 70, rather than a winning-votes-total-strength of 70) which satisfies the Equal Vote Criterion.
Since the proposed method handles equal-rankings fractionally in its RCV stage, I’d have to imagine you could mathematically construct a situation where adding in two opposing ballots could change the results, though it’d have to be something where one of the rounds has no Condorcet loser and the candidate with the fewest 1st choices then shifts because of the added-in ballots.