Forum Council

@wolftune I'm not sure this is just an Allocated Score thing. I think that sequential methods that start by electing the candidate with the highest total score and go from there are always likely to have this sort of thing happen. Non-sequential methods may avoid it more easily, but that's obviously more expensive computationally. The other option is to not go by total score.

@psephomancy said in GPT and I invented a new voting system metric?:

@toby-pereira said in GPT and I invented a new voting system metric?:

I think if a measure isn't cloneproof it's probably not a good measure.

Why would that matter for a measure?

Because you can have a candidate that is the closest to being the Condorcet winner but not the Copeland winner. E.g.

14: A>B>C
4: B>C>A
12: C>A>B

A>B - 26:4
B>C - 18:12
C>A - 16:14

A has the biggest winning margin and smallest defeat, and is the nearest to a Condorcet winner by any reasonable measure. But then you can clone C and have a C1 and C2 but where C1 is always ranked above C2.

In this case A now has two defeats (against C1 and C2) so loses to both B and C1 in Copeland. But A is still the nearest to a Condorcet winner in terms of defeat sizes, so I would say they are still the "most Condorcet" winner.

@psephomancy Yes, I agree about the resistance to strategy. But I see a failure of independence of clones as separate from strategy. I think if a measure isn't cloneproof it's probably not a good measure. Also, Copeland is very low resolution anyway in that it just looks at number of defeats rather than the size of any of them.

To speak to Adam's point itself-

Exhausted ballots in RCV are not only "votes of no preference" ie. ballots that intentionally didn't rank any of the RCV finalists, but also:
a) votes that were unable to transfer to a viable alternative because of the order of elimination, and that could have made a difference if their other rankings had been counted.
b) voters who were unable to rank all the candidates due to ballot limitations.

STAR always counts all the ballot data. Every ballot is included in the final round. A vote of No Preference is explicitly a voter who chose to score those candidates equally, and that neutral runoff vote IS counted. No voters will have their votes unable to transfer.

Note: Spoiled/voided ballots are often included in RCV exhausted ballot stats, but are sometimes counted separately. In any case, spoiled ballots in RCV are common. Accidentally spoiling your STAR ballot is much harder to do.

To clarify, the doc shared above, and the image at the top, is Adam's work. It's not not an Equal Vote graphic. The part in color is a crop from one of our scorecard infographics. The other part is from a volunteer's DRAFT graphic she's working on and shared privately asking for feedback. Adam has these two element combined in the doc he created.

@masiarek I expect this wasn't your intent, but taking her draft work that she shared in a private conversation (the STAR Slack is a private group for our volunteers) and then reposting it for critique without context in a public forum (votingtheory.org) is problematic, especially because you've modified the work and collaged it with other graphics in a way she didn't approve.

Interesting. So by starting with a random committee, the method is not deterministic. Not that that is necessarily bad in all cases (see COWPEA Lottery), but I do wonder if it's worth the extra complexity over SPAV in what seems to essentially be a box-ticking exercise.

@jack-waugh OK, there are obviously various competing Condorcet methods, and on the EM mailing list there's discussion and debate going back years, but a good starting point is ranked pairs, which has good criterion compliance (such as being cloneproof) and is relatively simple to explain. It's also far more likely to pick a unique winner than Copeland.

After that, the rest is just tie-breaking, so if you're not using Copeland any more, it's not such a worry.

With the Borda count, although it's normal to give points to positions in a linear manner, it is equivalent to just summing the margins of the head-to-head wins so it looks equivalent to this Dasgupta/Maskin score. It's rare you'll reach that stage if you use ranked pairs. You'll almost never have to use your scores. So at this point, it's a case of just picking your favourite Condorcet method.

@toby-pereira said in Name For This System:

@jack-waugh It seems very convoluted. Also just three methods stuck together in order of tie-breakness.

Why Copeland in particular of the Condorcet methods? It's not cloneproof, which I see as a major weakness.

Because I don't know any better. Do you want to suggest a different Condorcet method?

Then you have the Dasgupta/Maskin scores. Is that not just equivalent to the Borda count?

I don't know. My thinking was that if Copeland ties, D/M gives every voter equal power and it considers only how many voters prefer one member of a pair of candidates over the other, ignoring the intensity with which the voters so claim to prefer them.

And then you have for the final tie-break score, but score under this very particular ballot that only allows certain scores.

The reason to limit the vocabulary of scores to only seven possibilities is to make it easier to tally ballots by hand. They can be sorted manually into seven piles, then run through simple mechanical (so unhackable) counting machines. This is repeated for each candidate.

If there are seven possibilities, do they have to be spaced evenly? I think that putting the finer grains near the top and leaving the middle relatively coarse could make the most economic use of the few possible scores better toward fueling strategies for voters who want to push the system from where a Nader gets 1% approval to where he gets a serious chunk and people don't think he's fringe anymore.

What's the thinking behind this method?

I thought the level of complexity (just three stages) might be justified for building a system that would likely be seen as fair, as compared, say, to STAR. And since the final tiebreaker is based on a ratio scale rather than just an ordinal scale, that could help take into account more information from the voters, and again, provide that strategy to edge out the duopoly gradually.

@jack-waugh It seems very convoluted. Also just three methods stuck together in order of tie-breakness.

Why Copeland in particular of the Condorcet methods? It's not cloneproof, which I see as a major weakness.

Then you have the Dasgupta/Maskin scores. Is that not just equivalent to the Borda count?

And then you have for the final tie-break score, but score under this very particular ballot that only allows certain scores.

What's the thinking behind this method?

The voter gets to grade each candidate from:

• A (100)
• B (98)
• C (80)
• D (0)
• E (-80)
• F (-98)
• G (-100)

This distribution of possible scores comes from a logistic curve.

Leaving a candidate out is the same as giving her a G.

It is permissible to grade any number of candidates the same.

The count of ballots yielding preferences in each direction with respect to each pair of candidates is collected at each polling place and published. Also, each candidate's sum over the scores, and the total count of voters.

Once the summary data are available from all the polling places, calculate each candidate's Copeland score. This is the count of other candidates that that candidate beats minus the count of other candidates who beat that candidate. A candidate A is said to beat a candidate B exactly under the condition that strictly more count of ballots show a preference for A over B than the count of those showing a preference for B over A.

Eliminate those candidates who do not tie for the top Copeland score.

If only one candidate remains, she wins.

Otherwise, from among the candidates still in the running, calculate their Dasgupta/Maskin scores. This is, for each candidate still in the running, the sum over the other candidates still in the running, of the count of voters who preferred the first candidate over the other candidate minus the count of those having the opposite preference between the two of them.

Eliminate those candidates who do not tie for the top Dasgupta/Maskin score.

If only one candidate remains, he wins.

Otherwise, from among the candidates still in the running, elect the one with the highest total over the grade scores they received from the voters.

This is quite like what Rob Brown suggested as Reverse STAR, but with the added stage to consider the Dasgupta/Maskin scores.

Score After Dasgupta/Maskin After Copeland -- SADMAC