All that said, if we know we’re using proportionate representation ranking, the CIVS docs really can’t be clearer about which mode is applicable wrt to best-candidate vs. combined-weights in a multiwinner election:
Combined weights. In combined-weights mode, the voter gives a nonnegative weight to each candidate instead of ranking the candidates. The voter’s goal is to maximize the sum of weights of selected candidates. This is an appropriate criterion for elections where the quality of all the candidates is important to the voters, such as the election of an actual committee that will be voting on some issues.
The SC is “an actual committee that will be voting on some issues”. In which case, the CIVS docs plainly say “use ‘combined weights, don’t use best-candidate’”. I’m concerned that the decision to use best-candidate mode was just an oversight, and I have no idea what impact it will have on the outcome, because it seems like a nonsensical choice for any multiwinner election.
Best-candidate makes sense for representative bodies, in which case it optimizes for the likelihood that a given voter gets a member of the elected body that they feel is aligned to their interests.
I think that is the function of the “proportional” part of “proportional best-candidate,” but not the “best-candidate” bit. “Best-candidate” is for single-winner elections.
No, I don’t think that’s right. The best-candidate versus combined-weights decision is about how to ‘lift’ (CIVS’s term) a voter’s ranking of individual candidates to a ranking of sets of candidates. CIVS’s proportional elections work by conducting a ‘single-winner’ election over potential winner sets—if there are 20 candidates for a 7-winner election, there are (20 choose 7) 7-candidate winner sets to be ranked in this way. This is true of both best-candidate and combined-weights modes; the only difference is how the ranking of the (20 choose 7) sets is computed from a voter’s ranking of the 20 candidates. Best-candidate will prefer set A over set B if the best candidate from A is better than the best candidate from B, and combined-weights will look at the entire composition of A and B to compare them.
The subgroups are sort of dynamically calculated from the votes; see the CIVS docs for details, but the short version is that voters’ preferences for a winner set are truncated past the number of candidates approximately equal to the size of that voting bloc, so that 60% of the voters who more or less agree can only influence about 60%-ish of the final committee.
Ah. I don’t have an opinion then. Well, I have one- CIVS is misusing terminology and causing confusion. Elections shouldn’t be this hard to understand.
OK, I won’t quibble about it. I’d still like to know how the result will be computed, because I’ve been told that OpaVote doesn’t support this particular best-candidate mode.
The CIVS recommendation has been considered, but the level of confidence of this recommendation was found not high enough for overruling other considerations.
The tallying systems considered were those supporting some emergent proportionality, not proportionality among explicit electoral blocks. This is unlike e.g. Kubernetes who choose top-Condorcet-ranked candidates.
Note to the people who haven’t voted yet: Please be aware that if you leave the voting page open for more than an hour, you won’t be able to submit the ballot without reloading and losing what you entered, so you might want to make a screenshot beforehand!
While testing tallying of ballots using CIVS this weekend, I’ve noticed a potentially major problem: CIVS might be infeasibly slow to compute the result. In a test with randomised ballots, the result is still being computed after ~40 hours, with absolutely no progress being reported!
Since closing the election would give us access to the ballots, I’m making the case to the EC that we need to delay closing until we are sure that using CIVS is feasible, and otherwise decide on an alternative tallying mechanism in advance.
As such, I’m holding off closing the election myself for now and am bringing this topic up for internal discussion in the EC’s meeting later today. So you can still vote for now if you haven’t already, but should expect it to be closed later today.
It’s not even optimisation, as much as exploration — random ballots mean that basically all combinations are plausible, and life is full of pain because one needs to apply tie-breaking rules to thousands of thousand-committee preferences cycles (which need to be enumerated first). Have some people with a large support base and some with a much smaller one, and there are suddenly much fewer plausible committees to even compare.