[EM] [ESF #1547] True Ranked Choice - for Condorcet

[Kristofer Munsterhjelm]

Jameson Quinn wrote:
> This thread has touched several points.
> 
> *Branding*
> 
> I'm not particularly fond of "TRC" as a name for Condorcet. Ideally, a 
> name should give some idea of how the system actually works. That was 
> where my "VOTE" branding idea came from (Virtual One-on-one Tournament 
> Election). Other ideas along those lines:
> 
> True Ranked Choice
> Preference Playoff Voting (Note: given a winner of a condorcet system, 
> it is easy to make a post-hoc "playoff seeding" which results in that 
> winner. For some condorcet-tiebreakers, it may be possible to give a 
> logical seeding algorithm which actually gives the right winner, instead 
> of first calculating the winner and then using that for seeding.)

To have a post-hoc playoff seeding seems like cheating. Opponents will
ask why the actual method, the things happening behind the curtain, is
different from the final approach (using a certain after-the-fact seeding).

> Beats-All Voting
> Best Runoff-Winner Voting
> Victory Square (VS.) Voting (This refers to the matrix, but it is mostly 
> an attempt to get an acronym that would be pronounced "versus voting".)
> Pairwise Champion Voting

I've also given the MMV term, but it may be too technical. How about
"candidate round robin"? It may not fit methods like RP, though, as
those appear to be sequential in the way they lock defeats one by one.

> (In this regard, Range/Score Voting and Approval Voting already have 
> good names. The Borda count doesn't, but it's not a good system for 
> politics, so I'm not going to waste time on it. Bucklin could use a 
> better name, here are some ideas:

What do you consider the best name for cardinal ratings - Range or Score?

> American Preferential Voting (This name was actually used in some cases 
> where Bucklin was implemented in the Progressive era.)
> Approved/Preferred Voting (Shares an acronym with the above, best for 
> the 3-rank equality-allowed Bucklin which I support)
> Majority Choice Approval (Not my proposal, but I like it. Refers to 
> 3-rank, equality-allowed.)
> Expanding Approval Voting
> Approval With Compromises
> Fallback Approval Voting (FAV - not a bad acronym)

I'm not sure about the ones that refer to "Approval". Approval voting,
as a description of the method, is reasonable enough (strategic concerns
notwithstanding), but if you refer to Approval when describing Bucklin,
you're no longer referring to the concept of "approve all you like", but
to the mechanics of Approval, the method, itself.

> I'd like to set up web-polls on these naming questions. *What 
> openly-available web-poll system supports at least Range, and also as 
> many as possible other, scoring outputs? *(Range is certainly the best 
> system for such questions, where honest voting is likely to dominate 
> strategy.)

I know of CIVS for Condorcet, but it doesn't handle cardinal weighted
pairwise or even simple ratings. Brian Olson's betterpolls.com handles
cardinal ratings.

If you want to be truly accurate, you would have the site hide the
outcome and the ballots of those who have already voted, until a fixed
date has passed, after which nobody can vote further. That would avoid
feedback effects.

I also think the ballots should be made public after the deadline. Then
anyone who wants to can calculate the result according to their own method.

> *Criteria*
> 
> Bayesian regret is absolutely a fundamentally-important criterion for 
> evaluating voting systems. However, it is not the only criterion. 
> Neither is it entirely objective, since any actual Bayesian regret 
> measure depends on a true-preference model and a strategy model, both of 
> which are inevitably debatable. I don't think it's helpful to try to use 
> some specific set of Bayesian regret measures as a be-all-and-end-all 
> argument. They may be decisive for you, and it's helpful if you say so; 
> but they won't, nor should they, stop other people from taking other 
> positions.

If we want to maximize Bayesian regret and still have Condorcet (or more 
advanced aspects of the same, like Smith or Landau), we should try to do 
to Borda, Warren's eigenvector methods, or "sum of victories"[1] what 
Schulze did to Minmax. As you say, however, I'm unsure of the strategy; 
perhaps it would be easier to perform strategy against these methods 
than against Schulze or RP.

> *"Let's just pick one system we can all support"*
> 
> Let's pick all of them. As a practical matter, I support Bucklin > 
> Approval > STV-PR > Condorcet > Range >> TTR > IRV > [Plurality] > 
> [non-democracy], in approximate order of expected benefits (potential 
> improvement * probability of implementation / opportunity cost of 
> support).  I don't ask you if your ordering of that list is the same as 
> mine before I join you on the barricades, and you shouldn't ask me.

It only makes sense to "eat our own dogfood" if we're to make a 
Condorcet group; for the members (or society, if we can ask them) to 
rank the different name proposals to find out which is better - or for 
the members to rank the different Condorcet methods, and then to use the 
Condorcet method or methods to find the best outcome.

For a grand compromise, I think we'd need something more like 
parliamentary procedure, with arguments in favor of each method and an 
up/down vote to determine which to choose. Ideally, that would happen 
within the context of a voting reform proposal itself, such as with the 
Citizen's Assembly of British Columbia. If it is in that context, the 
participants would have to know about the differing methods, and that 
can only happen if there are groups promoting each. However, there is a 
question of balance. If there are too few groups, then good methods are 
lost. If there are too many, then they may weaken each other, or make 
the discussion of voting methods appear to be a tiny issue with factions 
arguing over minutiae. What is the optimal number? I don't know; perhaps 
one for Condorcet methods, one for non-Condorcet ranked methods, and one 
for rated methods. Even that sounds a bit much.

> 
> JQ
> 
> 

[1] "Sum of victories", "condolr" in my simulator, simply consists of 
summing up the victory magnitude on the winning side of any pairwise 
contest. It is Condorcet because the CW will lose to nobody and so gain 
at least a majority from each contest. It seems to have relatively good 
Bayesian Regret. I wouldn't imagine it to be cloneproof.