[EM] Coombs method and typical RCV hybrid, River

[Kevin Venzke]

Hi Daniel,

Le samedi 22 janvier 2022, 14:50:54 UTC−6, Daniel Carrera <dcarrera at gmail.com> a écrit :
> Yeah, RP is notoriously expensive. It was my favorite method until I discovered
> River. RP is very strong and I find it a million times more intuitive than
> Schulze, which is its best known competitor. I suspect that you'll get a lot of
> buy-in for River in this list. Yes, it is less widely known but I think everyone
> in this list would love to fix that. It has nearly the same intuition as RP but
> it is vastly cheaper (computationally), and if I recall correctly, it meets all
> the criteria of RP and Schulze plus one more... I forget which one... Maybe
> independence of Pareto-dominated alternatives.
> 
> So... if you like RP but you find it slow... River is its faster, better, but
> obscure cousin.

River in its default WV form is certainly the sort of method that I think is
good. WV gets you the Plurality criterion, and experimentally very low
compromise incentive and few FBC failures (i.e. where compression is an
inadequate strategy to be able to safely vote a favorite candidate at the top).

But I feel like I ought to say, that I am pretty sure that any analysis which
concludes that Condorcet//IRV (or similar) is a good method is not going to
agree that River is a good method. A surface-level analysis will probably show
River and similar methods have a lot of burial incentive.

To me at least the principle of Schulze is intuitive. It's sort of a de-cloning
of MinMax.

One property unique to Schulze(WV), which perhaps I alone find interesting, is
that it always elects from the CDTT, which is the Schwartz set defined using
only the full majorities.

For example, it could be that some candidate "X" has a majority-strength
beatpath to every other candidate, and there are no majority-strength paths back
to X. It could be that River tosses out X's wins because they defeat candidates
already locked as defeated (by an even larger majority). And then X is
subordinated to another candidate via a sub-majority win.

Now does this really matter, probably not, in this context, where the
alternatives are RP(WV) and River, which I think can be trusted not to mess
anything up too badly when electing from outside the CDTT.

Kevin