James replying to Mike... Mike: >I'd written a partial reply to James' posting, and was trying to send the >partial reply to myself, to save and work on tomorrow. But somehow, >instead >of sending it to my e-mail address, I posted it to EM with my e-mail >address >as its title.
It's just as well that you didn't complete the message, because you seem to have significantly misunderstood my previous message, and it would be rather wasteful for you to continue replying to the misunderstood version of my ideas. Hopefully this current message will clear up the misunderstanding. See below. James: > Let's say that there is a man named Irving, who sometimes punches you in >the face with no provocation. However, he doesn't do any more damage than >that, and if he doesn't feel like punching you in the face, he won't do it >even if you try to provoke him. > Let's say that there is a woman named Marge, who is quite gentle if you >don't provoke her, but if you do provoke her, she can do some terribly >violent things... crack open your head with a baseball bat, cut your >throat, cut off your fingers. And yes, she'll punch you in the face >sometimes, too. > Let's say that there is a man named Waverly. If you don't provoke him, > he >is gentle. What is he capable of, if provoked? Will he ever do anything >worse than punch you in the face? How often are people likely to provoke >him? > (Provocation = strategic incursion. Violent acts = undesirable results. >Hopefully you can guess who the three people are.) Mike: >That's really unfair to Marge & Waverly. WV & Margins won't elect a >Condorcet loser, in their better versions, and are good by Pareto. >It's scurrilous to imply that Marge and Waverly will hurt you more than >Irving will. >I'm talking about non-PC versions of wv & Margins. Yes, so am I. Let's assume that we're using a Smith-efficient base method, e.g. SD, SSD, beatpath, ranked pairs, or river. You've misunderstood me. I'm not saying that margins and winning votes will ever produce worse results than IRV with respect to expressed preferences. What I'm worried about is the possibility of their producing worse results than IRV with respect to sincere preferences. Obviously ranked pairs(margins) will never violate the Smith set defined by the votes, but it may well violate the sincere Smith set (as any method might). Obviously ranked pairs(margins) will never elect the last choice of 90% of the voters as expressed on the ballots, but I've described scenarios where it could elect the sincere last choice of 90% of the voters. One of my open questions is whether WV is capable of doing anything similar. This is what I mean when I say that Marge and Waverly are gentle if not provoked. If votes are sincere, then of course they'll give you a good result (in their better versions). But I want to know what kinds of bad results (with respect to sincere preferences) might happen as a result of strategy, and how likely those bad results are. So what I'm saying is that some Condorcet methods might be capable of giving worse results (with respect to sincere preferences) under strategic conditions than IRV can give under any conditions. I'm not making any conclusions about WV yet. I'm just telling you what I'm interested in studying. As for margins, I've pretty much made up my mind that it is unusable (although of course I may be proven wrong). James: >It is possible that WV performs >well in all or nearly all of the scenarios where IRV performs well. Mike: >Yes, in the case of the 2-viable-candidate case. And in 1-dimensional >spatial examples, or anytime the CW is between the others, it takes a >plurality to succeed with offensive order-reversal. So, then, so far it >looks as if wv doesn't have problems where IRV doesn't, for the most part. > This is one of the questions that I'm very interested in. When I have time, I'll see if I can cook up some counterexamples. If I can't, then I won't argue any advantage for IRV over WV. > > Sincerely, James ---- Election-methods mailing list - see http://electorama.com/em for list info