[EM] Mike: WV (was: SFC)

[James Green-Armytage]

James replying to Mike...

>I can't prove that offensive order reversal won't happen on a scale 
>sufficient to change the outcome, and you can't prove that it will. 

	Correct. However, I'd like to try to demonstrate that even if order
reversal is inclined to happen on a large scale, WV will be able to
maintain stability. If this can't be demonstrated to my satisfaction, I
will remain somewhat uneasy about WV being used on a large scale. If it
can be demonstrated, then I will do my best to promote WV for public use.

>You could argue that in the polls people didn't care enough to 
>order-reverse, and hesitated to do so because voting wasn't anonymous. 

	Yes, I would argue that. I don't think that EM polls are a good source of
evidence for predicting voter behavior in large, contentious elections, in
which there is a lot riding on the outcome and a lot of discourse
surrounding the process. 
>
>Anyway, in public elections, where the outcome matters, offensive 
>order-reversal is well-deterred. 

	That's an interesting assertion. How can we establish that burying (OOR)
can be reasonably well-deterred while people still vote enough of their
full rankings to elect a Smith set member, in all or nearly all situations
(including situations with imperfect information)?

James:
>For example, wouldn't we expect a Bush backer to rank Kerry in
>last place even if there were other candidates on the ballot whom that
>voter would like even less than Kerry if that was the only choice?
Mike:
>That will backfire badly unless the Kerry voters have ranked Bush. 

	They don't all need to rank Bush. All that is needed is enough to make
the Bush>Third defeat is greater than the Kerry>Bush defeat. If Third
doesn't have much support, then this wouldn't take too many votes.
	If there are only two strong candidates, then there can be an obvious
deterrent counterstrategy (no one should rank both of them). However, if
there are more than two serious contenders, the correct deterrent
counterstrategy does not seem so obvious.
>
>Anyway, all this misses the main point of my posting: Truncation will
>always 
>happen on a large scale, and it will be nonstrategic truncation. A whole 
>different problem from offensive strategy, on a vastly larger scale.

	I'm agreeing with you that truncation should generally happen with much
greater frequency than order reversal. And yes, I see how that makes GSFC
compliance a good quality for a method to have.
>
>But that's a good question: I believe we should all be a lot more
>particular 
>whom we vote for. And that means that in rank-balloting we should be very 
>particular whom we rank. Not so much for strategic reasons, but just
>because 
>we shouldn't rank people who don't deserve a vote. It's important to show 
>such candidates that they don't deserve a vote, and show other voters
>that 
>too.

	In methods that use both ordinal information and cardinal information
(whether a 0-100 scale or an approval cutoff), voters can clearly express
disapproval of candidates while still ranking them above candidates who
they like even less. I prefer this to truncation.
>
>But, if there were no undeserving candidates, would I suggest truncation 
>just to deter OOR? No, not unless OOR became a problem. I wouldn't bother 
>people about that problem that will probably never be a problem.

	For the purpose of our discussion, I'd like to assume that we are in an
electorate where burying (OOR) is a fairly well-justified worry, i.e. a
devious electorate, if you like. The question, then, is what general rules
people can follow to give effective deterrence while still voting out the
rankings that are important to the election.
>
>But I would always ask people to not rank people who don't deserve a
>vote, 
>such as someone like John Dean or Kerry, etc.
>
	Do you mean Howard Dean? Are you mixing up the names of the candidates on
purpose to make some kind of political statement?

James:
>By the way, we should try to solidify this a bit. Do you think that
>"always truncate after your approval cutoff when a burying strategy
>(offensive strategy) is a possibility" is good universal advice for voters
>in WV elections.
Mike:
>
>For the purpose of that question, let's assume that anti-OOR strategy is
>the 
>only reason to truncate, though I've stated a lot of other reasons to 
>truncate.

	Good, yes, let's assume that.
>
>In a devious electorate that would be one thing that could be
>recommended. 

	Okay, that's interesting.
>
>Or just don't rank anyone but your favorite if you believet that your 
>favorite is CW and that the electorate is devious. 

	To consider the more difficult scenarios, let's assume that the election
is close enough that no one knows who the CW is, i.e. that there are at
least one or two key pairwise contests that are too close to call.

>Your 
>offensive order-reversal can only steal the election from people who are 
>trying to help you.

	Not necessarily. It could be people trying to help themselves by ranking
a candidate whom I like better than their favorite candidate, but whom
they like more than a candidate whom they believe to have a significant
chance of winning. (That is, they are trying to help themselves by ranking
a compromise candidate.) 
	In some situations, it seems that voters may have a hard choice
truncating before a candidate who may be the CW, and opening the door for
a potential result-changing burying (OOR) strategy. I'd rather that voters
didn't have to face this choice.

>Or just don't rank anyone 
>whose voters are likely to attempt OOR against your favorite.

	Careful here. Let's say my sincere preferences are C>B>D>A, and I'm
considering whether I should rank B. Now, the people planning a burying
(OOR) strategy could be voters with preferences A>B>C>D (i.e. they could
vote A>B>D>C, to bury C). But by your definition, are these really the "B
voters"? No. So, should I rank B or not. If not, what general rule can we
define that would dictate that I not rank B?
>
>I emphasize those suggestions are only for a devious electorate. I
>wouldn't 
>suggest those anti-OOR strategies with our electorate unless it turns out 
>that there is an OOR problem.

	Yes, I realize that. For the sake of this discussion, let's assume that
the electorate is devious.
>
>Good, then you understand that for a pairwise-count method to not have a 
>problem when nonstrategic truncation takes place makes a very big 
>difference. 

	Yes, I do. The first voting methods example I worked with (before I had
ever heard of Condorcet's method) helped to illustrate this principle for
me.

Sincerely,
James