[EM] minmax is not a good public election method

[James Green-Armytage]

James replying to Mike.

>MMPO meets FBC, WDSC, and SFC. 

	(All criteria that Mike made up, I think.) If we're comparing MMPO to
beatpath(wv), then the only new compliance listed above is FBC. More on
the significance of that in a bit.

>FBC is the most basic guarantee to reassure the timid, and SFC is the
>most 
>ambitious guarantee for strategy-free voting, under the conditions for
>which 
>that can be guaranteed. Those make a good combination. Even unenhanced,
>MMPO 
>looks good in coimparison to Cardinal Pairwise.

	I wasn't really focusing on the cardinal pairwise-MMPO comparison.
Instead, I was primarily comparing MMPO to things like ER-IRV, SD(wv), and
beatpath(wv). However, since he brings it up, here's a quick summary:
* pro MMPO: 1. LNHarm, 2. zero compromising-reversal incentive (as opposed
to an extremely small compromising-reversal incentive), 3. maybe very
slightly less vulnerable to burying-compression strategies (?)
* pro CWP: 1. Smith, 2. Condorcet, 3. MMC, 4. CL, 5. less
compromising-compression incentive, 6. much less vulnerability to
burying-reversal strategies, 7. much less incentive for deterrent
counterstrategies that distort voter rankings 

James:
>If we're able to get beyond IRV, then why
>not use SD, Smith//minmax, beatpath, ranked pairs, river, etc.?
Mike:
>1. They aren't nearly as easily and simply introduced and defined to the 
>public as MMPO. They don't even come close in that regard.

	I don't think that Mike has been involved enough in advocacy to make such
a firm conclusion on this point. Few of us have, including myself of
course, but I'm not making the same kind of sweeping statements.
>
>2. They don't meet FBC. MMPO's FBC (and, when enhanced, Strong FBC) gives
>to 
>the more timid voters a guarantee that they have shown that they need.
>As I said, Australian voters often bury their favorite, because they're 
>using Plurality strategy. And no one can assure them that IRV won't make 
>them regret that they didn't do that--because IRV can make them regret
>that 
>they didn't do that.
...
>Lesser-of-2-evils. With Cardinal Pairwise I couldn't absolutely guarantee
>to 
>that person that that person could never regret not burying Nader lke
>that. 
>With MMPO I could absolutely guarantee that.
>That's a guarantee that voters need, as has been shown by experience.

	FBC. Favorite betrayal criterion. A very dramatic name. I prefer to talk
about "compromising-reversal incentive", because that tells us what's
actually at issue. (The name "favorite betrayal" in itself doesn't tell us
whether ranking a less-preferred candidate equal to one's favorite
constitutes "betrayal".) MMPO has zero compromising-reversal incentive,
which is good. However, beatpath(wv) and SD(wv) already have a very minor
compromising-reversal incentive, so this doesn't seem to be a strong
argument for MMPO over SD(wv) and beatpath(wv).
	Mike's argument above is strange. He cites compromising-reversal behavior
in methods with high compromising-reversal incentive (i.e. plurality and
nER-IRV (that's IRV with no equal rankings allowed)) in an attempt to show
that compromising-reversal will continue to be a problem even in methods
with much less compromising-reversal incentive.
	(Aside: I personally haven't seen evidence that Australia voters
frequently engage in compromising-reversal strategies, but I accept that
they should do so in theory, so I won't argue the point. Still, if anyone
is aware of such data, please let me know.)
	Why do plurality and nER-IRV have a higher compromising-reversal
incentive than SD(wv) and beatpath(wv)? For one, they *don't allow
compromising-compression*, which means that compromising-reversal is the
only available form of the compromising strategy. Also, in SD and
beatpath, there is no need to compromise at all unless there is a cycle.
	So, Mike argues that because voters frequently engage in
compromising-reversal in methods that have high compromising-reversal
incentive, we should expect voters to frequently engage in
compromising-reversal even in methods that have low compromising-reversal
incentive. This is not logical. Thus, his conclusion that FBC compliance
is a strong MMPO>beatpath(wv) argument is not well-supported.
>
>As I said, I recently observed someone voting in an Internet presidential 
>poll, and that person agreed that Nader was more honest than the
>Democrats 
>and had better polices. But that person ranked all the Democrats over
>Nader. 

	Is one person in an internet poll a sufficient sample size to make
sweeping conclusions about voter strategy in Condorcet elections? Mike
doesn't even say what the tally method was. Anyway, I wouldn't expect
voter strategy in internet polls to be a good indicator of voter strategy
in public elections, because (1) the outcome is totally unimportant, and
(2) the electorate is more technically inclined.
>
>Aside from all that, though LNC isn't important to me, the fact that MMPO 
>meets it will help in discussions with IRVists.

	It meets LNHarm, but it still doesn't meet LNHelp. IRV supporters won't
be impressed. LNHarm compliance by itself does little or nothing to
decrease the method's vulnerability to burying strategies. 
>
James:
>There is no need to implement MMPO instead of SD. SD is just as easy to
>explain
Mike:
>I consider SD a good proposal. But not as good as MMPO. 

	Okay, maybe MMPO is easier to explain than SD(wv). But I expect that the
aesthetic impact of its MMC failure would be a substantial impediment to
public selling. When it comes to merit, I believe that SD(wv) is clearly
ahead, and that SD(wv) is much less likely than minmax to seriously
embarrass pairwise count methods in general.

Sincerely,
James Green-Armytage
http://fc.antioch.edu/~james_green-armytage/voting.htm

>
>