[EM] Why care about later-no-harm or prohibiting candidate burial?

[Kristofer Munsterhjelm]

Jonathan Lundell wrote:
> On Feb 21, 2011, at 4:06 PM, Kristofer Munsterhjelm wrote:
>> There might also be a trade-off. If you have a certain election
>> where a candidate wins, that election might be made up of honest
>> ballots (in which case it's good that the candidate wins), or of
>> strategic ballots (in which a metod that resists strategy should
>> elect another candidate); but the method can't know which is the
>> case because all it's got are the ballots themselves, free of any
>> context. Looking at the structure of election methods may let us
>> know more about where that trade-off actually resides, though, or
>> in simpler terms: how strategy-resistant a method can be and still
>> be a good method.
> 
> On a related note, one of the problems with burial-prone methods is
> that burial is a simple, intuitive and attractive strategy that can
> be easily employed by relatively naive voters. It's a perverse
> incentive precisely because of KM's point above: the method is faced
> with an unknown mix of sincere and strategic ballots. GIGO.
> 
> By contrast, non-monotonicity is relatively benign (in this sense),
> in that it's very hard for a voter to come up with a practical voting
> strategy to take advantage of it. Among other things, voters don't
> have enough information about the detailed preferences (and
> strategies) of the other voters to successfully strategize
> themselves; there's little to be gained by voting other than
> sincerely.

I'd say they are different sorts of properties. Burial resistance is 
something that makes it less tempting to try a certain strategy, 
(preferably) because it won't work, because it won't work any better 
than less distorting strategies, or (less preferably) because it will 
almost always backfire. There may be other strategy resistance 
properties, like compromising resistance or the favorite betrayal criterion.

Monotonicity, however, is a property whose failure implies that the 
method is confused about itself and that there's something suspect about 
its value system. Raising a candidate shouldn't make him lose, because 
all other things equal, if a candidate gets a a higher rank, that should 
be evidence he's doing better. To put it another way, if you have a pair 
of ballot sets, the latter of which raises a candidate in the former, 
and the method decides to award that candidate the win in the first case 
but not in the second, that seems like implying the method is wrong 
about one of the two ballot sets. Either the candidate should not have 
won in the first case, or he should also have won in the second.
I would put criteria like Plurality and reversal symmetry in this 
category, too -- or Pareto, for that matter.