[EM] Chain Climbing

[Kristofer Munsterhjelm]

On 04/26/2014 01:32 AM, Forest Simmons wrote:

> The approval cutoffs allow voters greater ability to make their will
> known.  In some cases they help to discern strong preferences among
> ranked clones that would otherwise be impossible to express on mere
> ordinal ballots.  Which is a stronger refutation of putative clone
> status?  A new candidate D being ranked between C and C' on a few
> ballots. or an approval cutoff being placed between them on the same
> ballots?
>
>
> Where do we go from here?

The next obvious step seems to be ratings or grades. Grades could be 
considered either quantized ratings or "the ability to skip a rank", the 
latter being more of a Bucklin interpretation.

But if we go to grades or ratings, that raises the question of what the 
ratings *mean*. And if they mean whatever the voter wants it to mean, 
i.e. they're neutral devices that can be employed in whatever way the 
voter sees fit to maximize the impact of his vote, then we're right back 
to plain old Approval (by detour of Range). Approval could possibly be 
improved by runoffs, but absent that, it's pretty much a stable local 
optimum: given the assumption of votes as impact-maximizing devices, and 
of every voter as a strategist in that sense, any "reasonable" extension 
of Approval itself will simply reduce to it.

If, on the other hand, we suppose the voters will compare candidates to 
an independent scale, then we get MJ.

Perhaps we could get further by creating randomized strategy proof 
methods for each model. Then a honest vote with respect to a certain 
model is what the voter would put in the model's corresponding 
randomized method. The strategy-proof method for Plurality-style voting 
is Random Favorite, and the strategy-proof method for ordinary ranked 
voting is Random Pair (but is that with or without truncation?).

Maybe the strategy-proof method for certain ideas of what rating means 
is Hay. But what's the strategy-proof method for ranked voting with 
approval cutoffs? I don't know.

Beyond this, I would again suggest that one sets up experiments to test 
to what degree voters would strategize under ranked voting. Since 
ordinary ranked voting doesn't let us distinguish a honest situation 
from a strategic one by the ballot set alone, find out if it is going to 
be a problem in reality. There's no need to overengineer a method and 
add more armor than it needs. OTOH, it should be resilient enough to 
weaken strategy so that the number of strategizers decreases rather than 
increases with time. Where this balance should be doesn't seem to be 
something we can deduce by logic alone.

Jameson said he was doing something data-gathering of that sort. Did he 
get useful results?