[EM] Manipulability stats for (some) poll methods

[Kristofer Munsterhjelm]

On 2024-05-16 02:33, robert bristow-johnson wrote:
> 
> 
>> On 05/15/2024 2:30 PM EDT Kristofer Munsterhjelm <km_elmet at t-online.de> wrote:
>>
>>   
>> On 2024-05-15 17:04, Filip Ejlak wrote:
>>> śr., 15 maj 2024, 10:47 użytkownik Michael Ossipoff
>>> <email9648742 at gmail.com <mailto:email9648742 at gmail.com>> napisał:
>>>
>>>>      Yes, then, as you suggest, “manipulability” doesn’t tell us anything
>>>>      of interest. I agree.
>>>
>>>>      Then how much do those manipulability numbers mean, in regards to
>>>>      the strategic merit of the methods. Nothing?
>>>
>>>
>>> I can't agree at all. IMO the primary goal of a good voting method is to
>>> make voters not regret voting honestly.
> 
> Yes.  But it also has to be perceived by Joe Sixpack as fair and "right".
> 
> The method needs to be principled and these principles few in number and concise in expression.

If it weren't for this part, my fpA-max fpC method would've been quite 
good. Its strategy resistance is around that of Condorcet,TTR, but it's 
monotone. But alas, its scoring function is pretty hard to understand 
because it was found by a computer search.

It might be easier to deal with complexity if/when better methods gain a 
reputation for being good. Some countries use very complex methods, but 
it works because the people generally trusts them. E.g. Meek in New 
Zealand, or the greedy top-up seat algorithm used here in Norway for 
national proportionality (or Pukelsheim's in Switzerland for that 
matter, which, IMHO, is better than the greedy method).

>>> While it's useful to be able to
>>> use a defensive strategy after analysing expected poll outcomes,
>>> frontrunners etc., the best voting method would be the one that does not
>>> create the need to take these things into the account at all.
>>> Chances of being able to vote honestly, with no strategic burden to
>>> bear. That's what the manipulability numbers are about.
>>
>> Thank you for saying that much more succinctly than I did.
>>
> 
> That says something.  (I have credited Kristofer before for succinct analytical judgements and statements.)
> 
>> Although I would say that winner quality given honesty also matters :-)
>> At least to avoid the kind of outcomes that lead people to repeal the
>> method.
> 
> I dunno how to measure, in a non-partisan manner, "winner quality".

In a real election, we can't. But the spatial model I use for the 
manipulability test gives each voter a quality measure of each 
candidate. So we can just use those utilities directly.

See 
https://en.wikipedia.org/w/index.php?title=Social_utility_efficiency&oldid=1222909293 
for more information.

James Green-Armytage has also done some investigations into the 
trade-off between strategy resistance and utilitarian efficiency. 
https://www.jamesgreenarmytage.com/strategy-utility.pdf Figure 1, e.g.

Though for the models derived from real sources, we should keep in mind 
that if the voting method gets better at protecting candidates, the 
candidate landscape may change.

-km