[EM] Election Day causes stress

[Kristofer Munsterhjelm]

Dave Ketchum wrote:
> Really a trivial question, and brings us back to looking closer at 
> Condorcet.
> 
> IRV also does ranking, but has a different order of looking at ballots:
>       Vote for minor candidate?  Likely discarded when seen, thus of 
> little effect.
>       Vote for third party before major?  May prevent major being seen 
> when most effective.
>       Vote for major before third party?  This could be the day this 
> third party needs seeing quicker to win.

This seems to be a more general covering problem, now that I think of 
it. You have it in IRV, where voting for major before third "covers" the 
third party so it's only seen later if at all - and you have it in 
strategic Approval, where voting for both major and third means third 
can't overtake major - at least not by your vote alone.

The difference is that in Approval, you can take it into consideration 
when you decide upon your strategy, but in IRV, it's much less opaque. I 
hadn't really thought of these problems to be similar before, though...

> Ranking methods only require deciding which candidate is better, while 
> range also asks how much and for voter to be understood when expressing 
> that "much".

If I remember correctly, the Majority Judgement paper makes that point 
when it says that for the method to work well, the voters should have a 
common conception of what "Poor" or "Good" means in terms of rating. The 
authors suggest using named ratings (like Poor and Good, or A/B/C/D/E/F) 
instead of numbers to better make use of these common conceptions or 
standards when they do exist.

We could thus consider three kinds of ballots: first, the labeled rated 
ballot, like MJ, where the question is "according to a common standard, 
what grade would you give X, Y, and Z?"; second, a numeric rated ballot, 
where the question is "how many points do you give X, Y, and Z, 
considering the tradeoffs between awarding points to each of the 
candidates?"; and third, a ranked ballot that just asks "Who do you want 
to win? Who do you want to win if you can't have the first one? Who do 
you want to win if you can't have the first two?" etc.

The advantage of considering each ballot type to be answers to a certain 
type of question is that one can ask how well a method behaves with 
respect to the ballots. If each ballot type only is defined in context 
of the method with which it is associated, then every method is perfect 
from its own point of view. I think I've seen this kind of reasoning 
elsewhere, e.g. that IRV is a good method because the ballots specify 
contingency programs, not "who would I prefer to have win" type answers, 
and therefore, IRV faithfully follows the programs given by the voters, 
giving results entirely consistent with the programs.

So there's more complexity to the apparently trivial question than meets 
the eye!