[EM] Re: Range Voting

[Brian Olson]

On Dec 31, 2004, at 7:00 PM, MIKE OSSIPOFF wrote:

> Someone wrote:

That'd be me.

> Being able to express ratings on a ballot is more expressive than only
> being able to express rankings. I think more expressive is a good
> thing.
>
> On the back end, we could silently collapse the ratings into rankings
> by sorting the candidates, then apply Condorcet, IRV, etc.
>
>  I reply:
>
> Sillently? Secretly? Letting people think they're voting ratings, and 
> then counting them as rankings doesn't seem quite right.

Ah, um, "could" not "should" in most cases. One hypothetical reason to 
do it is if all the computers die and a hand count is needed. It might 
be the best thing to fall back and do a Condorcet hand count on ratings 
ballots.

> You continued:
>
> Straight rating summation is vulnerable to strategic voting. Perhaps in
> this study people voted honestly because it obviously didn't matter and
> so there was no incentive to vote strategically. In a real election the
> stakes would be higher.
>
>  I reply:
>
>   Do you know of a nonprobablilistsic methods that isn't vulnerable to 
> strategy, that doesn't sometimes give incentive for strategy? If so, 
> congratulations--you've proved Gibbard & Satterthwaite wrong.

I was just considering the zero-info strategy of maximizing and 
minimizing your ratings, degenerating to the Approval ballot. I think 
we can do better than that. I think "Instant Runoff Normalized Ratings" 
(IRNR) does better.

>  Seriously, all nonprobabilistic methods can have strategy incentive. 
> One chooses which  method's strategy incentive one wants. Some methods 
> minimize it. For instance, Approval & Range-Voting (also called CR), 
> never give anyone any incentive to vote someone over their favorite. 
> No other nonprobabilistic method can make that claim.

Right. I choose "maximize social utility", "be fair to all voters", 
"encourage honest voting" (also, dishonest/strategic voting decreases 
social utility because if successful some victorious minority is 
happier while the majority suffers). Sorry if those goals are too fuzzy 
to be amenable to proof.

> I also defined WDSC  in my previous post. Approval, CR, and wv 
> Condorcet meet it.

>> If a majority of all the voters prefer X to Y, then they should have 
>> a way of voting that ensures that Y won't win, without any member of 
>> that majority voting a less-liked candidate over a more-liked one.

I think IRNR has this. I think IRNR is a strict improvement on straight 
Cardinal Rating summation, except for the computational burden.

Brian Olson
http://bolson.org/