The problem with "utility" (Re: [EM] Re: Election-methods Digest, Vol 15, Issue 1)

[Abd ulRahman Lomax]

At 08:18 PM 9/2/2005, Rob Lanphier wrote:
>On Fri, 2005-09-02 at 15:17 -0400, Abd ulRahman Lomax wrote:
> > At 04:07 PM 9/1/2005, Rob Lanphier wrote:
> >
> > >[...] I'd just as soon not favor a system that favors those prone
> > >to hyperbole. [...]

> > This is the core problem with higher-granularity Range. It is [...]
> > avoided with [...] Approval [....]
>
>You made the case here, very well, I might add:
>http://lists.electorama.com/pipermail/election-methods-electorama.com/2005-August/016811.html 
>[...]
>
>In response to a point you made there:
> > So to go the distance, I'd suggest that Range ballots be analyzed
> > pairwise, and that they be normalized within the pairs.... I have not
> > considered all the implications, for sure.
>
>James Green-Armytage has detailed this pretty extensively:
>http://wiki.electorama.com/wiki/Cardinal_pairwise

This is not what I was suggesting. I was suggesting that *each* 
ballot be normalized such that the lowest rating be 0 and the highest 
1. So, in Range 10 (ratings from 1 to 9 explicit and 0 is blank -- 
some details of Range remain controversial), if a voter rates A, B, 
and C as 2, 4, 6, and leaves no blanks, then the votes would be 
analyzed as 0, 4.5, 9.

(It is my opinion that raw ballot data should always be available. 
Ballots are used to make statements other than those that are 
effective in elections. A voter who votes 2, 4, 6 as I mentioned is 
likely stating that, while C is favored, the voter is not fully 
satisfied with C. But "fully satisfied" is a very personal decision. 
Some people may, quite sensibly, recognize that no candidate is 
perfect and may therefore rate the best candidate with less than the 
highest possible rating. And the effect of that, without 
normalization, is to give more weight to the votes of those who 
exaggerate, which is undesirable.)

(It is also my opinion that if higher-order range is used, the top 
two ratings should be considered rating-equivalent, with the actual 
top being "favorite" and the second highest being equally rated for 
aggregative purpose. This would allow the supporter of a relatively 
unpopular candidate to indicate the fact, while not diluting the vote 
for an approved frontrunner, thus making lower-granularity Range less 
vulnerable to the need for strategic voting; this avoids the 
Favorite-Betrayal problem. In high-granularity Range, 
Favorite-Betrayal is less of a problem, but it is quite likely that 
Range, as practically implemented in large public elections, would be 
limited to ten or fewer options; and, without a method of indicating 
Favorite, this would force the voter to compromise: either rate the 
favorite equally with the approved front-runner, which loses the 
expression of favorite and is thus a form of favorite betrayal, or 
lose 11% of rating strength if voting, say, 9 for the favorite and 8 
for the approved frontrunner.

This procedure might also reduce the need to normalize, for most 
voters might well choose a "favorite," which would then create a 
maximum rating automatically. Voters who really can't choose could 
either choose two favorites (equivalent to Approval Voting), or rate 
the candidates equally at lower ratings, with no weakening if the 
next-topmost rating is used.

Allowing the expression of a Favorite would also solve the problem of 
the distribution of public campaign money, which might only go to the 
parties of candidates designated as Favorite. (And, in this case, if 
a voter votes for two favorites, the money would be split.)

It is entirely possible, if this is done, that the runner-up earns 
more campaign finance money than the winner.... which might be a 
salutary effect.