It came back to me yesterday what I -- and some others -- have seen
as the biggest problem with Range voting, and also a solution that I
think I also expressed somewhere. But I tend to write way too much
and sometimes I think good ideas have been buried in fluff.
Range rewards those who exaggerate, it may weaken the vote of those
who do not use the full range of ratings.
Consider two candidates within an election, A and B, and two voters.
Voter 1 rates A and B as 99 and 20. One might analyze this vote as
indicating that the voter clearly approves of A and considers B as
possibly better than Genghis Khan.
Voter 2 rates A and B as 0 and 99.
In Range, voter 2's ratings outweigh those of voter 1.
This is essentially a kind of violation of one-person, one-vote.
It is fixable by using granularity-2 range, i.e. Approval, or
partially by normalizing the votes. I think basic normalization
should be done regardless in Range, that is, the maximum vote cast
should be normalized to 1 and the minimum vote cast to 0. However, in
an election with more than two candidates, this would not solve the
problem, because there might be a third candidate who was truly
awful, and Voter 1 might rate that candidate as zero, leaving the
same problem in the pairwise race between A and B.
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. But if this problem cannot
be solved, it could be fatal for higher granularity Range.
There is another problem which has been noted: if Range averages only
expressed ratings, a dark horse candidate may win by being rated at
max by a relatively few voters. Again, one election outcome like this
(unless the majority was pleasantly surprised) could kill Range
forever (in public elections).
The public is indeed, as Warren points out, familiar with Range, but
the common application is in rating athletes or contestants, as in
the Olympics, and the rating is done by supposedly neutral judges who
do try to follow some kind of common standard. Such judges are not
supposed to have a "favorite" whose rankings they would highly
elevate above the normal for the actual performance. I'm sure that
there is some level of violation of this, judges being human, but
nothing like what would be routine in public elections.
So Range should total ratings, not average only expressed ratings.
And if ratings are normalized, the system really does resemble
Approval with relatively harmless intermediate options. I kind of
like Mr. Ossipoff's suggestion of votes of -1, 0, and +1. I think
people would very clearly understand that. (And, again, I have not
examined all the implications; for example, should granularity-4
range (Ossipoff's suggestion is that, because there is also the "no
rating" option) be normalized?)
But another option is granularity three, which could be labelled -1
and 1. No vote would be counted as zero.
----
Election-methods mailing list - see http://electorama.com/em for list info
