[EM] Re: The "Turkey" problem and limited ranks

[Forest Simmons]

On Mon, 19 May 2003, [iso-8859-1] Kevin Venzke wrote in part:

<snip>

>
> The point of obtaining priorities is that we may be able to avoid a
> situation where the strict-ranking CW polarizes people more than the
> three-rank CW would.  Neither candidate need be a turkey.  The situation,
> quite simplified, looks like:
> 55: A>B|  (A preferred to B, but given same rank)
> 45: B|A   (B preferred to A and placed in a higher rank)
>
> These may be in the middle of everyone's ballots instead of at the
> top.  In other words, both A and B may well be turkeys.  In any
> case, I want to see B elected here over A.
>

It would be nice to have B win in this case because the choice is between
55 mildly disappointed voters and 45 devastated voters.

In this case the cruder resolution would bring victory to B in the zero
information case, for example.  But in the case of near perfect
information, the 55 majority would probably (insincerely) demote B to the
middle category so that the ballots would look like

55 A|B...
45 B|A...

allowing A to win after all.

Is there an antidote to this kind of problem?

In the past I have argued that this is an improvement over greater
resolution methods because at least in the zero information case the
results are better.

My main reservation about plain Approval and other low resolution methods
is a nagging feeling that methods that give better results in the zero
information case than in the perfect information case can be manipulated
too much by fake information.

Can anybody elucidate this?

Forest