[EM] proposal: weighted pairwise comparison

[Dgamble997 at aol.com]

Hello James

You wrote regarding the examples:

45 A 100 > B 70 > C 0
10 B 100 > A 70 > C 0
5  B 100 > C 70> A 0
40 C 100 > B 70 > A 0

45 A 100 > B 10 > C 0
10 B 100 > A 90 > C 0
5  B 100 > C 90 > A 0
40 C 100 > B 10 > A 0

>Yes, that is correct. In both of those examples, B is a Condorcet winner,
>and is therefore elected by my method. Hence yours is a criticism of
>Condorcet methods in general, rather than my method in particular. This
>criticism, i.e. the possibility of a low-utility Condorcet winner, has
>been debated before.

and continued:

>So, the standard pro-Condorcet counter-argument is to say that if we were
>doing an IRV count that was repeated over and over again, you would
>probably get to the point pretty quickly where B would win every time.
>That is, in the first round, perhaps, people would give their sincere
>rankings, and A would win the first IRV tally. But in subsequent rounds,
>the C>B>A voters would be likely to realize that they could get a better
>result for themselves (B instead of A) by voting B in first place. So C
>would be eliminated first and B would win the tally. At this point, it
>seems to me like the situation would be largely in equilibrium, in that
>there wouldn't be a way for any of the voting blocs to get an immediately
>preferable result.

In the first example where B is a high utility generally preferred option 
this quite probably would happen -C voters could get a much better result by 
abandoning C in favour of B.

In the second example B is barely preferable to A. C voters are left in the 
unenviable position of abandoning their favourite C and supporting somebody bad 
to prevent the election of somebody slightly worse. I don't think they'd do 
it. They'd continue supporting C since neither A or B are acceptable 
alternatives.

Collecting ranking and rating data and then only considering the rating data 
in the event of a Condorcet cycle seems a potentially perilous thing to do. 
People can look at the ratings data after the event and say well B was a 
generally preferred choice but 85% of people didn't like him/her. Related criticisms 
can be made of all ranked ballot methods. What your method seems to be doing 
is gathering additional information and then ignoring it in all but certain 
selected circumstances.

David Gamble







-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20040610/92c432a4/attachment.htm>