I am more convinced than ever that the best way to measure defeat strength in 
Beatpath (aka CSSD) is 
by giving the covering relation the highest priority, and where neither 
alternative covers the other, falling 
back on winning votes.  This is a natural way to extend the covering relation 
(which is a partial order) to a 
total ordering of the candidates.

So if A covers B, then that beatpath with one link is stronger than any 
beatpath that B can have to A.

If neither A nor B covers the other, then all beatpaths in both directions have 
links (defeats) that are not 
coverings (because the covering relation is transitive).  The strength of a 
beatpath that has at least one 
link that is not part of the covering relation is the wv strength of the 
weakest such link.

If we say that A>>B whenever the strongest (in the above sense) beatpath from A 
to B is stronger than 
any beatpath from B to A, then the >> relation is a total order barring exact 
ties at weakest links. If 
range style ballots are used to infer the voter rankings, then these ties can 
be resolved without use of 
randomness.

Among the tied candidates give preference to the one with positive ratings on 
the greatest number of 
ballots.

If the tie is still not resolved, break it by giving preference to the still 
tied candidate rated above one on 
the greatest number of ballots.  If still tied, give preference to the still 
tied candidate rated above two on 
the greatest number of ballots, etc.

If we use the above method of defining defeat strength, we can still use the 
CSSD algorithm to find the 
Beatpath winner.  The method retains its clone free property, and its 
compliance with the Monotonicity 
Criterion.  Beyond that it always elects an uncovered alternative.
----
Election-Methods mailing list - see http://electorama.com/em for list info

Reply via email to