Random-fill, majority, cancellation

Mike Ositoff ntk at netcom.com
Wed Oct 14 19:26:54 PDT 1998

Just a few more things to add to my reply about "random-fill":

Even if the insincere extension is concerted rather than radom,
it can't cause VA to violate majority wishes by avoidably
electing a candidate over whom a majority have expressed preference
for a particular other candidate. That's true whether "majority"
is defined as Blake defines it (majority of those expressing an
opinion between a particular pair) or as the term is usually
used (more than half of all the voters taking part in the
multi-alternative choice).

The only exception is where Schulze (and almost as often)
SD will let that happen in a subcycle in order to avoid the
clone problem, on the reasonable asumption that what happens
in a subcycle is less important.

But Margins, though it, like any Condorcet Criterion method,
won't violate majority rule as Blake defines it, fails majority
rule as it's usually defined. If a majority indicate that they
would rather have A than B, Margins will  sometimes pick B.


To finish what I was saying about the sincerity standard,
I talked about Blakes' criticism of VA, involving random-fill,
but I didn't talk about Margins' failures, though I've
mentioned it before.

It would be nice to not be forced by defensive strategy need,
to be so insincere as to vote a less-liked alternative over
a more-liked one. That happens in Margins, but not in VA
or Approval.

It would be even nicer not to have to vote a less-liked alternatiave
equal to a more-liked one. That happens in Margins anytime
lots of people truncate, as experience suggests will happen 
quite a bit in public elections. That kind of forced insincerity
is necessary in Margins, but not in VA. 


Cancellation: Let's not skip by the fact that if the
insincere extension is really random, as befits the name
"random-fill", then in a public-size election, it will cancel
out, and won't affect who beats whom, and won't affect a CW's


More information about the Election-Methods mailing list