# [EM] Independence from Clones Criterion

MIKE OSSIPOFF nkklrp at hotmail.com
Thu May 26 21:29:04 PDT 2005

```Yesterday I tried to briefly summarize Independence from Clones, and I
didn't say it well. Let me start over: It should take nothilng other than
sincere voting by everyone to ensure that no faction can gain or lose
advantage by running several identical candidates.

More completely, most would probably agree that this is the preference
definition of a clone set:

Set S is a clone set iff, for any candidate X outside S, everyone who
prefers X to someone in S prefers X to everyone in S; and everyone who
prefers someone in S to X prefers everyone in S to X; and everyone who is
indifferent between X and someone in S is indifferent between X and everyone
in S. A voter is indifferent between two candidates if s/he doesn't have a
preference between them.

[end of preference definition of a clone set]

Of course nothing in that definition says that a clone set must have more
than one member. So a set containing one individual candidate is a clone
set.

Independence from Clones Criterion (ICC):

If everyone votes sincerely, then adding a new candidate to a clone set, so
that that new candidate is an additional member of that clone set,
shouldn't change the matter of whether or not the winner comes from that
clone set, if before  that candidate-addition there was only one winner, and
if after the candidate-addition there is only one winner.

[end of suggested definition of ICC]

There are votes-only ICC definitions met by Plurality. Like other such
votes-only definitions, they're a great way to show a way in which
BeatpathWinner/CSSD and RP are as good as Plurality. The same remains true,
nothing is really changed, when such criteria are written to explicitly say
that nonrank methods fail, or that the criterion doesn't apply to nonrank
methods.

The whole purpose of criteria is for comparing methods, and the best
criteria compare all methods. When you want to replace Plurality with a
better voting system, wouldn't it be nice to have criteria that compare it
to Plurality? ...other than by saying "Nonrank methods don't pass because I
say so"?

The above ICC definition is probably the briefest and simplest universally
applicable ICC definition that is consistent with what we expect and intend
from ICC, and which doesn't mention methods' rules.

I was saying before that I agree with the desirability of ICC, but that I
didn't consider it important enough to spoil PC. Likewse, now I don't
consider ICC imporant enough to spoil MMPO.

There are other lesser-of-2-evils guarantees that don't apply only to
identical candidates with everyone voting sincerely. Some of the best of
those guarantees are available with MMPO.

Mike Ossipoff

_________________________________________________________________