[EM] Re: Condorcet's strategy problem, ICA

Kevin Venzke stepjak at yahoo.fr
Sun Sep 18 14:00:14 PDT 2005

```Rob,

--- Rob Lanphier <robla at robla.net> a écrit :
> > ICA satisfies this:
> > "If there is one candidate who doesn't lose to any others after certain losses
> > are disregarded (due to being reversible by voters using equal-top ranking),
> > he must win."
> >
> > ICA scales Condorcet back only as far as is necessary to satisfy FBC.
>
> Hi Kevin,
>
> Let's take a look at the example posted on the ICA page on Electowiki
> ( http://wiki.electorama.com/wiki/ICA ) (probably your example):
>
> 40 A>B
> 35 A=B
> 25 B
>
> The head-to-head pairwise score in this election is 40-25 (35 abstain).
> I don't like the idea of inferring that all of these voters "approved"
> of B to declare B the winner.

In my opinion, you can't get away from this anyway. The best WV methods
still have compression and truncation incentive.

>  Those 40 voters would rightfully expect
> that "A>B" be the same as saying "A",

No Condorcet method satisfies Later-no-harm. It is always possible that
by listing an additional preference, you will unnecessarily hand the election
to that preference.

>and would feel ripped off that
> their 40 votes actually counted both for and against A in the A-B
> comparison.

No, they didn't. The 40 votes counted as normal in the pairwise comparison.
What happened was that the method went to the "second stage," which is
approval. A did not have adequate pairwise votes to decisively beat B in
the first stage.

>  They might also complain one person, one vote has been
> violated, since their 40 votes for A over B don't count for as much as
> the 25 votes for B over A.  Actually, the 40 votes don't count at all.

In the approval stage, no, they don't count more towards A than B. In the
pairwise stage, the 40 A>B votes counted more than the 25 B(>A) votes. Just
not more than the 25 B(>A) votes plus the 35 A=B votes.

The alternative, which you can see in the other example on that page,
is that voters will have to dump their favorite to secure a compromise
to avoid electing their last choice.

It seems to me that this is a worse problem than having to truncate a
compromise so that you can get your favorite instead of the compromise.
At least in this case, the ballot information is relatively sincere.

> My understanding based on your previous descriptions of ICA is that
> explicit approval is problematic in ICA, since there's a lot of gaming
> that can be done below the cut line.

This is the difference:
Implicit approval: Burying strategy is always too risky.
Explicit approval: Burying strategy only backfires when the other voters
try to use burying strategy also.

The latter would probably lead to some equilibrium that we could live with.
If it's really easy to bury, then everyone will try, and then it will
always backfire, so fewer people will try.

My feeling is: Why not throw out burying strategy altogether?

> All things considered, I'm not sure how this doesn't count as "favorite
> betrayal".  A voters gave the election away by "approving" B.  Their
> favorite (A) lost the election as the result of a lower ranking.

"Favorite betrayal" means lower ranking *of your favorite*, not higher
ranking of a *different* candidate.

The issue here is truncation incentive, which WV methods have also.

Example:

49 A
24 B
27 C>B

B wins. Now the B voters add another preference:

49 A
24 B>C
27 C>B

C wins. Has "one person, one vote" been violated? If you say no, I guess
I haven't understood what you mean by "one person, one vote."

Kevin Venzke

___________________________________________________________________________
Appel audio GRATUIT partout dans le monde avec le nouveau Yahoo! Messenger
Téléchargez cette version sur http://fr.messenger.yahoo.com

```