[EM] encourage dishonesty / range / WDS reply to robla

Warren Smith wds at math.temple.edu
Fri Aug 12 05:23:35 PDT 2005


>Robla: As I recall from discussions on this list, there are plenty of
simulations that show that IRV elections tend to resolve to two clusters
away from the political center (i.e. two-party system) thus not bucking
Duverger's Law any better than plurality.  The fact that this happens in
the real world is not a surprise, given the theoretical results.  In
fact, if I recall correctly, it's been shown that IRV tends to reward
more extreme candidates over time than plurality does.

However, the simulations show that this doesn't happen for Condorcet
methods.  They tend to resolve to the center, as do Approval and Range,
for that matter.  I can't remember who it was that ran the numbers, but
hopefully they are reading now (or someone who remembers is reading now)
and will chime in with more details on this.

Regardless, the fact that real-world IRV elections result in two-party
systems is unconvincing evidence that the same would hold true for
Condorcet methods.

--WDS:
This is a very interesting claim.  I would be very interested to learn about
whatever lies behind this claim that Condorcet does not lead to 2-party
domination while IRV does.

In IRV it often is not the case that the best strategy is to exaggerate
and rank the 2 perceived frontrunners max and min.  But nevertheless it
sometimes is the case, and a lot of members of Joe Public will do it
regardless of what the best strategy is.  If enough members of Joe
Public do this, then 3rd parties can never win under IRV.
I conjectured that this was the reason, or at least part of the reason,
that IRV countries are 2-party dominated just like plurality.

Now the above paragraph is completely true if everywhere you see the word IRV
you substitute the word "Condorcet Method", except for the caveat that there are
no Condorcet countries.  This suggests to me, but does not fully convince me,
that the exact same thing would happen under Condorcet.  This is an important question.

I in fact am the author of one theoretical study re IRV and 2-party-ization,
   http://math.temple.edu/~wds/homepage/works.html  #76.
However that study did not consider Condorcet since at the time I was not under the impression
anybody was seriously proposing it for actual use - I was trying to focus on Borda,
AV, and IRV.  If anybody reworks my arguments to put in Condorcet also,
then I will be interested to see that.



You are warned it is a bit tricky and I managed to make some stupid errors in
this study, which in fact were pointed out to me by some people on EM.  The
known errors are now fixed in the URL I just gave and the pointer-outers are
credited.
This theoretical study lived within a rather simplified model of the world
and hence maybe is not so relevant to reality, but we do what we can.



More information about the Election-Methods mailing list