[EM] Condorcet and other authors on Condorcet (and how does range voting fit in?)

[Abd ul-Rahman Lomax]

At 08:50 PM 5/17/2010, Kevin Venzke wrote:
>--- En date de : Lun 17.5.10, Abd ul-Rahman 
>Lomax <abd at lomaxdesign.com> a écrit :
>
> > Here is the issue. Is the "pairwise contest" some *other
> > election*, where the two candidates face off against each
> > other? But this is a completely different election! It's a
> > theoretical construct, not an actual procedure to follow
> > with ballots.
>
>Are you asking me or are you arguing that multiple definitions are
>possible? I don't dispute the latter.

It was a rhetorical question. Then I gave a point 
to be considered. You can still make different 
definitions, as long as it is understood what this means.

>If you're asking me, the answer is that "pairwise contest" has nothing
>to do with two elections, and only applies to rankings or things that
>can be interpreted as rankings. (That's a definition, not an attempt to
>justify the concept.)

In other words, there is a "virtual pairwise 
contest" in which vote-for-one full votes are 
cast, based on preferences shown in the original 
ballots. Just so it's understood that in a *real* 
pairwise contest, the same results might well not 
be obtained, and especially if it is a real 
contest that is, say, a runoff election between 
the Condorcet winner and the other candidate.

It's been said that Condorcet methods are 
"instant round robin." The interpretation being 
proposed above, however, causes the voters in 
this virtual election to vote *differently* than 
they voted in the original election. In the 
"instant round robin" virtual election, their 
votes are what is called, when cast in the 
original Range election, "exaggerated votes," and 
voters who vote this way are often denigrated as "selfish."

>You go on to argue that "pairwise contest" may not be a very relevant
>idea. That's a reasonable argument. I don't want to change my definitions
>as I change my opinions though. If necessary I'd rather coin new terms.

I've used pairwise contest to mean the same 
thing, Kevin, I'm just pointing out that the 
concept was a bit fuzzy, and that Smith is quite 
likely correct: Condorcet himself might have 
agreed that Range satisfies the criterion named 
after him, there is some evidence for that.

The pure preferential interpretation, to be sure, 
still has a utility. My own view is that almost 
always this Condorcet winner is the best choice, 
but when the exceptions involve choosing, 
instead, a Range winner, there are two 
possibilities: the Range winner actually does 
have higher social utility, which argues for 
choosing this one, or the Range votes are somehow 
distorted, with two reasonably possible sources: 
normalization error (where a voter votes full 
strength no matter what the absolute preference 
strength might be), and poor assessment of the 
strategic situation, i.e., the identity of the 
frontrunners. A good method would consider these 
errors as possible, and, when needed, test them with a runoff.

Unless experience shows that an actual error in 
the result, as defined by the primary method, is 
so low in probability or in magnitude that it is 
not worth the trouble. To know the truth about 
this, in real elections, will require collecting 
real range data in elections, and real experience 
in runoffs. Simulations can be valuable, but real 
data overrules theory; the problem is, generally, 
that we don't have real data. I'm pointing out 
that Bucklin using a Range ballot to drive the 
"voting machine" would very likely collect real data.

It's clear, though, that the assumption that the 
Condorcet winner is the best winner is based on a 
set of assumptions about how voting works that 
don't necessarily correspond to reality. Almost 
nobody disagrees that, with commensurable 
utilities, sincerely disclosed, the Range winner 
would be best, so the question boils down to 
whether or not *actually disclosed utilities* 
will provide any usable preference strength 
information, sufficiently related to true 
commensurable utilities to improve results.

The concept of a majority preference being 
overridden by a stronger preference of a minority 
is tricky. In small groups, it is very clear that 
social choices get made this way, within limits. 
Small groups don't choose a common pizza by 
straight numbers, not without looking for a more 
widely-acceptable choice. Not if they are 
functional, anyway! Same with restaurants for a 
group to dine at. The norm isn't the choice with 
the most group members having it as favorite, but 
generally the one with the least number of 
serious objections. In the FairVote restaurant 
choice election, recently discussed by them, the 
kicker, hardly mentioned by them, is that the 
choice wasn't objected to, there was a defacto 
"acceptance vote." And since the voters accepted 
a choice that they had voted as if it was awful, 
clearly their votes were distorted, and clearly 
their *real* intention was not to "win," but to 
have a nice lunch that all could enjoy. And, 
apparently, they enjoyed it. And if they hadn't, they'd have been fired.

Well, I don't know that the last is true, and 
probably it wasn't but .... this is the problem 
with conducting test elections like this with 
your boss or "leader" being one of the voters and 
a bit "involved" with the idea that one of the 
methods is the best. I don't recall. How did they 
decide which result to accept?