[EM] Condorcet vs unspecified point system

[MIKE OSSIPOFF]

Joe's objection to Condorcet, which I've quoted below, is something that's 
been much discussed here, on numerous occasions, since the
inception of the EM list. We should have made a FAQ about it at the
outset.

Yes Condorcet doesn't record how much more you like one candidate
than another. Yes, it's neglecting some information. Is that always
bad? Do we really want to use all the information available? For
instance, do we want to use the information about whether you'd
rather fully vote Favorite over Middle, or Middle over Worst (because
you can't do both)? Saying you want express and use that information
is like saying that you want to express & use the information about
whether you'd rather give your wallet to a mugger, or face the consequence 
of refusing.

I suggest that it's better if you don't have to express the information
about whether you'd rather fully vote Favorite over Middle or Middle
over Worst--because you can do both.

You see, that "information" is about how you deal with the lesser-of-2-evils 
dilemma. We don't want that dilemma. That's why we prefer rank
balloting. That's why we count it by pairwise-count. And that's why
we solve circular ties by Condorcet's method.

As we always reply when this issue comes up, as it always keeps doing,
in an ideal society, where everyone is completely honest and altruistic,
and everyone's only goal in voting is to sincerely express his/her ratings, 
to help the system make the best decision about the greatest
good for the greatest number, the highest social utility--and where
no one is out to optimize the outcome for himself/herself--then sure,
Cardinal Ratings would be the best voting system. We don't live in that
society. Cardinal Ratings would work fine in our actual society, but
it wouldn't work as Joe would like--it wouldn't measure people's sincere
quantitative preferences. Joe is right to like Cardinal Ratings, but
it's for an incorrect reason.


Yes I know, other point systems have been proposed that would moderate
the strategy of Cardinal Ratings in some way. But none of them
can claim the criteria compliances that Condorcet has. And some of them
even lose those that Cardinal Ratings can claim.

That being said, of course I don't object to Cardinal Ratings, because
Cardinal ratings is strategically equivalent to Approval. Approval,
in fact, meets a criterion that Condorcet doesn't strictly meet: FBC.
Under no circumstances could anyone ever have incentive to vote someone
over their favorite. Condorcet's violation of FBC isn't common or
important or blatant, it seems to me.

But because Approval is an excellent method, so is Cardinal Ratings.
But it certainly isn't true that we want Cardinal Ratings because it
measures all of our preferences quantitatively. We only want it because
it's like Approval.



>As a matter of fact I have no enthusiasm for Condorcet - or for any other
>pairwise comparison methodology which simply looks at qualitative
>preferences but ignores their strengths (i.e., preference is taken as one 
>of
>just three alternatives:  positive, negative or zero, and it doesn't matter
>how strong a positive or how strong a negative).
>
>To be sure, one can do - and get lots of professional credit for - some
>arcane math by investigating various qualitative pairwise comparison 
>schemes
>and issues.

>For me, Condorcet (as most methods treated here in the EM-list) has much 
>the
>same problems noted specifically for Irv about six weeks ago. Namely, the
>method gratuitously (indeed, from my viewpoint, stupidly) disregards
>preference info given or readily retrievable from the voter.  For its part,
>at each stage, Irv disregards all voter preferences other than between
>(currently) top and lower places.  For its part Condorcet ignores different
>strengths of preference.  All preferences are classed as either positive
>(never mind with what strength), zero, or negative (again, never mind
>strength).
>
>For me, Condorcet breaks down at the very beginning, in the simplest case, 
>a
>2-candidate contest between A and B, where Condorcet Winner A beats B over 
>a
>majority of the voters by very shallow preference, but B beats A over the
>remaining voters by heavy preference.  E.G.,  suppose (on a grade scale of
>100) the voters grade the candidates as follows:
>
>    55% of voters:  A(80) B(75)
>    45% of voters:  A(40) B(80)
>
>To me it is clear that the deserving winner is B, the candidate with a
>higher grade average (which some people equate exactly, and I roughly, to
>higher 'social utility').  Of course, one might argue (in my view,
>correctly) that - for purposes of true social utility - a grade of 40 is
>surely worth more than just exactly half of a grade of 80, etc.  However,
>even if grades were in some way convex-transformed to reflect this view,
>before being averaged, B would still be the clear winner.
>
>In order to 'justify' Condorcet (or another qualitative-only preference or
>ranking method), its partisans could insist that the election method allow
>only qualitative preferences to be expressed on the ballot.  Their one
>credible material excuse would be that the computers requisite for more
>quantitative input and tabulation would be too expensive or slow or
>unreliable.  Nowadays (and prospectively) this excuse is utterly mistaken.
>However, if one sincerely believed it, one should be more consistent and
>demand use of Approval (i.e. just pass-fail grading of each candidate):
>compared with Approval, such methods as Condorcet (not to mention Irv) are
>far more complex and demanding computationally.
>
>High-res grading provides a direct way to express quantitative preference
>strengths, and (through grade averaging) provides readily tabulated and
>consistent scoring of candidates.  By comparison, methods like Condorcet or
>Irv both disregard voter information (on preference strengths) and are
>computationally more intensive.
>High-res grading makes so much sense that an Irvie cultist to whom I
>explained it could think of only one rejoinder: it's too 'corny'.
>
>Yes, direct grading of candidates is 'corny', and makes a far less
>interesting research-game topic than complex info-discarding pairwise
>comparison setups like Condorcet or Borda.  Similarly, today's money-based
>price-based economy is far too 'corny'.  Economic life and research could 
>be
>a lot more interesting if we ditched money and went to a system of pairwise
>product barters.  Yes, and we could ditch Kepler's elliptical planetary
>orbits in favor of Ptolemy's more interesting and far less corny system of
>epicycles.
>
>I leave to a forthcoming posting a full discussion of the issue of Hi-res 
>vs
>Low-res grading, = Approval.  (But hint: the old argument - that hi-res
>'strategy' boils down to that of Approval' - is less than decisive, because
>- as some of Approval's staunchest supporters here have lately noted - the
>utility of a vote embraces not only the direct outcome but also the ability
>to send a message, so the above 'strategy' is not all there is to true
>strategy: in fact, it's only a small part.) At least Approval, like
>higher-res, is part of the family of unconstrained grading methods.
>Considering criteria which I also plan to post and discuss soon, any one of
>these methods is superior to the bother of Condorcet(let alone Borda, Irv,
>etc.).
>
>THANKS FOR YOUR HEED
>
>Joe Weinstein
>Long Beach CA USA
>
>_________________________________________________________________
>Get your FREE download of MSN Explorer at http://explorer.msn.com
>

_________________________________________________________________
Get your FREE download of MSN Explorer at http://explorer.msn.com