Problems with Condorcet

[Bart Ingles]

Mike Ositoff wrote:
> 
> On Tue, 4 Aug 1998, bartman wrote:
> 
> > While Condorcet-type elections favor a compromise candidates in a closed
> > setting such as a parliament (as favored in Roberts Rules) where
> > candidates are relatively well-known peers of the voters themselves,
> > Condorcet is not well suited for use in large public elections.  Given
> > the realities of political advertising and voter psychology, full
> > ranking can not be relied on to give meaningful enough results to
> > support pairwise counting.
> 
> Though it isn't clear what you mean, people I've talked to would
> very much welcome the opportunity to express all their preferences
> & have them counted. Where do you get your information?

How about the voter turnout figures?  Most eligible voters can't be
bothered to go to the polls to vote for one candidate per race, let
alone rank every candidate.  I don't know how many people you have
talked to outside of these election reform lists, but I don't think you
will find many for whom "all their preferences" include every candidate
on the ballot.  Do you have any studies or surveys to the contrary?

> People like the idea of voting for their favorite over everyone,
> without giving up their opportunity toi vote a lower choice
> compromise over less-liked alternatives.

I would agree, except for the idea that a lower choice is necessarily a
"compromise" (where the word is taken to mean "negotiated solution" or
"middle ground").  The lower choice is simply the most palatable
remaining alternative.  

> > For example, a vote which ranks a major candidate last may simply mean
> > that he is perceived as the greatest threat to the other major
> > candidate.  This is reinforced by negative advertising by major
> > candidates, who tend to demonize their main opponents while ignoring the
> > rest.  This does not necessarily mean he is an extremist or unqualified;
> > it may simply mean he is the second most popular.
> 
> Were you here when we were discussing order-reversal?  I wrote
> at great length, a month or 2 ago, about why it won't be a
> problem in the votes-against methods that we recommend.
> Are you saying that you have a proposed method that's
> strategy-free? Are you saying even that you have one that's
> as strategy-free as the various votes-against Condorcet versions
> we've recommended, and are discussing? Maybe you'll propose one
> further down in this letter that I'm replying to.

I don't know that "strategy-free" is even a worthwhile objective,
compared to solving some of the grossest problems present in the current
system, and I don't see why my pointing out a flaw in one system
requires that I propose another in its place anyway.  You're not just
setting up a straw-man that you plan to knock down later in your reply,
are you?

> > Middle-rankings are also suspect, since not all candidates are equally
> > well known.  A vote which ranks a candidate somewhere in the middle may
> > mean the candidate is an acceptable compromise, or it may simply mean he
> > is an unknown who is not perceived as a threat to a more favored
> > candidate.  Even if he is a reasonable compromise in one sense, he may
> > be an extremist in other areas.  He may also be otherwise unqualified
> > for some reason which could easily be overlooked, since a voter is not
> > likely to use the same care in selecting lower-ranked candidates that he
> > would use for his first choice.
> 
> So you feel that voters might not know all the candidates well,
> or that they'll overlook something. And, for that reason you'd
> deny them the right to express as many preferences as they want
> to, and to have them counted reliably & fully. Maybe it's not
> for you to judge the voter's judgement? I just wanted to establish
> just what it is that you were saying.

I don't object to counting all of everyone's preferences, I only object
to drawing false conclusions from the results.  I don't think I am
calling the electorate incompetent by claiming that aggregate results
for lower rankings will contain increasing amounts of "noise" the
farther down you go, and that at some point the data becomes
meaningless.

> > If anyone wants to assert that the above scenarios will be rare
> > exceptions, I would like to see the argument.  Personally, I would
> > suspect that these scenarios would be just as likely as not; the real
> > problem is that there is no way to predict one way or the other.  As a
> 
> The real problem is that you haven't pointed out a problem.
> 
> > result, the only rankings with relatively unambiguous meaning would be
> > the first, or possibly the first couple.  This may not matter much with
> > IRO, since the lowest rankings are the least likely to be used, but it
> > would render Condorcet the equivalent of pulling names out of a hat.
> 
> People really do have opinions about the relative merit of
> candidates for an office. You only care about voting for your
> top 2 or 3 choices? Fine. I encourage you to only rank 3 or 4.
> With our methods you an do that without worry. It won't
> adversely affect the results. But it certainly isn't for you
> to say how many preferences others have, and how many they should
> be allowed to express.

Nonsense.  I have just as much right to an opinion as you, whether to
express my guess as to how many choices the majority of people would be
happy with, or to say where I think a ballot should be truncated. 
Ultimately it will be up to the public or their representatives to make
a judgement call.  You may have an opinion on every candidate in a race,
but since when does having more opinions entitle someone to more votes?

> IRO encourages the expression of long rankings too. It just
> doesn't reliably count the preferences that it allows you
> to express. It's rather like a car, up on blocks. You get to
> have fun running the engine & turning the steering wheel back
> & forth, but nothing more is guaranteed.

Kind of like sitting in Condorcet Hell debating endless variations on
the same theme.

> > While limiting the number of rankings may rule out some aesthetically
> > "ideal" election methods, at least the remaining possibilities are
> > simpler to implement.  Besides, there are other (probably more
> 
> I've often talked about the easy implementation & easy explanation
> of Approval. But we agree that our time is better spent working
> on & proposing the best. The best methods are rank-balloting
> methods. You may disagree, because you have different standards.
> Fine. Your standards aren't wrong, just different. But we're
> interested in which methods are best according to some standards
> important to many people.
>
> > important) reasons to replace FPTP elections than the prospect of
> > occasionally electing a 3rd-party candidate.  These should be of major
> > concern to anyone interested in electing "compromise candidates".  Some
> > examples:
> >
> > 1)  When an FPTP vote is "spoiled", the winner is the candidate that the
> > 3rd party voters would have voted *against* in a runoff (i.e. worst
> > possible outcome).
> >
> > 2)  In FPTP there is a tendency for centrist 3rd parties to push the
> > major parties away from the center by devaluing swing votes (in a
> > two-man race, swing votes are worth twice as much as "fringe" votes,
> > since in gaining a swing voter, the opponent loses the same voter, for a
> > net difference of two votes).  When a third candidate absorbs a large
> > number of centrist votes, the major candidates have less reason to
> > compete with each other for these votes and instead pay more attention
> > to extremist voters.  This is in addition to any shift of the major
> > candidates' bases of support due to displacement by the other parties.
> >
> > Bart Ingles
> 
> You've gone and disappointed me, Bart. You didn't tell us what
> method is better than the ones that we recommend & propose.

Why did I know that shot was coming?  It doesn't refute anything I have
said, anyway.

I think any reasonable method that addresses the FPTP problems and can
be adopted sooner than yours is by definition "better".  It could be
IRO, truncated IRO, or even a simple runoff election -- unless you
believe that none of these is a vast improvement over FPTP.  I keep
seeing Condorcet billed as a way to promote compromise winners, but I
suspect much of the interest in this method is due to the way it favors
third-party candidates more strongly than other methods.  If you want to
support it for that reason, fine, but it will be a tough sell getting it
through a two-party legislature.  At least the "--runoff" methods can be
presented as a win/win situation for both major and minor parties, with
the major parties becoming immune from vote-splitting, and minor parties
being freed of the "wasted vote" stigma;  I know of at least one
Republican representative who is in favor of a runoff.