[EM] Many candidates (Re: language/framing quibble)

[Kristofer Munsterhjelm]

Juho Laatu wrote:
> Is the target here to have a method
> that would allow and encourage having
> multiple candidates? (to allow the
> people of Owego to select the winner
> themselves instead of others/parties
> telling them what their choices are)

The target here, I think, is to have a method that uses another method 
of discovery than that of having the candidates push that information 
onto the people. This is based on that the usual method favors the 
candidates who have greater strength of dissemination, which translates 
to expensive campaigning budgets, which translates into that the 
candidates that do appear viable are either very rich or have the 
backing of external forces that demand quite significant favors in 
return (be those forces the rest of the political party, or lobbyists).

Or in short, the rationale for finding another method is that the 
current method favors those in power. Hence those in power gain more 
power, which is undesired positive feedback.

> This can be taken as an independent
> challenge. Which methods / systems
> lead to having numerous candidates?
> 
> (I limit the scope of discussion to
> single-winner elections, and exlude
> primaries and other party internal
> candidate selection and hierarchical
> proxy based methods.)

Fred's method could be used to select a single winner. Would you call it 
a hierarchical proxy? There is somewhat of a proxy thought in that the 
continuing candidates from each council "represent" the councils below, 
but it's not as thorough or direct as with say, delegable proxy, because 
the structure is fixed and one may argue that the lower councils select 
"good candidates" rather than "candidates that represent the council".

> Plurality certainly is not the method.
> It typically has only two candidates
> with chances to win, and others are
> easily spoilers.

I think one could make Duverger's law more general. If the method limits 
the voter to ranking k candidates, then the system tends towards a 
k-party state. For plurality, k = 2, since if you vote for A and only A 
and B are relevant, then that in essence is A > B.

However, the grip of that law is weaker as k increases. Consider a 
country like Canada, for instance. In it, different provinces have 
different strong parties (e.g. BQ). Local support keeps the system from 
degenerating into a two-party state. As k increases, the possibility 
that each local area will have different strong parties also increases: 
with k = 2, each local area can only have two strong parties, but with k 
= 3 they may have three, etc.

> Approval discourages nomination of
> more than one candidate per party
> or section.

It also has the Bush-Gore-Nader problem (if Nader is relatively 
popular). Both of these problems disappear if voters use strategy and 
know the others' sincere votes, but I've mentioned before why I don't 
like Approval (let's not have the entire VNM debate again).

Incidentally, Range supporters say that Range would be a method such as 
you're seeking. The idea is this: if the candidate is viable or really 
matters (McCain or Obama, for instance), then voters would max-min 
strategize, but if the candidate doesn't (Baldwin, McKinney, Barr), 
voters would vote honestly; thus third parties receive more support than 
one would expect if everyone voted honestly, since their significant 
competitors would be rated disproportionately low.

> IRV also carries some risk of early
> elimination of potential winners if
> one party has several candidates.
> Also exhausted ballots may be a
> problem if some section has numerous
> candidates. IRV is however probably
> better than the previous two.
> 
> Condorcet seems to work a bit better
> than IRV.

IRV has troublesome discontinuities. More significantly, it seems to 
lead to two-party domination (as in Australia). There are two possible 
explanations: either single-round single-winner elections in general 
increases the strength of the two most powerful parties, or IRV in 
particular distills the ballots badly enough to give a bias to the two 
major parties.

Condorcet might produce an opinion monopoly if there is only one 
political axis. The winner would be a centrist every time, so everybody 
would try to become centrists, as close to the center as possible to 
capture as many votes as possible. If there are more than one axis, 
there could still be competition, if on no other axis than "good 
centrist" (honest, fair, etc) versus "bad centrist" (corrupt, populist, 
etc).

> All methods that expect the voters
> to evaluate (rank or rate) large
> number of the candidates will be
> in trouble when the number of
> candidates gets high. At some
> point methods with shorter ballots
> become useful.

Most serious ranked and rated ballot systems would allow the voter to 
rank or rate only a subset of the candidates. The other candidates are 
either considered "below all ranked/rated" (Condorcet, plain Range, 
etc), or as if the voter didn't specify any information at all about 
them (Warren's average Range with quorum, but could also be used in 
Condorcet if so desired).

> One approach is to use a candidate
> tree where the votes (to individual
> candidates) are summed up in all
> the branches to see which branch,
> sub-branch and candidate wins.
> This would allow very high number
> of candidates.

One could also have a series of runoffs. A possible way to do that would 
be to eliminate, after each stage, those candidates where one is sure 
the people don't want them (e.g a Condorcet loser where a majority 
specified a preference regarding this candidate). Another way would be 
to differentiate the runoffs based on different areas of the country: 
the council democracy concept is a bit like this, where "runoffs" merge 
in the later rounds. A council-like merging runoff system could work 
like this:
	- first level, local elections of people in each district
	- second level, regional elections with winners of the district 
elections as candidates.
	- third level, national elections with winners of regional elections as 
candidates.

It would probably involve some sort of campaigning between each phase. 
It might also be weak in this respect, because if a very nationally 
well-known candidate would stand for first level election, he might make 
the first and second levels irrelevant by being easily elected in both 
instances.

> Discussion above covered only the
> part of making nomination of
> multiple candidates possible. In
> addition to this the method should
> also discourage nomination of only
> few candidates (parties may have
> some interest in doing so because
> this way the "inner circle" may
> better determine who will win
> instead of leaving that to the
> voters).

It's not technically that hard to make such rules. For instance, demand 
that the president were a Senator or Representative before becoming 
president, then allocate Senate or House seats according to a rule that 
favors many candidates (that is, that exhibits teaming); or, for that 
matter, use a single-winner method that favors teaming, directly.

However, doing so would lead to a situation where the parties would try 
to outcrowd the other, leading to immensely long lists of clones. Clones 
do no good, because they don't give the voters any additional variety. 
In Borda, for instance, the party with the largest slate wins (unless 
it's really unpopular). Therefore, using teaming directly is pointless.

One may argue that good ranked vote methods would encourage the 
nomination of multiple candidates, in particular if the party is unsure 
where the median voter is positioned. On the other hand, parties might 
benefit more from uniting behind a candidate (and throwing all support, 
campaigning, etc, towards that candidate) than they would by having 
multiple candidates run.