[EM] To ws, re: Condorcet vs Approval

[Abd ul-Rahman Lomax]

At 09:08 AM 7/18/2007, Michael Ossipoff wrote:
>As I've been saying here, in a series of elections in which voters 
>base their strategy on previous results, Approval quickly homes in 
>on the voter median candidate, the CW, and stays there.
>
>I certainly don't disagree about the great merit of Approval.

It *has* great merit. And it is important to understand its shortcomings.

Approval is Yes/No voting. Potential problems with Approval have 
mostly to do with its use in purely aggregative, single-step methods. 
Of course, that's what we usually mean by an "election method."

It is also important to understand that simple Plurality on a Yes/No 
question, where the question is formed through deliberative process, 
will settle on the Condorcet winner, when there is one.

But, of course, deliberative process is far more complex than the polling part.

Ultimately, to have full democracy, the result of an election 
requires that the electorate *ratify* it, a Yes/No question. This is 
what is done in small democracies and it only becomes a problem when 
the scale is large. So one solution we have proposed is Asset Voting, 
which is essentially a form of Delegable Proxy, and which makes 
possible the multiple-election process, i.e., many runoffs, but also 
even more than that, such as the introduction of new candidates 
midstream. (Asset Voting creates electors; in some implementations, 
electors are also candidates, but if enough candidates holding enough 
votes agree, why not allow a new person to be drafted?)

However, there is a serious shortcoming of Approval, and it also 
applies to ranked methods in general, Approval being a ranked method 
with only two ranks (just as it is a rating method with only two 
ratings), and that is the lack of preference strength information on 
the ballot. Because it is not on the ballot, preference strength 
can't be used, except in some methods that infer it, rather 
imperfectly, from rankings.

And preference strength is clearly important in decision-making. A 
group of friends will rationally maximize their collective 
satisfaction by placing greater weight on strong preferences. To a 
priori conclude that strong preferences are signs of fanaticism or 
ignorance is not only rude, it is often false (just as it may be true 
sometimes). Rather, strong preference may just as easily be a result 
of clear knowledge, as can weak preference, more rarely. (But I 
assume that clear knowledge would tend to make decisions more clear.)

The only method on the table that collects preference strength 
information is Range, at least Range 2, with three rankings. (I'm now 
using the convention that Range N has N preference increments, which 
means that it has N+1 ratings.)

And the best practical method in Warren's simulations is Range+2, 
with a top-two runoff.

Given that Range (like Approval) will usually choose the Condorcet 
winner, I have suggested that pairwise comparison from the Range 
ballot be used to detect failure to elect a pairwise winner, and, in 
that case, a runoff would be held between the Range winner and any 
candidate who beats the Range winner pairwise. I have not resolved 
the question of what to do if there is a Condorcet cycle of 3 
candidates, all of whom beat the Range winner, nor have I analyzed 
that situation in detail. It should be extraordinarily rare, but the 
method should address it. I expect that simulations will show similar 
results to Range+2, since if the Range winner is not the pairwise 
winner, the pairwise winner will usually be the runner-up under Range.

Not only would this extra feature, quite in line with existing 
practice when election results do not show majority victory, in many 
places, increase SU performance of Range under real-world conditions 
(that is, some strategic voting, which in Range is generally only 
"exaggeration," voting Approval style, or what I've called 
"truncation), but it could also make Range fully Condorcet-compliant, 
not to mention satisfying the Majority Criterion.

(A similar method can be used with Approval; specifically, where no 
candidate receives a majority, there is a top-two runoff, which 
should function much better under Approval than under Plurality+2, 
but also a runoff if more than one candidate gets a majority, thus, 
quite likely, encouraging broader approval in the first vote, since 
it becomes less harmful to one's favorite to also approve another.)

In Range, in particular, the actual need for a runoff should be 
uncommon. But the fact that it would be there should make, at least 
for knowledgeable voters, it more strategically wise to rate in more 
detail, less in Approval style; at least one would want to show some 
distinction for first preference. If the Range method allows 
sufficient resolution, nothing else need be done, but if the Range 
method is restricted, such as Range 2, then having a preference 
indicator on the ballot, to be used in preference analysis but not in 
the Range vote, would be advisable. In Approval, I've called this A+, 
and that preference indicator has a number of uses, even if not used 
for purposes of determining the winner. For example, it can steer 
vote-based public campaign funding as well as serving as a poll for 
better determining the real support for political parties, which 
Approval by itself can somewhat suppress (not as badly as Plurality, 
but still....)