[EM] Preference criteria vs votes-only (was " CP: guilty till proven innocent")

[MIKE OSSIPOFF]

James--

You said:

	Votes-only criteria are just easier to state
I reply:

True. But, as you say later in this posting, the preference criteria tend to 
be better, universally seamlessly applicable, and give answers that are more 
in keeping with our intent.

You continue:

, and when the method in
question is a ranked ballot method with sufficient slots for all
candidates, I think that there's no difference between votes-only criteria
and preference criteria. But when applying a criterion to non-ranked
methods, I think that the preference version is the most meaningful
version to use.

I reply:

Absolutely.

In addition to SFC, GSFC, WDSC & SDSC, and now PMC, that may well be 
generally true, though I don't have preference versions for Participation 
and IIAC. Who knows, though, maybe preference versions of those too could be 
written, and would be better in some ways.

But even if it turns out that Participation and IIAC are rightly votes-only, 
that what they're for, what they're intended to say, is votes-only, I notice 
that the criteria that are important to me are preference criteria. And 
Participation, Consistency, and IIAC are much less important to me.

Well, in fact I'm going to admit that, in spite of what I've been saying 
during the most recent few days, Participation, Consistency and IIAC aren't 
really important to me at all. Yes, having lately made such a big fuss aout 
those criteria, I'm now contradicting myself in a big way.

Participation, IIAC, and Consistency are only about, well...consistency. 
They aren't about some kind of wretched, abominable or blatantly unjust 
result, but only about results in 2 elections that don't really agree with 
eachother. Furthermore, those 3 criteria, when failed, don't involve 
strategy problems of any importance. And important strategy problems are my 
interest in voting system reform.

So why did I make such an issue about those 3 criteria? Because you 
emphasized caution above all else, with the concern that new methods might 
be worse than Plurailty. I wanted to go along with what you said, and 
sometimes the best way do that is to say it as if you believe it. I haven't 
used that tactic before, and don't expect to again. My point was just that 
that caution would apply to Cardinal Pairwise too, and on that we agree.

But, speaking truthfully, I haven't the slightest doubt that wv, AERLO/ATLO 
wv, and Cardinal Pairwise are better than Plurality. That's because of what 
I've said earlier about Participation's, Consistency's, and IIAC's lack of 
importance.

Of course it's true that it's necessary to convince voters of that. And so, 
for that reason, Approval and CR have an advantage over ranked methods. Then 
it isn't just that, as I've been saying, there are innumerable ways to count 
rankings, and only one obvious way to count Approval or CR ballots, and that 
Approval is an obvious minimal change in Plurality, and that CR is extremely 
well-known and liked. There's also the fact that Approval and CR are better 
than Plurality in _every_ way, and with Approval and CR we would avoid the 
criticism that there's some way in which they could be worse than Plurality.

So, though I don't really take that criticism seriously, it must still be 
taken seriously in regards to public acceptance.

You continued:

The basic template I use for a preference criterion is to
(1) define sincerity for the method

I reply:

I disagree. Have one sincere voting definition that applies to all methods. 
Defining sincere voting separately for each method would mean, in effect, 
that the critreria are defined differently for different methods. I use one 
sincere voting definition for all methods, and my criteria, as I always say, 
apply universally, uniformly and seamlessly to all methods.

You continue:

, (2) require that the voters vote
sincerely

I reply:

Not necessarily. Yes, for Condorcet's Criterion, Smith's Criterion, 
Condorcet Loser, and maybe Mutual Majority. But it depends. The majority 
defensive strategy criteria, all 5 of them, including the new PMC, tell of 
things that can be guaranteed for a majority even if others don't vote 
sincerely. Sometimes it's more meaningful to speak of a majority having a 
way to accomplish something without some particular specified degrees of 
insincerity, as with WDSC and SDSC.

Only SFC and GSFC say "If that majority votes sincerely...", because SFC & 
GSFC are about conditions under which it can be guaranteed that a majority 
can make someone lose without doing other than voting sincerely. But those 2 
criteria don't stipulate that other voters vote sincerely.

So SFC & GSFC are saying something much stronger than what WDSC & SDSC are 
saying. Instead of saying that that majority should have a way to accomplish 
something without some specified degree of insincere voting, SFC and GSFC 
say that they should accomplish it by _any_ sincere votng, that they should 
accomplish it merely by voting sincerely. Under certain plausible 
conditions.

You continued:

, and then (3) ask whether the criterion is met by the method,
given these restrictions on how preferences translate into votes, for all
possible sets of voter preferences. This seems fairly straightforward to
me. If you don't require that voters vote sincerely, or place some similar
restriction on the preference-->vote translation, then pretty much any
method can fail pretty much any preference criterion.

I reply:

Absolutely. Preference criteria must say something about how votes are based 
on preferences.

You continued:

	In approval voting, if a majority approves X and doesn't approve any
other candidate, then of course X will win. But does this really tell us
anything about whether approval voting can be relied up to deliver
majority rule in this sense?

I reply:

True, MP is a very weak criterion. PMC, written as a majority defensive 
strategy criterion, gives answers more in accord with the intent of MC,  as 
compared to the original votes-only MC. But MC & PMC are both very weak 
criteria. Much too easily-met to be useful for comparing most proposed 
methods.

Their only real use is for showing a way in which Borda is worse than any 
other proposed method. But that's a worthwhile use, because Borda's 
worseness in that regard is serious, and results in strategy problems.

You continued:

	We assume that there is a group of voters who constitute a majority of
the electorate, and who prefer X over all other candidates. But does this
necessarily mean that they will all approve only X? No, not necessarily.
Surely they will approve X, but there is no guarantee that they will not
approve other candidates as well. And if some of them do approve another
candidate, then X might not get elected. Which, from the standpoint of
majority rule, is significantly bad.

I reply:

Yes, but what that overlooks is that the people who will vote for more than 
one candidate in Approval are often people who won't vote for their favorite 
in Plurality. Look at the vast majority of progressives, who voted for Kerry 
instead of Nader, knowing that Kerry was a sleazy flip-flopper and 
enthusiastic Iraq-butcher, and that Nader was the only honest candidate. 
Having Plurality didn't mean that they voted only for their favorite. The 
difference beween Plurality and Approval is that those people didn't vote 
for their favorite at all, and implied, by their favorite-burial, that they 
like Kerry better than Nader.

So, when Plurality passes FHC and Approval fails FHC, that's very misleading 
and doesn't reflect anything genuinely better about Plurality as compared to 
Approval.

When a criterion gives results that seem wrong, like that, that suggests 
that a better criterion be foiund. I suggest PMC.

You continue:

If you still don't like using preference criteria to evaluate approval
voting

I reply:

I do prefer preference criteria, for all methods, including Approval. Yes, 
Approval passes Condorcet's Criterion, Smith's Criterion, Mutual Majority, 
etc, when they're votes-only (unless we write rules-criteiria that don't 
apply to nonrank methods, or that explicitly say that nonrank methods fail, 
or say something that means that &/or turns into an awkward patchwork 
contrivance. My criteria are not rules criteria in any way, to any degree).

So I have no objection to preference criteria for all methods, including 
Approval. But I claim that PMC gives answers more in accord with the intent 
of a majority criterion, as compared to FHC, and that FHC isn't showing any 
genuine way in which Plurliaty is better than Approval, when Plurality 
passes FHC and Approval fails FHC.

You continue:

..., consider the following method: "All voters must either vote for
all of the candidates, or vote for none of the candidates. The winner is
the candidate with the most votes, or if two or more candidates are tied
with the highest score, the tie is broken randomly."

Which votes-only
criteria does this method meet? I think that it meets pretty much all of
them. Does this mean that it's a good method? No.

I reply:

You don't need to convince me about that. I object to nearly all votes-only 
criteria for that reason. For instance, Plurality passes votes-only 
Condorcet's Criterion, Smith's Criterion, Mutual-Majority, etc., unless we 
make them into contrived and awkward rules-criteria that don't apply to 
nonrank methods, or that explicitly, lexicographically say that nonrank 
methods fail, etc.

You continue:

The fact that approval
passes so many votes-only criteria (again, nearly all of them, I think) is
similarly meaningless.

I reply:

Yes.

I agree that Participation, Consistency and IIAC are so unimportant as to be 
near-meaningless. (In spite of what I was insincerely saying during the past 
few days). And you know that in general I apply preference criteria, not 
votes-only criteria, to Approval.

I do that even though it means that Approval fails Condorcet's Criterion, 
Smith's Criterion, and Mutual-Majority. As an Approval and CR advocate, I 
use those preference criteria even though Approval & CR fail them, because, 
as you said, the votes-only versions are meaningless.

You continued:

[referring to FHC]

	Right. I don't know how new it is, but I guess I'll take credit for it if
nobody else wants to.

I reply:

I claim that my PMC gives answers more in accord with the intent of the 
majority criterion than does FHC or the original votes-only MC.

As I said, Plurality isn't better because a majority-favorite candidate wins 
if we suppose that everyone votes sincerely. That's because Plurality makes 
people not vote sincerely. It makes them bury their favorite. Approval 
doesn't. Plurality's better showing in FHC is completely unrealistic. FHC 
isn't measuring anything genuine when it gives Plurality a better answer 
than Approval.

FHC needs replacing. Replace it with PMC.

You continued:

I'd said:

>But PMC's [I'd called FHC "PMC" before I wrote my own PMC]  indication that 
>Plurality is better than Approval isn't very convincing. Those voters who 
>vote for X and Y even though they prefer X
>to Y are presumably doing so because they're making an informed strategic 
>decision that they need to help Y beat someone worse. Approval gives them 
>the freedom to vote for compromise in addition to favorite. Plurality 
>doesn't. Does it make any sense to call that an advantage of Plurality
>over Approval?  PMC penalilzes Approval for giving voters more freedom than
>they have in Plurality.

You reply:

	I'm convinced that approval is quite clearly better than plurality. And
yes, any benefit that plurality has comes as the result of a restriction
on voters. However, the PMC criterion (as you call it)

I reply:

I now call that criterion the Freedom-Hater's Criterion (FHC). I've written 
a better Preference Majority Criterion (PMC).


is not meaningless.
It makes an important point: There could be some candidate in an approval
election whom >50% of the voters prefer to all other candidates, and yet
that candidate may not be elected, even if all votes are sincere. Indeed,
people may never learn that the candidate was the favorite of >50% of the
voters.

I reply:

The only difference with Plurality is that those same people (nearly all of 
them, if our elections are any indication) won't vote for their favorite 
under the same circumstances). Plurality's better answer from FHC is an 
artifact of a not-meaningful criterion rather than a genuine indication of 
merit.

You continue:

That's a legitimate problem with approval.

I reply:

Wrong. Giving voters the freedom to vote for their favorite in addition to a 
perceived needed compromise is not a problem of Approval. The lack of that 
freedom is a problem of Plurality.

That Plurality does better than Approval at FHC is an artifact of a 
meaningless criterion, rather than a genuine indication of merit-difference.

You have made a choice to look at the situation in a particular way, and you 
don't want to consider that that might not be the meaninful way to look at 
it.

You continue:

Whether it's a problem
in plurality as well depends on how we define sincere votes in plurality.
Using the definition we have agreed on, it's not a problem in plurality.

I reply:

It's a problem in Plurality because, in real life, when there's perceived 
need to compromise, you won't get sincere votes in Pluralty. You'll get 
favorite-burial.

I'd said:

>Maybe we should rename PMC the Freedom-Hater's Criterion (FHC).

You replied:

	What a strange idea, especially when so many ranked ballot methods pass
it, methods that give voters freedom to express many levels of
preferences, rather than only two.

I reply:

No, we were comparing Plurality to a method that adds some voter-freedom to 
that of Plurality.

I'd said:

>
>Maybe I should have said that there's no criterion that Plurality passes
>and Approval fails, except maybe for new criteria that could be designed 
>for just that purpose.

Mike defines a preference majority criterion:

[I've replaced what you wrote with the improved final version]

If a set of voters consisting of more than half of the voters prefer X to 
each one of the other candidates, then they should have a sincere way to 
elect X.

[end of PMC]

You reply:


	So a method passes this type of preference criterion if one can construct
a criterion-passing set of sincere votes based on any set of sincere
ranked preferences.

I reply:

Yes, if "any" means "every". A method fails if, with that methd, it's 
possible to contrive a configuration of candidates, voters, and voter 
preferences for which the criterion's requirement is not met. But that's 
typical of preference criteria.

You continue:

It's kind of interesting, but it's kind of like saying
that approval passes majority in the best case scenario of every
preference profile, which isn't a very bold statement.

I reply:

It's kind of like saying (in fact it _is_ saying):

If a set of voters consisting or more than half of the voters prefer X to 
each one of the other candidates, then they should have a sincere way to 
elect X.

[end of PMC definition]

You continue:

	There are dozens of methods that are superior to plurality. Approval is
one of many. Perhaps all of these method have some slight drawback
somewhere, e.g. in approval's case failing my version of the majority
criterion while plurality narrowly passes it

I reply:

That's a completely illusory advantage, an artifact of a criterion that 
rewards Plurality's lack of voter-freedom and assumes sincere voting in a 
method that forces favorite-burial.

You continue:

, or failing participation and
consistency in the case of Condorcet and IRV, etc.... but those drawbacks
are more than made up for by very significant advantages.

I reply:

Quite so. I don't consider Participation, Consistency or IIAC to be 
important at all.

I'd said:

>
>But people are leery about Approval because they think it gives some
>voters more power than others, or violates a very important standard called 
>1-person-1-vote. It can be shown that Approval doesn't give as unequal
>power as Plurality does, and that 1p1v is a worthless rules-criterion, but
>whether that can be shown to the average voter is another matter.
>

You reply:

Unfortunately, it's not just the voters that matter, but also the judges.
Judges can quite possibly strike down voting methods that they deem not to
meet 1p1v, since 1p1v is a principle established by the US supreme court.
But judges are humans who exist in particular political contexts, and they
can be influenced by the popular will. Hence it may be easier to convince
judges that approval meets 1p1v if it is widely believed to do so by
academics, politicians, and average folks.

I reply:

That's the difficult part. I don't criticize optimism, but don't assume that 
the voting system academics are ever going to take their heads out of their 
ass. The haven't so far, and there's no reason to expect any different in 
the future.

Politicians are mostly bribetaknig sleazes who are useless for reform.

As for average folks, though Approval can be shown to not have a genuine 
1p1v problem, demonstrating that to the average voter may not be possible. 
And that's not even counting the fact that reformers aren't likely to have 
the media money that opponents will surely have.

But CR isn't subject to the 1p1v illusion. By proposing CR instead of 
Approval, we completely avoid that public misperception. And CR is already 
very well-known and popular, whlle Approval is unheard of, new, and an 
object of suspiciion.

Mike Ossipoff

_________________________________________________________________
Don’t just search. Find. Check out the new MSN Search! 
http://search.msn.click-url.com/go/onm00200636ave/direct/01/