[EM] Re et al Chicken and Egg

[Kristofer Munsterhjelm]

On 12/21/2011 05:10 AM, David L Wetzell wrote:
> Happy Holidays, I reply to RBJ, Ted Stern, Dave Ketchum and Kristofer M
> below.

(...)

>> DK:But when marketers lie and get caught, potential customers get
>> suspicious as to future marketing.
>
> dlw: To simplify is not to lie.

"IRV finds the majority winner". Given that IRV doesn't even do that by 
its own standards when it's limited to three ranks, is that a lie or 
simplification? "IRV solves the spoiler problem". Given that IRV 
exhibits center squeeze, which by its nature involves spoilers, is that 
a lie or simplification? "In IRV, voters just have to vote honestly". 
Given the Burlington example, is that a lie or simplification?

Where do you draw the line between separating lie and simplification? Do 
you just fit the line or curve so that all the counter-IRV objections 
get clustered on the disingenuous side of the line and the IRV marketing 
arguments get clustered on the "oh, merely marketing simplifications" 
side of the line?

> Now to Kristopher M.
> On 12/14/2011 09:59 PM, David L Wetzell wrote:
>
>>>     if we push hard for the use of American Proportional Representation
>>>     it'll give third parties a better chance to win seats and they will
>>>     prove great labs for experimentation with electoral reform.
>
>
>> KM: You keep on saying that. We can keep stating our priors until the
>> cows come home, but that won't do anything. Instead we should find some
>> information that would resolve the uncertainty.
>
> dlw: What uncertainty?  These are not simply priors.

The uncertainty within the parameters that you call p_irv and x_irv, and 
the uncertainty within the truth that the party dynamics will behave the 
way you do.

In other words, because we disagree, and because there is only one set 
of facts, one reality, there must be some discrepancy in our parameters 
wrt their "true" (God's value, to use Quinn's metaphor) values.

If we had the same priors, i.e. the same judgement of either the likely 
range of these parameters or of the data from which we make the 
conclusion about the likely range of these parameters, we would be in 
agreement. We're not, so we don't.

> PR does help 3rd parties(we agree right?) and American forms of PR would
> tend to help LTPs since there'd be fewer seats in plenty of "more local"
> "super-districts", this tends not to encourage nation-wide 3rd parties.

PR does help third parties; or rather, we can say that systems that give 
political minorities representation help the organizations that 
represent those minorities. If your American PR does just that, then it 
will help third parties.

Of course, it is possible to twist PR so that it is no longer PR. Using 
a divisor rule with an enormous large-party bias would be one way of 
doing it, for instance. Having a very low number of seats per district 
is another way, as that imposes an effective threshold below which a 
group gets no representation.

> 3rd parties are great places to experiment with electoral reform.
> This would especially be true with LTPs since any org that is smaller
> and consequently has less hierarchy is better at innovating.  And, when
> you got a whole lot of 3rd parties, as with LTPs, there's great scope
> for experimentation.  The kicker is to get them to share with others
> about the experiments so that the results will trickle out and up.

But that's the whole point. You have constructed an elaborate ecology 
where small parties compete on a local level, somehow build up large 
enough momentum, and then challenge the big parties. You think that if 
we institute STV+IRV, then the sort of system we're going to get is the 
system you detail, that it will be good, and that it's going to succeed 
at what you say it is going to do.

However, the system has never been tried. Therefore, we don't have any 
exact data from which to draw conclusions. Where we do have somewhat 
similar data, you say it's not applicable -- even local AU elections 
that you'd think would have different "local third parties" but don't.

So, in the absence of all of that, the only thing we have is theory. 
Well, theory and your Illinois example. I think that it's a very risky 
assumption to put the Illinois example against Yee, criteria, etc. and 
then say that the Illinois example wins. Furthermore, I think it's silly 
to take that risk when the opportunity of reform presents itself: better 
do it right and have something that *will* work, than rely on limited 
examples and push for something that only might.

You differ, so we've reached our disagreement. Which brings me to this...

>> What kind of information would resolve the difference? What observation
>> would make you think IRV is more likely to pull harder than STV pushes,
>> or that advanced voting methods are so good that IRV is a net loss?
>
> dlw: I don't expect LTPs woud resolve "IRV is more likely to pull harder
> than STV pushes".  I do expect that the use of 3-5 seat forms of PR in
> "more local" will give a greater return than IRV or any other
> single-seat election, due to de facto segregation. I expect a variety of
> voting methods to be tried out, sometimes multiple at the same time with
> the results then processed by the LTPs.  I also expect we'd find that
> different election rules are better for different sorts of elections,
> which is something that'll also emerge from case-studies of experimental
> results.
>
>> KM: You have suggested using Bayesian regret as a heuristic. I have been
>> away for some time. I haven't replied to your Bayesian regret post, and
>> I will likely be more idle in the future than I am now, as well. But all
>> that said, if you want to use Bayesian regret as a heuristic, how much
>> weight would you put on it? If the improvement of IRV over Plurality is
>> 100%, how much do you need for the advanced methods over IRV? 50%? 100%?
>> 1000%?
>
> dlw: My heuristic arg is that typically comparisons of election rules
> have presumed 7 candidates.  I argue that 4 candidates is more
> realistic, since there are limits in the number of serious candidates in
> single-winner elections.  In a 4 candidate election, the select
> candidates at random rule would do quite a bit better for BR than in a 7
> candidate election.  And if that rule does better so would all of the
> other rules between it and the Score Vote/Approval Vote rule.   Then,
> you can also add the effect of hybridization between AV and IRV
> IRV3/AV3, which is bound to redress some of typical args made against
> IRV by its electoral analytical critics. Thus, |Xirvav-Xoth| gets
> lowered and consequently the Ps become more important for deciding what
> is the right election rule alternative to FPTP to push for right now.

You still haven't given me any numbers. If we're going to resolve 
anything, we'd have to find some kind of agreement as to what data would 
be accepted.

Say, for instance, that I run the Bayesian regret calculations, and that 
I decide to limit myself to four candidates (even though I think there 
should be more), so that you can't dispute that aspect. Say, further, 
that I get a result that the best Condorcet rule improves upon IRV about 
10% as much as IRV improves upon Plurality. Then you could easily say 
"see, Condorcet isn't worth it". On the other hand, if I got a result 
that the best Condorcet rule improves upon IRV 10x as much as IRV 
improves upon Plurality, then you could also claim "yeah, but that's a 
static simulated result under conditions that aren't realistic, so in 
reality X_other - X_irv is still small".

After the fact, it would be simple for either of us to readjust the 
rules of the game, as it were, so that we get off free. If the Bayesian 
regret heuristic is going to solve anything, it must have power, and it 
doesn't have power if we can just step around the result no matter what 
it might be.



I suppose, then, that what I'm really saying is this: you discard 
theoretical points by saying theory isn't practice, that you're 
middle-brow so it doesn't matter anyway, and even *if* they showed the 
other rules are better, they don't show the other rules are *that much* 
better. You discard what little practical (experimental) data we have by 
saying that it's inapplicable (AU) or that the conclusion was just 
because of interference from scheming Plurality advocates (Burlington). 
At that point, very little remains. Thus I ask: what would it take to 
change your mind? What demonstration, what experiment would give you the 
data needed? What sort of argument would meet your "middle-brow, true 
test of IRV" standards? If your answer is "nothing", then we're done and 
this is just text on a screen.

(Incidentally, I didn't see you reply to the 36% backsliding rate for 
IRV. Were all of those due to scheming Plurality advocates?)

>> KM:Only if STV pushes harder than IRV pulls. It doesn't in Australia.
>> You disclaim Australia because you say the data can't be generalized,
>> and you consider repeals of IRV to be merely victories by Plurality
>> advocates won by incomplete or flawed presentation, whereas incomplete
>> presentation in the other direction is simply "marketing" and thus
>> nothing to be concerned about.
>
> dlw: AU uses IRV in "More local" elections where it is less likely to
> help due to de facto segregation by characteristics correlated with
> political preferences.  It uses PR in "less local" elections where it is
> less needed.  And so yes, it's not generalizable. I consider IRV to be
> reliable improvement on FPTP and two-round elections, moreso when
> coupled with the strategic use of PR that takes the edge off of how IRV
> does not tend to end effective 2-party domination.

Would IRV + PR be better than Plurality + PR? If you've constructed all 
of this from the Illinois example, which did use Plurality, why IRV? Or 
is IRV just an expedient, something one has to swallow to get the whole 
FairVote package, PR and all, through?

>> KM: So, in summary: I don't trust that IRV will give the necessary
>> changes. You do. We can keep on stating our claims backed by those
>> positions, but as long as we disagree on a more fundamental level, those
>> claims won't do anything but highlight our own positions yet again.
>
> dlw: I recommend that you not push for the use of IRV in Norway.  I also
> hope that you and others do not stymie the coupling of IRV(hopefully in
> hybrid form) + Am. Forms of PR that is emerging as what
> progressives/centrists/activists are going to be rallying around in the
> US.  When smart people like you and others here state unequivocally that
> Xoth>>Xirv it lowers Pirv without increasing Poth.

I hope that the momentum (to what degree it exists) can be turned in the 
right direction. Condorcet methods are being used in organizations (as 
well as a political party) right now. The United Nations uses Approval.

Furthermore, I would not say that my statements that X_other >> X_irv 
"lowers" P_irv as much as that it brings P_irv in line with its true 
value, should it persist. That is, I think IRV has several unappetizing 
properties that, when discovered, will have people leave it; and it's 
better they see those aspects now than later, so that they don't pull so 
much of the general idea of electoral reform down with them when they 
*do* see the bad parts of IRV.

True, claiming that X_other >> X_irv might not do much to P_other, 
except perhaps by making it less likely that IRV's flaws will taint the 
other methods. Raising P_other is a separate concern. It can be done by 
people signing the declaration, or when the parties and organizations 
currently using other methods serve as momentum of their own.

> Ethically, the burden of proof is on those who oppose the working
> consensus proposal for reform to show that their preferred approach
> is considerably better. I believe I have been holding to this ethical
> principle in my emails on this list.

You have given proof based on your assumptions. You have then argued, 
based on those assumptions, that the counters were not applicable (such 
as that the theory is too theoretical or doesn't impact enough).