[EM] More non-altruistic attacks on IRV usage.

[Kristofer Munsterhjelm]

David L Wetzell wrote:

>>     KM:I think this is where we differ, really. On a scale from 0 to 1,
>>     you think their relative merit is something like:
> 
>>     0: Plurality
>>     0.7: IRV3/AV3
>>     0.72: Condorcet, MJ, etc
> 
>>     while I think it's something like:
> 
>>     0: Plurality
>>     0.25: IRV
>>     0.3: IRV3/AV3
>>     0.7: Condorcet, MJ, etc.
> 
>>     (Rough numbers.)
> 
>>     If you're right, of course arguing for Condorcet seems like an
>>     angels-on-the-head-of-a-pin thing - and it's even harmful because
>>     X_IRV * p(IRV) >> X_Condorcet * p(Condorcet) . The step from
>>     IRV3/AV3 to Condorcet is only 0.02 after all, and the momentum
>>     difference is huge! But if I'm right (and this is why I keep
>>     bringing up the examples where IRV has been used), then going for
>>     IRV is much more likely to rebound on you later.
> 
> 
> dlw: This is a good statement of our diffs.  I'd say I rank the partial 
> use of PR to get a contested duopoly (or to prevent a contested 
> monopoly) very high, while you rank it rather low, especially relative 
> to the development of a multi-party system, not unlike what you have in 
> Norway.  This is likely a matter of political cultural differences, 
> which makes my valuations more likely to be prevalent in the US.

I base my low confidence of PR's capacity to pull stronger towards 
competition than IRV does towards consolidation in that IRV pulls 
stronger wherever it's been tried. You say they aren't applicable. You 
may have that opinion, but then there's little I can show that will help.

So there are two disagreements. Ultimately, I think that multiparty 
democracy would be better than your contested two-party rule. I could 
pull market analogies for this (oligopolies and cartels), or I could 
simply say "it's harder to buy off ten than to buy off two". Here you 
may claim that this is because of my political difference if you want to 
do so.

However, stronger is this: even if you wanted a contested two-party 
rule, I think IRV would pull too strongly. Again I take my evidence from 
other countries where that is the case - where the major two are the 
same as they have been a long time ago, and again, you say that's not 
applicable. Of course, nothing is absolute: even with Plurality, your 
own major parties have changed a few times since the time of the 
Founding Fathers. I just don't think IRV will make a difference.

So if we boil down our disagreement further, I think that we *can* 
generalize from other IRV nations. You think we can't, because your 
rules are different. There have been many IRV elections (so there are 
samples to pick), but not very many different systems of government in 
which IRV is placed. If I pull 100 local Australian elections and 
NatLibs or Labor win in 95 of them, you could say that's because of the 
Australian rules so they only count as one sample.

> I think you're judging IRV too harshly on account of Burlington VT.  The 
> sample size is too small to make such a strong judgment.

Well, Burlington just confirms things. The simulations say IRV can fail 
to pick the CW, and may squeeze the center out, and the less minor the 
minor parties are, the worse it gets. As Burlington agrees with the 
simulations, that doesn't count in IRV's favor.

>>>       4. This is why I pick away at how the args in favor of other
>>>       election rules get watered down or annihilated when you make the
>>>       homo politicus / rational choice assumptions more "realistic" or
>>>       you reduce the number of effective candidates, or you consider
>>>       how perceived biases/errors get averaged out over time and
>>>       space, or you focus on the import of marketing and how IRV has
>>>       the advantage in that area of critical importance to the
>>>       probability of successful replacement of FPTP.
> 
> 
>>    You try to do so. From my point of view, when I give you examples
>>    from the real world, you say that it'll be different here (re
>>    Australia on the one hand and France on the other, for instance).
>>    When I pull from theory, you say that the theory doesn't apply
>>    because it assumes too much; and when I pick examples where theory
>>    and practice seem to agree (Burlington), you say that that's just
>>    because the status-quo-ists put pressure to bear on IRV.
> 
> 
> dlw: 1. Well, the sample of IRV uses is small, which makes it hard to 
> render verdict on it.

So why would IRV improve things enough over Plurality? That verdict, 
too, has to come from somewhere.

> 2.  AU does use IRV/PR in the opposite from ideal mix if the goal is to 
> increase the number of competitive elections.   
> 3. WRT France, we disagreed on matters of taxonomy.  I classified their 
> top two as a hybrid.  You classified it as a winner-take-all and used it 
> to show how IRV has been improved upon and could be improved upon further.  

Let me try your pragmatism for a minute. You say that our disagreement 
about top-two is taxonomy. Why should taxonomy matter, though? If I have 
a "tacs"-type voting method, and an "intar"-type voting method, both 
elect winners to single positions, and the voters know both, but the 
difference is that the "intar" method produces a greater diversity of 
winners than does the "tacs" method, then why not use the "intar" method?

Or, moving back into reality, if we're comparing TTR and IRV, and TTR is 
known in the US and shows you can go beyond a dominant two-party system 
even without PR, why not use it? Whether TTR is a "proper" single-winner 
rule doesn't matter if you're pragmatic.

I say top-two (TTR) is known, since runoff elections are used many 
places in the US. FairVote tends to market IRV as "runoffs without the 
runoff" there, and as my list showed, that particular form of marketing 
carries with it a risk of backfiring.

 From what I understand, the answer is "because IRV is linked to STV, so 
it's a way of getting PR in the door". But PR can be implemented, as the 
proportional representation organizations have shown in the past, 
without even mentioning IRV. So if that's the answer, it can essentially 
be rephrased as "because that's the way FairVote sells STV, and 
FairVote's got all the momentum these days". If *that* is true, I don't 
see that we should abandon other methods simply because FairVote made a 
mistake about how to market PR.

Also note that I didn't use it to show how IRV "has been improved upon". 
Top-two got there first. The most significant point I was making with 
top-two was that it *is* possible to have multiparty democracy with 
single-member districts. The multi- may not be as multi as with PR, but 
by showing this, I can counter statements of the form that "IRV leads to 
two-party domination within IRV seats, but so do all single-member 
rules". You might use that kind of statement against Condorcet, but you 
can't use it against runoffs.

> 4. Rational choice theory is unrealistic (at least for all political 
> elections).  This makes a lot of the Condorcet et al stuff be much ado 
> about nothing.  You acceded that fuzziness in the perceptions of voters 
> and the positioning of candidates muddied the waters considerably.  

Condorcet regards not just a single very precise result, but a whole 
class of them. Therefore, it is resistant to perturbation, so I don't 
think there's "much ado about nothing".

To clarify by taking an extreme, you could argue in the same manner 
against majority rule. Say something like "chaos means there may be 
majorities when the *real* majority isn't in favor, so Majority 
compliance doesn't matter". That sounds absurd - it is - but it's absurd 
because the majority criterion is robust.

As for "muddying the waters", I said that to the extent it does so, it 
cuts against IRV. First, I introduced the Yee diagram, where IRV has 
much more comple win regions than Condorcet. The thinner and more 
convoluted the "border" between win regions, the greater the chance that 
an election result will fall in the wrong "country". Second, I pointed 
at IRV's amplification dynamic, where near-ties in one round could lead 
the method on a completely different path in a later round. Third, I 
showed Brian Olson's graphs, which seem to show that IRV handles noise 
less well than the other methods.

So I was saying "Alright, you think that rational choice is too 
simplistic on account of fuzziness. Well, here's what happens if you 
take noise into account, and it's not favorable to IRV".

> 5. You can't divorce what happened in Burlington from the real-politik. 
>  It's not a slam-dunk, because the opponents of electoral reform are 
> well aware of the divide and conquer strategy.

Are they? I don't think the opponents of electoral reform know about 
Condorcet, much less Majority Judgement, Range, Approval, or the likes. 
The greater you think the order of magnitude in p_IRV >> p_Condorcet, 
the less of a point you'd think there would be for the opponents to even 
care about Condorcet.

To me, it seems more plausible that they said "okay, we want to repeal 
this. What can we throw at IRV and have it stick?". Then they might have 
looked at pages of people like Warren and thrown nonmonotonicity at IRV. 
This would have had a much lesser chance to actually stick if IRV had 
behaved properly. When FairVote advertises IRV as the way you can vote 
without having spoilers distort the outcome, then people vote, and 
spoilers distort the outcome, then that's not just the opponents of 
electoral reform using massive marketing machines to convince the 
people. Draw attention to, perhaps, but attention or not, there was a 
majority who preferred Montroll to Kiss, and that majority got its wish 
denied in favor of the Kiss-over-Montroll minority.

> All of which is to say that | X_IRV-X_Condorcet | is likely smaller than 
> the electoral analytics purport, while you concede p_IRV is considerably 
> greater than p_Condorcet.

If FairVote continues on its marketing of IRV and we do nothing, yes, 
IRV is more likely to be adopted than Condorcet, at least in the short 
term.

However, I think that would be unfortunate in two ways. First, I don't 
think IRV improves Plurality enough that it'll matter. It'll keep major 
parties safe as long as minor parties are minor enough, but not beyond 
that point. Therefore, if you do end up with IRV, you keep your 
uncontested two-party rule.
Second, you may not even keep IRV. If IRV gets it wrong often enough, or 
reproduces Plurality's results often enough that it doesn't seem to be 
worth it, then the option of reverting it can seem quite tempting. If 
FairVote claims that it's a runoff-without-runoff or that you can vote 
as you wish without fearing spoilers, and that turns out to be false, 
then IRV may not last; and if IRV is considered equal to ranked 
balloting, then the immediate reform chance is lost. You might have to 
go a far more circuitous route involving augmenting Plurality with MMP - 
and that wouldn't help executive positions like Governor or President.

>>    KM:How can we go anywhere from there? If you can say every
>>    application is a special case that doesn't apply in the situation
>>    you have in mind, and if you can say that the theory that remains
>>    has no verification in the form of practical results, then we're not
>>    left with much except restating our relative merit numbers to each
>>    other.
> 
> 
> dlw:  There is a small data set for IRV apps and an even smaller one for 
> the infinite array of alternative electoral rules.

There's a small number of rulesets. There's a substantial number of IRV 
elections, but most of them (local Australian elections) have the same 
overarching ruleset that you would probably argue taints the results.

> My point is that when you argue "falsely" that | X_IRV-X_Condorcetetal | 
>  >> 0, with the latter over the former, you risk lowering 
> | p_IRV-p_Condorcetetal |  by making p_IRV drop a lot more than 
> p_Condorcetetal rises.  p_Condorcetetal does not rise because there's no 
> heir apparent, likely because the Xothers vary relatively little among 
> real world voters.  

The lack of a heir apparent might not be that bad. Committees like the 
Rhode Island one might pick the best among near-equals according to what 
they deem important. This kind of approach has worked in New Zealand and 
led to election reform referenda in parts of Canada (though the voters 
there decided not to go for it in two cases, and had a majority but not 
a supermajority in the third).

If you want electoral reform to happen from the bottom up, you don't 
need the national government to set up the committee. A state can do so, 
or a more local area like a city.

> We can't say it's just a matter of opinion, cuz it's probably not such, 
> and so what makes sense to me is to rally around IRV3/AV3 and trust that 
> when it's use is prevalent that it'll be the basis for choosing among a 
> wider set of electoral reforms, which will have a further ratchet effect 
> in expanding upon what is democracy.  

It's a matter of data either incomplete or considered incomplete. From 
your position on what does and does not count as a valid distinct 
sample, and from your relative merit ideas, your conclusion follows. I 
see that. I can put myself in your shoes, as it were. From mine, my 
follows. I don't want to risk that IRV turns out to make no difference 
or that p_IRV turns out to be hollow (to collapse when enough scrutiny 
is brought to it).

> dlw: And I'm saying the real life sample sizes are too damn small to 
> justify rhetorically torpedoing FairVote's marketing/bundling strategy. 
> Why not, push for more experimentation with other election rules in 
> Norway or elsewhere and trust that the US will find its way in its own 
> way, hopefully with some critical learning from across the pond.

I don't think there's any support for that here. Modified Sainte-Lague 
(with two-tier proportionality - a sort of double list MMP based on the 
same counts) is good enough, or so goes the opinion of most of the 
voters. A vote is only "wasted" when it doesn't change the party's 
rounded number of seats, which happens much less with 169 MPs than with 
a single president. Therefore, most voters (who don't vote for parties 
below the threshold) feel their vote matters, and we don't get 
lesser-evil strategy problems.

In an ideal world, perhaps we'd be using STV with Schulze's adjustments 
for the second tier of proportionality. Or perhaps we'd pick our 
"representatives" completely randomly with different advisory bodies 
giving options for different sorts of legislation, so the representative 
sample of the population determine "what" and the advisory bodies 
determine "out of which options" and "how". Who knows? In any event, 
that's very far off: people are basically satisfied.

(And you may find this of interest: After independence, there were two 
main parties in Norway: the Conservative Party and the Liberal Party. 
While the Conservative party has remained strong (it's currently third 
largest by support), the Liberal Party's support has been below 10% 
since 1969. Major party status can indeed change in PR. I find the image 
of an US equivalent funny: a Democratic Party at 6-7% support?)