[EM] Re et al Chicken and Egg

[Kristofer Munsterhjelm]

I think we're reaching the end of this thread, as I'm not all that 
interested in continuing further. It's relatively clear that changing 
your position will take a lot of work, and to put it simply, "I'm not 
getting paid enough for this" :-) I have recently had other things to 
focus on, as you may have noticed in my lack of posting to EM of late.

Let me reply to one of your concerns, and then I'll finish with an 
argument against IRV, based on the point of view of the voters. If you 
want to reply with your own "closing statement", go ahead and do so; I 
don't think I will continue.

You say there is something to American exceptionalism, so it isn't 
unreasonable to put the US on one side and everything else on the other 
side of the PR line.
There might have been a case for American exceptionalism in 1787, but 
the United States is now hardly unique in having a presidential system, 
or even a strong presidential system. Other nations have been inspired 
by the United States' form of government, and so presidentialism has 
spread (although, to my knowledge, there are still more parliamentary 
nations than there are presidential ones).

Let's look at the Wikipedia list of republics considered to have a 
presidential system of government 
(https://en.wikipedia.org/wiki/Presidentialism). From the beginning, we 
have "Afghanistan, Argentina, Armenia, Azerbaijan, Belarus, Bolivia, 
Brazil, Chile, Colombia, Costa Rica, Cyprus, Dominican Republic, 
Ecuador", and so on.

Intersecting Wikipedia's list with the PR list of my previous posts 
gives an intersection of Brazil, Colombia, Costa Rica, Cyprus, Dominican 
Republic, El Salvador, Guatemala, Paraguay, Peru, Sri Lanka, Suriname, 
and Uruguay. Thus, I don't think you can justify putting the United 
States on one side of your historical specificity line and all the PR 
nations on the other.

These are not weak presidential republics either. In Colombia, a favored
approach of past presidents (like Lleras Restropo) was to construct 
autonomous quasi-governmental organizations, and then give them power. 
These organizations were designed to be accountable to the executive, 
but not to the legislature. A similar approach was used in Brazil. 
Again, quoting Wikipedia: "In Brazil, presidents have accomplished their 
objectives by creating executive agencies over which Congress had no 
say". So I do not think you can say they only have figurehead presidents 
and so don't count, either.

Where they do differ (whether they're Latin American, like Colombia, or 
not, like Cyprus), is that they don't have two huge parties and a bunch 
of tiny parties looking for scraps at the table. I think that is a point 
where the United States is genuinely different, but I also think that 
the reason they are is because the US has neither PR not a good 
single-winner method.

To claim that the United States can't use PR or advanced voting methods 
because it has two+tiny parties while the rest of the world does not is 
to confuse cause and effect. The reason there are two parties is because 
so few people want to throw their vote away in such an important contest 
as a Presidential one is. That is an artifact of the Plurality system. 
As a consequence, it becomes vitally important for parties and 
candidates to appear to be electable, and the way they accomplish this 
is to spend lots of money on very expensive media coverage and on the 
whole primary and electoral circus. Thus, it takes a lot of money just 
to overcome the barrier of he-won't-be-elected-anyway. Any method that 
goes wrong when third parties enter the scene would keep this bug (to 
the extent that it does go wrong with third parties present), because 
the candidates still have to spend a lot of money to signal the 
strategic voters that they are electable -- that the strategists should 
rank at least him honestly.

In concluding the above: Presidential, even strongly presidential 
systems with PR do exist. That they don't have extremely expensive 
primary and presidential races should not be used to claim historical 
specificity, because that is specificity of the influence of monied 
interests - part of the what we're trying to change. Instead, it should 
be used as evidence that strongly presidential systems don't need to 
waste money on a grand scale; and that multipartyism helps keep the need 
for such waste down.

=

So, my voters' argument. I'll focus on the people in general and the 
people-as-voters in particular, because it is for the sake of the voters 
that we're doing this. It is the people that benefit when the system 
elects representatives that represent them. It is the people that 
benefit when a single-winner method picks a good winner; and it is the 
people that ultimately benefit from better governance less encumbered by 
plutocratic influence, as well.

I'll argue that multiparty democracy is more in line with what the 
voters actually want. To add strength to my claim that the voters do not 
prefer two-party situations, I'll go in two directions. First, that 
non-official results show that people "vote" (express their opinions) in 
a way significantly less polar when they don't have to compromise to 
elect the lesser of two evils; and second, that where proportional 
representation methods have been used, both nationally in other 
countries and locally in the United States, the result has been a growth 
of many parties, consistent with multipartyism.

Since IRV does not give multiparty democracy (and you have said as much, 
and that it isn't what you want anyway), and the voters express their 
desires in a way as to support multipartyism, that counts against IRV.

Then I'll refer to that the advanced methods don't have the center 
squeeze problem of IRV. I'll go further and state more general 
conditions where the advanced methods do well, so as to support that the 
advanced methods don't act as a patched-up IRV ("IRV 1.1") that just 
hides the most immediate problem of IRV the way IRV acts as a "Plurality 
1.1" that hides the most immediate problem of Plurality.

Finally, I'll conclude that, based on the above, IRV requires that 
voters twist their votes into an IRV shape merely to get an acceptable 
result in just the setting where third parties are growing large, 
whereas the advanced methods do not; or that third parties have to 
intentionally decide to stay minor to spare the voters of this, whereas 
that is not the case for the advanced methods. Further, I'll point out 
that IRV can't justify demanding this of the voters and parties, because 
it doesn't produce results closer to the multipartyism the voters want 
than does the advanced methods. Therefore, IRV is worse than the 
advanced methods twice over, and so I cannot support it even if the 
other arguments not detailed here were to be invalidated.

Without further ado, the meat:

1.1. According to their preferences, the people prefer multipartyism to 
two-party rule.

This is where my reference to the Orsay exit poll, the other French and 
German studies, as well as the two United States exit polls, all shown 
on Rangevoting, come into play. The Orsay exit poll (but also the other 
exit polls) show that the people of France and Germany vote in a manner 
consistent with multiparty preference, while the United States exit 
polls and studies show that while the United States voters don't vote 
for third parties to the extent that those parties would win, they 
nevertheless vote (or express preference) for third parties and 
candidates to a much greater extent than in the official count.

When I direct your attention to the Approval voting polls (telephone 
polls, etc.) in the United States, my point is not that Anderson came 
second. As you say, you don't get a seat for coming in second. My point 
*is* that the voters vote in such a way as to show that candidates not 
officially blessed also have some popularity. Stripping away the 
strategic distortion caused by Plurality, we see that "no-hopers" have 
some support. What is keeping them down is a catch where they have to be 
seen as electable to win or gain further support, but can't be seen as 
electable unless they have a great chance of winning.

In my other posts, I have quantified this tendency in terms of effective 
number of candidates, at least for the European polls.

1.2. According to the outcomes, the people prefer multipartyism to 
two-party rule.

Internationally, I need only point to the countries that have 
proportional representation. If the people wanted a two-party situation 
in a PR nation, they would only have to vote for the two largest parties 
to establish that situation. That does happen in some places; most 
notably, in Malta. Malta uses STV, yet has only two parties. In Malta, 
the voters only make use of 5-seat STV's capacity for intra-party 
competition, but not of the capacity for inter-party competition. See 
more at http://aceproject.org/ace-en/topics/es/esy/esy_mt .

Even in elections for single seats, this pattern appears to be true. 
When the voters are given a single seat method that can elect candidates 
from multiple parties, the voters make use of that capacity. To avoid 
appearing to construct a tautology or to beg the question, I will 
specify that more clearly: the method is being used in more than ten 
nations (so you can't claim one-nation micronumerosity), and in more 
than two thirds of the nations where it is used, that nation has a 
multiparty democracy.
I am, of course, talking about top-two runoff voting. According to 
http://rangevoting.org/TTRvIRVstats.html, there are 27 runoff-using 
nations, and 21 of these are multiparty. (If you wish to argue Warren is 
wrong, go ahead and do so.)

Within the United States, I point to New York under STV. In New York 
under STV, the voters made use of the capacity of STV to provide for 
multipartyism and voted (and got results) which let multiple parties get 
on the assembly. This significantly weakened the party machines, who 
immediately countered but only got STV repealed after battering down 
public opinion by the card of red-baiting. This is why the New York STV 
example matters more than your proverbial hill of beans, and this is why 
I am surprised you do not look at it when you talk about historical 
specificity. It shows that American voters also avail themselves of the 
capacity for multipartyism when they have the instruments by which to do so.
Indeed, the leaders of the New York county Republican party had no 
illusion about this as they worked with Tammany Hall against PR, calling 
PR "a threat to the two party system" ( 
http://www.mtholyoke.edu/acad/polit/damy/articles/kolesar.htm ).

-

2.1. The advanced methods (most Condorcet methods, as well as MJ, 
Approval, and Range) do not suffer from center squeeze under honest voting.

This is very easy to check. One may simply go to Ka-Ping Yee's 
one-dimensional Gaussian vothing method visualization and push the wings 
closer to the center. In Plurality, the center vanishes quickly. In IRV, 
the center lasts a little longer, but first it splinters 
nonmonotonically, and eventually it fails altogether when the center 
does not have enough first place votes to outlast the wings. That 
happens even in a completely symmetrical situation where the major 
parties are equally far from the center, "clustered around it".

In contrast, both the rated/approval methods and Condorcet methods elect 
the candidate closest to the middle-most voter (in ratings, possibly as 
weighted by strength of preference). This is a reasonably fair standard, 
because one can imagine left-wing voters "canceling out" right-wing 
voters until the centermost remains.

In fact, Condorcet methods will always be impervious to center squeeze 
in a left-right situations like the above. I'll get to that later. Some 
Condorcet methods may behave strangely in higher dimensions (such as a 
contrived "CW if there is one, otherwise the Plurality winner" method), 
but most will degrade gracefully.

The theoretical result can also be empirically verified in two cases. 
First, the Burlington election of 2009: all the Condorcet methods picked 
Montroll, as would Range under certain assumptions (of monotonically 
decreasing ratings). Second, by ranked votes from a French exit poll in 
Faches-Thumesnil and Nord in 2007, Condorcet methods elect Bayrou (as 
does Range and MJ using the Orsay data from another exit poll in the 
same election), whereas IRV sides with
Plurality and elects Sarkozy. See 
http://rangevoting.org/French2007studies.html for more.

I will point out that I am not talking about dynamics. This part of the 
argument is about the voters not needing to learn to vote for the 
stronger wing, or the minor parties not needing to stay out of the way 
if the voters don't learn.

2.2. The advanced methods are significantly less impacted by adding or 
removing candidates than is IRV and Plurality.

For the rated methods of Range and MJ in particular, as well as for 
Approval, this is also simple to see. When the voters are honest and 
rate according to an objective standard, these methods pass a 
rated-version variant of IIA. This means that if you add or remove a 
candidate X, that doesn't alter the winner unless the winner used to be 
X (in case of removal) or X is now the new winner (in case of adding 
candidates). Thus, no amount of extra candidates can make the method 
misuse honest voters' votes to squeeze out the center. I think that this
particular variant of IIA is stronger for MJ than Range, since MJ 
suggests voters compare the candidates to a common graded standard and 
gives less incentive for voters to deviate from that standard, whereas 
in Range, normalization may be considered honest (and so the IIA variant 
would fail).

For the monotonicity-fixed SODA, as it is based on Approval, it passes 
an analogous version of IIA. If a new candidate appears, either the old 
winner will keep winning, that new candidate will win, or a candidate 
preferred to the old winner on the new candidate's delegation order will 
win.

For Condorcet, the details are more complex. In a left-right scenario, 
as one might expect when third parties are just getting off the ground, 
Black's single-peakedness theorem shows that the Condorcet winner will 
always be the candidate closest to the median as defined above, no 
matter the amount of other candidates and their positions. Since maximum 
likelihood estimation over relatively simple error models imply 
Condorcet while IRV can only be considered an MLE if you make the error 
model very complex indeed, I think voter uncertainty will drag IRV 
further from this in reality than it will drag Condorcet.

But perhaps that is not enough. In that case, I refer to James Green 
Armytage's paper on voting strategy, 
http://www.econ.vt.edu/seminars/seminarpapers/2011/jamesgreenarmytage10142011.pdf 
 In it, James proves that when there is a Condorcet winner, Condorcet 
methods are vulnerable to neither exit nor entry (candidates being 
removed or added). Furthermore, his computer simulation results indicate 
that minimax (which is not even cloneproof, but is Condorcet) provides 
nearly no incentive for strategic entry or exit, but IRV's incentive to 
strategic exit increases with the number of candidates, and eventually 
reaches Plurality's in the case of single-candidate exit.


Thus, I reason that the advanced methods are not simply patches to IRV; 
their improvements with regards to candidate exit and entry, and 
resistance to center squeeze, are general (not specific), and don't lose 
power as the number of candidates is increased further.

-

3.1. IRV requires that either votes twist their votes into an IRV shape 
to get an acceptable result, or that parties intentionally stay small to 
avoid the conditions where the voters have to do so, or both.

This follows from the center-squeeze vulnerability shown in 2. In order 
to avoid center squeeze, either the third parties can stay far enough 
away that their votes can't make the wrong winner win, or the votes can 
vote as if the third parties were that minor.

Burlington provides an example of what happens when the voters don't and 
the parties don't, either. You've said that it did, but you claim that 
dynamics will fix it. Those dynamics don't counter this point, since the 
dynamics take the shape of voters or parties altering their behavior to 
accomodate IRV in a similar, though somewhat reduced manner, to how 
voters or parties alter their behavior to accomodate Plurality.

3.2. IRV doesn't give multipartyism.

This is not contended by you, and it is a simple consequence of the 
nature of IRV. If IRV breaks to center-squeeze when third parties become 
too large, and the countermeasure is for either the parties to be small 
enough that IRV doesn't fail, or for the voters to act as if the parties 
were, then the parties can never by themselves grow large enough to lead 
to multi-party rule within the area of the IRV election in question.

Further evidence can be seen, though scarce as it is, by that all the 
nations that use IRV have two-party rule where IRV is used. Australia 
has two-party (or two-and-a-half party, since the NatLib coalition is 
pretty much a party) rule in its IRV body. You have not disputed this, 
but you have argued that this would be countered in the non-IRV bodies 
that would use PR but would not give multipartyism in general. Also, in 
Fiji before the military coup, the system was coalescing to two-party 
rule despite being in the context of a diverse nation, and despite 
ethnic quotas. See http://rangevoting.org/FijiPol.html for that data.

3.3. Therefore, IRV fails twice.

By 3.1., IRV imposes demands on voters and/or on parties that the 
advanced methods do not. That counts once against IRV, but it wouldn't 
by itself be so bad IRV could justify that demand by providing better 
outcomes in return. However, it does not. By 3.2., IRV fails to give the 
multiparty outcomes that voters appear to want (as established in 1.). 
Therefore, at the very best, IRV is no better than the advanced methods 
in bringing about multipartyism, and realistically (supported by the 
presence of single-office voting rules that actually do give 
multipartyism), IRV is worse than the advanced methods.