[EM] Why I think IRV isn't a serious alternative

[Abd ul-Rahman Lomax]

At 10:09 AM 12/4/2008, James Gilmour wrote:
>Kristofer Munsterhjelm   > Sent: Thursday, December 04, 2008 8:17 AM>
> > I'm not Abd, but I think the argument goes like this: in TTR, if a
> > (usually) third candidate gets enough FPP votes to make it to the second
> > round, that candidate has a real chance of winning, since the second
> > round will be focused on those two candidates alone, whereas, on the
> > other hand, if it's IRV, then IRV's chaos may deprive the candidate of
> > its rightful victory, and even if it wouldn't, people can only vote for
> > the third candidate that would become the winner as one of
> > many, not as one of two.
>
>I'm afraid I cannot follow your argument at all.  The whole point 
>about TTRO is that a strong third-preferred candidate, who with
>IRV or Condorcet might come through to be the eventual winner, is 
>dumped at the first stage by TTRO rules.  That is almost certainly
>what happened in the 2002 French Presidential election  -  and that 
>"defeat" ended the political career of the politician concerned.

But, of course, the same can happen with IRV, we need only posit a Le 
Pen with some broader support, such that vote transfers don't alter 
relative positions. In nonpartisan elections, this is actually the 
most likely scenario. But that election was very close in the first 
round, so, even if it had been nonpartisan, it could have gone either 
way, the difference was down in the noise.

The argument is not that TTRO is "better" than Condorcet methods; 
when the first round is Plurality, it's obvious that it can make a 
poor choice of the top two. (Essentially, it's making  a poor choice 
of the second candidate, once the frontrunner is eliminated -- and 
all the voters who voted for that candidate as well. Batch 
elimination IRV, exactly.)

IRV *sometimes* will find the compromise winner, but it's not 
reliable at all for this. It's really closer to TTRO than Condorcet 
methods. A Condorcet method *must* find Jospin in a situation like 
that which was faced, assuming that voters express the necessary preferences.

If the defeat in that election ended the political career of Jospin, 
it was ripe to end. However, it wasn't Jospin's fault that the method 
did what it did, and neither was it a fault that the French desire a 
majority result. Rather, they simply have used a limited method for 
the primary. Rather, consider using two-winner STV for the primary. 
If a true majority is found for one candidate, done. If not, then 
runoff between a better top two.

It can be done better and cheaper, but if that's what people want 
.... The sin is in leaving behind a majority method for one which 
would demolish the French multiparty system, probably, IRV.

IRV is being sold in the U.S. as finding a majority without the need 
for runoffs, but that is one of the most deceptive arguments in this 
field. It "finds a majority" by simply eliminating the vote of every 
voter who didn't vote for one of the top two. As I've pointed out, 
with a procedure like this, you can claim election by unanimity, just 
carry the IRV procedure one step further. That's not a "majority of 
votes cast," as the San Francisco voter information pamphlet for the 
proposition that brought them RCV, their name for IRV, claimed. It's 
a majority of ballots "containing votes for continuing candidates," 
i.e., excluding those who voted only for candidates other than those. 
Their sin? Assuming that they could vote sincerely, and only vote for 
candidates they liked.

(The latter problem still exists with full ranking allowed.)

>And just to be clear, in the examples I gave we are not dealing with 
>TTRO that started with only three candidates .  In the 2002
>French Presidential there 16 candidates.  And some of the mayoral 
>elections in England also have large numbers of candidates  -  one
>immediately to hand had 14.  I think in these circumstances I would 
>prefer the risk of some lower order chaos with IRV exclusions to
>the high risk of excluding of the most-preferred candidate with TTRO.

But the problem is not the real runoffs, it's the primary method. 
Want IRV? Use it for the primary method. But you could accomplish the 
same effect much more cheaply and more reliably. Bucklin.

Mr. Gilmour, you really should take a look at U.S. IRV elections, 
which have almost entirely been nonpartisan. With sometimes over 
twenty candidates on the ballot. IRV in that environment, functions 
like plurality. There have been *no* comeback elections before 
November 2008; and in this most recent election, I think there was a 
comeback, but it was a partisan election, where these things happen 
more often. French 2002 was very close in the primary, between all 
the frontrunners, actually, so you can argue that IRV would have done better.

However, if you also look at runoff elections in the same places, I 
looked at data for Cary, NC and San Francisco, runoffs were 
"comebacks" about one-third of the time. There are various 
explanations for this, but the lack of such elections with IRV is a 
strong indication of possible Condorcet failure. You may "prefer the 
risk," but you don't seem to have any idea of how strong a risk it is.

And you are making an artificially defective comparison. TTRO has 
major benefits that aren't present with IRV, which involve the 
reconsideration of the candidates by the electorate. There is no 
doubt that the final round is a true majority result, with the 
restricted set, and it's possible to *not* have a truly restricted 
set, but only ballot exclusion.

Essentially, TTRO has an obvious flaw, which can strike under 
conditions like that in France 2002. But that flaw can also strike 
with IRV, and it is not what would reasonably be considered rare. 
TTRO can be fixed or made better by quite a few different techniques, 
which include:

(1) Use an advanced method in the primary. In partisan elections, 
probably a method which avoids most strategic voting by voters, which 
could be Bucklin or 2-winner IRV. Both of these would terminate with 
a majority vote for any candidate. (I'm presuming that there isn't 
obligatory ranking and that the method does allow unrestricted 
bottom-ranking.) Approval is the simplest primary fix and, for sure, 
it would not make matters worse, and it's free. Same ballot can be 
used, except for the instructions. Any of these methods would reduce 
the need for runoffs, but not eliminate it. Bucklin would almost 
certainly be the most efficient at reducing runoffs.

(2) Use a better method for the runoff. Either allow more candidates 
into the runoff -- which would create the possibility for a third 
runoff -- or allow plurality election in the runoff, but with 
write-in votes. This allows a *major* Condorcet failure, accompanied 
by significant preference strength, to be corrected, and such 
corrections have happened in the U.S., which sometimes allows 
write-ins in all elections, including runoffs. If, of course, the 
preference strength involved is small, few will bother, and be 
willing to risk a spoiler effect in the runoff.

(3) Conduct the runoff using Asset Voting. I.e., each candidate 
receiving votes in the primary -- which could then be vote-for-one, 
though there is a neat interpretation that avoids considering 
overvoted ballots spoiled, then has those votes to recast in an 
"electoral college" runoff. The candidates are electors, holding 
various numbers of votes reflecting their primary vote strength, 
which they vote publicly. Jospin would have won, I'd predict, with 
practically no fuss. Of course, he'd have had to negotiate with some 
of the smaller parties. That's a good effect, and would give them 
real power, reflective of their true vote strength.

Of course, the best solution here is Asset; is it a down side that 
this could radically transform the entire system, shifting it toward 
direct democracy (i.e., democracy by chosen proxy)?

This would be, of course, using a form of proportional representation 
in the Presidential election process. Asset Voting was invented, 
first proposed as far as we know, by Lewis Carroll, as a tweak to 
STV, to avoid discarding exhausted ballots. The world would be a 
different place if the Australians had used this instead of an 
alternate, and quite undemocratic fix, requiring full ranking, which 
essentially extracts noise from many voters and uses it in 
determining the winners, thus making the voice of minor parties 
largely moot. That's a political choice, but certainly not a 
democratic one. Carroll knew that this was noise, but he realized 
that these voters knew who their favorite was, the candidate they 
most trusted, so he suggested utilizing that.

There have been proposals to use something like this for the U.S. 
electoral college, and it would shift the college toward its original 
intention, but that's politically very difficult. It's too bad that 
the National Popular Vote movement didn't realize that there was a 
better way than attempting to turn the U.S. Presidential contest into 
pure Plurality!

(I've described elsewhere how it would be possible to move in that 
direction *immediately,* state by state, without waiting for an 
electoral college majority before implementing it. Basically, a state 
would decide to support a representative electoral college, and would 
phase that in, using its full electoral weight to rebalance the 
college to reflect the popular vote. It would not be the mindless 
"proportional assignment of electors for the state," such as the 
Republicans attempted to set up for California, which would be 
political suicide for the majority party in the state involved. It 
would be a compact, of sorts, but one not dependent on other states 
for it to work. If we were serious about electoral college reform, 
we'd go for something like this.)




>James Gilmour
>No virus found in this outgoing message.
>Checked by AVG - http://www.avg.com
>Version: 8.0.176 / Virus Database: 270.9.13/1827 - Release Date: 
>03/12/2008 17:41
>
>
>----
>Election-Methods mailing list - see http://electorama.com/em for list info