[EM] FairVote on Robert's Rules of Order and IRV

[Abd ul-Rahman Lomax]

At 12:49 PM 12/14/2008, Steve Eppley wrote:
>Hi,
>
>I think Mr. Lomax missed the big point (though I agree he is right 
>to criticize Instant Runoff).  The big point is that the authors of 
>books on Robert's Rules showed zero awareness of the existence of 
>Condorcetian preferential voting methods--or perhaps they were aware 
>but their analysis was made before the technological age made it 
>easy to exhaustively tally all the voters' pairwise preferences--so 
>their "recommendation" of single winner STV preferential voting was 
>only relative to a few even worse methods.  Clearly, Condorcetian 
>methods have properties that are much closer to the properties of 
>the Single Elimination Pairwise method that RR advocates, because 
>Condorcetian methods are not subject to the criticism they made of 
>STV that it can easily defeat the best compromise.

This analysis is incorrect. Yes, they show no specific awareness, but 
the language they used was quite precisely crafted, surprisingly so, 
if they were not aware that other preferential voting methods did not 
suffer from the failure of the STV method. That is, they make it a 
criticism of the *specific method they have described*, which is STV. 
They have also mentioned that there are many forms of preferential 
voting. That they spent precious words -- this is a manual of 
practice, not a dissertation -- to make it clear that center squeeze 
was a specific problem of "this method," i.e., the one they describe, 
indicates to me that they were quite aware that this wasn't a 
universal problem with preferential voting.

You have missed something else. RR does not recommend single 
elimination pairwise. They recommend, indeed *require* by default, 
repetition of the election, until a majority is found. There is no 
candidate elimination. It's true, though. The RR method -- election 
repetition -- together with associated rules, is an approximately 
Condorcet compliant method. The deviation is, in fact, a Range-like 
effect. When a proposed candidate is "close enough," i.e., the 
general preference for the Condorcet winner is low enough, the 
process terminates. People would rather finish with the election than 
seek any more improvement in satisfaction with the result. If there 
is some group of voters who strongly oppose this, they will attempt 
to prevent it, they will attempt to wheel and deal to come up with 
some better compromise. It's when the remaining preference strength, 
possible improvement, is lower than the perceived cost of continuing 
the process, that it terminates. With the explicit consent of a 
majority for the result.

I'm told that the reason they didn't describe other voting methods is 
that those other methods, at the time, were not in common use, and 
they still are not. They are a manual of actual practice, and it's 
remarkable that they said as much as they did. In any case, they 
clearly think that the practice of repeated elections is superior to 
IRV, and that using this *even with a majority requirement* is 
deficient compared to repeated elections. That's because, if voters 
do fully rank, a majority may be found which is *not* the compromise winner.

But they don't seem to have realized that truncation is a reasonable 
voter strategy in Center Squeeze conditions. And when the election 
must be repeated, the top-two failure is irrelevant, or almost so.

>(Approval can easily defeat the best compromise too, because many 
>voters will fail to approve compromise candidates out of fear of 
>defeating preferred candidates, which in turn will deter potential 
>candidates from competing.  If Mr. Lomax likes Approval due to its 
>cheapness and simplicity, I'll point out that the family of voting 
>methods known as Voting for a Published Ranking are as cheap as 
>Approval, easier for the voters, some methods of the family are as 
>simple, and if I'm right about how candidates would behave would 
>tend to elect a good compromise.)

Published ranking is interesting, for sure, but Approval is far, far 
simple and far less radical. Bucklin, in fact, addresses that 
reluctance. Unstated here was how the published rankings would be 
used. Condorcet? Bucklin is simpler, but when we are dealing with 
published rankings, we need only collect those votes en masse, and 
then applying them to a Condorcet matrix would be simple.

However, politically, it's, shall we say, a step. Count All the Votes 
is a small step, *and* cheap. And quite surprisingly powerful, 
considering. Bucklin has been used, and this might make it easier to 
bring it back.

The behavior of Published Rankings is unknown. There are a *lot* of 
questions, some of them quite difficult to answer. I'd prefer pure 
Asset; candidates could certainly publish their own Range ballots 
regarding other candidates, but I suggest that encouraging voters to 
select for trustworthiness, which covers a lot, is the best way to 
proceed to reform elections, and Asset has legs. It should be able to 
walk, one step at a time, all the way to full, highly accurate 
proportional representation, continuous democracy (no fixed terms of 
office, but, naturally, regular elections for electors).

>It would be worthwhile, I think, to reach out to recognized experts 
>in Robert's Rules and teach them about better voting methods, and 
>then see what they recommend.

It's an error to assume they don't know. They are not voting systems 
theorists, they put together a manual of actual practice. It's quite 
possible that in the next manual, there will be some description of 
Approval, for example, because there are some major organizational 
implementations.

>Another deception by the IRVings is their widespread claim that IRV 
>eliminates spoiling.  It's an even bigger deception, much more 
>important.  A variation of IRV that permits candidates to withdraw 
>from contention after the votes are published, before the votes are 
>tallied, would be much better at eliminating spoiling and electing 
>the best compromise.

Sure. IRV eliminates, to a degree, the lower-order spoiler effect. 
I.e., minor party, no chance of winning, draws votes away from one 
major candidate, resulting in an election unsatisfactory to a 
majority. That, by the way, is an assumption. Nader, in 2000, claimed 
that voters who preferred him should vote for him because the majors 
were Tweedledum and Tweedledee, both shills for the corporations. If 
they believed him, then why would we think that they would add votes 
under IRV? However, in fact, voters are a bit more sophsticated and 
uncontrollable. Some of those who voted for Nader would have added 
ranked votes or additional Approvals for Gore.

Bucklin is what I recommend, as a first reform, beyond Count All the 
Votes (Open Voting or Approval). It addresses the big problem that 
most people give as an objection to Approval, but it is very much 
like Approval. It's roughly as efficient as Condorcet methods with 
social utility.

Ultimately, I prefer Range with explicit Approval cutoff, and 
pairwise analysis, and a runoff in the case of majority approval 
failure or a candidate who beats the Range winner by pairwise 
analysis. It's my contention, by the way, that a genuine, sincere 
Range winner would likely prevail in a direct runoff against a true 
Condorcet winner. And if you don't know why, ask!

When I first proposed this, some thought it preposterous, a result of 
single-ballot, deterministic thinking that the whole field of voting 
systems fell into.