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

[Abd ul-Rahman Lomax]

At 05:53 AM 11/26/2008, Greg wrote:
> > Oh, and actually it _is_ likely to be bad. See that first graph? See how
> > over thousands of simulated elections it gets lower social satisfaction?
>
>Brian, you're graphs are computer-generated elections that you made
>up. They aren't actual elections that took place in practice, which
>show a high unlikelihood of being bad. When your theory is a poor
>predictor of the data, it's time to change the theory, not insist the
>data must be different from what they are.

The Range simulations that Smith did are different from the Yee 
diagrams, though Yee diagrams, independently developed, show similar 
things. The Range simulations are based on voter preference profiles 
and candidate positions that are designed to be roughly realistic. 
And thus, with them, it becomes possible to make predictions about 
the frequency of occurrence of election pathologies.

The problem with "real election data" from "actual elections" is 
twofold: first of all, such data is only available for very few 
elections, comparatively, and it is quite tedious to collect. 
Secondly, we don't have the internal preference profiles on which, 
say, Condorcet failure would be judged. San Francisco data, for 
example, only collects the top three preferences. One might need to 
go deeper. There is no way to know preference strength data to know 
if what might be called normalized and summed social utility failure 
is taking place, i.e., failure to select the optimal candidate from 
the point of view of overall voter satisfaction.

We can do all this with the simulations. Simulations, properly done, 
should predict real election behavior, at least to some extent. That 
work is yet to be done. However, at this point *the simulations are 
all we have.* We don't have any other reasonably objective way of 
measuring election method performance.

Measuring. Basically, Greg, you've just got your mouth and a pile of 
opinions, no proof at all that IRV, in the U.S., replacing 
nonpartisan top two runoff races, is any better, and there is pretty 
good evidence that it is worse. It is choosing winners who, in a real 
runoff among the top two (and the IRV winner would be among the top 
two, we can be practically certain of that, since moving from third 
place in first preference to a win simply does not happen in IRV 
elections in Australia), would lose. I'd call that a poor result, one 
which, by the principles of majority rule as long understood and 
practiced, would be rejected.

So if you think we should look at real elections instead of 
simulations, why don't you. Take a look at them, take a look at all 
the IRV elections that have been held in the U.S. since San Francisco 
was duped into implementing IRV. (Yes, duped. The voter information 
pamphlet gave the voters false information, and, unfortunately, none 
of the opponents picked up on it. We have a lot of work to do! It's 
possible to argue that the voters would have approved IRV anyway, but 
what is really important is that they did not do what many were 
thinking they were doing. They did not implement a system that 
supports majority rule, they did not implement a system that 
guarantees a majority, far from it. The IRV code change actually 
struck the majority requirement from the code. But the voter 
information pamphlet said that the winner would still be required to 
gain a majority. Now, few people, encountering IRV for the first 
time, are going to realize the problem with exhausted ballots, they 
will have no idea that so many votes will be discarded and the 
ballots treated as if the voter hadn't voted. That's *highly* 
offensive to the concept of majority rule and the importance of 
majority decisions. The name of Robert's Rules of Order has been used 
in vain by IRV promoters, who claim that RRO "recommends" IRV for 
mail voting. In fact, what RRO describes is IRV, but with a true 
majority requirement ("or the election will have to be repeated").

Naturally, they don't call what they describe "instant runoff." 
Because it isn't. It's the *method* of IRV, but with that critical 
rule: a majority of ballots cast must contain a vote for the winner, 
revealed. What they are actually recommending, by the way, is 
"preferential voting," and they simply describe the IRV method "by 
way of illustration." Because there is a simpler and better method 
which is more likely to find a majority than IRV, because it counts 
all the votes if it's needed, I've been asked, "Why, then, did they 
not describe that other method, why did they describe IRV, smarty pants?"

Robert's Rules of Order is a manual of practice, reflecting actual 
practice. The better method of preferential voting mentioned hasn't 
been used in this country for many years, it's Bucklin Voting, which 
was very, very popular. It's a monotonic method, it is really instant 
runoff Approval, it finds majorities more efficiently than IRV, it's 
far easier to count, being a simple sum-of-votes method, and it has 
better performance in the simulations. Consider this: it was widely 
used. No examples of failure known, i.e., it was not rejected because 
of poor results. Much cheaper to count, can be run on ordinary voting 
machines. Fixes the spoiler effect. It's quite like Approval except 
it does allow preference ranking and lower ranks are only used if a 
majority isn't found with the higher ranks.

Bucklin is not in use. I'm unaware of any recent examples. The 
editors of RRO were obviously aware that there were many forms of 
preferential voting, but they were constrained to describe one 
example. Which they then comment is problematic because the method of 
sequential elimination can eliminate a "compromise candidate." For 
that, read "Condorcet winner."

This is, in other words, Center Squeeze, and that is *not* a rare 
phenomenon; we've seen it many times with Top Two runoff: the Lizard 
vs. the Wizard election in Louisiana, the French presidential 
election in 2002 where the probably Condorcet winner was narrowly 
bumped down to third place by Le Pen, into the decimal percentages, 
with both getting about 19% and Chirac only a few percent more. (Lots 
of candidates!). Jospin would almost certainly have beaten Chirac, 
the first-preference frontrunner. When the runoff came, Chirac 
slaughtered Le Pen, so to speak, with 80% of the vote.

IRV is sold as a simulation of runoff voting, and some 
implementations do exactly that: batch elimination of all candidates 
but the top two, then vote transfers from lower preferences of 
eliminated candidates.

Both IRV and Top Two Runoff are, on the face, vulnerable to Center 
Squeeze, and there is no reason I've ever seen to think it will be 
more rare with IRV than with TTR. So we have lots of examples to look 
at to see actual behavior. Center Squeeze takes special conditions, 
though. It takes three major parties, at least. Most IRV elections 
have been held under two-party conditions, and, in spite of the 
claims of IRV supporters, it appears that IRV supports and makes 
two-party systems more stable. Yes, you really ought to look at real 
IRV results and effects, instead of the pipe dreams FairVote has 
smoked up for you.

Greg, did you know all this? Be honest!

Did you know that Robert's Rules of Order doesn't recommend what is 
sold as IRV or RCV in this country?

That Top Two Runoff is much closer to what Robert's Rules of Order 
prefers? (They prefer no candidate elimination, and some variations 
on TTR allow write-in votes, thus they are closer to parliamentary 
procedure than any method limited to one ballot can be)

That Center Squeeze is not uncommon whenever there are three parties 
with the largest number of first preference votes, and that RRO 
specifically criticizes sequential elimination preferential voting 
for this? In other words, if IRV were to actually encourage third 
parties, it would then create the conditions for its own failure?

That much better, simpler, cheaper, and more reliable methods exist 
and one of them, Bucklin, has seen more use in the U.S. than the 
recent experiments with IRV?

Now, one argument you made was true, or at least represents the 
intention of FairVote. You argued that IRV would pave the way for 
proportional representation. That's probably FairVote strategy from 
the beginning; they started as the Center for Proportional 
Representation, morphed into the Center for Voting and Democracy, and 
then to FairVote. The original goal was PR, for most people involved 
in the early meetings. IRV is a band-aid, PR is a serious change, a 
major reform. IRV in the U.S., in nonpartisan elections, only rarely 
differs in result from Plurality. It differs in result from Top Two 
Runoff, in possibly about one election out of ten.

But the common (and generally good) method for doing PR is Single 
Transferable Vote, with its complex and difficult vote-transfer 
schemes. FairVote realized that they couldn't get this in one step. 
At some point the term "Instant runoff voting" was suggested to 
Richie -- we have the testimony of the person who suggested, who 
later came to regret it, he does not support IRV -- and he realized, 
I'm sure, that this gave him a toehold. IRV could be sold as a 
cheaper, more convenient substitute for real runoff voting. Did he 
care that it is *not* a simulation of real runoff voting, that by 
making this change, the right of the people to choose was being 
lessened? I don't think so. I'm not sure that, at that point, anyone 
was aware of the difference. But we know now. Real runoffs are far 
more powerful a reform than *any* deterministic method!

But FairVote's goal isn't better single-winner elections, not really. 
If it were, there are much better methods! Haven't you noticed, Greg, 
that those who really understand voting systems, who have studied 
them for many years, almost universally don't like IRV? IRV uses a 
complicated system of candidate elimination and vote transfers that 
makes it violate quite a number of voting system criteria, and, at 
the same time, satisfy a highly questionable criterion -- read 
Woodall in his original description --, Later No Harm, which strikes 
at the heart of the principles of democratic compromise.

FairVote's goal is now a bit unclear. I think it is largely to win. 
Whatever. The long term goal may remain PR, and there is, indeed, an 
STV-PR element to the Minneapolis situation. That is, if it passes 
constitutional muster (on the face, it most certainly does not, 
several independent attorneys gave that opinion, against an attorney 
affiliated with FairVote, plus, hey, I've read the whole Brown v. 
Smallwood decision and have studied it carefully. FairVote has 
extensively lied about it, misrepresenting it by quoting it out of 
context. If the attornies for FairVote and the City of Minneapolis 
rely on that argument, they may be sunk. However, Brown v. Smallwood 
was deeply flawed, self-contradictory, and not supported by any other 
similar decision elsewhere. They were aware that they were violating 
precedent, that other courts had found preferential voting to be 
valid. B v. S should be reversed. FairVote, though, isn't going for 
that. Instead they are trying to argue that B v. S is valid, and was 
really about Later No Harm -- this very recently invented criterion 
that they find dicta supporting in B v. S -- and because the method 
involved was Bucklin voting, not single transferable vote, they argue 
that IRV isn't covered by B v. S. In other words, they are arguing 
that all other advanced voting methods, including Condorcet methods, 
Approval, Range, Bucklin, should remain illegal and that only 
sequential elimination be allowed. Poisonous. Narrow self-interest, 
for narrow political advantage. It's what we've come to expect from 
FairVote, unfortunately.

(Brown v. Smallwood actually outlawed any form of alternative vote, 
allowing voters to vote for more candidates than there are winners 
for the office. They mention this not just once, and it is reaffirmed 
in the court's response to the request for rehearing. In order to 
take the position they take, they have to seize on a possible meaning 
of one sentence and ignore the rest of the decision. I wish I could 
be confident that the Minnesotata courts will see through this.

Note that I support allowing Minnesota to proceed. I'm worried only 
that a precedent will be established in Minnesota, as a result of how 
FairVote is playing this, that could prevent further reform there. 
IRV is no reform at all, apparently, when applied to nonpartisan 
elections where runoff voting was being used. It's a step backwards. 
(Literally: in some places in the U.S. that used IRV for a time, it 
was replaced with top two runoff. Do we ever think of looking at 
history so we don't make the same mistake more than once? I'm not 
sure it was a mistake, but at least we should look! Runoff voting was 
long considered a major reform. IRV, in actual practice in 
nonpartisan elections, regresses to give Plurality results, that's 
what real elections are showing. I estimate something like one 
election out of ten, Top Two Runoff would give a different result, 
and it is almost certainly a better one. The cost is roughly three 
runoffs to get one "comeback election." The other two confirm the 
plurality result from the first round.

Again, Greg, did you know this?

However, there are better ways of doing PR. One of them was a tweak 
on Single Transferable Vote invented by Lewis Carroll (Charles 
Dodgson) and published in a pamphlet in about 1883. Warren Smith, of 
the Center for Range Voting, reinvented it early in this century and 
called it Asset Voting, a metaphor that was also used by Carroll. The 
big problem with STV is exhausted ballots, and the problem is 
intrinsic. The Australians mostly deal with it by requiring full 
ranking, which will never fly in the U.S. and it is actually quite an 
undemocratic idea, coercive. You *will* vote for all candidates 
except for one, thus legitimizing the election by an absolute 
majority -- sounds good, eh? -- even if you detest the candidate and 
would much rather see the whole election thrown out. When a real 
majority is required without coercion, that's exactly what happens if 
a majority of voters don't cast a vote for the winner. Preferential 
Voting systems don't change that, they merely make it possible for 
voters to cast alternative votes.

Single Transferable Vote is a pretty good method for PR, but it gets 
much better with the Asset tweak, which, it turns out, if I'm 
correct, opens a whole new door into democratic possibilities. But at 
the start, it is this simple: if a ballot is exhausted, the vote it 
represents comes, as Carroll put it, "as it were, the property of the 
first preference candidate on it," who may then reassign this vote at 
will. It turns the resolution of the election into a deliberative 
process, and deliberative processes are far more democratic than pure 
aggregation, relying on aggregation alone is a method for keeping the 
masses disempowered.

Immediate effect: no wasted votes. You can vote with absolute 
sincerity for the candidate you most trust. It doesn't matter if that 
candidate will win or not. You don't have to rank more than one, 
unless you want to. (If you rank, then your vote is transferred 
unless and until exhausted, as you have directed.)

Used for PR, it can result in a totally and fully and accurately 
representative peer assembly, where every voter (almost, with some 
tweaks it can be just about every) represented by someone they 
*chose* or someone chosen by  someone they chose. The only way to do 
it better -- and it's not much better, if it is better at all -- is 
with delegable proxy and thus variable voting strength in the 
Assembly (that was also an idea which came up in the U.S. early in 
the last century).

There are other methods, such as Reweighted Range Voting or 
Proportional Approval Voting, but Asset is such a spectacular idea, 
yet so simple, that we should be looking at it.

You could get very good PR, using Asset Voting, with a standard 
Plurality, vote-for-one ballot. Carroll invented Asset because he 
realized that most voters couldn't rank a whole list of candidates 
based on knowledge: in Australia, they use Robson Rotation to even 
out the effect of the very common "donkey voting," where voters just 
mark down a list of candidates in the sequence listed, to satisfy the 
requirements of law -- otherwise their vote is voided as "informal."

Where they use optional preferential voting in Australia -- i.e., 
more like what we do here, except that full ranking is allowed but 
not required any more, they change the election rules to no longer 
require election by an absolute majority, but by a "majority of 
continuing ballots containing a vote for a candidate not yet 
eliminated," or some such language. And they see *lots* of exhausted 
ballots. Voters do not want to rank everyone, nor should they be 
required to. Carroll realized that most voters had little trouble 
identifying their favorite. So why not let them vote that way, and 
use the favorite as a proxy for the voter?

(The same idea was proposed by Mike Ossipoff in the 1990s, under the 
name Candidate Proxy. As to myself, when I first heard about Single 
Transferable Vote, I asked myself, who controls the transfers, and I 
thought, must be the named candidate! In other words this isn't a 
really obscure, difficult concept, it's rather easily thought of. So 
why hasn't it been tried? I'll tell you. It's the same reason Bucklin 
Voting was outlawed in the U.S. The same reason that STV-PR was 
eliminated in New York. The same reason why the IEEE dropped Approval 
Voting. It's not that it wasn't working! There are powerful forces 
that do not want democracy. There are even election reformers that do 
not trust democratic process, that's why you'll never see FairVote 
open to democratic process and a change of direction and strategy. 
Democratic process is messy, allegedly inefficient, and might decide 
to abandon my fabulous ideas. Why should I build a successful 
organization and let the mere will of the people change it? It's a 
common phenomenon with political activists....)