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

[Abd ul-Rahman Lomax]

At 06:25 PM 11/26/2008, Ralph Suter wrote:
>To Greg Dennis:
>
>I appreciate your efforts to express your arguments clearly and 
>defend them with good data. Nevertheless, I find them mostly unpersuasive.

Yes, we noticed. That they were unpersuasive. That Mr. Suter comments 
on this is significant, given his history.
>Data from previous elections won't settle the IRV versus Condorcet 
>debate. There have not been enough of them in the U.S.

My sense from my study so far is that, actually, we can find a lot of 
information about IRV in the elections so far. But Mr. Suter is 
correct that it is unlikely to be persuasive. Even if we find, for 
example, Condorcet failure, it will be argued that the sample is 
small and that this was just a fluke. I've found strong evidence that 
IRV is failing to elect Condorcet candidates as tested in runoff 
elections, and it will be argued that runoffs are defective because, 
well, everyone knows that they have poor turnout which is obviously 
bad, right? (Wrong, but it's a common idea.)

However, there is already enough data from U.S. IRV elections to 
raise serious concerns. There is also now quite a bit of evidence 
about the difficulty of counting IRV when an "instant runoff" is 
needed. It is far, far from "instant," and it's expensive, especially 
if there is a need to audit the results.

>  More important, there haven't been any major federal or state 
> elections (presidential, senatorial, or gubernatorial) and very few 
> major local elections (mayoral or other) using IRV. These would be 
> far and away the most important kinds of test cases - i.e., the 
> kinds of elections that would matter the most and where voters 
> would be most familiar with all the candidates and therefore would 
> find it easiest to rank them.

IRV probably will look a little better from such elections, 
particularly where it replaced Plurality. The tragedy is that IRV is 
replacing Top Two Runoff, an older reform that actually works better 
than IRV. Both of these methods suffer from Center Squeeze -- for the 
same reason -- but TTR makes better choices in the last round (and 
not only rarely, in roughly one-third of runoffs). Further, it's 
fairly easy to fix the Center Squeeze problem, both so that there are 
fewer runoffs and so that a compromise winner is more reliably 
detected. In a word, use Bucklin for a primary. IRV with a true 
majority requirement is better than IRV without it -- this is what 
Robert's Rules of Order *actually* describes, not the IRV that is 
sold by touting that as a recommendation. (And RRO also notes the 
Center Squeeze problem.)

>I also must reject your contention that IRV is easier to explain. 
>Condorcet, or what I prefer to call IRRV (Instant Round Robin 
>Voting) is every bit as easy to explain as IRV. IRV and IRRV both 
>use the same kinds of ranked ballots. The main difference (setting 
>aside problems involved in permitting or disallowing equal ranking 
>and unranked candidates) is that IRV uses the ranking data to 
>simulate a series of runoff elections whereas IRRV uses the same 
>data to simulate separate 2-person contests between each candidate 
>and every other candidate. There's no need to talk about matrices 
>and other technicalities about data storage and calculation. Using 
>the same kinds of simple examples, it's just as easy to explain how 
>IRRV works as it is to explain IRV. And although the possibility of 
>cycles makes explaining IRRV more complicated, other kinds of 
>problems make explaining IRV similarly more complicated.

It's easy to explain the basic Condorcet principle. It's not so easy 
to explain about Condorcet cycles and resolution methods. It's also a 
problem that Condorcet methods -- like *all* single ballot methods -- 
can elect by a plurality, even a small one. There is no doubt but 
that a Condorcet winner is better than an IRV winner, but, of course, 
IRV advocates argue that IRV chooses the Condorcet winner most of the 
time. What's "most?" In partisan elections, in a strong two-party 
system, nearly always. (Two-party systems are what makes Plurality 
work reasonably well.) But in the nonpartisan elections where IRV is 
replacing TTR, my estimate is one election out of roughly ten, there 
is Condorcet failure.

Again, there are only two ways to fix the election-by-plurality 
problem, and one of them is not only a Bad Idea, it's impossible in 
the U.S., fortunately. That way is to require full ranking. This is 
amount to requiring all voters to vote for every candidate but one. 
So the "majority" is coerced. The other way is further process. Very 
little study, to my knowledge, has been done on this among voting 
systems experts, who have focused on deterministic methods, seeking 
the Holy Grail of the best one. It's been found, or at least a major 
characteristic of it has been found, but the single-ballot 
restriction is serious. Why don't direct democratic bodies use 
advanced methods to decide multiple-choice questions? Because the 
single-ballot restriction is a terrible limitation, sometimes, when a 
majority hasn't been found. In a deliberative body, they will 
continue working on the problem until they find something better, 
something a majority will accept. Or no decision is made.

Tell me, should we boil the Republicans in (1) Oil, (2) Milk, or (3) 
Water? Let's use IRV. We've decided that the Australians really know 
what they are doing, so, (a) It's against the law to not vote, (b) If 
you don't rank all the candidates, your ballot is discarded. All, 
right, folks, what are the results! Amazing, isn't it? We have a 
majority for one of the choices.

In a decent election system, if you dislike all the candidates, you 
can write in another and cause majority failure. So, perhaps, you 
might, if you are a yellow-bellied Republican-lover, or maybe for 
other reasons, write in "Nothing." Might even win. But if you haven't 
been able to communicate with others who think the same way, you 
might choose different other options, or at least different names. It 
doesn't matter. There will be majority failure. (And this points out 
why write-ins should be allowed in runoffs, as they are, by default, 
in California and maybe in some other places.)

Runoffs with write-ins present the voter with a reduced set of 
candidates, but they do not actually eliminate any candidates. 
Determined voters can write the name in and may even win (as the 
Mayor of Long Beach, California, did in a recent runoff election). 
Usually the top two, even if not the best, will include the 
compromise winner, and, if not, the preference strength violated is 
likely to be small. Only in unusual situations will voters be 
exercised to participate in a write-in campaign.

Runoff voting is a very important reform! Yes, it has a cost. 
Democracy has a cost: participation.

>I realize that many IRV supporters are fond of dismissing compromise 
>candidates as "bland" and with "little core support."

Yeah. Imaginative rationalization. "Little core support" is correct, 
but means very little. "Core support" is a standard that was made up 
to justify IRV, though it is also used to justify Plurality. If Core 
Support is so important, why do we need IRV? *Usually,* it turns out, 
"core support" correlates with overall support. But as has been 
pointed out by another writer here, there are reasons why it could 
happen otherwise, and it is not a defect of the candidate, it is a 
product of the overall configuration of the political system and 
political parties. Political parties create a lot of inertia, and 
when one party moves too far to one side of the spectrum, and the 
other does not respond by moving toward the center, a centrist 
candidacy becomes possible, one where the candidate is in better 
agreement with the majority of the electorate; but this candidate is 
at one end of the spectrum in both major parties, and may easily fail 
to win a majority in either. So the candidate must run as an 
independent or third party candidate. And this party is almost 
certain to be not as well organized as the major parties. Further, it 
faces a huge problem: very many voters vote habitually by party 
affiliation. Getting all the way to the top, in an environment like 
that, is extraordinarily difficult. That's what's necessary with IRV. 
With Top Two Runoff, the candidate must make it only to second place.

This is why top-two runoff is, outside the U.S., associated with 
strong multiparty systems. IRV isn't.


>By the way, I happen to be fairly knowledgeable about the history of 
>FairVote, the leading organizational advocate of IRV, having 
>attended the organization's founding meeting in June 1992 when it 
>was originally named Citizens for Proportional Representation (CPR). 
>At the time, reforming single winner elections was barely on the 
>organization's radar screen. I'm not sure it was even discussed at 
>the founding meeting. However, a short time after the meeting, 
>former presidential candidate John Anderson got an op-ed published 
>in the NY Times arguing in favor of "majority preferential voting", 
>the name he then used for what is now called instant runoff voting, 
>and soon after that he joined CPR's board. (He wasn't at the 
>founding meeting.) You can read Anderson's op-ed at:
>http://query.nytimes.com/gst/fullpage.html?res=9E0CE7D91F3FF937A15754C0A964958260
>
>A short time later, the organization's name was changed to Center 
>for Voting and Democracy (CVD). Anderson no doubt had something to 
>do with getting CVD more focused on reforming single winner 
>elections, but it was not until 1996 or so, four years after the 
>founding meeting, that CVD began taking the single winner election 
>issue very seriously, and it took another year or two for it to 
>decide on Instant Runoff Voting as the name for its favored single 
>winner method.

Mr. Suter has explained this to us before, but it's good for it to be repeated.

>I'm much less troubled, however, by the slowness of CVD to focus 
>seriously on single winner elections than by how it decided, with 
>virtually no public or even inter-organizational debate, to go with 
>IRV and to reject alternative voting methods, even though CVD at the 
>time had some notable supporters of other methods on its advisory 
>board. (That board included a number of highly respected voting 
>methods experts and political scientists, including Steven Brams and 
>Arend Lifphart, whereas the FairVote advisory board today has no 
>such people.) I personally asked CVD's executive director Rob Richie 
>to allow a debate at its 1997 national meeting between IRV 
>supporters and supporters of alternative methods, but he declined to 
>do so, and he has continued to refuse to sponsor or (to my 
>knowledge) participate in any serious public debates.

How is it that one person -- or a tight group -- could make a major 
decision like that? Essentially, because nobody in the group knew how 
to put together an alternative to the CPR, which was organized quite 
traditionally. The people with the money were used to being able to 
control what happened with it, and using the traditional self-elected 
board form allowed them to insure that the initial direct was as they 
chose. It is a totally common situation.

Richie began discussing alternatives, to some degree, with supporters 
of other methods only after, I think, these people started having 
some impact on IRV implementation campaigns. The internet is changing 
things. There are lots of discussions taking place in places where 
Richie can't control. What he does, though, is pretty much what could 
be expected. He knows a whole series of "sound bite" arguments: 
brief, easily swallowed. He, or a surrogate, shows up where a 
discussion is taking place, and he makes sure that the deceptive 
arguments are presented. To refute these arguments takes more words, 
you have to actually explain stuff, not just rely on knee-jerk 
assumptions that people hold. (That's what spin-doctors do, they know 
how to touch and use these assumptions.) And then, of course, Richie 
can say he doesn't have time for this endless argument, and he 
disappears. Each time, though, more people start to suspect that they 
are being conned.

I predict that IRV implementations will become increasingly difficult 
to get. And we may start to see some other experiments. There is an 
obvious one: Bucklin. Wouldn't it be interesting if Duluth decided to 
try again? They *loved* it when they had it. It's cheap, it's easy to count.

And, really, it's Open Voting, or Approval, turning totally into 
Approval if a majority isn't found from higher preferences. So it 
fixes the most common objection to Approval.

>Finally, for me there are bigger questions right now than whether 
>IRV is superior to IRRV. Although I lean strongly toward the latter, 
>I doubt that either will be really feasible for major national and 
>statewide elections until problems with how elections are conducted 
>are much better resolved so that elections will be more secure and 
>accurate and able to reliably handle voting methods as complex as 
>IRV and IRRV. Given the current state of U.S. election systems and 
>equipment, there is only one alternative to plurality that is now 
>really feasible across the U.S., and that is approval voting, which 
>is nearly as simple to implement as plurality voting.

Or simpler, actually. It's Plurality minus a rule. Period. The only 
people who need to vote extra votes are minor candidate supporters. 
But if people still want to express a favorite and avoid the failure 
of the Majority Criterion -- that's an other knee-jerk assumption, by 
the way, that Richie uses well -- there is Bucklin, which would work 
pretty well as two-rank, and three-rank, it can handle many more 
candidates than three-rank IRV -- the only thing we've had in the 
U.S. so far, in the recent implementations.

Yes, I've found that nearly all people familiar with voting systems 
agree that Approval makes a good first step, and, given that it costs 
nothing to speak of, and is highly unlikely to do harm, compared to 
the status quo, it doesn't impede further reform.

I personally favor using Range and analyzing it to see if any 
candidate beats the Range winner by pure preference. If there is such 
a result, or if the Range data shows majority failure (it must have a 
means of indicating "acceptance"), there is a runoff which includes 
the Range winner and a candidate who beats the Range winner. (This is 
quite unusual, by the way, Range usually chooses the Condorcet 
winner, from what we know in the simulations.)

This is a Condorcet-compliant method. If there is a Condorcet winner, 
that winner will be in the runoff, unless voters deliberately conceal 
their preference (and needlessly conceal it, by the way, it can be 
made cost-free to specify all preferences, but that's a detail). And 
all it takes for the Condorcet winner to prevail is that the voters 
in the runoff confirm it. But usually, I predict, they will not. For 
a number of reasons I won't mention now; basically, a Range winner 
has a natural advantage in a runoff, because this is the winner by 
preference strength, which correlates with voter turnout as well as 
with the likelihood of a voter to change his or her mind. On the 
other side, there are anomalies with practical Range voting that can 
sometimes distort, to a degree, the real utilities extracted from the 
votes. So sometimes the Range winner -- it's rare! -- isn't actually 
optimal. And if that is the case, the Condorcet winner is likely to 
prevail. The simulations show that runoff Range has better results 
than Range alone.

The Condorcet Criterion is intuitively satisfying, but the intuition 
turns out to be incorrect. The Condorcet winner is not necessarily 
the ideal winner, this is merely the most common case. "Condorcet" is 
based on pure ranking, which entirely neglects preference strength, 
and it's easy to show that this neglect makes it impossible to detect 
the optimal winner, the one who will most satisfy the voters, 
overall. It's easy to show examples where every reasonable person 
would agree that the Condorcet result is not only wrong, but 
seriously wrong, harmful.

That result is quite unlikely in real political elections, because 
pre-election process *usually* doesn't allow the situation where it 
would happen to arise, compromises are made before the election, etc. 
But given that it is actually easy to address these problems, if only 
we recognize the problem and address it, and that "Condorcet winner" 
is a valuable concept which can be incorporated into a good -- but 
still simple -- method, I see no reason to risk bad elections. The 
cost of them could be tremendous. Our unwillingness to address these 
problems brought us a disastrous eight years, now ending, where even 
conservatives recognized that a serious mistake had been made.

Open Voting. (Vote for as many as you choose. -- Approval Voting.)

Don't Eliminate Candidates, Consider Them All.

Every Vote Counts, Count All the Votes.


>Another of my concerns is that virtually all advocates of voting 
>reform, including supporters of IRV, IRRV, approval, and range 
>voting, have neglected the question of which methods are most 
>appropriate for different voting situations. IRV advocates sometimes 
>argue for using IRV in all kinds of situations, even in small groups 
>to decide such things as  what kind of food the group should order. 
>But for most small group purposes, approval voting would be much 
>easier and quicker to use and would produce more satisfactory results.

Yes. I've seen it used. Very easy, very efficient, didn't even need 
explanation. Just a series of options read out, and people were asked 
to raise their hand if they would accept the option. The Condorcet 
winner, from the outside, was one option, the status quo. Probably 
over 50% of people preferred that. And about 70$ accepted it. 
However, there was another option that was *unanimously* acceptable, 
except for one person, and this group valued group unity. A motion 
was immediately made to accept the most-widely accepted choice, and 
the vote on that unanimously accepted it. I think the person who had 
voted against it, at first, accepted it. This was really an example 
of how the hybrid system I've designed would work. The "Range winner" 
-- most broadly approved -- wasn't automatically selected. Open 
Voting was used, but it was known at the outset that the majority 
favored the status quo. The result was put for ratification, which 
proved that *after the result was known*, this had become a unanimous 
choice. (There were maybe seventy or eighty people at the meeting, it 
was packed. It had been expected to be highly contentious.)

The key in any small group is to verify that *any* election method 
has produced a decent result. A majority should always approve the 
outcome, or it is not a democratic one. If people stand aside or 
abstain, no problem. They don't care strongly, and the result isn't 
going to displease them. (If so, they should not stand aside!)

>  If more precision is needed, some form of range voting would still 
> usually be easier and better than either IRV or IRRV.

Bingo.

>It may also be that approval voting or range voting would be better 
>for most but not all kinds of public single winner elections, 
>especially ones for lower level offices and primary contests where 
>most voters have difficulty acquiring enough information to do 
>confident rankings. It may be best to reserve IRV or IRRV or range 
>voting for a few of the highest level offices.

And the larger the election, the more difficult the IRV counting.... 
IRRV is easier, though with a lot of candidates, it gets tricky.

In the hybrid Range/Condorcet system I've described, the counting is 
easy. And it is not necessary to complete the matrix, all that is 
needed is the set of pairwise votes between the Range winner and the 
other candidates, or maybe just the top few.

By the way, the very concept of single-winner public elections for 
high office is flawed. It is far safer to elect such offices 
deliberatively, using a representative system; and the representative 
systems can, actually, be made fully and accurately representative, 
Lewis Carrol made the basic suggestion almost 120 years ago, we now 
call it Asset Voting. This could create a virtual Electoral College 
that totally represents all the voter, who then deliberate and 
negotiate and do whatever it takes to get a majority. And then the 
same college can undo what they've done if they, pending the next 
election, conclude that they made a mistake or conditions have 
changed. No term of office. Really, this is a parliamentary system. 
Do we imagine that any large business would hire a CEO for a fixed 
four-year term? The U.S. system created a restrained monarch with 
very substantial personal power and a term. It was just a tweak on 
what we were used to. Strong executive, not continuously responsible 
to the people. We can't fire him, impeachment is way too difficult. 
When the board of a corporation decides that a CEO isn't working out, 
they don't have to try and convict him -- unless they want to avoid 
paying on his contract -- they just hand him his pink slip.

I'm certainly not proposing corporate practice as idea, but ... if 
what we have as a public system were a good idea, wouldn't it be a 
good idea for businesses as well? It isn't, period. Bad Idea. But 
better than what we had before. We've just seen some of the damage it 
can cause. It could have been worse. Of course, Bush isn't gone yet. Pray.

And lets try to make sure it doesn't happen again. This was not 
Bush's fault. To the extent that it was anyone's fault, it was *our 
fault.* We let it happen, by tolerating a poor political system.

It's not about the Republicans, or the special interests, or the 
wimpy Democrats, or even Ralph Nader.... It's about the system, which 
grew like Topsy, and, big surpise, it's never been optimized, we 
don't even think, most of us, about what "optimal" would look like. 
That's part of what I'm trying to change.

>A new voting methods reform organization that seriously considers 
>these issues and allows for and encourages discussion as well as 
>research and experimentation about different methods and their 
>relative advantages and disadvantages in different kinds of 
>situations could conceivably transform the debate about voting 
>methods. It could result in much more effective efforts to get 
>better methods adopted, both for public elections and for all kinds 
>of other purposes.

I start all kinds of things, but I have a terrible time keeping 
things going, part of the same psychic structure that allows me to 
think outside the box and to see certain things that have generally 
been overlook also makes it very difficult to follow up on the ideas. 
In any case, I started the Election Methods Interest Group, some time 
back, and a number of experts and other interested parties did join, 
including some IRV supporters. The goal was to be such a forum, and 
to try to come up with consensus reports, or at least to measure the 
support for various ideas among a community of informed and interested people.

http://groups.yahoo.com/group/electionmethods/

This is an FA/DP organization. FA stands for Free Association. While 
EMIG may report the results of polls on controversial positions, EMIG 
itself will not take any controversial position. Joining it is Free 
in many senses. Further, DP means that a delegable proxy structure is 
set up. There is a Proxy Table. Naming a proxy on that table does not 
bind the member in any way, it merely is a means whereby members can 
allow another member to loosely "represent" them, we may use this in 
estimating a preliminary consensus, or, in the other direction, the 
proxy may inform them if there is some activity the member should 
know about, in the judgement of the proxy. It's a way for, among 
other things, someone who is too busy to watch mailing list traffic 
can still join and participate in an indirect way. The goal was to 
get some of the major experts to join, even if they don't have time. 
They would name a proxy as someone they trust to more or less get it 
right, or at least to let them know if something warrants their 
attention. We did find that some experts joined. Now we need to 
actually start using it....

This is an example of what a large consensus-negotiating and advisory 
organization might look like, a small and very low-cost experiment in 
that direction. Please don't depend on me to make it work. I'll try, 
but I am *very* unreliable.