[Election-Methods] RELEASE: Instant Runoff Voting

[Abd ul-Rahman Lomax]

At 07:06 PM 7/26/2008, James Gilmour wrote:
>Kathy Dopp  > Sent: Saturday, June 21, 2008 5:20 AM
> > "Later-No-Harm", however, is incompatible with the basic
> > principles of majority rule, which requires compromise if
> > decisions are to be made. That's because the peculiar design
> > of sequential elimination guarantees -- if a majority is not
> > required -- that a lower preference cannot harm a higher
> > preference, because the lower preferences are only considered
> > if a higher one is eliminated.
>
>The meaning of the second sentence isn't completely clear to me, but 
>I am fairly sure there is a perverse interpretation of
>"majority" in the first sentence.

Actually, the term in the first sentence is "majority rule," which, 
in actual operation, makes decisions always between two alternatives, 
minimized to Yes or No on a single question. Compromises are made for 
efficiency, with various degrees of damage to majority rule. The 
traditional method is very simple: a motion for an action, seconded. 
Discussion. Amendment, each amendment being treated as if it were its 
own motion, which it is. A calling of the question, typically by 
supermajority (2/3 for Robert's Rules, but it is always possible for 
a majority to bypass this, with some damage to collegiality and a 
general sense that the majority is playing fair). And then vote, and 
if a true majority of those voting is not obtained, the motion fails 
and the status quo continues.

Deliberative process, as described, breaks down if every voter 
insists on their first preference, without compromise until it has 
been proven beyond doubt that the first preference cannot be 
obtained. Generally, in deliberative process, there is no such 
elimination. What was rejected before can be recalled and considered again.

For Later-no-Harm to function, candidates must actually be 
eliminated. That doesn't happen under deliberative process. Now, for 
efficiency, there are voting methods that allow more than one 
candidate to be considered at a time. Plurality is one such. In 
deliberative process, there is no harm if the winner obtains a 
majority, in a Plurality election where voters are properly informed 
and make the necessary compromises as they vote. But it's tricky to 
do that optimally, so voters might compromise when they don't need to, etc.

Efficient methods would allow voters to indicate preference strength, 
and Range Voting is really the method that does this most accurately. 
If voters do this sincerely, the method will predict which candidate 
would enjoy the maximum overall satisfaction as rated by the voters 
themselves. (The question of sincere voting in Range is a complex 
one, with common assumptions being made that are actually 
self-contradictory, i.e., weak preferences expressed strongly. Why? 
If it is not merely an accident, it would be because the voter wants 
the favorite to win, enough to risk loss from not giving some 
preference strength to another pairwise election. But that's a strong 
preference!) But I would not consider a decision made by Range Voting 
to be a democratic decision *unless the approval of a majority for 
that outcome were explicit.*

IRV, as implemented everywhere I've seen it, will elect by plurality. 
That is, it will elect even though a majority of voters voted for 
someone else and not for the winner. That is, in case it needs to be 
said, a majority  who showed up and voted, who did not care to vote 
for the IRV winner, who, by the only means allowed that didn't 
require some other vote offensive to the voter, voted *against* the 
winner. IRV, as used, is incompatible with majority rule.

It could be made compatible, and the method is obvious, and is 
precisely what Robert's Rules of order describes as how it would be 
used. A true majority is required to win. IRV then becomes a method 
of finding majorities, provided that enough voters add enough ranked 
choices. If all voters rank all the candidates, a majority is 
guaranteed. But when full ranking is optional -- and it must be 
optional for the method to be fully democratic -- majority failure 
can occur, and thus, Robert's Rules notes, "the election must be 
repeated." They have in mind repetition with no elimination. What 
I've realized -- and I never saw this in print anywhere -- is that 
top-two runoff, in some places, actually doesn't eliminate any 
candidates, it merely restricts who is on the ballot and, it has been 
proven, this doesn't prevent a write-in from winning. So there is, 
technically, no elimination, and IRV under such circumstances would 
be a clear improvement over plurality, by eliminating some 
unnecessary runoffs. Turns out, though, that it would only eliminate 
maybe one-third of runoffs in a place like San Francisco. Bucklin 
voting, using the same ballot, and probably seeing the same voting 
patterns, would probably eliminate about half of them. Approval would 
probably do about the same.

>   An IRV election is an Exhaustive Ballot election contracted into 
> one voting event, instead of
>being spread over several rounds in which the one candidate with 
>fewest votes is eliminated at each round.

Exhaustive ballot is quite rare. Sure, IRV seems as if it would 
simulate exhaustive ballot, but this assumes fixed preferences and no 
shifting of preferences due to re-examination by voters of the 
reduced field. And the same electorate.

>   It is no surprise that
>the numbers of voters participating varies from round to 
>round  -  usually a progressive (or severe) decline.

The decline is common, but by no means universal. And there is a 
factor in the decline which has been overlooked, particularly by IRV 
promoters who are trying to disparage runoff voting, which is, by 
far, the most democratic voting system in common use. That extra 
election makes a very substantial difference in results, on the order 
of ten percent of elections, under nonpartisan conditions, the runoff 
will elect a different candidate than will Plurality or IRV, which 
almost always elect the same candidate.

What has been missed is that when voters don't have a strong 
preference between the remaining candidates, they will almost 
certainly turn out in lesser numbers. When they have a strong 
preference, they will turn out in increased numbers. Sometimes with a 
runoff election where there is strong preference for many voters, 
even though it is a special election (not with the general election, 
which would ordinarily depress turnout just on that basis), turnout 
can actually increase.

Generally, it could be said that the overall preference strength, 
usually, between the top two in a many-candidate field, is going to 
be lower than the range in the original election. Thus lower turnout 
would be no surprise.

Consider this: one of the major arguments against Range Voting is 
that supposedly voters will "exaggerate," voting strong preference 
when their preference is actually weak. I consider this an oxymoron, 
but when there are factors which encourage realistic expression of 
preference strength, we can expect more accurate expression. If an 
election is inconvenient, those who have a strong preference will 
vote more frequently than those with weak preference. This introduces 
a Range-like effect. Differential turnout will tend to favor 
genuinely strong preference strength and thus the results, we can 
expect, will be overall better than without this effect.

In any case, there is no evidence that voters are disatisfied with 
the results of top-two runoff when they would be satisfied with 
Plurality or IRV results (which are the same, almost always).

FairVote has concentrated, in its explanations of IRV, on partisan 
elections, where vote transfers tend to take place in a biased way, 
and which can then reverse the first-round preference and elect the 
runner-up. (IRV, in Australia, even with full ranking, almost *never* 
elects a candidate who was third place in the first round, it is 
extraordinarily rare, maybe a couple of examples in a century, out 
of, what, thousands of elections?)

But then actual implementations of IRV have been in places with 
nonpartisan elections, without such preferential vote transfers.

Take a sample of voters and ask them if they prefer A or B. Then ask 
these voters if they prefer C to both A and B. Look again at the 
preferences of the voters who preferred C and those who did not. 
Apparently, the ratio of A to B preference is about the same for 
those who prefer C and those who do not. In nonpartisan elections. 
Thus vote transfers in IRV from the elimination of C have little 
effect on the relative vote count for A and B. That's why IRV 
reproduces, in nonpartisan elections, Plurality results.

Bucklin would probably do the same thing, by the way, though I 
suspect that there would be a small increase in exceptions.

But top-two runoff introduces a new effect. When the main field is 
reduced to two candidates, voters take another look. Once it is down 
to A and B, maybe they preferred C and never really looked carefully 
at B. Maybe they didn't vote in the original election because they 
didn't think B could win -- but now B is in the runoff, even though 
trailing A in the first round. So they vote. For lots of reasons, 
voters will now express a preference that they did not express in the 
first election. Robert's Rules mentions this, I'm not just making it 
up! Preferential voting "deprives voters" of the ability to base 
later votes on earlier results.

Replacing top-two runoff with IRV will bring questionable savings in 
cost (highly speculative, it turns out, because of the generally 
substantial increase in voting costs with IRV), but clearly is a loss 
in democratic values. If we were to follow Robert's Rules 
recommendations, we might use IRV -- or another form of preferential 
voting that doesn't have certain IRV problems -- and hold a runoff if 
a true majority is not found. I'd say, though, that IRV would 
definitely not be the best choice for this, because of the expense. 
Bucklin is cheap, Approval is even cheaper (almost free, actually, 
just count all the votes), and both would avoid some runoffs. I 
predict that Bucklin, "instant runoff approval" would be the most 
popular. It *was* quite popular in Minnesota, the town of Duluth 
fought mightily to keep it, but, my guess, there were powerful forces 
arrayed against them. With the Minnesota Supreme Court ruling it 
unconstitutional, they'd have had to get a constitutional amendment, 
and that apparently wasn't going to fly.

(Note, by the way, that Brown v. Smallwood, the decision involved, 
was against any kind of alternative vote, and the FairVote argument 
that it only applied to Bucklin is a self-serving one, which has been 
repeated by no neutral legal opinion, whereas two city attornies in 
Minnesota, in formal opinions given to Minneapolis and another town, 
that B v. S did apply to IRV, and, after extensive study of the 
decision -- it's on the Range Voting site -- I've concluded the same. 
FairVote's attorneys seized on a piece of dicta as if it were the 
basis of the decision, which it was not. It is possible, though, that 
the Minnesota Supreme Court will reverse B v. S. Or, as well, that 
they will rule that it does apply to Bucklin and not to IRV, though 
this latter decision, like the original B v. S decision, would be 
unique, my guess is that they aren't going to do that. They will toss 
the whole thing out, or they will leave it in place. Since the 
original decision was a poor one, in my opinion, I'm, here, rooting 
for IRV, as it were. I think that voters have the right to set this 
system up, foolish though it might be.)



>   The votes in an
>Exhaustive Ballot election might look like this:
>
>Round 1
>A  4,000
>B  3,000
>C  2,000
>D  1,000
>Total voting 10,000
>
>Round 2
>A  3,500
>B  2,500
>C  1,500
>Total voting 7,500
>
>Round 3
>A  3,000
>B  2,000
>Total voting 5,000.
>
>A is the majority winner in Round 3, that is to say, the majority 
>winner of those voters then voting.   And IRV satisfies that
>criterion  -  and the Exhaustive Ballot is the valid comparison for 
>IRV (because that is the origin of IRV).  The only difference is
>that to ensure the integrity of the count (accounting for all ballot 
>papers at all stages of the count), the ballot papers (votes)
>of those who opt out at the later stages (rounds) are recorded as 
>"non-transferable".

These figures were, of course, simply made up. They are somewhat 
realistic, though they don't show the kinds of shifts that actually 
occur in real repeated ballot elections. The preference order was maintained.

Exhaustive ballot is not the origin of IRV. IRV originated as STV 
applied to single-winner elections. The "runoff" analogy is very 
modern; while it was used by an election official in Ann Arbor to 
describe how the method worked, that analogy wasn't taken up; 
"instant runoff voting" was invented in the early 1990s as a way of 
promoting preferential voting by the predecessor of FairVote.

"Opt out" at later stages is a gross assumption. In San Francisco, 
there are sometimes more than 20 candidates for a single office. 
There are three ranks on the ballot. If voters simply vote sincerely, 
many of them, even if they fully rank, will have an exhausted ballot. 
Further, as Lewis Carroll noted, more than 120 years ago, in an STV 
election, many voters really won't have much idea of how to rank 
less-known candidates. They know who they want to elect, presumably. 
Second choice, maybe. Third choice, less likely.

In top-two runoff, though, they simply vote for their favorite, first 
round. Then, if there is no majority, they have two major candidates 
to choose from. It was only decided quite recently, while IRV was 
pending implementation in San Francisco, by the way, that it was 
legal for the City to not allow write-ins in the runoff. It took a 
special ordinance to do that, in any case, because the default in 
California law was that write-ins were allowed, and that this wasn't 
merely a dead letter, Long Beach re-elected their mayor, bypassing 
term limits that prohibited her from being on the ballot. She did not 
gain a majority in the primary, though she was the plurality leader. 
Again, she wasn't allowed to be on the runoff ballot, there was only 
one name on it, the runner-up from the primary. She again got enough 
write-in votes to win by a plurality. (There was another write-in 
candidacy that took away enough votes to deprive her of a majority, 
but a majority was not required for the runoff.)

When voters have two candidates to choose from, they learn more about 
the candidates and they may then make an informed choice. And the 
reality is that in about one-third of real runoff nonpartisan 
elections (or party primaries, same thing), the runner-up from the 
primary wins. *This does not happen with IRV, in nonpartisan elections.*

This really must be emphasized. This is real behavior of real 
nonpartisan runoff elections, and real behavior of real IRV. IRV does 
not find majorities, most of the time, and it reproduces, in the 
conditions where it is being introduced, nonpartisan local elections, 
Plurality.

The interpretation that exhausted ballots don't count is directly 
opposite to Robert's Rules, which considers such ballots valid. They 
contain legal votes. The interpretation that is being given here has 
never been validated by any court, and those who promote IRV in real 
legislation know this. There used to be, in the SF election code, a 
provision requiring a majority for election. The IRV proposition 
eliminated that provision, even as it was promised to voters that 
"candidates would still be required to gain a majority of votes." If 
that were true, why was the provision struck?

Because the method doesn't require a "majority of votes." It is that simple.

They know this, quite well, in Australia. In most of the country, 
they use Preferential Voting, and full ranking is obligatory, or the 
ballot is spoiled. So the election counting rules use "absolute 
majority" as the criterion. (This is different from our usage of 
"absolute majority" here, which would refer to a majority, usually, 
of all eligible voters, whether they vote or not, or which could be 
restricted to refer to a majority of all ballots cast -- which would 
include ballots voided because of incomplete ranking. But there it 
simply means a majority of unspoiled ballots.)

And then, with Optional Preferential Voting, they use a different 
term, I think it is "majority of votes cast for continuing 
candidates" or something like that. Definitely, it is not a "majority 
of votes cast," or they would be needing to have some runoffs and, 
apparently, they don't want that.

Runoffs will give underdog candidates a better chance. I'd expect to 
see an occasional Green win if they had OPV and a true majority 
requirement, with runoffs, in Australia. As it is, Greens never win 
House seats, though they get, with STV, a few Senate seats.

What would happen is that occasionally a Green would get enough votes 
to get to second position, and then would have a running chance to 
win, to convince the electorate that he or she would be a good choice.

Single ballot? It is far too much of a leap to accomplish in one ballot.

> > But many think that
> > later-no-harm is undesirable
>
>"Many" on this list may think that, but it is my experience of more 
>than 45 years as a practical reformer explaining voting systems
>to real electors, that 'later no harm' does matter greatly to 
>ordinary electors.  If they think the voting system will not comply
>with 'later no harm', their immediate reaction is to say "I'm not 
>going to mark a second or any further preference because that will
>hurt my first choice candidate  - the one I most want to see elected."

And if those voters were, say, Nader supporters, and the election 
were, say, Bucklin, you are saying that they would not vote for a 
second preference? Like Gore? Because that would hurt Nader's 
chances? Are you saying that they are utterly insane?

Look, in a two-party system, and if we are ever going to have 
something other than a two-party system, it sure isn't going to be 
because of IRV, which is associated with strong two-party systems, it 
basically makes the world safe for them by damping the spoiler effect 
without, in return, giving third parties a true shot, then we need to 
realize that "voting for the favorite alone" is the norm, and there 
is no reason for most voters to add another preference. Those who do 
not study history are condemned to repeat it.

IRV was used in the U.S., for quite some time, for certain party 
primaries. It was abandoned. Why? Not enough voters adding second 
preferences. It's been claimed that Bucklin was abandoned, in its 
party primary usage, for the same reason, only ten percent of voters 
adding second preferences. Now, ten percent is enough to deal with 
the ordinary spoiler effect. Above that level, it's a bit 
questionable if it is a true spoiler effect, in fact. And the real 
problem is allowing election by a plurality. And IRV still allows that.

Ask the same voters if election by a majority is an important value. 
What do you think they will say? What did the Vermont legislature say 
in its findings preceding the failed IRV law there? They said that 
election by a majority was a fundamental value. And then they 
proceeded to assume that IRV did just this, which it does not. The 
word "majority," as in "majority of the vote" has a very clear 
meaning and it has had this meaning for centuries. It means that if 
there is a single ballot taken, a majority of voters, voting in the 
election and casting valid ballots for at least one legitimate 
candidate -- under the most stringent interpretation -- voted for the 
winner. The basic principle of majority rule is that no decision is 
made except upon the approval of a majority of those voting.

When an IRV ballot is cast, and *especially* if the ranks are 
limited, as almost all U.S. implementations have required, voters do 
not know which votes will be "continuing candidates" and which ones 
will not. They went to the polls, they voted, and they voted for an 
eligible candidate. They have a right to be counted as having voted 
*against* all candidates they did not actually vote for. Preferential 
ballot allows them to vote for more than one, that's all, so that 
there is more possibility of finding a majority of votes cast.

This redefinition of majority is one of the most pernicious aspects 
of the IRV campaign.

Here, Mr. Gilmour is making a virtue of what is really voter 
ignorance, about Later No Harm. The inventor of the criterion, 
Woodall, in his paper describing it, notes that one referee said that 
the criterion was "disgusting" or some word like that. Woodall wasn't 
promoting the criterion, just describing it. What that criterion 
means, for any method which satisfies it, is that the voters' 
alternate votes will not be revealed until the higher preferences are 
totally eliminated. This, then, makes it impossible to discover if 
there is a good compromise winner, because the votes will *never* be 
revealed. Later-no-harm thus requires Condorcet failure. The 
equivalent in personal negotiation is a negotiator who holds cards 
extremely close and who gives up the favorite outcome, and reveals a 
lesser preference only when it has become totally apparent that it is 
impossible; this, simply, makes negotiations much more difficult, 
because the other parties don't know what alternatives might be 
reasonably acceptable. It is, in essence, purely selfish, seeking 
only personal maximization of satisfaction without seeking social 
maximization. That's why the referee was disgusted. It wasn't just 
something he'd eaten.

>   And of course, if you once depart from 'later no harm' you
>open the way to all sorts of strategic voting that just cannot work 
>in a 'later no harm' IRV (or STV) public election with large numbers of voters.

STV is a totally different animal, let there be no mistake about 
this. STV satisfies later-no-harm, and this is a problem only with 
the last selection, where there are losers. Otherwise, voters are 
picking their representatives, essentially unopposed, until that last 
election. It should be noticed that Delegable Proxy and Asset Voting 
fully satisfy later-no-harm, particularly because there is, in most 
proposed implementations, no "later" at all. There is a first 
preference vote, there is no reason to compromise on that vote, and 
it is effective. It "wins." Always.

"Strategic voting" is a bugaboo that afflicts a lot of election 
theory. If you ask voters about Later No Harm, sure, they will tell 
you that it's important. But how they will actually behave is, I'd 
predict, a different matter. First of all, with Optional Preferential 
Voting, failure to rank is common. Apparently Later-No-Harm is 
inadequate as a motivator. The same was true for IRV in the U.S., and 
it continues to be true. There are lots of exhausted ballots. I think 
the main reason is that voters simply run out of preferences, they 
really don't have more than one or two reasonably strong preferences. 
Some of them don't even have one; these voters are there for the 
general election, if it is San Francisco, and maybe vote in the Board 
of Supervisors election because it is there. They are lucky if they 
recognize one name, much less three.

Strategic voting? The most common strategic voting still applies to 
IRV: vote for a candidate who can win. That means, with Plurality and 
a two-party system, vote for one of the top two. The same is still 
strategically required with IRV: vote for one of the top two. Even if 
that means betraying a candidate you prefer, because of the limited 
ranks. Gilmour is dreaming. IRV requires strategic voting, if the 
voter's vote is to be maximally effective. But how many voters 
actually do this? My guess is, with IRV, few. My strong suspicion is 
that if the elections in San Francisco were Bucklin, the votes would 
be the same. The ballot is the same. But Bucklin should occasionally 
produced a different result, a better one. Probably rare, though. 
What Bucklin would definitely do is to find a few more majorities, 
because it counts all the votes at any rank reached in the process. 
It counts them all simultaneously and not conditionally, so it fails 
Later-No-Harm, but the practical consequence of that is minor. 
Basically, it only counts lower preferences if the first preferences 
don't show a majority. The lower preference expressed doesn't exactly 
"hurt" the first preference, but what it does is to put the other 
candidate on a par with the first preference. Once the votes are 
counted, they become equal. It's like Approval Voting.

Now, as to strategy: Approval Voting was invented as a strategy-free 
method. In order to claim that Approval wasn't strategy-free, 
opponents had to redefine strategic voting. It used to be that 
"strategic voting" meant that the voter, to gain an improved outcome, 
would vote opposite preference. I.e., the voter really prefers A to 
B, but votes B over A. It's ironic that Gilmour, above, is using the 
shifted meaning. IRV rewards strategic voting under a number of 
different circumstances, according to the original definition. 
Approval never does. By definition abstaining from adding additional 
preferences as "strategic voting," Gilmour has radically distorted 
the meaning of the term. It now means failure to express a preference 
that exists. But voting systems in general require such failure. 
Approval Voting, of course, allows voters to vote for more than one, 
but when they do so, they give up, and do not express, the preference 
that they may have between the two. Bucklin allows the expression of 
that preference, but then, under some conditions, moots it. But 
neither of these is strategic voting under the original definition.

It's been interesting to see James Armytage-Green struggle with this. 
He had to define "sincere vote" in order to determine if Approval 
Voting satisfies the majority criterion. Anyway, it's late and I've 
got to move on tonight..... Basically, he ended up defining sincere 
vote as not reversed. That's a double negative, with a lost middle.



> > But many think that
> > later-no-harm is undesirable because it interferes with the
> > process of equitable compromise that is essential to the
> > social cooperation that voting is supposed to facilitate. If
> > I am negotiating with my neighbor, and his preferred option
> > differs from mine, if I reveal that some compromise option is
> > acceptable to me, before I'm certain that my favorite won't
> > be chosen, then I may "harm" the chance of my favorite being
> > chosen. If the method my neighbor and I used to help us make
> > the decision *requires* later-no-harm, it will interfere with
> > the negotiation process, make it more difficult to find
> > mutually acceptable solutions.
>
>This is all irrelevant because in a public election there is no 
>negotiation between voter and voter or between voter and candidate.

There should be. And it could be made so.

>I know that there are proposal for voting system that would 
>incorporate "negotiation" of various kinds, but none of those was under
>discussion here.

The relevance here is that deliberative process is proposed as an 
ideal to emulate as closely as possible. The step we can take which 
most closely emulates it, as a simple step that is often actually 
taken, is to require a majority or hold a runoff. Gilmour may not 
realize the situation in the U.S., where top-two runoff, in 
nonpartisan local elections, which does guarantee, with rare 
exceptions, a majority, and which clearly produces more democratic 
results than Plurality and IRV and which likewise encourages sincere 
voting, is being replaced by IRV on some very flimsy grounds.

I'll repeat this: the best common election method we have, albeit 
flawed in certain ways (similar to some of the IRV flaws) is being 
replaced by an inferior method, based on a series of carefully 
crafted and highly deceptive arguments. And hardly anyone is looking 
at this. There is no study of IRV results except for what I've done, 
to my knowledge. There is simply, among IRV advocates, an assumption 
that if the elections are being held, and in spite of massive 
counting problems and delays, if voters say that they liked it, must be good.

IRV is duplicating, quite clearly, Plurality. If voters simply vote 
for one, in these nonpartisan elections where IRV is being 
implemented here, they will get the same result as if voters have a 
preferential ballot and add additional preferences. Exceptions appear 
to be rare, there have been none in 32 elections here. Apparently the 
phenomenon is known in Australia, so it happens even with partisan 
elections, usually. (I know of an historical exception, Ann Arbor, 
MI, where an unusually strong third party had been splitting the 
Democratic vote for Mayor. Preferential voting allowed the Democrat 
to win on second-preference votes from that party. Apparently, 
though, this is not an ordinary occurrence. In very close elections, 
though, there is no telling what will happen.