[Election-Methods] A Better Version of IRV?

[Abd ul-Rahman Lomax]

At 02:01 AM 7/13/2008, Chris Benham wrote:
>Forest,
>"The voter ranks all she wants to and the remaining candidates are 
>ranked (later, i.e. below) by the voter's
>favorite or perhaps, as Steve Eppley has suggested, by the voter's 
>specified public ranking.
>
>Since IRV satisfies LNH, what's the harm in this?".
>
>The harm is that voter's votes are used to help candidates that the 
>voters may not wish to help.
>It offends the principle that the voter should be fully in control 
>of his/her vote.
>Giving some voters (candidates) the power to fully control their own 
>vote and also to complete
>the rankings of some of the truncators offends the principle that as 
>far as possible all voters
>should have equal power.

First of all, if we are talking about elections of representatives of 
some kind, the voter isn't going to be  "in full control of his/her 
vote" no matter what. At the point of the election, or later, when 
the representative casts votes, individual control is lost.

The equal power issue is spurious. The voting power is in the hands 
of those who cast ballots, originally, and they may choose to 
delegate that power or not. More about this below. The original 
"candidate proxy" or "Asset Voting" proposal was actually an STV 
proposal by Lewis Carroll, aka Charles Dodgson, in 1994.

>"In Australia, where (in single winner elections) most of the voters 
>copy candidate cards, this would save
>them a lot of bother."
>
>In Australia the only significant "bother" stems from compulsory 
>full strict ranking (for the vote to be
>counted as valid).  How many or few voters choose to exercise their 
>right to not follow their favourite's
>ranking advice is no argument for removing that right.

Compulsory full ranking, Dodgson noted, was a problem for voters who 
may not be sufficiently informed to understand how to rank *all* 
candidates. Obviously, full ranking only works when candidate count 
is limited, and even then donkey voting seems to be fairly common. It 
would be interesting to see statistics on straight sequence voting 
(which wouldn't be visible in Australian results because of Robson 
Rotation, one would have to look at actual ballots or true ballot images.)

> > And what do you have in mind as "Australia's worst problems
> > with their version of IRV"?
>
>"It has degenerated into a defacto second rate version of Asset Voting."
>To the extent that that is true it can (and should) be fixed by 
>simply allowing truncation.

That is done in Queensland and NSW, it's called "Optional 
Preferential Voting," but, of course, in that there is no remedy for 
ballot exhaustion.

Dodgson was studying the problem of ballot exhaustion, and hit upon 
the idea of allowing voters to vote for one.


>
> > Why do you want to "stop" IRV? Do you agree with Kathy Dopp
> > that IRV is worse than FPP?
>
>"I would stop IRV if we could get a better method in its place.
>
>If we cannot stop IRV, why not search for acceptable tweaks that 
>would improve it?"
>
>The short answer is because IRV isn't really amenable to 
>"tweaks".  In terms of  positive
>criterion compliances it isn't dominated by any other method, and 
>has both good and quite
>bad properties (averaging in my judgement to a "good" 
>method).  "Tweaks" generally muck
>up its good properties  without enough compensation in terms of 
>fixing or patching up its
>bad properties.

Problem is that the "good" property, Later No Harm, is actually a 
*terrible* property, see Woodall's original paper that coined the term.

However, for proportional representation, there is, in fact, a very 
simple tweak that is what Dodgson invented. He noted that many 
voters, not being specialists in politics, wouldn't have sufficient 
knowledge to rank all candidates, and therefore, if truncation were 
allowed, would indeed truncate and would therefore be at risk of 
having their ballot exhausted and thus their vote wasted. Why not 
allow voters to vote for one only, with that one then being able to 
recast those votes, if exhausted, in order to create quotas for 
election? I'm not sure that he specified it, I don't have a copy of 
his full pamphlet yet, but I'd assume that if one ranked more than 
one, but the ballot were exhausted, the first preference would get 
the vote for reassignment (but there are other ways to do it, I'm 
sure, and I haven't considered all the contingencies). It's a simple 
tweak, but it turns STV into Asset Voting, with voter control over 
transfers possible, to the extent that the voter exercises the right. 
And then fallback to deliberative reassignment of exhausted votes, 
based on the candidate the voter most trusts, the first preference.

There are other possible tweaks: for example, allow multiple votes in 
each rank. This is similar to Bucklin, it makes each round an 
Approval election. Voters don't have to do it, so they have the 
choice of whether or not to vote to avoid Later No Harm. If it's an 
Asset election, then, and the ballot is exhausted, the vote is 
fractionally distributed, as in Fractional Approval Asset Voting, my 
old proposal. (Which is really ordinary Asset but with a simple 
ballot, unlike the original Warren Smith Asset that allowed votes in 
the range of 0-1, any set of fractions such that all votes add up to 
one full vote. With FAAV, I'd expect, most voters would simply vote 
for one, there really isn't, in my view, a lot of value gained by 
splitting up the vote, but it is also harmless to do so, and allows 
voters the flexibility, plus it removes on possible reason for 
considering a ballot spoiled.)

>I think Smith (or Shwartz),IRV is quite a good  Condorcet method. It 
>completely fixes the
>failure of Condorcet while being more complicated  (to explain and 
>at least sometimes to
>count) than plain IRV, and a Mutual Dominant Third candidate can't 
>be successfully buried.
>But it fails Later-no-Harm and Later-no-Help, is vulnerable to 
>Burying strategy, fails
>mono-add-top, and keeps  IRV's failure of  mono-raise and (related) 
>vulnerability to
>Pushover strategy.

Once again, LNH compliance is a mark against a method, in my view, 
and apparently in the view of at least one of Woodall's referees. 
It's the kind of thing that sounds good, until the implications for 
democratic process are considered. It treats signaling a possible 
compromise as a weakness. Only compromise if you are going to die 
otherwise, would be the equivalent. In order to avoid "betraying" a 
favorite by making it possible for our total vote to help someone 
else to win, we choose a method, if we insist on LNH, that kills the 
candidate, instead of leaving the matter open for a broader 
determination. That eliminated candidate, my favorite, might have won 
if not for the LNH compliance of the method.

Further, LNH cannot be satisfied by any method that requires a 
majority, unless the majority is artificially created, either by 
eliminations *of votes* or by requiring full ranking, which amounts 
to coerced votes.

>"It is better than FPP in some ways and worse in others, especially 
>in complexity."
>
>With separate paper ballots for each race, I don't accept that IRV 
>is all that "complex".
>I think that you have somewhat dodged my question.

It is far harder to audit. That's what the election security experts 
think. Sure, that can be overcome, but why is it worth the effort? 
There are better, easier to count forms of preferential voting.

What is missing with much of this is that IRV, in nonpartisan 
elections, almost always mimics plurality. Even in partisan 
elections, it tends to do this strongly, but "comeback" elections 
become more common.

>"Do you think that Asset Voting is worse than FPP?"
>No, on balance.

We don't really know, since we have only theory. Asset hasn't been 
tried in political contexts. But Asset is a form of proxy voting, 
which has been used for centuries where property is involved.


>"Just to clarify, I think that Condorcet Methods and Range, though 
>better than IRV, share this complexity
>defect with IRV to some degree.  I have suggested the same tweak for them.
>
>In fact, that is the essence of DYN, wihich is simply carrying this 
>tweak to its logical conclusion in the case
>of Range, which is the only one of the three (Range, Condorcet, and 
>IRV) that satisfies the FBC."
>
>I find your  DYN method  less offensive than your "IRV tweak" 
>suggestion because it is an "opt in" system
>and to the extent that voters don't opt in it is just plain Approval 
>(a not-too-bad method).

Right. In fact, take Approval and require a majority for election, 
you have an *excellent* method, better than Plurality with majority 
required. (Simply because it avoids runoffs some of the time. Same 
thing with Bucklin, which is really a ranked Approval, especially if 
multiple votes are allowed in all rounds, unlike in the original form 
(Duluth is what I have in mind), where multiple votes were only 
allowed in third rank.

The irony in the U.S. is that, in nonpartisan elections in 
jurisdictions where a majority has been required (or else there was a 
top-two runoff), IRV is replacing this, with utterly spurious 
arguments based on the false analogy between IRV and real runoff. 
What is probably the most sophisticated election method in use in the 
U.S. is being replaced with a less democratic method.

In real practice, IRV in nonpartisan elections here is reproducing, 
closely, the results of Plurality while pretending to gain a 
majority. There have been some thirty or so IRV elections here, and 
there have been nine "instant runoffs." In seven of the nine, no 
majority of votes cast was found. I perhaps one additional one of 
those, there would have been a majority found if counting had not 
been discontinued because the false majority of IRV had been found. 
In *none* of the nine was the winner any different from the first 
round plurality result. On the other hand, in the real runoff 
elections I studied, one out of three resulted in a "comeback," where 
the runner-up in the first round ended up winning. In a series of 
federal primary elections, I think it was in Texas, FairVote found 
29% "comebacks."

IRV is changing election results, it does not "simulate" real 
runoffs, and the reasons are pretty clear: real runoffs give the 
voters a new look at a reduced field, plus there will be differential 
turnout, which tests preference strength, probably pushing results 
toward what sincere Range would produce.

Further, there is a little-known fact about *some* runoff elections 
in the U.S.: write-in votes may be allowed, thus providing a safety 
valve for the situation, fairly unusual *in nonpartisan elections* 
that a Condorcet winner (from underlying preferences) isn't in the 
runoff. In Long Beach, California, there was an election where the 
mayor won as a write-in. This was an incumbent, and had been 
prohibited from being on the ballot by term limit laws, which did not 
prohibit her election, which would have been unconstitutional, but it 
prohibited her being placed on the ballot. So she ran as a write-in. 
She was the leader in the primary, but did not gain a majority. In 
the runoff, there was only one candidate on the ballot, the 
runner-up. Again, with write-in votes, she won that by a plurality, 
which was all that was required. (There was another write-in with 
significant votes in the runoff.)

Top-two runoff with write-ins allowed is much more sophisticated than 
any single-round system. Robert's Rules of Order prefers repeated 
balloting with a majority required, and no eliminations, and if a 
majority continued to be required, it would *be*, in reality, what 
Robert's Rules recommends, since candidates aren't actually 
eliminated, there is only the "suggestion" of which candidates are on 
the ballot.

Voting systems theorists have, as far as I can tell, mostly ignored 
all this, and voting systems that aren't deterministic with a single 
ballot aren't even considered voting systems by many theorists, just 
as Arrow dismissed cardinal ratings as not being a real voting method 
since it didn't completely rank and, allegedly, there was no 
"meaning" to the ratings. (Hint, Arrow: the "ratings" are votes, 
fractional votes. That's what they "mean." Weight placed on an 
election possibility. These ratings might be "related" to voter 
utilities, but that isn't what they "mean" because there is no fixed 
relationship.)

What we consider "election methods" are shortcuts for basic 
democratic process, which restricts itself to single questions which 
can be answered Yes or No, with the questions themselves being 
designed by a series of questions all of that kind. From my point of 
view, good election methods are simply a way of making democratic 
process more efficient without sacrificing the democratic value of 
majority approval of any result. Good methods will also reveal 
possible improvements over mere majority approval, and all of this 
*requires* that Later-No-Harm be unsatisfied. The simplest such 
reform is Approval Voting, which is a very small tweak to Plurality. 
Range is likewise pretty simple and easy to count, particularly if 
the resolution is low. To be what I call "Majority Rule compliant," 
Range must have a specified Approval cutoff, probably an absolute one 
(such as midrange or above, or simply above midrange). Likewise 
ranked methods would need an Approval cutoff, mere ranking is not enough.

All voting methods which require completion of the election in a 
single round are not Majority Rule compliant, with the possible 
exception that full ranking is required, and any ranking above bottom 
is considered approval, which coerces votes, and if votes can be 
coerced, Plurality satisfies it. (You will vote for Our Supreme 
Leader or we will discard your ballot as obviously containing an 
error. Remember, Saddam Hussein was re-elected in an election where 
nobody voted against him. As far as we know.)

I'm always amused by the argument that the last-round majority of IRV 
is a true majority, because, with this logic, we could always find 
unanimity for the IRV winner: just carry the elimination one step further.