[EM] IRV vs Plurality (back to the pile count controversy)

[Abd ul-Rahman Lomax]

At 01:55 PM 1/22/2010, James Gilmour wrote:
>Abd ul-Rahman Lomax  > Sent: Friday, January 22, 2010 5:53 PM
> > > At 03:57 AM 1/22/2010, James Gilmour wrote:
> > >This
> > >second set of rules are those that prescribe the transfer of votes
> > >"to the bitter end", i.e. even after the winners have all been
> > >determined.  Under this rule a ballot marked "A" would be treated
> > >differently from a ballot marked "A>B": at the last possible
> > >transfer, the "A" ballot would become 'non-transferable
> > >(exhausted)', but the "A>B" ballot would be transferred to A.
> >
> > You mean transferred to B, of course.
>
>Apologies - my example was incomplete.  To illustrate this stupid 
>rule properly, I should have posited two candidates, A and B, (or
>just two left after all others have been eliminated), with A the 
>winner.  Then consider two ballots, one marked B and the other
>marked B>A.  In the last round of a count under the "to the bitter 
>end" transfer rule, the ballot marked B would be
>'non-transferrable (exhausted)', but the vote on the B>A ballot 
>would be transferred to A.  It is illogical to treat these ballots
>differently in an STV (contingency choice) election and it offends 
>the underlying concepts of 'Later No Harm' to transfer the B>A
>ballot to A.

If truncation is allowed, there is a difference, as you know. 
However, if a plurality of ballots is sufficient for victory, it's 
irrelevant to the result. The real difference shows up when a true 
majority is required.

In Australia, they use the term "absolute majority" as the quota that 
must be reached, but that is with mandatory full ranking. So there is 
never majority failure, absent a tie, and a majority is always found 
when there are only two candidates still standing. Where truncation 
is permitted, which is in a few places in Australia, they change the 
quota to "a majority of votes for candidates not eliminated." That, 
too, never requires that last counting step.

But we have been discussing the general case, and that case most 
notably includes elections as described in Robert's Rules of Order, 
Newly Revised, RRONR 10th edition, where a single-transferable vote 
method is described for single-winner (and a multiwinner variation is 
also described, a detail I won't address). RRONR never permits 
election without a majority unless a special bylaw has been passed 
allowing election by plurality. Which is strongly discouraged.

FairVote managed to confuse nearly everyone with their description of 
what is in Robert's Rules. They have slightly modified their rhetoric 
since I started nailing them on this, so that generally they aren't 
actually lying any more, but they still cherry-pick and create 
deceptive implications. If a majority is sought, and full ranking is 
optional, and the ballots are ones on which the voter writes 
candidates in order of preference, going to the last elimination is 
quite proper, for one has thereby found all the ballots containing a 
vote for the leader. If that is not a majority of all ballots, the 
election fails.

And, yes, this violates Later No Harm. If only a plurality is 
required, Later No Harm is not violated. LNH is incompatible with a 
majority requirement, unless voters are coerced or misled, that is 
one of the dirty little secrets of IRV.

In RRONR elections, the voters are not constrained to a list of 
candidates. In the normal procedure, the ballots are blank, and the 
voter writes down the names of candidates, ranking them. The voter 
may vote for *anyone*, including ineligible candidates or Donald 
Duck, or, more importantly, Mr. None of the Above.

Why does RRONR even propose the STV method? Good question! They 
propose it in cases where repeated balloting is not considered 
practical. But they think of it as a way to find a majority, and they 
advise the voters to rank "all the candidates," cautioning that if 
they don't, it is possible that no candidate will get a majority, 
thus requiring the election to be repeated.

Now, this is what I've found in studying U.S. elections with IRV. In 
partisan elections, IRV sometimes works and finds a better winner, 
clearly more democratic, than FPTP or Plurality. In nonpartisan 
elections, however, at least in these public elections studied, IRV 
simply reproduces the results of Plurality. There is enough evidence 
to come to the conclusion that exceptions would be rare and typically 
close elections.

We have been discussing the election in Burlington, Vermont. There, a 
naive impression can be created that Plurality would have elected 
Wright, the Republican, he did get the most first preference votes. 
However, prior to IRV being adopted there, they used top two runoff, 
with a 40% requirement for election, otherwise a runoff was held 
between the top two. The runoff would have been between Kiss and 
Wright. Likely result would have been the same as with IRV.

The problem is that, while Kiss was a better winner than Wright, the 
eliminated Democrat, Montroll, was a beats-all winner, based on the 
expressed votes. Burlington politics are complicated. In this 
election, there were four major vote-getters (some analyses neglect 
the fourth, Smith, but he got 15% of the first preference vote). The 
more there are significant candidates, the more IRV can break down. 
IRV works to reduce the first-order spoiler effect in a two-party 
system, but, with three parties at rough parity, plus a fourth that 
is significant, it's a mess.

And the shame is that there is a much better method handy. It doesn't 
fully satisfy later-no-harm, but if you want real majorities, as do 
most jurisdictions that have been sold IRV as a top-two-runoff 
replacement, later no harm is broken anyway.



> > >This second rule is, of course, a stupid rule but that does not mean
> > >it has not been implemented in some jurisdictions, including,
> > >sadly, Scotland.
> >
> > Not stupid, precisely because of the difference between A>B and A.
> > The former is an acceptance of the last listed preference, the latter
> > is not. It makes a difference if a majority is required. Not if it is
> > not, though it might make a difference with some methods. But not IRV.
>
>But my comments were exclusively in the context of STV elections 
>(IRV, STV-PR, RCV).

And I'm talking about STV elections, as described in the most popular 
parliamentary procedure manual in the United States.

What's your problem with the full counting, however? It doesn't cause 
LNH failure in the context you describe, because the uncovering of 
those last votes does not change the result, it merely provides a 
kind of blessing. Or not!



> > >  It is also a highly undesirable rule because it means that my vote
> > > could, in some circumstances, be transferred to
> > >the candidate I deliberately ranked last in the lowest possible
> > >place, e.g. 12th out of 12 candidates.
> >
> > Basically, if there are as many ranks as candidates, don't vote for
> > that last one! That's your choice, unless full ranking is required,
> > in which case you *can't* vote the truncated vote and it is
> > irrelevant if it's counted or not.
>
>That's why when running an STV election where we can use "write in" 
>boxes for all preferences, I always provide one fewer preference
>box than the number of candidates (as I see you recommended in a 
>later part of  your post).  But all of our ballots for public
>elections have the candidates names printed on them.

Here, write-in votes are allowed in all primary elections. Sometimes 
they are allowed in runoff elections.

In general, preferential voting encourages the expansion of declared 
candidacies.

When it comes to the bottom of the candidate barrel, I would not 
personally vote for the last few unless I had a strong preference 
between them. With IRV, that could be a tough choice.

But the vistas open up if we use and advanced voting system for a 
primary, still requiring a majority or some threshold that is 
reliable for predicting that it will become a majority in a runoff. 
(40% isn't that, but it is better than raw plurality: in the 
Burlington election, for example, we can assume that if the voters 
didn't alter their behavior and voted sincerely, the 40% requirement 
did suffice to detect a bad plurality result.)

See, James, I start by considering as an ideal election method 
full-on deliberative process, majority required. That is only not 
done on the excuse of efficiency. Can you imagine a Parliament using 
IRV to elect the Prime Minister?

No, they simply use repeated ballot if they must complete an 
election, or, if I have it right, the incumbent remains if there is 
majority failure. Or does something happen on a loss of a vote of 
confidence? (I'm not a student of parliamentary systems even though I 
advocate electing officers deliberatively in a legislature, 
parliament, or assembly, instead of direct elections by single 
ballot; the officers become, then, like public employees, hired and 
fired at will.)

Now, to optimize voting systems, I then look at what will simulate 
repeated ballot. IRV absolutely does not do that, because IRV is STV 
and  STV is sequential elimination of candidates, and repeated ballot 
does not eliminate candidates at all. Each ballot is a new election 
and could possibly have completely different candidates.

So, first of all, I want to see a winner gain an uncoerced majority. 
Now, in the study of Approval Voting, the concept of repeated 
balloting with declining approval threshold has been considered. In 
pure repeated balloting, the voters eventually make compromises (or 
sometimes they do even better, they actually revised their opinions 
or bring in a new candidate who actually is more satisfactory 
overall).  We can't do the new-candidate thing with a simulated 
series, but we can imitate the lowering of approval thresholds.

And that is what Bucklin does, and particularly Bucklin-ER, or pure 
Instant Runoff Approval Voting. It could even become more accurate: 
the voter would provide a range ballot, and the method would start 
out by looking at max-rated candidates, seeking a majority, then 
sliding down the scale to add approvals as they appear, until a 
majority is found. At each step we have an approval election with a 
particular approval threshold. If we want a "majority of votes," we 
would specify some level on the Range ballot (in advance, the voters 
must know it to make their decisions well) that is "approval." And 
votes below that level might be used in selecting runoff candidates, 
if there is majority failure, but would not, in themselves, ever 
determine a winner in a primary. The might in a runoff, and this 
would be a compromise for efficiency, if it is considered that a 
third ballot is impractical.

Bucklin, as I assume you know, was called American Preferential 
Voting at the time it was rather widely adopted (almost a hundred 
jurisdictions adopted it), and it was simpler than the Range-Bucklin 
I describe, but the basic idea is there, and the only problem was 
that, like IRV here, it was sold as a way to find majorities without 
runoff elections. Naturally, it failed to do that, because people 
would, in fact, truncate with Bucklin just as they do with IRV. (And 
probably more so, a bit. Not as much as FairVote claims.)

> > >   Following on from the
> > >concept of 'Later No Harm' (which underpins the whole of contingency
> > >voting, as in IRV and STV-PR), it is very important to be able
> > >to give a voter the absolutely assurance that under no circumstances
> > >will her vote ever be transferred to the candidate she has
> > >ranked 12th out of 12.  Sadly, the stupid "transfer to the bitter
> > >end" rule undermines this.
> >
> > Only because of voter ignorance, an ignorance which has sometimes
> > been encouraged by activists.
>
>No, not at all.  This is a piece of nonsense that some have 
>introduced into STV counting, especially since electronic counting
>became available.  It does not feature in any of the 
>long-established versions of STV counting rules promoted in the UK.

Are there any applications of STV counting rules for single-winner 
elections in the UK, where a majority is required or there is a 
repeat election?

Bottom line, though, what actually simulates repeated elections well 
is Bucklin. It allows voters to classify candidates into three ranks, 
which might be called Favorite, Preferred, and Acceptable. While it 
doesn't satisfy Later-No-Harm, strictly, it doesn't actually help the 
lower-ranked candidate beat the more-favored one, because what 
happens is that the ranked votes collapse into multiple approvals, 
which are abstentions, in effect, from that pair-wise election. But 
they aren't true abstentions, rather they contribute a vote to all 
supported candidates toward gaining a majority, either in the second 
or third round of counting. (or a plurality in the end, if that is 
allowed to complete.)

Instead of "your lower ranked vote won't hurt your more-favored 
candidate because it won't be counted until your more-favored 
candidate is eliminated" it becomes, "your lower ranked vote won't 
hurt your more-favored candidate, ever, because it won't be counted 
until it is known that your more-favored candidate won't win with a 
majority, but your vote for your favorite will only help your 
favorite over your lower preference up to the point that the majority 
hasn't been found and lower preference votes are counted. It isn't 
actually Later-No-Harm violation, it is Later-No-Help. If you add a 
lower preference vote, you may end up with the preference being lost, 
but both candidates are then preferred over all others.

I've noticed that if you have a relatively strong first preference, 
you could skip the second rank and only add additional approvals in 
the third rank. In original Bucklin, multiple voting was only allowed 
in the third rank, but this, by the way, is a counterexample to the 
claim that Approval Voting hasn't ever been used for a public 
election. Bucklin is really a close form of Approval, it simply 
phases in the votes instead of counting them all at once as with 
standard Approval. In many Bucklin elections, all the ranks were 
counted, so it became pure Approval.