[EM] Legal brief vs. San Francisco limited IRV

[Abd ul-Rahman Lomax]

At 11:21 PM 3/7/2010, Kathy Dopp wrote:
>On Sun, Mar 7, 2010 at 9:08 PM, Abd ul-Rahman Lomax 
><abd at lomaxdesign.com> wrote:
> > At 12:39 AM 3/7/2010, Kathy Dopp wrote:
> >>
> >> I've posted the latest plaintiffs' legal brief here. Plaintiffs
> >> Francisco rank only three version of IRV.
> >>
> >> http://kathydopp.com/wordpress/?cat=8
> >
> > It is actually brilliant. Yes; limited. My view is that IRV is generally
> > constitutional. I.e., if top two runoff is constitutional, full-ranking IRV
> > is constitutional.
>
>Top-two allows *all* voters the opportunity to participate in the
>final election of who will govern, so is IMO clearly more
>constitutional than IRV which fails constitutionality in at least
>three ways:

First of all, there is no controversy over the constitutionality of 
TTR. There likewise is no controversy over the constitutionality of 
vote for one Plurality. And, except for some narrow exceptions, there 
is no finding that IRV is not constitutional anywhere, if full 
ranking is allowed. The instant case is over the constitutionality of 
IRV with only three explicit ranks allowed (and assumed 
bottom-ranking if the voter does not list the candidate).

>1. fails to allow all voters to participate in the final
>decision-making process by excluding voters from the last round

That's correct. But if full ranking is allowed, the "exclusion" is 
debatable. I agree it's a problem, and so does Robert's Rules of 
Order. But, notice, the issue might be decided on general public 
policy grounds, i.e., does IRV make the situation worse for these 
possible excluded voters? If it doesn't, I can anticipate that the 
defendants might argue, and the court might agree, that the right of 
participation in the decision, over the situation with plurality, is 
enhanced,  not reduced. Courts will not always decide on narrow technicalities.

IRV is definitely problematic in many ways, but whether or not those 
rise to a constitutional issue is not so obvious. I do agree that 
there is an issue here, with twenty candidates and three ranks. With 
that many candidates and only three ranks, there are quite likely 
real voters who have been prevented from both voting sincerely (as 
promised) and participating in the final decision. Note that if there 
is a majority required, as the initiative pretended, there would be 
no problem. There would be a runoff, and the voter would then not be 
deprived. So the real source of the problem is that the method 
pretended to find majorities, in order to gain approval, but doesn't. 
The ballot information pamphlet promised that the winner "would still 
be required to gain a majority of votes," but did anyone notice that 
the provisions themselves removed the requirement for a majority?

Not enough attention was paid by the plaintiff attorneys to this 
blatant deception of the voters in the initiative that implemented 
IRV. Of course, if they had left in the requirement for a majority, 
as they pretended still existed, then it would have become quickly 
visible that IRV was not avoiding most runoffs, since whenever 
runoffs were required, it is rare that the additional ranking found a 
majority, and this is a well-known behavior of IRV in Australia, in 
the places where full ranking is not obligatory. Australian voting 
systems experts consider using IRV without allowing full ranking to 
be insane. And I think they might question the use of IRV in a 
20-candidate situation in any case! IRV works, sort of, with partisan 
elections, much less well with nonpartisan ones.

>2. fails to treat all voters' ballots equally, counting the 2nd and
>lower rank choices of some, but not all, voters

I agree that this is a problem, but its speculative that it will fly. 
Bucklin, of course, has no such problem, and can use the same 3-rank 
ballot as RCV, and, if a true majority is required, there is no 
penalty for declining to rank more than one candidate. Either a 
majority will be found that differs from your opinion, or you get to 
choose again in a runoff. It is also possible to use a full-ranking 
Bucklin ballot, and it remains fairly simple to count. If Bucklin is 
to be used as a plurality method, I'd say this would be required on 
the same principles as it is not being suggested that it's required 
for IRV. But I'd also allow equal ranking, it does no harm and 
reduces ballot spoilage, and is actually a more accurate vote if the 
voter has no significant preference between two candidates.

(Equal ranking is already allowed, only it's restricted generally to 
bottom rank. The proposal is to allow it at all ranks. Classic 
Bucklin, Duluth variation, allowed equal ranking in third rank, the 
lowest approved rank. It required exclusive ranking in first and 
second rank, but I see no reason to require a voter who has no 
preference between two candidates to decide which one to support. If 
the voter is supporting one, it would clearly indicate that the voter 
also supports the other. And the analogy is to multiple conflicting 
ballot questions; the common rule is that, if both pass, the one with 
the most Yes votes prevails.)

>3. fails to provide voters with knowledge of the effect (positive or
>negative) of their votes on the candidates they rank due to its
>nonmonotonicity.

Yes. With IRV, you can cause a candidate to lose if you vote for the 
candidate. For clarity, here is how it happens. Suppose there are 
three candidates, A, B, and C. Your favorite is A. Your vote for A 
could cause C to be dropped. So then it's between A and C. But the C 
voters prefer B, in their transfers, who beats A as a result. 
However, if you did not vote for A (and someone like you, or a coin 
toss does this), then B is dropped, and A wins over B.

It gets complicated because not all voters will add additional votes. 
This is not the only problem, because of another effect, Center 
Squeeze, where a more strategic vote would be to skip over your 
favorite and vote first preference for the compromise candidate whom 
you prefer over the IRV winner.

Voting sincerely in IRV seems safer, but can produce worse results, 
so bad that you'd have seen a better result if you stayed home and 
did not vote.

And that appears to have actually happened in Burlington. If fewer 
Republicans had voted for Wright, and instead decided to support the 
compromise candidate that they clearly preferred from analysis of the 
ranked votes, they'd have produced that result.

>Top-two runoff treats all voters equally, allows all voters to
>participate, and allows all voters to cast a ballot with a positive
>effect on a candidate's chances of winning, (I.e. voters know which
>candidate they are helping to win each election.)

Yes. The overall system still suffers from center squeeze, but it's 
possible to fix that. Use Bucklin in both rounds! And allow write-in 
votes in the runoff. Because of the use of Bucklin in the runoff, 
those preferring a compromise candidate can still make a choice 
between the primary winners. Theoretically optimal methods (in terms 
of decision quality) seek a majority and don't stop until they find 
one, but the only possibly practical method for public elections that 
could do this would be Asset Voting.

> > They've hit on a technicality, the loss of equal
> > treatment of voters if the voter can't rank enough candidates, but as they
> > point out, the deprivation of equal treatment can be quite small and be
> > unconstitutional.

That's correct. But it must be very clear. I think they've shown it.

> >
> > And there is a quick fix, of course. Instant Runoff Approval Voting. Allow
> > the voters to mark as many candidates as they choose in each of the three
> > ranks. Then use Bucklin procedure. And make this a primary round, hold a
> > runoff if there is no true majority found. What this will do is 
> to eliminate
> > *most* runoff elections. Not all. It's just a more efficient way of finding
> > a majority in the primary. The same trick could be done with IRV, but, note
> > this: IRV no longer would satisfy later-no-harm, and IRV does not count all
> > the votes. Unless all the votes *are* counted, which would mean that one
> > would count lower ranked votes against non-eliminated higher-ranked
> > candidates. Bucklin does it much better and because there are no
> > eliminations in the primary rounds, it finds a candidate who is 
> supported by
> > more voters than any other. Because of the runff, voters need not, in the
> > primary, support someone who is the "least evil." They can vote sincerely.
> > They can bullet vote if they want. They can add alternate preferences if
> > they'd prefer these to a runoff (or if they'd want to see them get into a
> > runoff). They choose.
>
>I agree with you and find no objection to Bucklin method or most other
>alternative voting methods that treat all voters' equally and is thus
>precinct-summable, but don't have time to thoroughly study it now.
>Interesting that this idea allows putting as many candidates as
>desired into a limited number of slots. I like that on first glance.

Right. And if it is used with a majority requirement, it is simply a 
more efficient method of finding a majority than plurality. It allows 
voters to decide how much they care, would they rather make a 
compromise now, and possibly avoid a runoff, or would they prefer to 
see a runoff between the top two? Let the voters decide! Count All the Votes!

>The Condorcet method would still be easy to count in that case too
>since an n x n matrix (where n = # candidates) would still work just
>fine.

Well, easy compared to what? Compared to IRV, yes. Compared to 
Plurality or Bucklin, no.

Condorcet methods suffer from the same problem as afflicts many 
complex voting systems: it requires voters to have much more 
knowledge than is necessary to express a Favorite vote. That's the 
strength of Plurality, and is why plurality, but with a majority 
requirement, is standard in direct democracies. Top-two runoff works 
much better, because of the new, separate focus on a limited set. 
(There are top-three runoff methods in use, by the way, for example 
the Vermont gubernatorial election. The "runoff" is before the legislature.)

The big problem with TTR is that it can miss a compromise winner. 
Using a preferential ballot with good rules there might be able to 
avoid this. The sign that a compromise winner is being missed is 
majority failure! With good rules, two-round runoff would hardly ever 
fail to find a majority, and could only do better than the status 
quo, or it would default to it. That's why Bucklin being inexpensive, 
easy to understand and vote, and quite good in performance (by 
theoretical analysis and actual history of application) is considered 
so important to me.

I'd prefer Asset, which can be as simple as vote-for-one, though 
there are obvious interpretations that would allow counting and using 
overvotes. Your vote gets split between the candidates supported. 
(That's a bad idea with Approval Voting, single-ballot, but Asset 
doesn't waste any votes.)

> > From my analysis, IRV, however, would not find a majority in most of the
> > elections that went to instant runoff in San Francisco, and Bucklin would
> > find it it in maybe half. But Bucklin is far less expensive to count, it's
> > precinct summable, just count all the votes in each rank (typically three
> > ranks were used) and report them separately. They can then be added
> > together.
> >
> > Bucklin was used in a lot of places in the U.S., at one time, pushing 100.
> > The actual method was only found unconstitutional in one place, Minnesota,
> > is a decision that was aware it was idiosyncratic. It was popular, the
> > Minnesotata Supreme Court decision was quite unpopular, the court 
> notes that
> > in its reconsideration. My guess is that political forces were operating.
> > Some people did not want a good voting system, it's more difficult to
> > manipulate.
> >
>
>I like Bucklin. Too bad the MN Supremes shot it down. MN Supremes seem
>to make consistently poor decisions re. voting systems thus far,
>although perhaps with an "as-applied" case they'd do better.

The MN decision was not replicated anywhere, and, as it was written, 
it did also prohibit IRV. The decision was bad, and I wrote, years 
ago, that voting systems activists should support the reversal of 
Brown v. Smallwood, which, as written, prohibited all forms of 
preferential ballot. FairVote clouded that over, pretending that the 
decision was only based on the Later-No-Harm failure of Bucklin. If a 
majority is required, any preferential ballot system, including IRV, 
will fail later-no-harm....