New measure would eliminate IRV in all but the supervisor races in SF
Clay Shentrup
Supervisor wants to eliminate ranked-choice voting for citywide offices
San Francisco’s ranked-choice voting system could be replaced with run-off elections for citywide offices but not for the Board of Supervisors, under a proposal Supervisor Mark Farrell introduced to the board Tuesday.
Farrell’s charter amendment would return to run-off elections for the citywide offices of mayor, district attorney, public defender, sheriff, treasurer, assessor-recorder and city attorney. Supervisors, however, would continue to be chosen through ranked-choice voting.
“This is still a significant step toward eliminating ranked-choice voting here in San Francisco,” Farrell said.
Farrell’s proposal is his second in as many months to eliminate ranked-choice voting, which he said “has confused and disenfranchised voters” in the city. His last proposed ballot measure, to do away with the system across the board, fell short of the number of votes required.
This time, however, Farrell appears to have six votes needed to put the measure before voters in November. Supervisors Carmen Chu, Malia Cohen, Sean Elsbernd, Christina Olague and Scott Wiener have signed on as co-sponsors.
“From my perspective, we would do away with all of it,” Farrell said, but acknowledged he wasn’t able to do so. “It’s about building consensus here and we were able to do that,” he said.
Under ranked-choice voting, voters choose up to three candidates in each race and rank them in order of preference. If no one candidate gains a majority when first choices are counted, the candidate with the least number of votes is eliminated and those voters’ second choices then go to the other candidates. The process continues until a candidate secures more than 50 percent of the vote.
Backers of ranked-choice voting say the system ends the need for costly runoff elections and has elected a wider diversity of candidates.
Clay Shentrup
Jameson Quinn
Jameson Quinn
Again, I strongly urge you to suggest Bucklin(3), aka Fallback Approval, aka MCA, with a runoff if there's no unique majority.
"Voters rate each candidate preferred, approved, or unapproved. Blanks count as unapproved. If there's only one candidate who is preferred by a majority (over 50% of those with any opinion in the race), or only one who is approved or preferred by a majority, then they win immediately. Otherwise, the two least-unapproved candidates go into a runoff."
Abd ul-Rahman Lomax
The idea that "overvotes" should be discarded comes from direct
deliberative process with secret ballot, such as in Robert's Rules
where it was *assumed* that votes for more than one were errors.
Because that process requires a majority *of ballots with any mark*,
it was harmless there. If your votes were not counted, your ballot
still counted, then, as a vote against every candidate. But in public
secret ballot elections with no majority requirement, it's harmful.
You are excluded from the process, whether your overvote was an error
(in which case you probably did intend to support one of the
candidates, at least), or deliberate. (Some voters will assume that
if they vote for more than one, the votes will be counted. Overvotes
are locked out from voting machines, though.)
(Most cases of error would still result in a sensible use of the
vote. Discarding the votes has done obvious harm in places where some
factor caused extensive overvoting, such as the U.S. Presidential
race in 2000 in Florida, and the "harm" there is not the election of
Bush, per se, but that election when, almost certainly, more voters
supported Gore than Bush.)
Count All the Votes has a direct appeal, entirely aside from issues
of voting system performance. Of course, one will *also* mention that
this happens to implement a system considered superior by many
experts and academics, "Approval Voting," superior even to IRV,
overall, (but also defective in different ways). It provides a way
for supporters of a minority candidate to vote effectively, while
still expressing their support for a frontrunner.
I'd also make sure that the supes know that Bucklin voting, which
could easily be called Instant Runoff Approval, or Ranked Choice
Approval, was passed by voters in San Francisco in the 1920s, but
then killed by the Board of Elections through an extremely weird
technicality. It might be useful to be familiar with that history.
There was a California Supreme Court decision on it, as I recall.
State law required that, if voting machines were used, they be able
to handle the voting method. So the BoE bought a single Edison voting
machine, and allegedly it could not handle Bucklin, or the version of
Bucklin that passed, so ... that was enough to stop Bucklin in San
Francisco. One of the big arguments in the debate over Bucklin was
that it had elected a Socialist in Grand Junction, Colorado. (Which
was true, but it was a nonpartisan race for mayor.)
There are and have been strong antidemocratic forces operating in the
U.S., as elsewhere in the world.
Abd ul-Rahman Lomax
A sound comment on Bucklin, and I agree that
Bucklin's characteristics and history should be
made known, but the simplicity of the Approval
argument, i.e., just Count All the Votes, should
not be buried with further detail.
The real key to anything beyond that is that the
two best reforms in U.S. election practice, Top
Two Runoff and Bucklin, can be combined. Bucklin,
then, *reduces* runoffs, and presents a better
choice of candidates if a runoff is to be held,
than either vote-for-one in the primary, or IRV,
which can easily eliminate a compromise candidate
who would beat all the others in paired elections.
Bucklin itself fails certain election criteria,
but using it in a runoff system almost entirely eliminates those problems.
Bucklin should be run as Count All the Votes,
even in first rank. In that context, equal voting
first rank simply means that the voter has no
strong preference between those two candidates.
(Un)FairVote activists will claim bizarre
scenarios where this can result in Majority
Criterion failure, but the reality is that voters
were quite clear, historically, with how they
voted. Multiple approvals in first rank would be
relatively rare if allowed, because voters, in
fact, often skipped the second rank and only
voted in first and third ranks, because they
wanted to hold out until the end. That indicates
stronger preference, and advanced forms of Bucklin can consider that.
Bucklin is a door into truly advanced voting
systems, accessible through gradual reform, where
each step can be observed for performance.
Bucklin can use a Range ballot, and Bucklin with
a Range-winner-detecting runoff can beat simple
Range in Baysian Regret (rather obviously,
because Top Two Range beats plain range, under
some conditions). There is more than that: Runoff
voting provides a kind of detection of "absolute
preference strength," by eliminating voters who
don't have a strong preference. This effect has
largely been neglected by those looking for the
mythical holy grail, a perfect deterministic
system, completing with a single ballot. Then,
when they do look at top two runoff, for example,
they assume that the runoff is simultaneous, taht
it's the same voters with the same preferences,
rather than at what really happens, *a new
election* with new voters and often shifted
preferences. Hard to simulate, but it could be
done, by simulating absolute preferences of the
*entire eligible voting population* and adding a
"GAS" factor. Ah, Give a ... a ... Care Enough to
Go And Vote. The low turnouts in 2-candidate
runoffs reflect lack of strong preference between
the candidates, and the proof is the large public
runoff elections where one of the top two was a
serious extremist, due to center squeeze in the
primary. Voters turned out in droves, whereas in
most runoff elections held as special elections, turnout is low.
And to really go to town with the simulation, add
a probability of discovery model. I.e., there is
something unknown about the candidates in the
primary, to the voters, but it becomes known
because of the runoff campaign process, by the
increased focus on a reduced set. This really
affects runoffs, in ways that almost certainly
improve results. Robert's Rules of Order suggests
an IRV-like method for elections where voters
cannot meet in person, as one possibility -- it
notes there are other methods and it is silent on
comparing them -- but it then notes that the
election will fail if voters don't add extra
preferences (UnFairVote seriously tried to cover
this up with all kinds of bogus arguments, but
the section in Robert's Rules is quite clear, UFV
was merely arguing from how a non-parliamentarian
might read the rules.) RRO also notes that
completing with a single ballot deprives the
voters of the chance for a new consideration, and
it notes Center Squeeze. It does *not* suggest a
runoff election if there is majority failure, it
*requires* a totally new election, with new
nominations, the whole thing. Sure, organizations
can create a bylaw allowing election by a mere
plurality, but that is *strongly* discouraged,
for quite good reasons. There are much better
ways. And, in reality, those better ways are
sometimes used. I'll leave off this detail at this point.
Combining Approval or Bucklin with runoffs, if
needed, is the obvious next step beyond Count All
the Votes. Compared to top two runoff, it costs
nothing, and will reduce runoffs, Bucklin will
reduce the need more than Approval, for obvious
reasons, the ability to rank as well as approve.
Count All the Votes and Bucklin are trivial to
canvass, they are precinct-summable, and only
require some special care if one wants the sum of
votes to equal the total valid ballots. (i.e,
provide a count of the voting patterns, but with
voting machines or scanned ballots, this is trivial.)
(The case that dumped Bucklin in Minnesota, Brown
v. Smallwood, noted that it was the number of
voters that mattered, not the number of votes,
but the court then proceeded to decide oppositely
to their own argument. Again, the history of
voting systems in the U.S. is full of examples
where our public institutions betrayed democracy.
That decision was very unpopular in Duluth, as
the review by the court noted. They did not care.
They were the SUPREME Court. Their decision was
one of a whole class of decisions where courts
were obviously operating from attachment to
results, not to actual arguments, and their
arguments against Bucklin were not accepted by
any other court, anywhere. And they knew that, they were proud of it.)
(And it happened in California, too, not widely
noticed. California law requires write-ins be
allowed in elections. San Francisco, with top two
runoff, allowed write-ins. Write-in candidates
were required to register, as I recall, and, if
I'm correct, this was allowed in runoffs also. --
and this can allow voters to fix a problem in the
primary, and *it sometimes happens*. But San
Francisco disallowed the write-in in the runoff,
when there was a viable candidate running as
such. And this went to the California Supreme
Court, which ruled that the city could do that,
because ... because ... the write-in was allowed
in the "election," i.e., in the first round, and
the second round was not an "election." Eh?
Forest for the trees, guys. Voting systems
activists utterly failed to defend top-two runoff
and democratic process, I don't think we even
noticed. We were all in a tizzy about Ranked
Choice Voting, which had just been approved. That
San Franscisco decision was about a method being
dropped, it was the last election. The harm of
allowing the woman to have her votes counted? Uh
... uh ... I know! It might have caused majority
failure! that's it! But, of course, IRV causes
the same thing, *usually*, in races with a pile
of candidates. What were they thinking?)
(Unfortunately, I think I know what they were
thinking. They were thinking that they did not
want this woman elected, and this was how they
could accomplish that. What else?)
IRV does *not* find true majorities, in races
with many candidates, generally. Bucklin did, in
all the major public elections run with Bucklin
that I've been able to find. IRV only finds
majorities by effectively discarding ballots with
no vote for either frontrunner. IRV, then, can
not only choose the wrong top two candidates --
real runoff can do this too, for the same reason
-- but it deprives those voters of the right to
participate in the final election (by eliminating the real runoff).
Except for the first rank, in the historical
implementations, Bucklin is a Count All the Votes
system. You could vote for as many as you wanted
in second and third rank. Given what we now know,
there is no reason to *prohibit* overvotes in the
first round. We know that most voters won't use
them, but they do no harm, and they do provide a
possible way for voters to fix certain rare
problems, as well as to sincerely express equal preference.
(The last Bucklin implementations standing in the
U.S. were party primary elections, and most
voters bullet voted. UnFairVote has claimed this
is because of Bucklin's Later-No-Harm failure,
but that's highly unlikely. Rather, it would be a
reflection of the small numbers of voters in
primary elections tending to be more highly
partisan for their candidate, at that point in
the process, this phenomenon wasn't seen in
public elections. Bucklin was replaced, I think,
in those elections, with top two runoff. In
hindsight, it would have been better to keep
Bucklin in the first round, and hold a runoff if
needed. Reduced cost is then possible, if enough
voters add additional preferences, it is simply
not guaranteed. As I recall, (Un)FairVote cited
9-11% of ballots with more than one candidate
approved, as if that were a bad thing. In fact,
the spoiler effect frequently operates with
margins smaller than that. A few percent of
additional approvals can complete an election,
more fairly than without them. Or it will have no effect. It does not do harm.)
I readily assume that Jameson noticed this from
me, though, of course, he could have discovered
it independently. I'm glad that the knowledge is getting out there.
What is missed in many discussions of RCV in San
Francisco, and where UnFairVote has been pushing
to replace Top Two Runoff with IRV, is that these
are almost all nonpartisan elections. IRV works
in a partisan environment where there are only
two major parties. It breaks down if there are
three, as in Burlington. IRV does almost nothing
in the San Francisco supervisor races, it almost
always comes up with the same results as would
Plurality. In these races, the voters are almost
completely unconcerned with a possible "chicken
dilemma," they don't think that way, only
political activists think that way. With
extremely little exception, the IRV races where
the winner "flipped" from first round plurality,
there was rather obvious partisan voting based on
ethnic affiliation, probably simply operating
through the names. That would function the same
way as would party affiliation being on the ballot.
UnFairVote has been damaging the most effective
and widely used voting system reform, which was
frequently hailed as such, when it was
introduced, top two runoff. All for a very narrow
strategic purpose that will probably fail, to
move toward proportional representation. There
are much more direct methods, but they made this
political decision in the 1990s, and invented the
name "instant runoff voting" to ride on the
popularity of top two runoff, and make it seem
like IRV would produce the same results. It does
*not* produce the same results, and that has been
demonstrated. TTR in nonpartisan elections flips
the winner about one-third of the time, whereas
it almost never happens with IRV. IRV is an
expensive form of plurality, as to real results.
In nonpartisan elections, where vote transfers do
*not* follow partisan patterns.
IRV does fix the spoiler effect in partisan
elections, but then the effect returns with a
vengeance if a third party approaches parity.
That's completely unnecessary. Bucklin is Instant
Runoff Approval, is far simpler to canvass, and
doesn't produce those nasty outcomes, obviously
unfair, as in Burlington. It fixes the center
squeeze problem of Top Two Runoff and IRV. And
it's a door to further reform. And it simply
starts with Count All the Votes, i.e., Approval.
Then we add ranks, so that voters can specify
order of preference. That's all. That's what a
Bucklin ballot does. It's very simple, and easy
for voters to understand, and optimal strategy is
quite close to simple sincere voting. That is, I
assume, why Bucklin was wildly popular with
voters, wherever it was implemented.
Abd ul-Rahman Lomax
>Oh, by the way, you can run that system on accu-scan voting machines
>(or any others; even on the most old-fashioned.)
Speaking about Bucklin. Of course. (Below, I get into delegable
proxy. Maybe I should adjust my meds. Be warned.)
The junk in San Francisco, that Bucklin could not be run on Edison
voting machines, was probably bogus. It might have been a bit
cumbersome, though. I'd guess that the system could handle
vote-for-N, because that's a common system used in places for
multiple-member boards (where it's a lousy method, by the way).
So a Bucklin race would be a series of questions:
1. Vote for first preference: A B C D E (write-in)
2. Vote for second preference: ....
3. Vote for approval, preferred to runoff: ....
Write-ins with voting machines always required hand counting of
write-ins. They didn't need to be counted unless the aggregate
write-in votes were large enough that, if they were all for one
candidate, that candidate might affect the outcome. So they are first
aggregated as if a single candidate.
Technically, though, the Edison machine probably could not handle the
ballot appearance as specified in the referendum. The outrage there
was that the Board bought a single voting machine just to be able to
kill Bucklin, obviously. And the Supreme Court simply looked at a
tree or two, the exact language of the California law, and completely
neglected the forest, i.e., public policy, and avoiding that neglect
is why we have courts in the first place, with human judges.
Unfortunately ... the judges *are* human and may make decisions based
on desired partisan results rather than legal principles and the
protection of the public. Which is why the public must supervise the
courts, ultimately.
Advanced voting systems threaten established political interests, and
those interests, by definition, have enhanced real political power --
which includes what happens behind the scenes -- so they may be able
to effectively block them. It's been called the Lomax Effect, but
it's related to the Iron Law of Oligarchy.
I'm interested in it because I've come to think there is a way around
it, both the Lomax Effect and the Iron Law.
Those effects seriously damage society, long-term. The answer is not
killing the Oligarchs, or excluding them from power, that's an error
which itself led to great mischief and vast harm, the answer is in
creating process that can watch the watchers, without setting up a
new centralized power. There are lots of forces moving in that
direction, but because the danger of centralization is commonly
unrecognized, most of these forces simply set up new oligarchs,
replicating the problem.
Centralized intelligence is, in a word, stupid, if that is all there
is. Human individual intelligence is highly decentralized in the
nervous system. There is no Dictator Neuron. There is, quite likely,
a kind of Range voting system, where many different neural systems
vote for attention and action, and repetition and persistence win.
When there is no majority, we are "confused" or "uncertain" or
"undecided." People who act based on a plurality, inside, or even
based on single impulses, are mentally disturbed, quite
dysfunctional, if the decision is an important one. (It's fine and
even excellent for investigation and play and, thus, learning). Most
of us will wait for some level of consensus, until waiting itself is
a problem, in which case we'll go with a majority -- or, if the wolf
is just about to catch us, for a quick plurality. It usually works.
So, now, how to apply this in human social structure what we can see
in individual internal structure, to create social intelligence. As I
see it, it requires hierarchy through highly redundant linkages
between "neurons," now understood as individuals, with the synapses
being direct communication. It's best face-to-face, because of the
very high bandwidth. Voice is second choice, probably a hundred times
slower, and text is more efficient, in one way, as to documented
result, but is very slow for expression (while faster for reception,
for most people, who can read much faster than they can write.)
This is Delegable Proxy, or Liquid Democracy, but, for maximum
effect, in the broadest "organizations," there would be no
centralization of power, no imposition of central decisions upon the
members of the DP system. No collection of funds for action *in the
DP system* based on majority vote, but rather only advice,
recommendations, intelligence. This is the independence of the
judiciary, really, total independence from the executive. For the
supervisory "organizations," no direct power, only the power of
information and advice, other than the very limited decisions made
about the communications structure itself, with safeguards through
ready fracture. The true organizational power remains with the
inviduals, who may *choose* to delegate it, as well. That's what they
do when they give money or direct support to traditional
organizations or to other individuals to use or distribute. This is a
hybrid structure, it creates a formal supervisory organization, while
keeping traditional centralized-power organizations to handle power.
The process of intelligence is kept separate from the process of
power, because power corrupts intelligence. As to intelligence, it's
anarchist. As to power, it's traditional, but traditional power
coupled with supervisory intelligence, and distributed real power,
ultimate power, being reserved for the people themselves. The
executive power of the traditional organization cannot act to exclude
from the supervisory organization. The supervisory organization is
"factionalized" through the natural caucuses that will form around
individual proxies. Each proxy controls his or her own process with
those represented, and can exclude (decline to represent, decline to
directly communicate), but obviously can only exclude from their own
"caucus." Not from the overall organization.
The overall organization is not powerless against trolls, but it will
sensibly, because consensus is powerful, always provide paths for
legitimate participation, through chosen representation. Centralized
structures, such as a top-level mailing list, may have restricted
participation, but if the management of those lists is
over-exclusive, they will be reducing their own power. Natural
consequences, it all flows from the structure being only advisory.
Anyone can set up an independent advisory structure, if needed. And
if the central organization is following FA/DP concepts, it will not
be necessary to do this separately, it can simply be done through new
proxies and new natural caucuses.
I'm very gratified to see the Pirate Party using delegable proxy, and
at least some of these concepts. The Pirate Party itself is a legal
entity, with a legal power structure, it must be. The process that
has been set up, if it's optimal from my perspective, will *never*
actually control the party directly, i.e., require the Party to do
anything simply because of some voting in the Liquid Democracy
process. However, the Pirate Party, if it ignores or with poorly
considered reason, neglects the advice form the LD process, it has
(1) ignored its support base, and (2) is vulnerable, then, to
fracture, if the LD process has set up one or more truly
representative virtual caucuses. Each one of those has a natural leader.
Further, if the PP bureaucracy sensibly disgrees with the LD
conclusion, it has a ready structure in which to negotiate a better
decision. It doesn't have to take the "ignore the people, they don't
understand" position which is behind a lot of such activity in
parties. It can explain, to the people, back through the highly
representative proxies, it can negotiate reversal or compromise.
Intelligent process. It takes a lot of discussion, so what delegable
proxy does is to break the discussion down into smaller discussions,
with reduced participation set, making real discussion possible. Mass
discussion fails, generally, Wikipedia is an absolute proof of this.
Abd ul-Rahman Lomax
>http://blog.sfgate.com/cityinsider/2012/03/06/supervisor-wants-to-eliminate-ranked-choice-voting-for-citywide-offices/
UnFairVote has managed to control how RCV is
viewed in the press. They got the same lie into
the voter information pamphlet that voters had
when voting on RC in the first place.
>Under ranked-choice voting, voters choose up to
>three candidates in each race and rank them in
>order of preference. If no one candidate gains a
>majority when first choices are counted, the
>candidate with the least number of votes is
>eliminated and those voters’ second choices
>then go to the other candidates. The process
>continues until a candidate secures more than 50 percent of the vote.
Nope. Not 50 percent of the vote. It continues
until all candidates have been eliminated,
leaving only one, who wins, even if more than 50%
of voters have not voted for that candidate, even
if more than 50% of the voters would favor
another candidate, presented with the explicit choice in a real runoff.
>Backers of ranked-choice voting say the system
>ends the need for costly runoff elections and
>has elected a wider diversity of candidates.
There is no evidence for "elected a wider
diversity of candidates." Some years ago, I
carefully studied the results of IRV in San
Francisco. There are two situations.
(1) True nonpartisan races, where there is no obvious affiliation to follow.
In these races, IRV has no apparent effect on the
winner, who would have won under Plurality. Top
two runoff, though, flips the winner about
one-third of the time. The real encouragement of
diversity in San Francisco was Top Two Runoff,
and it led, rather obviously, to an explosion of
candidates. That created some problems, which
could have been addressed by using an advanced
voting system in the primary. In particular,
Bucklin finds real majorities, because, unlike
IRV, it doesn't discard votes. It accumulates
them. Bucklin was passed in San Francisco in the
1920s, but was taken out through a technicality
by the Board of Elections, which obviously did
not want an advanced system in place.
(2) Partisan races, or the equivalent. IRV works
to eliminate the spoiler effect, but where there
is a third party or clear faction, and it rises
to the level that it might actually win, the
spoiler effect returns and can do major damage.
Because supervisor elections in San Francisco are
theoretically non-partisan, we see that IRV
almost always results in the same as plurality
would, for these elections (based on
first-preference votes alone). However, there is
an exception in San Francisco, where ethnic
affiliation was helplessly displayed through
candidate names. In this situation, IRV functions
as if there is a partisan affiliation.
Sure, IRV "ends the need for costly runoff
elections," but at the cost of democratic choice.
There is a far better way to avoid *most*
runoffs, and it starts with something very
simple. Just Count All the Votes. If this is
done, multiple approvals become a *possibility*,
which *will* be used by some voters who favor a
no-hope candidate. They will add a vote for the frontrunner.
This is an obvious improvement over vote-for-one
Plurality. Only a few voters will add additional
approvals, but it can be enough to find a majority for the winner.
And then there is a method that allows voters to
rank candidiates, it could be called Instant
Runoff Approval. It's a very simple method to
canvass, and simple to understand how to vote,
and it was used widely in the U.S. for a time,
political scientists were quite excited about it,
back in the 1920s. It was very popular with
voters, and it was probably eliminated *because
it worked.* It fixed the spoiler effect, but, in
addition, it actually elected candidates who
would not have had a chance otherwise, because of major-party dominance.
IRV eliminates runoffs, sure, but at the cost of
finding majority approval of the winner. It would
have been cheaper, and possibly fairer, to just
eliminate the runoffs and use Plurality.
But much better is possible. Use top two runoff,
which was always considered an advanced voting
system in itself, but use an advanced system in
the primary to (1) find majorities if possible --
and Bucklin was able to find majorities in
elections with *many* candidates -- (2) make
better choices of candidates for a runoff, if
needed. IRV can make a *terrible* choice when
there are three viable candidates, for its
"instant runoff," and ordinary top-two runoff can
do the same. But Count All the Votes (Approval)
is less likely to do this, and Bucklin would
almost never do it, with good rules.
Both Count All The Votes and Bucklin (which also
counts all the votes) are very easy to canvass,
unlike IRV, where a single error in a precinct
can cause all other precincts to need to be recounted.
Count All the Votes is so obvious as a desirable
reform that, even if IRV were kept, all the votes
should be counted. That would allow IRV to become
ranked Approval, similar to Bucklin, but with a
weird rule that discards votes. Still, with full
voting allowed (i.e., voters can rank more than
one candidate in a particular rank), it would
greatly improve IRV. 3-rank RCV with multiple
votes allowed would become a more flexible
system, able to handle more candidates, voters
could rank a "candidate set" rather than just a
candidate. Just as an option! Nobody has to vote
that way, but the question is whether or not
voters should be *allowed* to do it. And why not?
Have I mentioned that we should Count All the
Votes, no matter what system we use?