[EM] Robert: Condorcet, IRV, Approval, Score

[Michael Ossipoff]

Robert said:

i'm opposed to too much technology in governmental elections.  at least
regarding the instrument of voting.

[endquote]

Automated balloting and counting introduces particularly easy
count-fraud opportunity.  The only secure ballot is a paper ballot,
and the only secure count is a handcount. The best proposals are
methods that are easily handcounted, such as Approval and Score. The
worst proposals are methods difficult to handcount, such as Condorcet.
We agree on that.

Robert said:

i think it's well worth it to kill
a few trees to insure the integrity of the democratic process.  i think
it's important that the physical instrument that voters mark has built
into it the candidate names so it is clear upon manual review what the
voter was looking at when the ballot was marked.

[endquote]

For rank-balloting, the ideal would be a voting machine that would
print out a paper ballot, for the voter to examine, and then deposit
in a ballot-box in the usual way.

Robert said:

and as far as election method, i am convinced that inherent simplicity
is important .

[endquote]

Yes, and that's why Approval and Score are the only proposals
deserving of consideration for official public elections.

Robert continued:

  and i will concede that First-Pass-the-Pole is the
simplest to vote and simplest to count and determine the winner.

[endquote]

...And Approval is 2nd best, and Score is next best.


Robert said:

and
FPTP is inherently precinct summable which helps insure transparency

[endquote]

Precinct-summability was something that we came up with in order to
argue that Condorcet has an easier and more secure count than IRV.

It doesn't.

Robert said:

, no
way to fudge the precinct results as they are transferred from precinct
to the central counting location where the winner is determined.

[endquote]

Every method requires that precinct results be sent to a central count
location. Doing so for one method is not more secure or insecure than
doing so for another method.

Precincts should securely record the results that they send to central
count headquarters, and broadcast them locally.When the central count
location completes its work (at each stage, if the method is
multistage), it should broadcast or post all the precinct reports, so
that each precinct can examine them and compare them to what it has
recorded for what it sent in.

Robert continued:

 and i
think that precinct summability is the simplest way to be transparent in
that regard.

[endquote]

No. There is nothing non-transparent about IRV's count.

IRV and Bucklin suffer from the disadvantage of being multistage
methods in which there are a number of counts at each precinct, whose
results must be relayed to central count headquarters--and a number of
counts at headquarters that must be broadcast. In IRV and Bucklin that
amounts to broadcast information about whether or not there is a
majority yet. In IRV, there is the additiional information about who
is eliminated at that stage.

It's a disadvantage because it slows down the count. But that isn't
entirely bad, because it divides the count labor into smaller parts,
giving counters a rest while they wait for the central count at each
stage. But there is no loss of transparency.

Robert said:

Instant Runoff Voting is not really precinct summable

[endquote]

As explained above, IRV's multistage nature slows down the count, but
also divides it into convenient smaller parts, giving rests for
counters. It doesn't result in any loss of transparency.


unless you limit the number of candidates to a very small number.  but
each precinct could be required to give a copy of the raw ballot
information which is what gets transported downtown to be counted, to
share that information at the precinct with candidates, media, and other
interested parties.  to *post* it.

[endquote]

Precinct results and central count results should be broadcast and
posted.  And securely recorded and stored. The ballots themselves
should be securely stored.

As I've said before, with N candidates, an IRV count is nothing other
than 20 Plurality counts.  ...except for the additional labor of
crossing eliminated candidates off of the ballots. That additional
task adds a squared term to the labor(cand#) function. For Bucklin and
SR, that function is linear.

For Symmetrical ICT, labor(cand#) is quadratic too. But, for any
Condorcet method, the count job has a more complicated nature than
does the relatively straightforward and familiar IRV count.

IRV's genuine disadvantage is its FBC-failure. With our current
electorate, an FBC-failure is unacceptable.

But the fact remains that, even now, IRV would be a tremendous
improvement over Plurality and our usual Top-Two Runoff (TTR).

IRV's compliance with Mutual Majority and Later-No-Harm are a powerful
combination. If there's a mutual majority, then that
properties-combination is as good as FBC. Polls have consistently
shown that most voters are more progressive than the Republocrats. I
believe that there _is_ a mutual majority of progressives.

Of course the problem is that that mutual majority doesn't know that
it's a majority. That's why we need FBC, at least for now.

Condorcetists are in denial about the consequences of traditional
unimproved Condorcet's failure of FBC and CD (a weaker counterpart to
LNHa).

As I've said, voters are (mistakenly) sure that the winner must always
be a Democrat or a Republican. They've also been conditioned that we
should be resigned to and accepting of the corruption of the
Democrats.  ...and that the election of the Republican would be a
terrible and unprecedented disaster, despite the fact that it happens
every few years :-)

Given those beliefs and assumptions, the optimal strategy in
traditional unimproved Condorcet (TUC) is to rank the Democrat alone
in 1st place. That's true even if you don't think the FBC failure will
happen often. I myself would favorite-bury in a TUC election.

Robert said:

but even IRV can be hand counted with the original paper ballots by
literally transferring ballots from one pile to another, which is what
the central election processor does with the raw ballot data.

[endquote]

Transferring ballots among piles isn't necessary, and isn't feasible
in a national presidential election.

As I said, with N candidates, it would be like N Plurality counts,
with the additional labor of crossing the names of deleted candidates
off of the ballots. Sure, it would be entirely feasible.

Robert said:

the problem, of course, with single-mark ballots, is that in a
multiparty, multicandidate setting, *voting* is the most complicated and
tactical voting can be "rewarded".

[endquote]

Many people don't understand that it's been shown that "tactical
voting" is an inherent aspect of all non-probabilistic,
non-dictatorial methods, including TUC. In fact, TUC's strategy
situation is particularly bad, because, even in a u/a election, its
strategy is unknown, for the general case.

TUC's "tactical voting" involves favorite-burial, as does that of
Plurality. IRV too. But at least IRV won't have that problem if
there's a mutual majority. Approval and Score have no favorite-burial
incentive. Neither do Symmetrical IDT, ER-Bucklin, or SR.

I mention SR as a possibility, in case people insisted on a
rank-method, and wanted Borda. Borda is often what people expect when
you mention rank-balloting. Though SR is the best Borda, SR obviously
isn't as good as Score. SR can be used as Approval, and offers
rudimentary, method-assigned fractional ratings. It's probably the
most easily-counted rank method.

ER-Bucklin, unlike SR, offers some genuine rank-balloting benefit,
especially when it incorporates the delay that brings MMC compliance.

But SR and ER-Bucklin fail CD. There would be little point in using a
rank balloting method that fails CD. Such a method wouldn't offer any
significant improvement over Approval and Score.

Three full-propertied rank methods are:

Symmetrical ICT
JITW IRV
ordinary ICT

the most common tactic is
"compromising" and it's not a very fun tactic.  it's when you forsake
your favorite candidate and mark a candidate that you dislike the
least.

[endquote]

That's a big disadvantage of Plurality, traditional unimproved
Condorcet, and IRV (at least with our current electorate's beliefs).




Robert said:

and your "reward" for using that tactic is that maybe you helped
prevent Mitt Romney from getting elected.  this is opposed to the
"reward" is discovering that you helped elect George W. Bush, because
you voted for someone you thought was better than Al Gore.

[endquote]

It wouldn't take much to be better than Al Gore.

 Look up "Al Gore, East Liverpool, Jim Hightower" on your web-browser,
to find out about Environmental Hero Al Gore.

If you "thought" that the Green was better than Al Gore, you were right.

As I've said, we can talk forever about better voting systems, and
argue which proposal is the best, but no one is doing a state
initiative for voting-system reform. Such an initiative is probably
prohibitively expensive and difficult. But the Greens offer a better
voting system, and a much more open media and democracy  through which
we could then enact whatever voting system the public prefer (your
favorite could be on the initiative-ballot). We can debate voting
systems forever,or we can just vote Green in the elections. If people
would read some platforms,and vote for what they want, our next
elections would elect a Green government.

It would be as easy as marking a box on a ballot.

Plurality's u/a strategy is to vote for the most winnable acceptable
candidate. That's the Green candidate. The Green presidential
candidate was on the ballot in 85% of the states in 2012.

Robert said:

because it makes it complicated for voters interested in being
effective,

[endquote]

There's nothing "effective" about giving it all away by voting for
what you don't want.

Suppose we find out what would happen if we actually voted for what we
want. Will you start?


Robert said:

that their vote really helps their political interest, an
election method that protects voters from tactical influence

[endquote]

Gibbard and Satterthwaite showed that incentive for tactical voting
will be present in any non-probabilistic, non-dictatorial method.

But at least we can avoid strategic need for favorite-burial. But, for
that we need to avoid Plurality and traditional unimproved Condorcet.

Robert said:

 and
promises to allow them to express their sincere preference without that
risk, that *simplifies* the voting system in reality.

[endquote]

Then you want ICT, Symmetrical ICT, or JITW IRV.

As I've said, for an electorate who could elect a Green govt by
Plurality, Ordinary IRV's FBC failure wouldn't be a problem.

Additionally, when it's known that there's a mutual majority, then a
method that meets MMC and Later-No-Harm (as does IRV) is as good as a
method that meets FBC.

Robert said:

i am convinced that the ranked ballot extracts the right amount of data
from voters.

[endquote]

Rank balloting would be great, if the count method is good, and if
we're willing to handcount it and don't mind a little more count-work.


it's simple in concept and easy to fill out, despite the
complaints of the anti-IRVers that i have had discussions with.  but the
IRV method of tabulating the vote and deciding the winner is not as
simple as the ranked ballot

[endquote]

IRV _is_ a ranked method. IRV's count is simpler than that of Condorcet.

To handcount Condorcet, you'd need to place the machine-printed paper
ballot in a special holder, a holder having two horizontally-sliding
markers, each moving on a horizontal track that slides vertically on a
vertical track. Such a device could be used to keep track of the
pairwise comparisons. The "ballot-person" would thus keep track of
which pairwise comparison is being done.

S/he would place marker1 on a candidate, and would call out "Pivot
[candidate-name]". Then s/he'd place marker2 on the first candidate
ranked below the pivot candidate,and call out that candidate's name.
For each candidate there is a chart of the other candidates. A
"marker" is at the pivot candidate's chart. When the ballot-person
calls out the pivot candidate's name, the marker begins work at that
candidate's chart. Then, when the ballot-person calls out names of
candidates ranked below the pivot candidate, on the ballot that's in
the holder, the marker adds a tally mark to that candidate's name, on
the pivot candidate's chart.

In that way, the V(i,j) numbers would be recorded. How many people
vote i over j.

ICT would additionally require a count of how many people vote i equal
top with j. It would be done similarly to V(i,j) but would be easier.

Likewise for Symmetrical ICT's equal-bottom count.

So: Condorcet could be handcounted, but it would be more complex and
unfamiliar than an IRV handcount.

If you don't like the above-described Condorcet handcount, then you
know why I don't advocate Condorcet (including ICT) for official
public elections.

Robert said:

 and i happened to be living in a town where
we literally experienced a classic failure of IRV and only because the
IRV did not elect the Condorcet Winner.

[endquote]

Neither does traditional unimproved Condorcet, in an official public
election, due to TUC's strategy-needs, which will drastically distort
people's preferences, making meaningless TUC's Condorcet Criterion
compliance.

By the way, in Burlington, there wouldn't have been a problem if the
Republicans and Democrats had 2nd-ranked eachother's candidate. There
wasn't a mutual majority.



Robert said:

 i am still considering how to sell Condorcet
to Vermonters after the 2009/10 Burlington IRV fiasco.  i have to first
separate the concepts of the ballot form (which is the nearly the same
with IRV, but equal-ranks are allowed) and the counting method.  and
then i have to point out that Condorcet would not have resulted in the
same anomalies that IRV had in 2009.

[endquote]

If you tell people that, with traditional unimproved Condorcet (TUC),
they won't sometimes regret not burying their favorite, then you'll be
lying to them.

Robert said:

by this point some people are
losing patience with the discussion and they say it's all too
complicated.

[endquote]

They wont' think that Approval is a complicated improvement on Plurality.

And Score is so familiar that people would probably accept it too.

There are innumerable ways to count rank ballots. No one agrees on
which one is better. Don't lead the public into that morass.

Robert said:

 i try to, very early in the discussion, suggest a simple
principle that i would hope everyone agrees with: "If more voters mark
their ballots preferring Candidate A over Candidate B, then Candidate B
is not elected."

[endquote]

Traditional unimproved Condorcet fails that principle. If you collect
rankings, there can be cycles, and your principle can be violated.


Robert said:

if we can get here, i am convinced that it's still pretty much organic
and hopefully have no other conditionals ("if" statements in code) are
needed.  *that's* simple and that Condorcet principle keeps the voting
simple and does not reward superficial tactical voting.

[endquote]

Not only does traditional unimproved Condorcet reward tactical
voting,but (like all methods) it makes tactical voting optimal.

Unlike Approval, Score, MJ, Symmetrical ICT, ICT, JITW IRV and
SR--Traditional unimproved Condorcet will create a strategic need to
favorite-bury.

Be sure to disclose that when you offer TUC to people.

Robert said:

so if a method requires *more* information from the voters (like Score
voting does or MJ might

[endquote]

Sincere rating is a lot to ask of voters. But it isn't the best
strategy in Score or MJ anyway. Voters in Score shouldn't be asked to
rate sincerely.



), i think that burdens voters even more than the
basic ranked ballot.  and if the method does not elect the Condorcet
Winner when such exists, then i am not sure when it's
one-person-one-vote

[endquote]

TUC won't elect the sincere CW when there is one, because of the
drastic preference-distortions caused by its failure of FBC and CD.



, and you want the election to turn out the same when
it was two of the many candidates, then i also do not know why you would
want it.

*but* if any such [nonranked] method should be debated, i think right
here is the
right place.  i'm just still a bit dubious of its utility in a public,
governmental election.  simply because it's not Condorcet.


[endquote]

At least Approval, Score and MJ meet FBC. That can't be said for TUC.

  (and i am
also very dubious of claims, like those sometimes made by Score/Approval
advocates, that their currently favorite little non-Condorcet method
does a better job of electing the CW

[endquote]

It would be difficult to do a much worse job than TUC will.

But electing CWs isn't part of my advocacy of Approval and Score. They
have unique social optimizations that I've described.

Mike Ossipoff