[EM] Proportional election method needed for the Czech Green party - Council elections

[Raph Frank]

On Wed, May 5, 2010 at 9:19 PM, Peter Zbornik <pzbornik at gmail.com> wrote:
> What is the advantages of Schulze proportional ranking to the simpler top
> down STV modified method described in
> http://www.votingmatters.org.uk/ISSUE9/P5.HTM?

The first problem with this one is that it will elect the President is
the IRV winner.  This gets you a non-centerist President.  Even if you
use this proposal, I would strongly recommend electing the President
with a condorcet compliant method for the first ranking position and
the proceed to the next steps.

The general problem is that protecting a candidate from exclusion can
break the proportionality process.

In fact, a candidate who reaches the quota could be defeated by a
candidate on a fraction of the quota.

>From the link:

"In exceptional circumstances it is possible that two candidates, not
previously elected may exceed the quota in the same stage."

For example:

16: L>C>R
17: Y>L>C>R
30: C>L>R
37: R>L>C

Rank Number 1 (IRV count):

Round 1:
L: 16
Y: 17
C: 30
R: 37

L is eliminated, all votes go to C

Round 2
L: 0
Y: 17
C: 46 (+16)
R: 37

Y is eliminated and all voted go to C

C wins.

Rank Number 2

Round 1:
L: 16
Y: 17
C: 30
R: 37
Quota is 34

R is elected with a surplus of 3, goes to L

Round 2:
L: 19 (+3)
Y: 17
C: 30
R: 34*

Y is eliminated, 17 goes to L

Round 3:
L: 36* (+17)
Y: 0
C: 30
R: 34*

Now both L and R have quotas, so they are entitled to the 2 seats.
However, C won in the first round, so C must be one of the 2 elected.

The rules would say that as soon as R was declared elected in round 1,
no further rounds would be attempted.

However, the same problem presumably applies to more complex methods.

> Otten ends his second article by stating: "I do not at this point advocate
> that a generalised Condorcet method is adopted. However, I think the idea
> has its merits, and I do believe the question of ordering demands further
> consideration. While a single rule may not be appropriate for all
> circumstances, it should be possible to narrow the field somewhat from that
> in section 5."

I think his second proposal is similar to Schule's ranking method.

It also naturally elects the condorcet winner in the first round.

However, it represents more counting.

> How would you respond to Ottens remark above, which stems from the fact (?)
> that Condorcet methods (and thus Schulze proportional ranking) violate of
> the principle, that "that later preferences should not be allowed to count
> against earlier ones"?

This is a basic criterion that IRV and PR-STV meet.  It is normally
called "Later No Harm".

It says that if you vote

A>B>C>D>.....

the method will not even look at your lower preferences unless you
higher preferences have been elected or eliminated.

So, while A still has a chance of winning, your full vote strength
goes to A.  This might get A elected, but also, by the time your vote
gets to B, it might be to late.

This is like digging your heels in.  One side says that they want
their candidate and will accept no other, and the other says the same.

They then vote, and the side which ends up slight bigger is happy, as
they get their candidate, but the other side is disappointed.

However, if they compromised and accepted a candidate that both sides
agree is good, rather than refusing to negotiate, then the winner
would have been acceptable to both sides, even if not their first
choice.

I think "Later No Harm" is actually a sign of potential problems with
a methods rather than a positive thing.

> A wikipedia article on Schulze proportional ranking for dummies might also
> be of use (I guess anyone with a good understanding of the method could do
> the dummies explanation).

There is one on the standard Schulze-STV method.
http://en.wikipedia.org/wiki/Schulze_STV

This explains how to compare 2 councils.  I think it is easier to
understand than the paper.  However, it potentially is not as
accurate.

Maybe Markus Schulze can comment on its accuracy or otherwise.