Proportional multi-winner ranked voting methods - guidelines?
I just note that it would be useful to have some numeric requirements available when the usefulness of different proportional voting methods is evaluated. Some useful numbers (and other requirements) could be the number of candidates per district (e.g. 200, maybe 30 for each party), number of seats per district (e.g. 30), number of districts (e.g. 15), variation in the size of the districts (e.g. from 3 to 30 seats), handling of "overflow" and "lost" votes (e.g. votes to small parties in a district that has only 3 seats), required accuracy of proportionality (e.g. droop or hare proportionality at national level, or separately at each district), requirement of equal treatment of parties that get their support from one district only vs. evenly from all districts, planned style and accepted complexity of the ballots or voting machines, ability to vote for individual candidates, ability to vote multiple individual candidates across party borderlines, accuracy of party internal proportionality, ability to vote for one's favourite party (e.g. vote to one candidate inherited by other candidates of the party).
Juho
On 23 Feb 2017, at 20:56, Stéphane Rouillon stephane.rouillon@sympatico.ca wrote:
I agree with Richard Lung: STV could very well fit for a country using many representatives in bigger districts. I do not understand Mr. Fobes critics'...
Envoyé de mon iPhone
Le 23 févr. 2017 à 13:43, James Gilmour jgilmour@globalnet.co.uk a écrit :
Richard Fobes (VoteFair) Sent: 23 February 2017 17:54
However, as I stated before, STV would not provide fair results if it were used for a full national legislature.
Recently, in Canada, some people have been promoting the idea of using STV to elect about 5 MPs (members of parliament) from
each district (which they call a "riding"). That would produce very unfair results!
That's what I had in mind when I referred to using STV repeatedly.
Please see the attached results of some STV-PR elections to the Northern Ireland Assembly.
If Canada had results like those from its Federal and Provincial elections no-one would be calling for electoral reform in Canada.
James Gilmour
Edinburgh, Scotland
This email has been checked for viruses by Avast antivirus software.
https://www.avast.com/antivirus
<NI Assembly 1998 & 2003.pdf>
Election-Methods mailing list - see http://electorama.com/em for list info
Election-Methods mailing list - see http://electorama.com/em for list info
---------------------------- Original Message ----------------------------
Subject: Re: [EM] Proportional multi-winner ranked voting methods - guidelines?
From: "Toby Pereira" tdp201b@yahoo.co.uk
Date: Thu, February 23, 2017 5:09 pm
To: "rbj@audioimagination.com" rbj@audioimagination.com
"election-methods@lists.electorama.com" election-methods@lists.electorama.com
I suppose it depends what you want from a multi-winner election. A faction of 51% could command all the seats if you did it that way.
�
yes, i know that. �but, OTHER THAN GEOGRAPHIC division of the constituency into �groups (we might call those divisions "districts" or "wards"), i can't see how a government can legitimately divide the constituency into groups based on race or ethnicity or
gender-preference identity, just to get proportional representation. �we can't have folks in Greenwich Village registering as gay or straight so that the gays get their allotted proportion and the straights getting their proportion.
the alternative is that we elect someone to office even
though more of us voters explicitly mark our ballots that we prefer someone else. �that's the whole point behind Condorcet compliance.
---------------------------- Original Message ----------------------------
Subject: Re: [EM] Proportional multi-winner ranked voting methods - guidelines?
From: "Kristofer Munsterhjelm" km_elmet@t-online.de
Date: Thu, February 23, 2017 5:20 pm
"election-methods@lists.electorama.com" election-methods@lists.electorama.com
On 02/23/2017 09:59 PM, robert bristow-johnson wrote:
i wasn't aware that any STV would be Condorcet compliant (except for
bottom-two runoff STV).
my only thought about a multi-winner method that is Condorcet
compliant is to simply run the rank-choice election according to some
Condorcet method (say Schulze or RP or MinMax or BTR-STV), pick the
Condorcet winner and assign that candidate to the first available seat,
decrement the number of available seats by one, remove this winner from
the candidate pool, and then see who the next Condorcet winner is from
the remaining candidates. rinse and repeat until all available seats
are assigned.
what would be wrong with that approach to multi-winner elections?
That method fails the Droop proportionality criterion and amplifies
majorities into unanimities.
E.g.
51: A>B>C>D
49: E>F>G>H
Four to elect gives a council of {A, B, C, D}. If one wants a
proportional outcome (which is what all the quota business in STV is
intended to accomplish), then a majoritarian multiwinner Condorcet
election doesn't help much. If you want a majoritarian method, then it's
okay, but the subject says "Proportional" :-)
yes, i know. �i just think, in reality (despite what i know about the Chittenden Senate District in the state of Vermont) that with D and E on the edge (and maybe even C and F), that there would be probablistic influences on the
vote that might elect people from the slightly smaller group.
the Chittenden Senate District of the state senate of Vermont is, AFAIK the largest (most seats) legislative district in the United States. �the district elects six candidates (the six biggest vote-getters in a mark-up-to-six
race) and, in the past, many people voted with bullet voting. �this was because there was one Republican (she is the daughter of a former governor who happen to die while in office that that was the beginning of Howard Dean's gubernatorial term) that was well-liked and perceived to be
guaranteed election. �then the six Democrats (or Dem/Progs) on the ticket were literally running against each other in a sorta musical chairs because one of them wasn't gonna get elected. �but Ms. Snelling was appointed to some administrative position and is not on the ticket anymore and
the Dems (or Dem/Progs) have locked out the GOP from that senate district and the GOP really want to slice up that district (after the next census) and i don't blame them. �proportionately, they should get at least one seat and there is likely a portion of Chittenden County that would elect a
GOP state senator. �now the Dems don't bullet vote because their incentive is different. �rather than try to elect the particular candidate they really want, they'll just vote for the whole slate of six candidates because there is no longer this remaining GOP shoe-in candidate that we used
to have.
even though many of us are a bunch of hippies and Vermont is the bluest (or second bluest) state in the U.S., we have our goofy electoral dynamics in the state (which includes having the most successful third party in U.S., in terms of getting people elected) and this mondo 6-senator
district is a little goofy. �i call it a "cluster fuck".
A more proportional Condorcet method could be accomplished this way -- I
think that would be the most simple somewhat proportional Condorcet
method. For n seats:
- Repeat lots of times:
- Randomly divide the voters into n groups
- Order the groups in random order.
- Determine the first group's winner according to the Condorcet method.
- Give the first seat to this winner and eliminate him from every
ballot (of every group).
- Determine the second group's winner, elect, and eliminate.
- Do so until you have n candidate assignments.
- Choose the assembly that you saw most often.
In the 51/49 example above, it's basically a coin toss as to whether any
given group will elect one of {A,B,C,D} or one of {E,F,G,H}, and so you
get a 50-50 split.
this is cool. �but the problem is in the lack of determinism (this "randomly" do anything will cause objection with some).
is there a totally deterministic way to come up with these ensemble averages to get a good estimate of the proportional
representation? �like the outcome should be that A, B, E, and F are winners if there are four seats in this multi-winner election. (and the simpler, the better. �the only way i can think of is to geographically subdivide the district into smaller and smaller into atomic communities or
"microcosms".)
--
r b-j � � � � � � � � �rbj@audioimagination.com
"Imagination is more important than knowledge."
Richard Fobes is saying that the STV glass is not half full, it is half
empty. He is right – sort of. But not that right! The proof of the
pudding is in the eating, as shown by the examples given by James
Gilmour. STV returning officers have long known that they take
short-cuts with the manual count, that are not strictly logical, but
keep the procedure manageable.
Gilmour advised the BC Citizens Assembly on the more sound version of
the Gregory method of transferable voting.
Essentially, proportional surplus transfer is a standard method in
statistics, known as weighting in arithmetic proportion. If transferable
voting is “extremely flawed” then so is basic statistics.
One has to keep things in proportion.
(Borda method is essentially what statisticians call weighting in
arithmetic [series, or geometric series or harmonic series] progression,
which they use as an estimate of the relative importance of categories
of data, when they do not possess actual information of their weights,
in proportion to each other.)
From
Richard Lung
On 23/02/2017 17:54, VoteFair wrote:
On 2/22/2017 3:45 PM, Richard Lung wrote:
Don't understand your remark about STV ...
... for a single national constituency.
...
The HG Wells formula is "Proportional representation by the single
transferable vote in large constituencies."
I assume you are referring to this statement:
STV (the Single Transferable Vote) and similar methods(!) are designed
for a small number of available seats, and it is a mistake to think
that such a method can simply be used repeatedly to achieve fair
results for a large number of available parliament seats.
Although the designers think that they designed STV to handle a full
national legislature/parliament, the method has the same flaw as
instant-runoff voting (IRV), namely the method looks at each voter's
currently top choice (after any candidate eliminations), and that
approach -- of assuming the candidate with the most such "votes" is
the most popular (or the inverse, assuming that the candidate with the
fewest such votes is least popular) -- is extremely flawed.
In other words, STV is like using single-mark ballots, except that
when a voter's marked candidate is eliminated, then the voter
automatically supplies an alternate single-mark ballot.
In spite of that major unfairness/flaw, STV would provide reasonably
acceptable results if only two seats were being filled. And if a
nation has 3 equally dominant political parties, then filling 3 seats
for each district would work.
However, as I stated before, STV would not provide fair results if it
were used for a full national legislature.
Recently, in Canada, some people have been promoting the idea of using
STV to elect about 5 MPs (members of parliament) from each district
(which they call a "riding"). That would produce very unfair results!
That's what I had in mind when I referred to using STV repeatedly.
Although Schulze-STV uses pairwise counting, it still has the same
flaw that it would not provide overall proportional results if it were
used repeatedly to fill more than 2 (or maybe 3) seats in each
district. And this method, and similar methods, would also provide
flawed results if it were used to fill all the seats in a national
legislature (and those reasons are explained in my book).
So, OK, I was not clear about the meaning of the word "designed." I
was referring to the effect of the design rather than the intention of
the design.
Richard Fobes
On 2/22/2017 3:45 PM, Richard Lung wrote:
Don't understand your remark about STV, the name given by Thomas Hare,
who invented it for a single national constituency. (Tho Mill was
prepared to be flexible about this, when he moved "Mr Hare's system" of
Personal Representation in parliament.
The HG Wells formula is "Proportional representation by the single
transferable vote in large constituencies."
Agree with you about preference voting essential for fairness. Do you
have any good source for your assertion that the lack of it in European
party-proportional methods makes them especially vulnerable to moneied
influence?
You could say the same, for instance about the nuclear lobby in Britain
(tho not Scotland).
from Richard Lung.
On 22/02/2017 00:24, VoteFair wrote:
On 2/20/2017 11:57 PM, Armando wrote:
...
I am looking for multi-winner election with fair proportional
representation.
...
I would very appreciate if you can help me giving “guidelines”,
explaining pros and cons, advising further readings. ...
I suggest that you look at VoteFair ranking, which is a method I
developed years ago, over a span of about a decade.
It is described in detail in my book "Ending The Hidden Unfairness In
U.S. Elections," which is available through multiple e-book reading
platforms. The book includes lots of illustrations to make the
concepts easier for "average" (non-math) readers to understand. (With
so many illustrations the file size is large and the low price
basically just covers the download fee.)
Near the end of the book I explain that the same system would work in
other nations simply by increasing the number of parliament members
who are elected using cross-district voting methods.
Based on your questions, here is what I think is the most important
concept for you to understand:
STV (the Single Transferable Vote) and similar methods(!) are designed
for a small number of available seats, and it is a mistake to think
that such a method can simply be used repeatedly to achieve fair
results for a large number of available parliament seats.
You seem to correctly understand that ranking candidates -- rather
than using single-mark ballots -- is essential for fair results. PR
(proportional representation) methods in Europe did not get this part
of PR correct. That's why it is easy for campaign contributions
(money) to easily control European politics.
With these concepts in mind, I suggest that you read the overview of
VoteFair ranking, which is here:
On 2/23/2017 10:43 AM, James Gilmour wrote:
Please see the attached results of some STV-PR elections
to the Northern Ireland Assembly.
If Canada had results like those from its Federal and
Provincial elections no-one would be calling for
electoral reform in Canada.
The referenced data supports my belief that the number of seats per
district is heavily correlated with the number of popular political parties.
In this case there are 6 seats per district, and there are about 6
popular political parties. Specifically the political-party count
includes 4 popular parties that apparently fill the first 4 seats, and
then another 4 mid-popular parties that compete for the next 2 seats (in
each district). (The least-popular parties rarely get seats.)
Although in this case the correlation -- between seats per district and
number of popular political parties -- is a simple correlation, other
correlations are possible.
As a hypothetical example, if there were 5 seats per district, two
dominant political parties might typically win the first four seats (two
seats per party), and less-popular parties might often win the remaining
5-th seat (with different parties winning in different districts).
The more political parties there are, the messier it is to create a
ruling coalition that really represents what the voters want. Too often
a coalition compromises on issues that are most important to the voters
in those parties. If anyone doesn't understand this disadvantage of
needing coalitions, which I've explained before, please ask.
One of the advantages of VoteFair ranking is that it favors fewer
political parties in order to let the voters form their own coalitions
-- in the form of very popular political parties. Yet the method allows
as many parties as the voters really want.
VoteFair ranking accommodates as many parties as possible, yet limits
the number of candidates in each race. Here is how:
-
In every election the popularity of parties is re-measured, and the
popularity rankings are used in the next election. -
Each of the two most popular parties can offer two candidates (per party).
-
The mid-popular parties can offer just one candidate each.
-
The currently least-popular parties are not allowed to offer any
candidates in the race -- in that district, based on party popularities
within that district. (The popularity of parties can vary from district
to district, and only that district's popularity rankings are used for
these cutoffs.) -
The limited number of candidates allows voters to get to know these
candidates better, which is important for ranking them meaningfully
(without saying "after my first and second choice I have an equal
dislike for the remaining candidates").
In a sense, VoteFair ranking does the opposite of what closed-list PR
does. Closed-list PR gives control to party leaders, instead of giving
control to the voters. In contrast, VoteFair ranking gives control to
the voters, so that "insider" candidates (who are liked by party
leaders) cannot get elected -- to either the district seats or the
nationwide seats -- unless they are ALSO the most popular candidates
according to the voters.
The latest development in the U.S. Democratic party serves as an example
of the unfairness of a party being controlled by insiders, rather than
voters. Specifically the new chairperson of the Democratic party was
the insider favorite. Most Democratic voters would have preferred the
candidate who was favored by Bernie Sanders and Elizabeth Warren, who
the voters prefer as the most popular Senators in the Democratic party.
Since the above PR-related comments are about closed-list PR, I'll add
that open-list PR is better, yet it uses single-mark ballots, which are
vulnerable to tactics that take advantage of vote splitting.
Another important point about STV is that it fails to consider all the
preferences of all the voters. Instead, like IRV, it is based on the
mistaken belief that the candidate with the most currently-first-choice
"votes" is most popular, and the candidate with the fewest
currently-first-choice "votes" is least popular. This flaw makes it
easier for the "wrong" (less-popular) candidate to win a seat.
This flaw was recently demonstrated in the U.S. Republican presidential
primary election, where single-mark ballots were used and there were
more than a dozen (12) candidates. The more candidates there are in a
race (not counting the ones who get very few votes), the more dramatic
this unfairness can be.
Expressed another way, the first priority should be to fill seats with
the most popular candidates, and the second priority should be to get
some degree of proportional results for the remaining seats.
VoteFair ranking follows this prioritization. Specifically:
-
1-2-3 ballots and pairwise counting are used to fill the first seat in
each district. -
The second seat in each district is filled by the most popular
candidate (using 1-2-3 ballots and pairwise counting) after
proportionally reducing the influence of the voters who are already
well-represented by the first-seat winner. -
Some remaining "nationwide" seats are allocated to political parties
based on the voters' party preferences (i.e. voters rank political
parties and the first choice is used for this purpose). The nationwide
seats are filled with candidates who were the most popular (in their
party) within their district (but failed to win a district seat), and
who got the most support when compared across district boundaries.
This approach is much fairer than allowing parties to choose their
favorite insiders.
In particular, it defeats the strategy of each party offering just one
"insider" candidate. (The second candidate from the same popular party
is optional, but a popular party that does not offer a second candidate
will become less popular because the manipulation becomes obvious.)
To repeat my original key point, any voting method that only looks at
one voter's currently-top candidate at a time cannot produce fair
results. Fair results can only be assured if all the preferences of all
the voters are considered pairwise (not simplistically one-at-a-time).
Just hiding the unfairness -- by making adjustments according to
political-party "quotas" -- is not solving the underlying unfairness.
Yes, proportionality is desireable, but not at the expense of ending up
with less-popular candidates.
Both can be achieved, but:
-
PR sacrifices fairness by not providing a fair way to control which
candidates win each party's seats. -
STV sacrifices fairness in several ways:
** Through its counting method.
** Through what I'll call "round-off errors" if it is used to fill more
than 2 seats per district (and if the number of seats does not happen to
match the current number of popular political parties).
** It does not handle adjustments in "nationwide" seats, after the
round-off errors have accumulated over all the districts.
Questions? Please ask.
Richard Fobes
On 2/23/2017 10:43 AM, James Gilmour wrote:
Richard Fobes (VoteFair) Sent: 23 February 2017 17:54
However, as I stated before, STV would not provide fair results if it were used for a full national legislature.
Recently, in Canada, some people have been promoting the idea of using STV to elect about 5 MPs (members of parliament) from
each district (which they call a "riding"). That would produce very unfair results!
That's what I had in mind when I referred to using STV repeatedly.
Please see the attached results of some STV-PR elections to the Northern Ireland Assembly.
If Canada had results like those from its Federal and Provincial elections no-one would be calling for electoral reform in Canada.
James Gilmour
Edinburgh, Scotland
On 2/23/2017 12:08 PM, Toby Pereira wrote:
Does the VoteFair method obey a specific proportionality property? For
example, if you insist on electing the Condorcet winner, this is
incompatible with Droop proportionality - e.g.
https://groups.google.com/forum/#!topic/electionscience/n__VhVgjJqg
In VoteFair ranking, the first seat in a district is won by the winner
according to the Condorcet-Kemeny method. The winner of the second seat
in the same district is determined by first proportionally reducing the
influence of the voters who are already well-represented by the first
winner, and then with the remaining preferences, calculating the
Condorcet-Kemeny winner.
So, yes, those results are Condorcet compliant. Obviously those results
alone do not assure proportional results.
For the proportional part of the results, some "nationwide" seats in
parliament (or the legislature) are reserved for proportionality
adjustments. The criteria for winning them is not mathematically
sophisticated (because it involves comparing the popularity of
candidates who were not in the same race), so I doubt that the result
meets any sophisticated proportionality criterion. And I doubt that the
results meet Droop proportionality because that calculation method is
quite different.
With VoteFair ranking, a greater number of nationwide seats increases
proportionality, but that reduces the number of seats filled by the
carefully selected -- Condorcet-compliant -- district-based winners.
It's a balance that needs to be adjusted according to the needs of the
nation using it.
As a clarification, VoteFair ranking does not attempt to achieve fully
proportional results for every political party.
Instead it attempts to ensure that the majority of elected MPs (members
of parliament) best represent the majority of voters, keeping in mind
that political parties are often not really representative of voters.
This concept relates to what's currently going on in U.S. politics.
Most U.S. voters are not well-represented by either the Republican party
or the Democratic party. In particular, if fair methods of voting were
used here, none of the candidates who were in the primary elections
would have won those primary elections.
Expressed another way, voters want problem-solving leaders, but unfair
election methods give us special-interest puppets.
Richard Fobes
On 02/24/2017 05:45 AM, robert bristow-johnson wrote:
---------------------------- Original Message ----------------------------
Subject: Re: [EM] Proportional multi-winner ranked voting methods -
guidelines?
From: "Toby Pereira" tdp201b@yahoo.co.uk
Date: Thu, February 23, 2017 5:09 pm
To: "rbj@audioimagination.com" rbj@audioimagination.com
"election-methods@lists.electorama.com"
election-methods@lists.electorama.com
I suppose it depends what you want from a multi-winner election. A
faction of 51% could command all the seats if you did it that way.
yes, i know that. but, OTHER THAN GEOGRAPHIC division of the
constituency into groups (we might call those divisions "districts" or
"wards"), i can't see how a government can legitimately divide the
constituency into groups based on race or ethnicity or gender-preference
identity, just to get proportional representation. we can't have folks
in Greenwich Village registering as gay or straight so that the gays get
their allotted proportion and the straights getting their proportion.
the alternative is that we elect someone to office even though more of
us voters explicitly mark our ballots that we prefer someone else.
that's the whole point behind Condorcet compliance.
---------------------------- Original Message ----------------------------
Subject: Re: [EM] Proportional multi-winner ranked voting methods -
guidelines?
From: "Kristofer Munsterhjelm" km_elmet@t-online.de
Date: Thu, February 23, 2017 5:20 pm
To: rbj@audioimagination.com
"election-methods@lists.electorama.com"
election-methods@lists.electorama.com
A more proportional Condorcet method could be accomplished this way -- I
think that would be the most simple somewhat proportional Condorcet
method. For n seats:
- Repeat lots of times:
- Randomly divide the voters into n groups
- Order the groups in random order.
- Determine the first group's winner according to the Condorcet method.
- Give the first seat to this winner and eliminate him from every
ballot (of every group). - Determine the second group's winner, elect, and eliminate.
- Do so until you have n candidate assignments.
- Choose the assembly that you saw most often.
In the 51/49 example above, it's basically a coin toss as to whether any
given group will elect one of {A,B,C,D} or one of {E,F,G,H}, and so you
get a 50-50 split.
this is cool. but the problem is in the lack of determinism (this
"randomly" do anything will cause objection with some).
is there a totally deterministic way to come up with these ensemble
averages to get a good estimate of the proportional representation?
like the outcome should be that A, B, E, and F are winners if there are
four seats in this multi-winner election. (and the simpler, the better.
the only way i can think of is to geographically subdivide the district
into smaller and smaller into atomic communities or "microcosms".)
You could in theory integrate out the randomness, basically just
mathematically find out what would happen if you could do it an infinite
number of times. But that tends to make the algorithm a lot hairier, and
then we might just be back to "what does this even do" levels of
complexity for most voters.
Random methods can be very simple. Sortition is really powerful for its
simplicity, for instance. So I guess the question would be, what would
hinder the adoption of a proportional Condorcet method the most:
complexity, or randomness?