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

[Kristofer Munsterhjelm]

Peter Zbornik wrote:
> Hi,
> 
> I am a member of the Czech Green party, and we are giving our statutes
> an overhaul.
> We are a small parliamentary party with only some 2000 members.
> Lately we have had quite some problems infighting due to the
> winner-takes-it-all election methods used within the party.
> I am sitting in the working-group for the new party statutes, and
> would like introduce proportional elections instead.
> So I thought I might find some help in this forum in formulating my proposal.
> 
> There are several practical different types of elections in the party,
> which need to be addressed:
> 1. election of council members
> 2. election of delegates to regional and national rallies, where the
> regional and national council members are elected
> 3. election of candidates to the ballot - primary elections
> 
> In this letter, I would like to ask you to propose good proportional
> election system for the election of the council members.
> A council exists at all levels in the party organization: national,
> regional and local.
> 
> SCENARIO 1: COUNCIL ELECTIONS
> We have to elect the following:
> 1. Election of the party president
> 2. Election of one or more vice-presidents in order of importance,
> i.e. first vice president, second, third etc.
> 3. Election of the rest of the council members
> Normally the council has five or seven members.

Elect the president and VPs by a good single-winner method. I would 
recommend Schulze or MAM - the former is used elsewhere so you can use 
organizational precedence as an argument, whereas the latter is quite 
easy to explain. The vice president/s would just be the second, third, 
etc on the ordering provided (i.e. if you want to elect president and 
VP, whoever finishes second becomes VP).

Elect the rest of the council by STV. With a size of five or seven, you 
could reasonably do that election by Schulze STV (with a computer 
count), but you don't want that complexity, so I suppose that is out; 
thus use ordinary STV. You can use OpenSTV to determine the winner: 
http://www.openstv.org

If lack of precedence is not a problem, you might use QPQ instead. This 
method provides results similar to Meek STV, yet without the iteration 
required by Meek. QPQ is also supported by OpenSTV.

I prefer STV and QPQ to RRV because the former two satisfy a 
proportionality by bloc criterion that RRV doesn't: if a group of at 
least k Droop quotas worth rank the same set of r candidates above the 
rest (not necessarily in the same order), then the outcome will contain 
at least a number of candidates equal to the minimum of r and k from 
that set.

If you need the single-winner method and the multiwinner method to be 
fundamentally the same, patch your version of STV to, when eliminating 
in an n-seat election, eliminate the loser of an (n+1)-way Plurality 
contest among the lowest ranked candidates (all other candidates removed 
from that count). For IRV, n=1, so this means that the "loser runoff" is 
for the bottom two so that the one that is ranked above the other most 
often is spared from elimination, thus ensuring the Condorcet winner is 
never eliminated. Using a proper Condorcet method is better, but if you 
must, it's better to have a somewhat Condorcet method than one that 
isn't at all.

> DELIVERABLES FOR IMPLEMENTATION:
> In the end, if proportional elections are to make their way into the
> party statutesm, then I have to deliver the following:
> 1. a proposal of a text to the statutes, describing the election rules
> and procedures
There's an unfinished statute concept for the Schulze method here: 
http://groups.yahoo.com/group/Condorcet/message/645 . The text is 
missing a definition of weakest defeat, which is the defeat of one 
candidate by another where that defeat is closest to not being one at 
all, i.e. a tie. See 
http://www.telematicsfreedom.org/en/project/15/ssd-and-cssd-condorcet 
for an example.

> 2. a motivation of the proposal which shows why it is better than the
> present one.

The single-winner methods are better because they don't polarize as much 
as top-two runoff. Condorcet methods tend to find centrists with wide 
support, which should help moderate the infighting you speak of.

Using STV (or QPQ) rather than bloc vote is better because bloc 
vote/SNTV is only proportional under strategy (and considerable 
coordination) whereas STV is proportional when voters vote honestly. In 
simple terms, STV and QPQ distort proportionality less than does 
plurality at large.

> 3. a vote counting computer program which works

See OpenSTV.

> 4. preferably a ballot scanning program

I am not familiar with these. I think the best approach would be to find 
a general scanning system (OCR or optical scan), then feed the output to 
the vote counting program.

> 5. preferably some good examples that the system works in real life.

There should be plenty of examples for STV. For Schulze, I don't know of 
any explicit such examples or statements, but organizations like Debian 
and Wikimedia use it and have not changed away from it. See the list at 
http://en.wikipedia.org/wiki/Schulze_method#Use_of_the_Schulze_method

> CRITERIA:
> I am looking for a approximately proportional election scheme, which is
> (i) simple for the party members to understand - this is the main
> criterion. A complex system like Schulze-STV has no chance of getting
> required political support

You can't really get much simpler than STV/QPQ and still retain bloc 
proportionality. The exception would be party list, but party list won't 
do you much good.

An unconventional approach would be to elect the five or seven 
Range/Approval/Plurality winners and then weight their vote according to 
the score they got. That is simple beyond simple, but weighted votes may 
make the council too complex.

> (ii) simple to use, i.e. where it is quick to vote and vote counting
> is also quick (max 400 votes cast)

With the ballot scanning decoupled from the actual determination of the 
winners, that should be the case for any voting method.

> (iii) gives results which leave most party members reasonably
> satisfied with the result

Hence the use of Condorcet. If you prefer Warren's utility-based point 
of view, you'd want Range (or some other such method) instead, but the 
party members might not like a method that could elect a candidate 
supported by a minority.

> (iv) votes are cast "secretly" on paper ballots, alternatively on some
> smart electronic voting system that is as secure as paper ballot
> voting.

See point (ii).

> (v) asset voting is excluded due to lack of political support

> Currently I am considering Re-weighted range voting and range voting,
> since it fulfills the criteria above, but other simple-to-understand
> methods could be used.
> Maybe the RRV system will have be reduced to approval voting for the
> high dishonesty scenario.
> This would lower its attractivity.

Again, I consider bloc proportionality important, because it means that 
if you prefer A and B to all others, you don't have to care (as much) 
about whether you vote A>B or B>A - both will count toward electing from 
the set of A and B; or, you don't have to vote the popular members of a 
wing top in order to have your vote count for that wing.

> A contender to the RRV-range voting system is the STV-IRV system used
> by the green party of the USA, since it evidently works, but is more
> difficult to understand than RRV: see
> http://www.gp.org/documents/rules.shtml#section7

If you do, consider the elimination runoff. It makes IRV at least a 
little better.

> QUESTIONS:
> Please propose a voting system fulfilling the criteria above with the
> given assumptions, and answer the following questions:
> 
> 1. if the proposed election system is as simple to understand as RRV
> and range voting,
> name what advantages and disadvantages it has to RRV and range voting.
> Alternatively, which specific variant of RRV and range voting do you
> recommend for the elections described above (normalization of voter
> scores, number of categories, given that 70-90% of the voters vote
> strategically)?
> To clarify: Asset voting is excluded for this election type, since we
> have problems with transparency and political support.

If you have many voters voting strategically and still want a 
score-based system, you could consider one of the Condorcet-Range 
hybrids: Score DSV (http://wiki.electorama.com/wiki/Score_DSV ) or 
cardinal weighted pairwise 
(http://www.mcdougall.org.uk/VM/ISSUE19/I19P2.PDF ). Cardinal weighted 
pairwise is a modification to a base method (like Schulze) where the 
direction of a defeat is based on rank, but the magnitude is based on 
the margin between the two on those ballots that rank the winner of the 
match-up above the loser.

> 2. In which order should the election of the board members be
> performed in order to insure that all the voters will be reasonably
> satisfied with.
> a] how should the elections be done, it the current election order
> should be preserved (i.e. first you elect the president, then the vice
> presidents etc.)?

First do an election for president and VPs. Run the ballots through the 
method of choice to get an ordering. The one who is listed at top on the 
ordering (outcome) becomes president. The one who is second becomes the 
first VP and so on down. That's one election.

Then elect the rest of the council. Run the ballots through STV (QPQ, 
...) to get the composition of the rest of the council.

> b] Is one election enough to give an unambiguous winner, even if the
> president is elected by the margin of one vote?
> The talk about RRV not electing the Condorcet winner makes me a little nervous.
> The election of the president has to be unambiguous, and several
> elections is not a problem.

The Condorcet methods I've mentioned, as well as the hybrids and the 
patched-up IRV version, all elect the Condorcet winner when one exists.

> c] if you reverse the election order, i.e. first you elect the board
> members, then the president and the vice presidents, and lastly you
> elect the president? This order of election seems to be more simple to
> conceptualize.

I think it's better to do the presidential election first. Consider the 
case where the council/board is elected first. Then those who support a 
candidate X has to be careful not to rank him too highly, because they 
may want him for president, but if they ranked him too highly he would 
be elected to the "rest of the council" first and so be excluded from 
standing for president (as he couldn't keep two positions at once).

Therefore, it makes sense to elect the higher positions first. The 
voters that don't get their candidates elected as president can still 
vote for them in the ordinary council election.

> d] what are the main advantages of the your preferred method to the
> current election system?

See above, "motivation of proposal...".

> e] (optional question) if a member of the board leaves his/her
> position before the end of the election period, and a new member of
> the organ has to be elected, how should this election take place in
> order to insure proportionality is retained?

Keep record of the ballots and internal state of the multiwinner method. 
When the candidate leaves, consider him eliminated and then continue the 
election to find out who will take his place. For STV, this is better 
explained here: http://en.wikipedia.org/wiki/Countback#Countback .

Another solution is for each candidate to create a list of successors. 
When the member leaves, the successor is picked from the relevant 
member's list.