[EM] How to combine list and candidate ranking based proportionality?

[Juho]

On May 23, 2010, at 10:40 PM, Aaron Armitage wrote:

> I've considered the question myself, although I've never described my
> ideas publicly. Now's as good an opportunity as any.
>
> I came at it from the opposite direction, so to speak; trying to graft
> lists onto STV to make it scalable, rather than adding candidate  
> rankings
> to a list system. The basic idea is to use the lists as a shorthand,  
> so
> that voting for a list is the same as voting for all those  
> candidates in
> an STV ballot. I think the political environment makes a difference
> between the two approaches. In a parliamentary system you probably  
> want to
> go party first, allowing the parties to know where they stand for the
> negotiations in forming a government even before the individual  
> candidates
> are elected. In America, the parties don't form a government in any  
> case
> (we don't even have a government in the European sense, the  
> administration
> and the majority in Congress being different things altogether), and  
> there
> is so much time between elections and inaugurations that speed isn't
> important, even though people like it. On the other hand, if the  
> counting
> is done quickly enough it may not matter, and a delay of a few days
> normally shouldn't matter much even to a parliamentary system.
>
> The first way of adding lists to STV is simple: you list your  
> candidates,
> and last you put a list, which fills out the rest of your preferences
> according to the predefined order of the list. I suppose you could  
> include
> more than one list, or a list then a candidate, but that would be
> pointless because your vote would be used up.
>
> A simple example: The Yellow list is A>B>C>D>E, and
> the Brown list is F>G>H>I>J. If you vote B>I>Yellow, it counts as
> B>I>A>C>D.
>
> The second way is more sophisticated, and much more complex to  
> count. The
> vote itself needn't be any more complex (you could always just vote  
> a list
> and leave it at that), but it can be, and depending the layout of the
> ballot it may look more visually confusing. It will really need to  
> be done
> with a touch-screen, preferably using a drag-and-drop interface. A  
> paper
> ballot should be printed out and kept as a check (or perhaps the paper
> count should be all there is, the computer interface being simply the
> means of generating the paper ballots).
>
> Instead of using regular STV, the second way uses CPO-STV. Since  
> it's a
> Condorcet method, it allows tie votes which amount to voting present  
> in
> the choice between them. Or, in the case of CPO-STV, between two  
> outcomes
> which differ only in electing one or another of tied candidates. The  
> lists
> are unordered and instead of representing a completed ordering  
> filled in
> at the end of the ranking, they are a tie between all list members.  
> The
> party lists will probably be mutually exclusive, but there's no  
> reason not
> to have other lists which overlap the party lists and each other. If a
> candidate appears in two lists which are ranked on the same ballot, he
> takes the higher of the two positions, but if he is ranked  
> individually,
> as a candidate, he takes that rank regardless of how any of his lists
> might be ranked.
>
> So, going back to the previous example, say you vote
> B>Vowels>Yellow>Brown>Purple>F>White. This becomes:
>
> B > (A,E,I,O,U,Y) > (C,D) > (G,H,J) > (K,L,M,N) > F > (P,Q,R,S,T)

This is a good generic approach to how voters could express their  
opinions.

In an open list system a vote to candidate A of Yellow party would  
mean A>Yellow.

In a tree based system (that is also one approach to combining list  
and candidate orderings) a vote to candidate A could mean  
A>Subgroup1>Subgroup2>Yellow>Partygroup3.

Ability to vote for overlapping groups can be a real benefit to  
voters. It is quite well possible that some vote for candidate X  
because she belongs to some certain ideological group and some others  
because she represents some geographic region.

One could also have tree based systems where the candidates may be  
included in multiple branches. The counting process of course becomes  
more complex then. One candidate could be both part of a candidate  
group that represents the northern region and a candidate group that  
oppose nuclear power. The voter could decide which "variant" of this  
candidate gets her vote.

Trees do limit the ability to use user generated rankings /  
inheritance order but votes are very simple, just votes to one  
candidate. Trees could be used also as default orderings if the voter  
doesn't express anything else.

>
> Notice that separating out a candidate can be used to bury as well  
> as help
> him. In the example above F is rated lower than anyone but the  
> detested
> White Party.

This feature is a bit dangerous, for the same reason as negative votes  
are. It is ok if voters know what they are doing but it could also  
lead to problems. Voters could e.g. generally vote A>Party>B>C>D where  
B, C and D are either the strongest competitors of A or just  
candidates that are well known and therefore many people also dislike  
them. As a result some very unknown or/and weak candidate(s) from the  
"Party" list might win.

>
> Registering seven ranks for only five seats may seem like overkill,  
> but
> every outcome is compared to every other outcome, and you may not have
> used up your vote by the time your grudging preference for F over  
> White
> matters.

Already single-winner elections with multiple candidates could make  
good use of multiple levels.

> Most voters won't have such detailed preferences, of course, and
> they can just vote Yellow (or Purple or...) or even bullet vote. If  
> that's
> all they care about, that's fine. But if a voter has more  
> preferences to
> express, the system should include that information and make use of  
> it.
> That allows the ordering of the lists to be entirely voter-generated,
> based on overlaps with other lists and individual rankings.

I just note that tree based default inheritance order is different  
from party given preference order since there are no "safe" seats but  
voters will decide which branches will get the seats. Trees provide  
thus also fully user-generated results although they set some  
limitations on what kind of orderings are possible. (Parties could try  
to put their "favourites" in the tree next to some candidates that are  
known to get lots of votes but voters will still decide if they will  
give that branch votes, and which candidates on that branch will get  
the seats.)

>
> The second system also allows for a kind of MMP to be added. The  
> larger
> district (or the whole country, or the state, as the case may be) is  
> split
> into single-member districts. The single members are elected  
> according to
> some suitable method -- Condorcet is best, both for the usual  
> reasons and
> because all the tie votes represented by list rankings can be given  
> their
> full weight, but if we just have to we can use IRV, Bucklin, or even
> plurality. Once the local winners are found, the larger CPO-STV  
> election
> proceeds considering only those outcomes which include the local  
> winners.

In MMP style systems one must be careful not to lose proportionality.  
(In the Czech Green Party discussions there was some discussion on  
this. The election of the president and the council have to be linked  
somehow to maintain proportionality (since the president os also part  
of the council).)

In summary I think there are many opportunities in this kind of  
extensions of the basic proportional methods. The biggest challenges  
are in complex calculation rules and complex ballots (or complex and  
tedious filling of the ballots even if the ballots themselves are not  
that complex).

Juho