[EM] Preferential Party-List Proportional Representation (PPLPR)

[Kristofer Munsterhjelm]

On 11/06/2014 10:26 AM, Vidar Wahlberg wrote:
> On Fri, Oct 31, 2014 at 10:52:06PM +0100, Vidar Wahlberg wrote:
> <snip>
>
> I'm curious, there seems to be very little interest for party-list
> systems on this mailing list.
> When it comes to governments, party-list is quite common in the more
> successful democracies. Still, even party-list systems tend to lead to
> 2-3 parties being the dominating and practically unchallengeable forces.
> The "wasted vote" mentality is still around in party-list democracies,
> and parties that are liked by most, but not as their first preference,
> will not gain much influence.
> I find little research done on improving party-list systems, and I find
> it unlikely that existing party-list systems would abandon party-list in
> favour of directly voting for candidates.
>
> So why doesn't this gain more attention?

Note that Statistical Condorcet, the multiwinner Condorcet method I've 
been thinking about lately, is a party list method. It is currently too 
impractical to use on a national scale, though it could be used on a 
county level for local assignments.

(In an ideal world, I would find a way of pruning the search space so 
that it reduces to a method that scales according to the combinations of 
number of parties, rather than the combinations of number of seats; and 
then I would find a way of incorporating biproportional representation 
into it. But I am very busy with real world things at the moment, so I 
don't expect that to happen any time soon.)

I would guess that at least part of the reason that there isn't much 
focus on party list PR is because many of the list members are from 
countries that were former British colonies, and thus have 
candidate-based election methods. In those countries, trying to change 
to a better candidate-based election method is probably easier than 
changing completely to party list PR. At least it's an idea that comes 
more readily to mind.

I also think that the focus on single-winner methods contribute to that 
candidate multiwinner methods are more discussed than party-based ones. 
Multiwinner PR is just a generalization of single-winner methods so that 
one elects a winning set (the proportional assembly) rather than a 
single candidate. On the other hand, party list PR either have the same 
"candidate" winning multiple times (if one generalizes based on number 
of seats), or each candidate has a different weight (if one generalizes 
based on number of parties).

Finally, if party list PR is a way of making elections work, and the 
advanced methods constitute another way of making elections work, then 
one wouldn't usually consider the idea of fixing elections in both ways 
at once. The distance from Plurality to (advanced + party list PR) is 
greater than the distance from Plurality to either alone. And some might 
see the "you can vote for a party or for none at all" quantization of 
party list as a necessary evil, in which case it's not very tempting to 
augment a multiwinner method with that constraint.

--

In a conventional (candidate-based) multiwinner method, the number of 
seats (and winners) is fixed.

In a closed party list method, the "candidates" that the voters vote 
over are parties. The number of seats is fixed but the number of winners 
are not; a few parties might win, or many might.

In a weighted multiwinner method, the number of winners is fixed but the 
number of seats are not. In particular, in an unconstrained weighted 
multiwinner method, the weights are continuous quantities, and so the 
number of seats could be considered to be infinite. The weights 
themselves are simply the proportion of seats held by each candidate in 
the limit.

In open party list, the candidates that the voters vote over are party 
members. The method is not that different from a candidate-based 
multiwinner method, except that all the party candidates must be clones. 
That is, if you list the candidates in order of preference of some 
voter, all the candidates belonging to a single party must appear next 
to one another.

Maybe one could generalize a method to serve all of purposes above. Then 
progress on, say, a multiwinner candidate method would also improve its 
party list PR aspect and thus provide an improved party list method.

Even this is incomplete in some respect. It doesn't implement 
biproportional apportionment, which could broadly be stated as "if some 
fraction x of the voters have property P, then the candidates they elect 
(an xth fraction of the council) should also have property P, unless 
this conflicts with proportionality as a whole". Traditional quotas 
(gender, race, ethnic group) would be similar: "if some fraction x of 
the voters have property P, then an xth of the council, plus or minus 
some specified margin, should also have property p, unless this 
conflicts with proportionality as a whole".

But I think I've wandered off-topic enough for one post :-)