[EM] a question about apportionment

[Kristofer Munsterhjelm]

⸘Ŭalabio‽ wrote:
> Note: I got swamp with work a few weeks ago, but since Kristofer
> Munsterhjelm went through the trouble of writing me, I should
> respond.
> 
> 
> 2011-04-18T18:46:16Z, “Kristofer Munsterhjelm”
> <km_elmet at lavabit.com>:
> 
>>> The ideas of districts is that the politicians are accountable to
>>> the people in their districts.  If a proposed dam would flood a
>>> district, the Representative would try to stop it.  With
>>> proportional representation, none in the legislature may try to
>>> save the district.
> 
>>> With district-based systems, groups which are not at least a
>>> plurality somewhere receive no representation.  It is good to
>>> have an house of proportional representation and a separate house
>>> of district-absed representation.
> 
>> If politicians are only looking out for their own hides (which
>> seems to be a prerequisite for what you're saying), then you have
>> two situations:
> 
>> - In a district-based system, you would have one representative
>> that is very concerned about the dam, and (n-1) representatives
>> that aren't at all (because the district's votes don't go to them).
>> 
> 
>> - In a PR system, you would have n representatives that are all
>> somewhat concerned about the dam, because the people who would be
>> in the district all affect the composition of the council.
> 
>> Thus, on the average, you would have the same result; only that in
>> one case, all the concern for the dam is concentrated on a single
>> candidate whereas in the other, it is spread throughout.
> 
> People are lazy.  Being somewhat interested probably would not make
> the politicians act.  A politician whose district is about to go
> underwater if a proposed dam comes to be will be very motivated.
> That politician might talk the other politicians into not building
> the dam.

The country I live in (Norway) has PR with multimember districts, and I 
haven't heard of problems like that. Large projects usually get the 
required analysis before they're built, even if they would only impact 
part of a district if they went wrong. (Actually, I'd say there are too 
few large projects of the kind I think is important, but that's another 
matter and related to the particular nature of Norwegian politics more 
than PR.)

>> So I don't see any advantage to single-member representation here,
>> and quite a lot of disadvantages: A Single-member district method
>> requires either an independent redistricting commission or
>> equivalent, a program that will draw sometimes-unusual regions, or
>> that the people endure gerrymandering, whereas multimember
>> proportionality gets rid of this problem.
> 
> We have algorithms which draw nice districts.

You have algorithms that draw contiguous equipopulous districts. 
However, they're completely arbitrary as far as natural communities go. 
The districts might split a city or town in two, or might pass right 
through a representative's house, or any number of such effects.

Now, that's probably unavoidable if you're dealing with single-winner 
districts, since we can't expect that the size of natural communities 
will fit well with the size of the district, but this problem is 
attenuated with multiple winners, because each district can fit more 
people. For instance, here in Norway, each multiwinner district 
corresponds to one of our natural "counties" (first level administrative 
regions), of which there are 19.

>> There are of course reasons for not having too large multimember
>> districts, such as that it's hard to rank 10 candidates (or to know
>> their positions),
> 
> ¿Why rank?  ¿Why not use assetvoting?

I think it's better that the decisions are close to the people than far 
away. The point of representative democracy is that we can't make every 
single decision ourselves, so we need someone to handle the daily 
maintenance - to reduce the variety, so to speak - but the 
representatives have their own objectives that may differ from ours. 
Usually, that's considered part of the cost of having representative 
democracy in the first place, but if we don't have to pay that cost, why 
pay it?

Asset is better than Plurality (and probably SNTV), but I'm not 
convinced it's better than ranked multiwinner systems. The closest thing 
seen to Asset itself - Fiji's voting method twist where each party 
provides a prespecified "negotiating order" through the ranks provided - 
often gave counterintuitive results.

(All of this might seem internally contradictory with that I prefer a 
parliamentary system to a presidential one, but there, I think the 
effect of the legislature continually checking the executive is of 
greater benefit than what is lost by going through a "middle-man" - the 
legislature - to determine the composition of the executive. The people 
can only directly affect the executive once every n years, but the 
parliament can do so every single day, and if elected by PR, it should 
at least be somewhat representative of the people at large.)

>> and that feedback becomes too weak, but the good news is that
>> gerrymandering runs into diminishing returns pretty quickly, so
>> small multimember constituencies would be good enough. AFAIK,
>> gerrymandering an n-member district using a Droop proportional
>> method would only let you swing 1/(n+1) of the vote in the very
>> worst case. Even that disproportionality can be handled - at least
>> on a party level - by something like Schulze's STV-MMP suggestion,
>> if exact proportionality is very important.
> 
> Assetvoting does not require ranking and is proportional.
> 
> I would prefer an house for the states (each state would get get a
> score (20) senators chosen proportionally via assetvoting), an house
> for the districts (chosen via scorevoting), and an house of
> proportional representation (chosen via assetvoting).  I would prefer
> the districts to be drawn without regard to stateboundaries (the
> senate represents the states), but if you are married to districts
> fitting in states, we can split the states into districts totally 548
> (Wyoming has 1 548th the population of the country) for the nation.
> The president would be elected via scorevoting.

It could *work*, but I still think it's too heavily engineered. The 
states also aren't just a bunch of interesting boundaries on a map, 
they're administrative units, and I think it makes sense to elect within 
the administrative units, so I wouldn't reshape the states or have the 
national districts cross their borders.

My own heavily engineered solution would probably involve Condorcet for 
the president (hey, I like Condorcet :) ), STV or some ideal monotone 
multiwinner method for the House, and then if I'm to have three houses, 
the third would consist of people picked randomly (because they're very 
hard to corrupt), refreshed say, a tenth every 1/10 of the cycle decided 
for that house, and that period could be coprime to the others (i.e. odd 
number of years). Perhaps I'd use Condorcet (or a cardinal method) for 
the Senate, though Gohlke's triad system would be even more interesting 
to use; but at that point it's getting really overengineered.