[EM] New simple kind of party-based proportionality, avoiding deweighting, and using range-style ballots

[Jameson Quinn]

2009/10/9 Warren Smith <warren.wds at gmail.com>

>  Quinn then
> mutters about using correlations to cause a psuedo-party structure
> without actually having parties, but I think that is a bad idea and it
> too presumably could be manipulated.
>

I don't see the basis for that presumption... but my muttered proposal was
simply a random idea inspired by your proposal, and I don't want to get
caught up defending it.

Here's some further muttering, if anybody's interested; otherwise, just tune
it out and skip to the next paragraph:

(My example was an attempt at a "select-and-punish-by-reweighting-the-votes"
proposal expressed as a summable process. While I can't vouch for the
validity of the math that quick attempt, I do think such a process is
possible - a way to adjust a summed matrix which is mathematically
equivalent to readjusting the votes, assuming that the votes for each
candidate are the sum of a constant term, a one-dimensional correlated
variance, and a perfectly-uncorrelated noise factor. Since that model has 3
DOF, you could choose the first 3 winners from the summed matrix exactly as
you would from the reweighted votes, only getting possible deviations from
the "true" reweighted winners starting with the 4th winner. If I'm right
about this, it would take advanced linear algebra as well as 3 dimensions of
low-error knowledge of voter preferences to manipulate such a system.)




>
> In particular, with my scheme all candidates from a 51%-top-rated party
> could
> each choose to be in their "own" party-of-one, and then they'd win
> 100% of the seats!
>

That's exactly what I said in my second objection to your proposal: "one
candidate per "party". That way, a solid, strategic 51% coalition can elect
their whole slate."


> A variant intended to overcome that objection is to demand the ratings
> ballots
> be like in "Asset voting" instead of range voting -- that is, all the
> scores are >=0 and SUM to 100.


> A party's share then instead is the sum of all the votes its candidates
> receive.
> Parties splitting up, or merging, will not work to increase their share.
>
>
That's a much simpler fix than the one I proposed, and it does address my
second objection (51% dictatorship, right above) very nicely. However, my
first example (Alice, Al, and Berenice - you're right, I mistyped the
winners, Alice wins in the example as given) still applies, with some
fiddling with the numbers: you can still have voters like the "independents"
in that example who have to choose between voting their favorite candidates
across parties, or making a more-partisan vote which could end up electing
one of their least favorite candidates. In order to avoid such a situation,
voters from all parties would have to agree about the relative strength of
different candidates within each party, and only disagree about the relative
strengths of the parties themselves. This is clearly violated in any
"n-dimensional ideology" model.

Jameson
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20091009/29557b2a/attachment-0003.htm>