[EM] Approval-based replacement for jungle primary

[Kristofer Munsterhjelm]

On 2018-12-04 08:11, Ted Stern wrote:
> Hi Rob,
> 
> What you're describing with your percent satisfaction sounds a lot like
> Proportional Representation, though the context of a primary is quite
> different.

If you're doing what is in essence proportional representation, then the
primary problem and the PR problem becomes the same - so use a
proportional representation algorithm. (Something like PAV.) That's
pretty much what James Green-Armytage concluded in his runoff paper (
http://jamesgreenarmytage.com/runoff.pdf) too.

The difference seems to be that in a primary, you want to elect however
many winners are required to keep voter satisfaction above a certain
level, rather than elect a fixed number of winners. Furthermore, it
might be necessary to adjust the outcome towards a proportional ordering
(in practice, only considering outcomes that include the Approval winner).

Using proportional representation also avoids having a bunch of bland
center-seekers, because if the center only satisfies x% of the voters,
it will only get x% of the seats, up to roundoff errors and the
center-skewing effect of including the Approval winner. In proportional
representation methods, a slight center-skew might not be too bad, as it
helps kingmaker scenarios favor the center instead of a wing. In a
primary, there's no kingmaker scenario that I can see, so the reason for
including the Approval winner is mostly to not make the method any worse
than if it had just been single-stage Approval.

So, the above seems to suggest that the cutoff shouldn't be on support
of the last candidate, but on satisfaction (what something like PAV is
optimizing). On the other hand, just *explaining* what that score is
might be hard... and the more seats you use PAV to allocate, the more
opaque the workings of a global optimization method might become in the
eyes of the voters.