[EM] New Hugo (Science Fiction Award) voting method

[Kristofer Munsterhjelm]

On 08/27/2015 01:43 AM, Jameson Quinn wrote:
> Yes, I was the designer of this system, and I was there at Worldcon to
> get it passed.
> 
> What was needed was a proportional system based on approval ballots.
> There are of course a number of options in this vein. Within these
> limitations, this system was designed for:
> 
> -Relative simplicity of explanation. I found that explaining STV-like
> systems which are top-down and so require keeping track of how "used up"
> a ballot is were too hard to explain.
> -Resistance to "bullet voting" strategy, since widespread use of such
> strategy would weaken the non-slate voters against slate voters.
> 
> It is like IRV in that it is a bottom-up elimination system. However, it
> was in no way "based on" IRV, and in fact it differs in one key regard:
> it looks at the whole of each ballot from step one, instead of ignoring
> all but one of the choices on each ballot at any given time.
> 
> I'd be happy to answer any further questions about it here.

In 2010, you suggested a very simple method that you said was
Droop-proportional and that went like this:

1. Count the candidates by approvals.
2. Elect the Approval winner and discard a Droop quota's worth of
ballots on which he's approved. Start with ballots that only approve
already elected candidates, then choose the rest at random. If there
aren't enough (i.e. the candidate was elected by less than a Droop
quota), discard all ballots that approve of the winner.
3. Repeat from 1 until you have enough winners.

This method seems better than cumulative elimination/SDV in these ways:
- It passes monotonicity.
- With high probability, it passes a criterion analogous to the DPC. (I
doubt SDV does since it doesn't redistribute surpluses.)
- It's simple (it takes a certain number of ballots to elect a candidate
and those ballots are then used up). It can be made more fair by
duplicating each Approval ballot n times if that's desired, or by
running the method n times and picking the mode.

(It doesn't handle the "1 seat, elect C; 2 seats, elect L and R"
scenario, but I'm not sure if that's possible with Approval ballots.)

Why did you choose cumulative elimination/SDV instead of that method?
You don't need to keep track of /how/ used up a ballot is: it is either
used up or not.

-

As a side note, the monotonicity failure I had in mind is this:

Suppose some voters have preferences

A > B >> C > D

and suppose that in the next to last round, the candidates remaining are
{B, C, D}. The other ballots are so that D is eliminated first, after
which C beats B. Now the voters raise C:

A > C >> B > D

and B, now having fewer approvals/points than D, is eliminated first,
after which D beats C; so raising C made C lose. This is the typical IRV
monotonicity failure pattern, and single-winner SDV-LPE seems to be
enough like IRV that it fails in the same manner.