New voting system proposed for Hugo award nominations

[Douglas Cantrell]

From what I understand, they've been using plurality-at-large until now, and it caused problems this year because most of the finalists were chosen by a coordinated minority faction.

Here's a link describing the new system that was proposed:

http://nielsenhayden.com/makinglight/archives/016262.html

They say the system is called "single divisible vote with least popular elimination." At a glance it seems to be some sort of instant runoff variant of cumulative voting.

[Clay Shentrup]

http://www.wired.com/2015/08/won-science-fictions-hugo-awards-matters/

[Jameson Quinn]

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 STV-like systems which are top-down and so require keeping track of how "used up" a ballot is, set off people's complexity alarms. The EPH proposal has less bookkeeping.

-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.

2015-08-24 3:14 GMT-04:00 Clay Shentrup <cl...@electology.org>:

http://www.wired.com/2015/08/won-science-fictions-hugo-awards-matters/

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.
To unsubscribe from this group and stop receiving emails from it, send an email to electionscien...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Warren D Smith]

This is an interesting multiwinner system, which is
(a) simple,
(b) approval-like ballots,
(c) obeys a proportionality theorem in the "racist" model where the
voters and candidates
come in some number of "colors" and voters only support those of own
color; then
theorem is that winner color fractions will be same as population
color-fractions
(up to roundoff bounds)..

It is somewhat like asset voting, but no "negotiations" are needed so
the candidates
need not be thinking+acting entities, but could instead be, e.g. pizza flavors.

If desired, one also could change the ballots to be
pie-like (i.e. voter scores each candidate, all scores>=0 and must sum to 1;
or the summation=1 demand is not "demanded" but rather automatically is
"enforced" by rescalings). Plurality-like ballots would be the
simplest but least expressive
and would leave voters hanging after their favorite got eliminated;
pie-like ballots would
be the most complicated and most-expressive.

It seems rather pleasingly simple and nice to me. How it reacts to
strategizing (and just what good strategizing is) seems not
immediately obvious.

Perhaps good strategy if I got to have a pie-like ballot, would be
something like this:
vote A^(N-1) * (1-A) / (1-A^C) for my Nth favorite among the C
candidates, where
A is a tiny positive number. If all voters acted that way, then the
elimination process would act just like IRV in the A-->0+ limit,
albeit it would end once
down to W survivors (in a W-winner election). This behavior
is pretty unpleasant and nonproportional (IRV used in this multiwinner
way is a nonproportional multiwinner system).
The reason I am speculating this perhaps is a "good" voting strategy: this
allows the voter to put virtually 100% of her juice on whoever her favorite
candidate is, among survivors, and do so every round.
If her goal is to elect just 1 person, the best one she can (ignoring
the W-1 other
winners) this seems a decent strategy, and although quite possibly not
optimal, it
is optimal in a sort of short-sighted "greedy" sense.
This also might be a decent voting strategy NOT in the A-->0+ limit,
but using instead the best value 0<A<1, whatever it is (which is not
clear, but could
be guesstimated by computer simulation). The basis for that guess is
self-similarity --
this sort of "exponential vote" behaves the same every round.

Are there good reasons (aside from simplicity) for wanting an
approval-style ballot?


--
Warren D. Smith
http://RangeVoting.org <-- add your endorsement (by clicking
"endorse" as 1st step)

[Warren D Smith]

Yet another possibility would be, the voter specifies a rank-ordering
and then the system gives F(n) of her assets to the n-th ranked candidate,
for some pre-specified decreasing positive-valued function F(n) [then
rescales to enforce sum=1 demand].

Here the function F would be specified by the voting system designer and is
pretty arbitrary.

[Nevin Brackett-Rozinsky]

Am I correct that the novel innovation here is the pairwise comparison for elimination?

Meaning, the part that says, “Look at the two candidates with lowest scores, and elimination whichever has fewer approvals” is new?

Nevin

[Jameson Quinn]

As far as I know, both parts are actually novel. It would not surprise me if either or both had been proposed before—they are not too unusual in terms of proportional mechanisms—but if so, I do not know of it. 

[Clay Shentrup]

I find it very hard to follow. A really simple explanation of just the voting algorithm would be great.

> If there is only one work with the least number of points, then all of the works with the second-least number of points will be also be selected for the Elimination Phase in addition to the lowest-point work. 

Why??

> Elimination Phase: Of the works identified in the Selection Phase, the one(s) that appear on the fewest number of nomination ballots will be removed from all nomination ballots for subsequent rounds as if they had never appeared on any ballots.

On first read, this seems insane. Your arbitrary binary cutoff is *any points whatsoever*???

[Clay Shentrup]

Why isn't it just, eliminate the option(s) with the fewest number of votes and renormalize?

[Andy Jennings]

On Thu, Aug 27, 2015 at 10:08 PM, Clay Shentrup <cl...@electology.org> wrote:

I find it very hard to follow. A really simple explanation of just the voting algorithm would be great.

Did you see the first question under the FAQs?

1. Can you explain the system in plain language?
The important thing to remember is that nothing changes in how you nominate. If you think a work is Hugo-worthy, then nominate it. That’s all. There’s no need to rank your choices at the nomination stage, and there’s no reason not to nominate something you think even might be Hugo-worthy. All we are doing at this stage is throwing names into a hat. The final Hugo voting system, which actually chooses the winner, is unchanged. We could, in theory, simply put everyone’s nominations on the final ballot, but that would make for a very long ballot indeed. We therefore need to narrow the nomination list down. This system narrows down the list by eliminating the least popular works until only five (under current rules) finalists remain. Here are the basic steps to the elimination process:
a. You have one nomination “point” for each category that will be divided equally among the works you choose to nominate in that category. So, if you nominate two works in a category, each will get half a point; if you nominate three works, each will get one-third of a point, and so on.
b. All the points given to each work from all nomination ballots are added together. The two works that got the least number of points are eligible for elimination. One of these works is the least popular and will be eliminated. (We call this the Selection Phase.)
c. To determine which of these two works is least popular, we compare the total number of nominations they each received (that is, the number of nomination ballots on which each work appears). The work that received the fewest total number of nominations is the least popular and now completely vanishes from the nomination process as though it never existed. (We call this the Elimination Phase.)
d. We start over for the next round and repeat the process, however, if one of your works was eliminated, then you now have fewer works on your nomination ballot. This means that each work gets more total points, since you aren’t dividing your point among as many works. For example, if one of your five nominated works was eliminated, your remaining works now get one-fourth of a point each instead of one-fifth of a point. If four of your nominated works are eliminated, your remaining work now gets your full point.

[Clay Shentrup]

That's much clearer. I'm still not clear on how the Approval Voting component is useful.

[Clay Shentrup]

Why not just use 0-1 RRV? It seems a lot simpler.

[Jameson Quinn]

I initially proposed essentially that. People didn't like it. I think that the bookkeeping aspect of "using up" votes was one of the hurdles. Also the divisors seemed arbitrary to people.

This system is easier to explain in terms of how the parts relate to the whole. You have the "points", which make it proportional. You have the bottom-up elimination, which is intuitive, as well as being familiar to Hugo voters from IRV. You have the "nominations" (ie, whole-number approvals), which ensure that each vote is counting fully at the moment of elimination, so that bullet voting strategy is as weak as it possibly can be. And it's pretty easy to tell the story of how this works just like the old system if there are no slates, but how any candidates on a slate will tend to end up eliminating each other (unless the slate has enough of a majority to legitimately win).

It could easily be extended to a range ballot. Warren has pointed out that, if used with a range ballot, there is a way of voting that would effectively reduce it to IRV. That's true, but it's far from clear that that would be strategic, at least, not except in very particular situations. It seems to me intuitively that in a zero-knowledge case or with a realistically small amount of information, and for somebody who cares more about having at least one candidate they like than about the rest of them, optimal strategy would probably be to vote an approval ballot, with the cutoff set so that the sum of the win probabilities of your approved candidates is as close to 1 as possible. Which is to say: like one-candidate score voting, it can be plausibly argued to reduce to approval.

2015-08-28 16:12 GMT-04:00 Clay Shentrup <cl...@electology.org>:

Why not just use 0-1 RRV? It seems a lot simpler.

[Ted Stern]

Hi Jameson,

I worked on something akin to this method 4 years ago, before I was lured away by your AT-TV proposal:

https://github.com/dodecatheon/Cumulative-Tranferable-Vote

In my experiments, I found that CTV was subject to some of the same problems as IRV, though it was summable at each elimination stage:  failure to satisfy participation or IIA; as many stages as candidates.

Given the nature of the Hugo competition, a PR voting method that satisfies participation and Immunity from Irrelevant Alternatives seems desirable.

Ted

[Clay Shentrup]

I thought your proposal also had bookkeeping. I.e if one of your candidates is eliminated, you have to re-scale your points to your remaining picks.

[Jameson Quinn]

In EPH (my proposal), the only "state" information needed to figure out what happens next is the set of ballots and the set of non-eliminated candidates. In RRV, you also need a number between 0 and 1 for each ballot to tell how "used up" it is. So yes, EPH has bookkeeping, but RRV has significantly more. 

2015-08-28 20:55 GMT-04:00 Clay Shentrup <cl...@electology.org>:

I thought your proposal also had bookkeeping. I.e if one of your candidates is eliminated, you have to re-scale your points to your remaining picks.

[Jameson Quinn]

(responding separately to Ted Stern's comments about a similar system he had looked at)

Your system is interesting. It definitely points the way to how I'd generalize EPH to score ballots. (Perhaps I'd also add a threshold and quota, a la BTV, for the pairwise eliminations; this would reduce strategic incentives towards extreme ratings, without changing the "points".)

I agree that EPH fails summability just as badly as IRV (and of course STV). I don't think that is really avoidable with most proportional systems besides asset (which obviously wouldn't work when the candidates are inanimate works of art).

As for participation and IIA: can you give any examples of proportional systems (again, besides asset) which do satisfy these? And can you give examples where your system failed these? I actually think that, because of the head-to-head nature of EPH, it would be hard to come up with a realistic case where EPH would fail these; and if we allow unrealistic scenarios, I think that it would be hard or maybe even impossible to come up with a ballot only (non-delegated) proportional system which passes them.

Note: I'm taking the side of defending EPH here, but it's not because I don't want to see it criticized; it's just that I want to get to the bottom of the criticism, and I think an adverserial-style debate is a good way to do that. Feel free to be as anti-EPH in your response as I am being pro-EPH; that's how we'll see how good a system it is.

Jameson

[Clay Shentrup]

On Saturday, August 29, 2015 at 4:27:11 AM UTC-7, Jameson Quinn wrote:

In EPH (my proposal), the only "state" information needed to figure out what happens next is the set of ballots and the set of non-eliminated candidates. In RRV, you also need a number between 0 and 1 for each ballot to tell how "used up" it is. So yes, EPH has bookkeeping, but RRV has significantly more.

In RRV, the only "state" information needed to figure out what happens next is the set of ballots and the set of winning candidates. I don't see where you get the idea that there's a distinction.

[Jameson Quinn]

You're right. I'm sorry.

--

[Clay Shentrup]

On Saturday, August 29, 2015 at 12:12:48 PM UTC-7, Jameson Quinn wrote:

You're right. I'm sorry.

Time for a victory lap.

But in all seriousness, it sounds like your proposal was a pragmatic way of satisfying a lot of human/psychological demands while still upholding some core election science principles. What do you think the odds are that this will be accepted?

[Jameson Quinn]

Since it already passed 3:1 this year, I think that the chances it will pass again next year are pretty good. And once it's in use, I'm even more sure that people will be happy with it. Using historical ballots, in the absence of slates, it gets 90% of the same answers as the current system (which I showed over thousands of bootstrap runs.)

--

[Warren D Smith]

that sounds good, if the Hugoites are happy with it for several years
that could be a stepping stone to further use -- and it might be one
of the best multiwinner systems out there, although I'm uncomfortable
at present trying to say how good different multiwinner systems are.
I would like to make a set of about 5 more web pages each explaining a
different multiwinner system proposal, and that would include a page
on this Quinn/Hugo system and/or its obvious variants,

[Jameson Quinn]

Minor note: this system is generally referred to as "EPH", for "E Pluribus Hugo". That name is not my idea, but it's what people decided to call it.

[Toby Pereira]

I think this looks like an interesting system, but I think it can produce some undesirable results. But I'm not sure whether anyone would propose it for use outside these Hugo awards so it might not matter that much. If we extend it to proportional elections generally, we could have the following example:

2 to elect, approval voting

21 voters: A

40 voters: A, B

39 voters: C

In terms of points, we'd have:

A: 41

B: 20

C: 39

B and C would go head to head for elimination and B would win 40-39. C is eliminated, and A & B are elected.

I'm not sure I like this result. Are there any other proportional systems out there (ranked, rated whatever) that in a two-winner election would fail to elect a candidate when 39% of the electorate bullet vote for them, and no-one else rates or ranks them anywhere?

[Leon Smith]

Many congratulations,  Jameson.   I think you did some excellent work on this.   I hope that it passes and that people like the result.    And I especially hope that it'll eventually be a feather in the cap for both yourself and the Center for Election Science,  and that it will help raise our profile and help the cause.

Best,

Leon