"No Award Test"

[jack]

Background

I thought I wrote this last year, but I can't find it now. The Hugo Awards are decided by Instant Runoff Voting between all the nominees and the "no award" option to not let anyone win that year at all.

They have an extra step at the end, the "No Award Test", which says that in addition to the normal instant runoff vote, at the end, after the eliminations, if the final winner is one of the nominees (not no award), then all the original ballots are compared, are there more ballots which voted for the winner at all (ie listed them on the ballot, and listed "no award" lower or not at all) than ballots which voted for "no award" higher than the winner.

There is a detailed description at: http://www.thehugoawards.org/the-voting-system

Although I found I needed to look up the WSFS constitution and an example (pdf) of the hugo results step by step from last year to make sure.

Exmaination

However, I got to wondering, when does this extra step make a difference. If ever?

First note that, despite the controversies about nomination this year, even if people vote "no award" above many or all nominees in many categories, I think "no award" will just win the vote in the usual way, this post examining the extra "no award test" will be completely irrelevant. So I'm just interested out of principle if it ever WOULD be used.

My impression is, it would make a difference if the vote is quite polarised (ie. different people voting the same book at the top and bottom of the ballot), and with people voting "no award" more than half way up their ballot. Both of which seem quite unusual.

My favourite example is to recall the Drazi from Babylon 5. Suppose the candidates for Dictator For the Year are Chief Green Drazi, Head Purple Drazi, and Susan Ivanova who is likely to smash the faction urns and do away with the faction system for all time. "No Award" means not declaring a dictator this year and waiting for next year. No-one wants the opposite-faction leader in charge. Traditionally all Drazi voted for their own faciton leader, but now nearly half don't want EITHER faction leader in charge.

40% of the Drazi are sick of the fighting and prefer the human to anything else. Ivanova, then no award, then the two colours.
30% vote along faction lines and are randomly chosen as Green supporters, voting Green, then "no award", then the interloper human, then the hated purple faction.
30% vote exactly the same but are randomly chosen as Purple, so vote Purple, then "no award", then Ivanova, then Green.

What are the results under IRV? What result SHOULD an election system give that voting? When the preferences are a bit circular is a time election systems often find it hard to give meaningful results, as often no result is really what the electorate wants.

According to IRV, "No Award" is eliminated immediately as nobody actually wanted that more than anything else. Then one of the two faction leaders is eliminated, and because all their faction supporters voted Ivanova over the other faction leader, Inanova wins. Which is good, but isn't really what the electorate wanted, as 60% of them voted that no dictator at all was better than her.

What the no award test says is that in this case, Ivanova won the run-off, but because she's behind "No Award" on the original ballots, there's no dictator this year.

Conclusions

So, is that useful? It makes "No Award" slightly more likely to win. I think in something like the Hugo Awards, where it's fairly plausible to not award one one year, it works. It makes "no award" slightly more likely to win. And it ensures that if the electorate hates a candidate, they can't somehow slip through a split vote and win.

So I think it's a mild positive which is unlikely to come up, but will probably do the something worthwhile if it does. And probably isn't worth the extra complication of it existing, but isn't worth trying to remove. I hear it may be a watered down version of an earlier proposal that would have given no award more teeth?

I'm not sure if it should be used in other contexts such as national political elections. I would like to see more "none of the above" votes, since many people's preference DOES seem to be "I hate all of them". But there's usually protest parties you can (and should) vote for in that circumstance. The "no award test" would only make a difference if "none of the above" didn't win but everyone voted for it highly. And I'm not sure if that's ever likely to happen, or if it did, if re-running the election is likely to help...

Asides on No Award in general

I admit, when I went to worldcon, it didn't really occur to me to vote any of the generally enjoyable novels below "No Award", I just voted for the nominees in the order I liked them.

But now I think, I'd vote for books that I think actively deserve a hugo, but when there are books that are perfectly enjoyable, but don't really feel stand-out to me, I WOULD put them below no award. (With the exception that if I feel the author deserves a hugo in that category but didn't get one yet, I would vote for a less-outstanding sequel they wrote as long as I enjoyed it.)

ETA: I found someone posted a similar explanation last year: http://www.kith.org/journals/jed/2014/08/15/14938.html

Comments

[forestofglory]

I think the no award test is really about the principle of the thing. The principle being if more people ranked no award higher than they ranked the prospective winner, then no award should be given.

ETA: Remember we are talking about a system designed by pedantic nerds.

[jack]

Remember we are talking about a system designed by pedantic nerds.

*snork* Yes, indeed.

And I mean, I can see that principle that "if 60% of the electorate rank no award higher than Ivanova, then Ivanova shouldn't win." But it sounds equally sensible to say "if 60% of the electorate rank Corwin higher than Ivanova, then Ivanova shouldn't win". And we DON'T have that in IRV, if everyone voted a fairly mediocre choice just above "No Award" then it would lose for the same reasons.

So it there a good reason to single out "No Award" for special treatment? I think maybe there is, but I'm not sure. And I'm not sure it's clearly _correct_, it's clearly some sort of compromise...

[forestofglory]

I'm not sure that "No Award" gets special treatment. The relevant bit of the WSFS constitution is:

Section 6.5: Run-off. After a tentative winner is determined, then unless the run-off candidate shall be the sole winner, the following additional test shall be made. If the number of ballots preferring the run-off candidate to the tentative winner is greater than the number of ballots preferring the tentative winner to the run-off candidate, then the run-off candidate shall be declared the winner of the election.

However people often talk about the "no award test" like it is separate thing. So I remain confused on this point.

[jack]

OK, now I'm a bit confused.

I'm fairly certain that the way this works is:

1. There is an IRV election between all nominees including "No Award" (as described in section 6.4).
2. The winner is additionally compared against "No Award" on all the original ballots (as described in section 6.5), and if it loses,

But apparently in the constitution #1 provided an abbreviated description of how the vote-counting shall be done, but doesn't use the (possibly more modern?) name of "Instant Runoff Voting". Whereas it DOES say "run-off" in section 6.5, which I think is usually referred to as "the no award test".

I guess that is a form of run-off election, but I've never actually seen run-off elections, only that they were explained to me in the context of IRV.

[forestofglory]

Yeah, I'm really not sure what "run-off candidate" is supposed. to refer to in 6.5

[jack]

I'm pretty sure what it means is that the "run-off candidate" is the candidate that won the IRV phase, and now has a run-off against "no award". But I'd not heard that terminology before. The constitution doesn't seem to really define the terms, I think it's unambiguous but not really clear.

[forestofglory]

"unambiguous but not really clear"

I think that describes it perfectly!

[jack]

:)

[ptc24]

In your Drazi example, "No Award" is the Condorcet winner, the one that could beat any other contender in a two-horse race, and "isn't really what the electorate wanted, as 60% of them voted that no dictator at all was better than her." seems to imply that in any situation where there's a Condorcet winner, a system that picks a different winner is failing to do what the electorate wanted.

Of course, the problem with Condorcet systems is that you need some twiddly bits to cope with cases where there is no Condorcet winner, and as such, they're even harder to successfully explain to people than IRV.

[jack]

Aah! Thank you, ok, NOW I think I get what the Condorcet criterion actual is, I think I never really got it before.

So in this case, there's a special hack to make sure that if "No Award" is the condorcet winner (or closer to it?), it's not outvoted in some circumstances, but that hack doesn't apply to the other candidates. I'm not sure if that's "what the electorate wanted" or not -- it sounds plausible, but so does the reverse. But I'm _really_ curious if there's a particular reason to apply this to No Award _only_.

I wonder if the Hugo IRV voting has ever produced a different result than the condorcet winner? (Or for second place etc assuming you calculate it like that in both systems.) My guess is that it hasn't, in which case it's never made any difference.

[ptc24]

On the No Award front, as far as I can tell in recent history the only nominee that ranked lower than No Award was Vox Day's work, so I don't think the No Award rule has made any difference (OTOH its mere presence may have influenced voting, who knows).

However it's possible that IRV might have selected non-Condorcet winners, hard to know.

[jack]

In principle, I think the ballots should be re-constructible from the voting results PDFs (at least for recent years), so you should be able to re-run the elections whatever systems you like and see if you get a different answer. But I don't know if you could do it with little enough effort to make it worthwhile...