ENnies discussion thread

[Mark]

In the first duma election under Gorbachev, approval voting resulted in a majority of the seats being unoccupied at the end of the election because the various negative campaigns were so successful that in most districts, none of the candidates managed to avoid majority disapproval.

That's a bad break, ain't it?

 

[Conaill]

Fusangite, I'm getting the feeling we're talking about two entirely different definitions of approval voting here. For one, I have never heard of any other "approval voting systems" then - well - Approval Voting itself. It's not a "class of systems" as you put it earlier, it's one very well defined and very simple voting method.

The other thing that makes me suspect we're talking about different animals is that Approval Voting does indeed avoid the type of strategic voting you seem to be worrying about. In particular, it is *never* beneficial to vote for one candidate but not for another you like better (monotonicity - a property IRV lacks!), and it does indeed discourage negative campaigning. Your mention of "majority disapproval" also seems to confirm my suspicion - there is no such concept in the method that has traditionally been called Approval Voting by voting system experts.

I am not familiar with the voting method used for the election of the Duma under Gorbachev, but I have yet to find anywhere online where this is referred to as "approval voting"...

 

[Umbran]

Keep in mind that Approval Voting encourages people to vote for multiple candidates, so the situation you just described will never happen. (The exact same situation can happen in IRV if everyone only ranks one choice per ballot - but that'll never happen in real life either.)

I gave teh extreme example as illustration of the pioint. The exact situation might not happen, something similar might. Specifically, the system you describe can easily find a winner without consensus of the voting populace. And if you want consensus, your Approval Voting system has no way to find it without having people go back to the ballot box.

Some more endorsements of Approval Voting;

- It is used by the American Statistical Association, the Mathematical Association of America, the Institute of Electrical and Electronics Engineers (>300K members), the US National Academy of Sciences, the United Nations, as well as numerous smaller professional organisations and universities.

Yes, but that's meaningless unless we know what they're using it for. If the application isn't the same, the endorsement means little. The Hugo and Nebula awards use IRV, for purposes very similar to ours, and those communities are also filled with mathematicians and tech-heads

I'm taking a couple of things out of order here, because it's a better way to get at the crux of the matter...

In statistics, that sort of estimated quality of an unknown entry is called a prior, and it's perfectly acceptable to have a prior estimate which is higher than one of your known datapoints.

It is statements like this that give truth to the old saw that "there's lies, damned lies, and statistics". In statistics, what is or is not perfectly acceptable actually depends upon the application. Why you want the numbers should influence what methods you use.

For a marketing firm polling to get the public's opinion before designing an ad campaign, the prior assessments are useful tools. However, I feel they are contrary to our goals with an awards program.

Seriously - why would it be unfair to assume that a product you are familiar with and have a strong dislike for is actually *worse* than another Ennie nominee you are not familiar with? After all, these nominations have been through an exstensive vetting process already, so you know *something* about their quality.

Well, now we run into a problem...

Personally, I think the judges already have a lot of input, and we shouldn't have the voters depending upon the judges even more in making their votes.

However, given that you already known somehting about the product through the judges, it becomes even more unfair to assume that an unknown is better than a known. After all, the judges chose this product that you detest, didn't they? That means that the other products they choose could likely be equally detestable. Given that the judges made one "bad" choice", you shouldn't assume that all their others are good ones.

Why would it be more fair to assume that the unknown product is even worse than the one you know you detest?

It isn't as if it's that simple.

In the end, what we want is a system that allows reasonable abstention. That would be most fair. For that, the numeric ranking system is actually preferrable, if you could depnd upon the voters to not try to finesse the system.

Your approval voting only allows two options for dealing with unknowns - either they are as good as the best known, or they are as bad as the worst known. There are no other options. Neither is really fair, and your ranking of unknowns is taken into account at the same time and with equal weight as the ranking of knows.

Consider, though, with IRV, that an abstention isn't quite as bad as you think, especially with a null or "no prize" option. IRV has leeway. If you know a given product, and think it is good enough to give a prize, you get to vote for it. The system doesn't get to your abstentions (or even your low-rankings) until after it has tried to count every single preference you had above that. What you chose to do with the unknowns may never be seen at all, if the contest is decided before they are reached. Thus, the opinions about unknowns are left out unless the race is particularly difficult.

And, in IRV, the voter does still have the option to rank an unknown over a known, if they really want to do so. But if they rank that bad known last, the unknown can still be fairly low, too, rather than as equal to the best thing out there. IRV allows some of the benefits of numeric ratings, while doing away with most of the weakness to finesse votes, while not forcing voters to vote on unknowns as equivalent to knowns.

Of course, this is more difficult to implement. In order to get the best of both worlds, you sometimes have to do a little more work.

 

[pensiv]

I wonder what the drawback would be to rate the votes according to a number of votes to ranking of votes ratio

So if product #1 gets 500 votes and a total score of 1000 and product #2 gets 250 votes and a total score of 500, they'd have an equal ratio and thus are equally rated.

I'm just throwing out an idea here without too much thought, but I wonder if this is a good way to account for products that the judges opt out of voting for.

 

[Macbeth]

I wonder what the drawback would be to rate the votes according to a number of votes to ranking of votes ratio

So if product #1 gets 500 votes and a total score of 1000 and product #2 gets 250 votes and a total score of 500, they'd have an equal ratio and thus are equally rated.

I'm just throwing out an idea here without too much thought, but I wonder if this is a good way to account for products that the judges opt out of voting for.

Only problem I can see with this is that then we could get 1 vote and a total score of 2, and still have the same ratio. Granted, nothing this dramatic is likely to happen, but it still ends up giving a preference to anything with only a few high votes out of a very small number overall.

 

[Conaill]

In the end, what we want is a system that allows reasonable abstention. That would be most fair. For that, the numeric ranking system is actually preferrable, if you could depnd upon the voters to not try to finesse the system.

Yes, a ranking allows the voter to express more information about their preference - including abstentions. Doesn't mean that information is actually used correctly though!

The standard IRV as you are proposing treats abstentions as bottom-of-the-barrel votes! Yes, it is as bad as that. Let's say there are two unknown products on a ballot. At least if you were to rank those #4 and #5 out of 5, one of them would get a miniscule chance of ever coming into play. But if you abstain on both of them, they will always be treated as worse than the very worst vote on your ballot. In the simple IRV case, an abstention is *always* a down-vote, just like in a standard winner-takes-all election.

Now, it is possible to come up with more sophisticated variants of IRV or other ranked voting methods, where the quality of the candidate is only measured across the voters who have actually expressed an opinion about it. Somewhat along the lines of what we were trying to achieve by looking at the mean score last year (whereas standard IRV looks at the sum of the first-choice votes each round). But that does get more complicated, and it's typically equivalent to the system extrapolating what you would have voted, based on what other people have voted for that product. In that case, as a voter I would much rather be able to make a ball-park estimate myself!

Your approval voting only allows two options for dealing with unknowns - either they are as good as the best known, or they are as bad as the worst known. There are no other options. Neither is really fair, and your ranking of unknowns is taken into account at the same time and with equal weight as the ranking of knows.

Spoken as someone who prefers a ranked system.

Of course, Approval Voting doesn't actually use ranking, so claiming that it "ranks" the unknowns together with all the good ones or all the bad ones is a little biased. Might as well complain that an exact ranking forces you to state that the differences between subsequent ranks are equal, even though you might have preferred to score them 10-3-2-1-1.

I would rather say that Approval Voting allows you to split the candidates into two groups: acceptable or unacceptable. If you want to say that any of those unknown products is acceptable over one you know is crap, that is an acceptable choice.

Yes, ranked voting captures more information, but IRV actually winds up throwing most of that information away anyway, since in any round it only takes the 1st-choice votes into account. Which is why compromise candidates with few 1st choice votes are often dropped early on. Plus it is still open to strategic voting, meaning that it may be in your best interest to rank your favorite candidate lower than a more widely popular one.


In the end, it may all come down to what we are willing to implement, and what features we are willing to live with...

- Very easy implementation, no bias against poorly known candidates: Approval Voting

- Harder to implement, allows the voter to state his preferences in more detailed way, optimal voting strategy may be different from your actual preference ranking, biased against poorly known candidates: standard IRV

- As above, but no bias against poorly known candidates, needs sophisticated statistics: some novel form of IRV

 

[fusangite]

I've read the Citizens for Approval Voting description. First of all, Conaill, what is being suggested here is not "approval voting"; it is a particular kind of approval voting process. Just as there are over 100 kinds of proportional representation systems in use in the world today, there are many kinds of approval voting. The definition of "approval voting" is not the domain of non-profit advocates of a particular variant of the system, nor is it under the control of random wikipedia volunteers. Approval voting, in mathematical discourse, refers to a family of systems, of which the one you recommend is a member.

I do not support your particular approval voting model for the ENNies because, like Borda counting, it deforms people's articulation of their non-first preferences. Here's what I mean. Going back to my example of the fictional five finalists, let's recap. The Edwards text and the Gygax text are frontrunners.

If I am a supporter of Gygax and Edwards is the closest competitor, I have no incentive to vote for it because doing so will effectively cancel my Gygax vote. Let's suppose there's a tie. Gygax has 454 votes; Edwards has 454 votes. If I just vote for Gygax, I break the tie and Gygax wins; if I also give a vote to Edwards, I effectively cancel my vote in favour of Gygax. Therefore, as with Borda, the only products for which I can safely vote without the risk of cancelling my vote for my favourite product are inferior and unpopular ones.

Similarly, if I am a supporter of Platform Sandals, it would be completely irrational for me to use any of my additional votes to vote for any other products at all because nearly all of them are likely ahead of the product I like and I would be effectively cancelling my own vote by voting for them.

This brings me to the important question of: who has an incentive to cast multiple votes? Answer: someone who hates a particular product and whose priority is for it to be beaten at all costs. So, a person who despised the Edwards book to the point where they didn't care who beat it or what the best book was would have an incentive to vote for every single product except for it.

If one looks at the approval voting site, we can see that this kind of political thinking is informing the association's ideology. What the association is concerned about is people who support fringe parties still being able to have their votes count against the candidate they really despise. What approval voting, as described by the association you support, Conaill, does is to give Ralph Nader supporters a way of giving their own candidate a vote without giving up the opportunity to vote negatively against Bush and Constitution Party supporters an opportunity to vote for their candidate without giving up the opportunity to vote negatively against Kerry.

Thus, while approval voting may be useful for solving the problem of strategic voting by fringe party supporters in American politics, it neither addresses nor solves the problems we might need to address here on ENWorld. First of all, there are no massive institutional frontrunners in the ENNies; nor are there protest candidates. Because there is no risk of Nader winning, there is no risk associated with voting for both him and Kerry. However, in my model, there is a genuine risk associated with voting for Edwards and Gygax if you actually really do like one better than the other.

IRV (formally Alternative Vote) does not have these problems. It allows one to indicate a first preference for your favourite product and a second preference for your next favourite product without any risk that indicating what your second preference is will result in the defeat of your first choice. The model of approval voting you suggest, on the other hand, may result in the defeat of your favourite product if you honestly indicate your second preference. As such, it should not be considered for ENNies voting.

 

[zenld]

I'm not too sure which voting system I like better, but I can't wait for these books Fusangite keeps talking about.



zen

 

[fusangite]

I would rather say that Approval Voting allows you to split the candidates into two groups: acceptable or unacceptable. If you want to say that any of those unknown products is acceptable over one you know is crap, that is an acceptable choice.

But this is assuming people are being "gentlemen" about who is acceptable and who is not. I would rather not assume that voters will vote in favour of a product they merely find acceptable if a possible consequence of doing so is that their favourite product loses. I certainly wouldn't be enough of a "gentleman" to vote for a product that I thought was second-best if I felt that doing so would effectively neutralize my vote for my favourite product Why bother voting at all then unless my primary motivation is to engineer the defeat of some product I loathe?

As with many models that style themselves "approval voting," the actual net effect of your system is to reward voting strategies that are based on defeating bad things and punish voting strategies that are based on rewarding good things.

Furthermore, I can't really detect how you have reasoned that your model of approval voting will somehow be more beneficial to products about which the voter does not know. There is no system we could design that would place little-known products on the same footing as well-known products. As in real-life elections, there is no way to do well if the voters don't know who or what you are. And that, in my view, is a good thing.

Yes, ranked voting captures more information... Plus it is still open to strategic voting, meaning that it may be in your best interest to rank your favorite candidate lower than a more widely popular one.

This criticism is true of Borda, which uses a ranked ballot but is completely false for the system Umbran and I are advocating.

If my favourite product is Monte Cook: Black Sheep and I know that my second choice, Gygax is significantly more popular. It is still perfectly rational for me to make Monte my first choice and Gygax my second choice because the only condition under which my vote will transfer to my second choice will be if my first choice has already been defeated. So, no -- IRV is the best system for choosing a single winner because I am punished, not rewarded for incorrectly representing my non-first choices on the ballot.

- Very easy implementation, no bias against poorly known candidates: Approval Voting

You have yet to demonstrate (a) that IRV is difficult to implement or (b) that approval voting results in unknown candidates doing better or (c) that unknown candidates should do better.

Finally, if you want to look at voting systems that are actually supported by a large number of academics, are in use all over the world in functioning democracies and are well-described, check out http://www.fairvote.org/. As you can see, plurality, majoritarian and proportional systems have vastly more public and academic support than the particular model of approval voting you are pushing.

 

[fusangite]

I'm not too sure which voting system I like better, but I can't wait for these books Fusangite keeps talking about.



zen

Thanks. I was hoping I would keep people from nodding off from all the math by putting forward these books as an idea. Glad to see someone is finding my sideshow entertaining. Maybe if this debate continues, I should come up with a new fictional category of product for which I can generate another five candidates. Any suggestions?