[EM] Why the concept of "sincere" votes in Range is flawed.

[Jonathan Lundell]

On Dec 2, 2008, at 5:55 PM, Abd ul-Rahman Lomax wrote:

> At 06:30 PM 12/1/2008, Jonathan Lundell wrote:
>
>> I don't really see a need for equal-ranking in a single-winner
>> election. As a voter, I'm answering the question "if you were
>> dictator, of this set of candidates, who would you choose?". I don't
>> really need the option of naming two candidates to the same office;  
>> if
>> I really have no preference between them, I can flip a coin, or  
>> choose
>> the tallest, or ugliest, or whatever.
>
> Odd, in fact, this is the way a "strategic voter" thinks. I.e., "if  
> this election turns out to be close, and my vote can decide it, how  
> should I vote?
>
> If you have no preference, why do you vote at all? By flipping a  
> coin, you are simply adding noise to the system, with no expected  
> value to yourself, but it might harm others.
>
> My question is not whether or not Jonathan Lundell needs equal  
> ranking, but whether or not we should *prohibit* voters from equal- 
> ranking. We do it now, in the U.S., and the result is a lot of  
> spoiled ballots, for starters. The result is spoiled elections, for  
> another effect. None of this explains why it is prohibited. Why? If  
> a voter decides not to make a choice between two candidates, but to  
> support both of them, why should it be prohibited? What's the harm?  
> The voter is adding voting power, yes, but at the same time the  
> voter is losing voting power. The voter is voting in more pairwise  
> elections (with one vote more), but is abstaining from a pairwise  
> election, between the two candidates. These are really alternative  
> votes, effective simultaneously. I.e.: If A isn't going to win,  
> please count my vote for B. And if B isn't going to win, please  
> count my vote for A. It's like preferential voting, but symmetrical.  
> And, of course, this action effectively abstains from deciding  
> between those two candidates, but does contribute toward a majority  
> for either, as the winner, if that is needed.
>
> Why should it be prohibited, why should we discard the expressed  
> votes of someone who votes for more than one candidate in a single- 
> winner election?
>
>
>>> I guess what I'm trying to say is that the problem of discerning a
>>> honest vote from a strategic (optimizing) one seems to be inherent
>>> to all cardinal methods, because we can't read voters' minds. That
>>> is, unless the external comparison can be made part of the ballot
>>> itself.
>>
>> I suggest that the problem is worse than that: that the voters can't
>> even read their own minds, in this sense. Suppose that I would have
>> ranked Edwards > Obama > Clinton in the recent US primaries. Fine, I
>> can make Edwards=100, but I really don't have the foggiest idea what
>> it would mean to make Obama=75 as opposed to Obama=50. Do I like
>> Edwards "twice as much" as Obama? What can that possibly mean? It
>> seems to me that range voting (including approval) immediately  
>> reduces
>> to a purely strategic exercise. And what I'd prefer to do is to
>> eliminate (to the extent possible) the motivation to strategize at  
>> all.
>
>
> We can tell that you have not really considered Range Voting much.  
> You've left out a totally critical piece of information. Two,  
> actually. You haven't described the candidate set, and you have not  
> described which candidates you thought were in range of winning. You  
> don't need that second piece of information to decide how to vote,  
> and certainly it doesn't need to be very accurate, but it makes  
> deciding how to vote much easier, and, at the same time, makes it  
> more effective.
>
> So I'll describe how to "sincerely rate" three candidates. When  
> those are the only choices, or, we might suggest, the only realistic  
> ones. Doesn't this simplify the problem? I.e., there are twenty  
> candidates, how the BFH do you rate them all? Answer. You don't,  
> except to classify the unimportant ones or the ones you don't know.  
> I'd either classify them bottom, in most Range methods, or maybe at  
> default (midrange) in some.
>
> You only need to determine a vote for realistic candidates. You can  
> add any others and how you vote exactly, will depend on the Range  
> implementation.
>
> I'll assume Range 100. But I won't use all that information.
>
> First of all, I'd vote for my favorite at max. I'm going to assume  
> that my Favorite isn't one of the three!
>
> Favorite 100.
>
> Now, I'll look at the "important" candidates. Do I want to  
> participate in the choice between them? If so, that's exactly what I  
> will do, with (almost) one full vote;
>
> Okay, Edwards>Obama>Clinton.
>
> Edwards, 99
> Clinton, 0.
>
> If I want to express that I like Clinton better than some on the  
> ballot, I might rate Clinton at 1. These are votes, the word  
> "rating" is misleading. In Plurality, remember, I'd vote for one --  
> which one? my favorite or Edwards. That is the same as max rating  
> one and rating at 0 everyone else.
>
> Range Voting is just Approval Voting with fractional votes allowed.  
> Allowed, not required. Just as Approval is Plurality with  
> simultaneous votes allowed. Again, Allowed, not required.
>
> Where to rate Obama. At 100%, jerk! -- just kidding!
>
> Which would I consider more important, that Edwards beat Obama or  
> that Obama beat Clinton, if it was a tossup which of these elections  
> was the real one, the one that counts, the one where my vote counts?
>
> In other words, roughly, *how much better is Edwards than Obama, or  
> Obama than Clinton*.
>
> You only have one vote to spread out, you want to put most of it  
> where it is most likely to do some good. That would be, zero- 
> knowledge, in the election pair where you care the most.
>
> But remember, this is only one vote. Vote approximately, that's all  
> that is needed. But, of course, if you know which of these pairwise  
> elections is the one that will really count, you can simply vote in  
> that one. With Range 100, I would always preserve preference order,  
> there isn't sufficient loss of voting power to outweigh the  
> satisfaction and other values to indicated whom I prefer, if I have  
> a preference. Even if my preferred candidate has zero chance of  
> winning.
>
> (Pure von Neumann-Morganstern utilities would show infinitesimal  
> difference in rating with infinitesimal probability of winning. It  
> would put all the voting eggs into real baskets.)
>
> (Somebody tell us how to calculate von Neumann-Morganstern utilities  
> from absolute ones, okay? I'm too lazy to figure it out at the  
> moment.)
>
> So: how could I estimate my preference strengths? I'd start by  
> midrating Clinton. That's a Borda count vote, and Borda is quite a  
> respectable method. To vote Borda, in Range, I'd suggest considering  
> clones identical, they will be rated the same because you have no  
> preference between them. By definition. This is one way that Range  
> is superior to Borda; Borda with equal ranking allowed *is* Range  
> (that is, as many ratings as there are candidates, so if one equal  
> ranks, there is a rank emptied each time one equally ranks. Then  
> distribute the candidate sets though the rating space. In this case,  
> three candidates, we've already placed one at max (almost), the  
> other at min, so Obama goes at 50%. You could leave it at that. How  
> does it look? Is Obama better or worse than "in the middle? If so,  
> nudge it until it seems better. Already, you are deciding less than  
> one-half vote. One vote isn't a big deal, in public elections, how  
> much is one half vote, or one-fourth vote?
>
> Now, I made it simpler by considering that the three were the only  
> viable candidates. If there are more, again, and you want to vote  
> zero-knowledge, sincere, starting with a Borda vote is quite  
> reasonable. Group together candidates that seem equally good or bad,  
> you will get more reasonable ratings. And spread them across the  
> space in preference order. And then nudge them if it doesn't seem  
> reasonably expressive of how you feel or judge them.
>
> Your vote is going to be averaged together with many other votes.  
> It's like a guess. Large numbers of people making a guess that is  
> averaged together can come up with surprisingly accurate results,  
> under the right conditions.
>
> You could get a little more sophisticated. How much would you be  
> willing to contribute to the election campaign of each of these. I'd  
> guess, for the primary, that you wouldn't send money to Clinton. 0.  
> If she was the candidate for the party, that's a different thing.  
> *Range ratings are relative, not absolute. Voting 0 for Clinton does  
> not mean that you detest her. It just means that, in this election,  
> you are not adding weight to her as a choice. You *certainly* are  
> not enthusiastic about her. So how much would you be willing to  
> contribute as a campaign donation to either Edwards or Obama,  
> suppose you can't tell which it will be, but it will be one of them.  
> Say the money goes to a Dump-Clinton PAC. Let's say, $100. Okay, you  
> actually are going to contribute this, right? This is sincere, and  
> the actual donation proves it. Except that, of course, you are going  
> to divide up the money between them. Where do you send your $100? If  
> it goes entirely to Edwards, you are an idiot. Sorry, got carried  
> away.
>
> But seriously, if you've decided to send it all to Edwards, that's  
> where your vote should go, too, I'd suggest. They are the same  
> thing, both help advance a candidate, and you prefer Obama to  
> Clinton, so, really what's that worth to you. Sending all the money  
> to Edwards means it isn't worth anything to you.
>
> Already already. $80 to Edwards and $20 to Obama. (Imagine, you'd be  
> able to tell your grandchildren that you did, in fact, send a  
> contribution to the Obama campaign, before he was the Democratic  
> candidate.) You'd want to assign your sincere vote to the same  
> relative value. But Edwards is at (almost) 100 rating. So both votes  
> would be increased proportionally from the relative dollar value  
> Edwards 99, Obama, 25.
>
> If that seems too low for Obama, maybe you should reassess your  
> contribution ratio.
>
> Look, there are many ways to do it, and they are all fine, except  
> one thing should be understood. To the extent that you don't vote  
> the full range for the set of reasonable winners, you are abstaining  
> from the real election. Range is just a more sophisticated version  
> of Plurality, after all. You want to exercise full voting power,  
> vote a full vote. Want it to count, to actually have a chance of  
> influencing the outcome, vote it for a candidate who can win. Think  
> that this is circular. Fine. Vote with pure sincerity, don't worry  
> about who will win. But, note, I had you vote 100% for your  
> Favorite. You could not do more for that candidate, except that you  
> might consider derating Edwards; the problem is that, by the  
> conditions set up, your favorite winning is really impossible. Polls  
> can be off, but not *that* far. So the vote would be merely symbolic.
>
> Probably the simplest method, if one has any difficulty, is to use a  
> Borda method, spread the candidates, in preference order, across the  
> Range. Saari will think you are a genius, and I'm sure he won't mind  
> that you, while you are at it, fix the little ICC problem that Borda  
> has. (Borda is vulnerable to manipulation through clones. It's fixed  
> with equal ranking. In fact, allowing equal ranking does nice things  
> to just about every voting system. It would make IRV much better.  
> Think about it: the voter could vote Approval style. Or IRV style.  
> Voter concerned about Later No Harm? Fine. Guarantee it. Only three  
> ranks on the ballot, but you'd like to show support for a fourth  
> candidate? Fine, add another vote in third rank. Or shove two  
> candidates into rank one or two.
>
> Most voters won't do it and won't need it, but the few percent who  
> do can sometimes improve election results greatly *and they will not  
> harm them.*
>

Phew. QED, I say.