[EM] Student government - what voting system to recommend?

[Abd ul-Rahman Lomax]

At 06:41 PM 4/24/2007, Juho wrote:
>The reason why I talked about learning is that Range is often
>described so that the first impression voters will get is that they
>"should" put their sincere ratings on the ballot (and they would not
>be aware of how to vote with full strength).

what ballot instructions have you read that so instructed voters?


> > First of all, we think that it will be common knowledge that if you
> > don't vote the extremes for at least one candidate on either side,
> > you are casting a weak vote.
>
>In most cases any use of intermediate values makes the vote weaker
>than it could be.

That is correct. You vote the extremes where you have a strong 
preference, you vote in the middle when you don't. And this also 
happens to be necessary where you face opposing risks. Range *never* 
requires you to vote "strategically" in the sense of reversing preferences.


> > Nobody is recommending that truly weak votes be cast. (But some
> > people may want to cast them anyway, and they should be able to.
> > Consider it a partial abstention, and many people abstain from this
> > or that race now.)
>
>That's ok. Weak votes and abstention can be options for the voters.

But they are not options in ranked systems.....


> > If A was the favorite, why in the world would the voter vote A=5 in
> > the first place?
>
>The voter didn't find him/her "excellent" but just "reasonably good".

Hey, I'd be happy with that! I wish! Instead, we get "truly awful."

We get a "uniter not a divider," who, it turns out, means by "unite," 
"you all do it my way or you are a traitor."

>With fully sincere (utility based) ratings maybe no candidate gets
>the max or min score.

That's correct if we are talking about pure utility. Indeed, there is 
no max or min score. But Range really asks voters to rate candidates 
*relative* to each other, with Best being max rating and Worst being 
min. It really should be explained that way.

We have sometimes suggested that Range Votes be normalized. That is, 
if a voter voted, in Range (0-10), 0, 3, 5, this would be normalized 
to 0, 6, 10. But I think it better to keep it simple.

A ballot could actually say, instead of numbers, Best, Good, 
Acceptable, Not-Acceptable, Worst. Range 5.

Basically, those who object to Range on the basis that voters will be 
confused and mistakenly vote weak votes are assuming idiotic or at 
least inadequate ballot instructions.

This is not to be confused with "weak" votes, meaning intermediate 
votes, which are another matter. These are votes where the voter has 
no strong preference, or, for a more sophisticated voter, is 
balancing opposing considerations: "I'd like to see A defeat B, and B 
to defeat C. Range requires voters to weigh the relative merits, the 
preference strengths of these pairs. If the first consideration is 
more important than the second, then the rating of B will move closer 
to that of C, and if the second is more important, then the rating of 
B will move up toward A."

There is really no alternative to this that makes sense. Either 
preference strength matters or it doesn't! If it matters, then it 
*must* follow that a weak preference will be expressed in a weak vote.

> > What is continually asserted here is that voters with weak
> > preferences will somehow decide to vote strategically.
>
>I assumed that voters with strong wish to win, or those that
>(strongly) want to counter the ones that vote with full power, would
>vote with full power ("strategically").

Yes. However, voting strongly *requires* you to vote weakly. Haven't 
you noticed that?

What is a strong vote? Well, bullet voting is strong, it's been 
presented that way in this thread. Okay, three candidates, and you 
vote A=10, B=0, C=0. Strong vote?

Actually this is a maximally *weak* vote in the B/C pairwise 
election. If you care seriously about that pair and you want to vote 
strongly regarding it, favoring B, you must vote A=10, B=10, C=0. But 
then you have cast a maximally weak vote in the A/B pairwise election.

If there are voters who are bullet voting for B, and I don't like B, 
I can vote zero for B. This *maximally* counters those allegedly 
strong votes. But suppose there is another candidate C, far worse 
than B. By trying to cancel out those nasty bullet voters, I force 
myself to abstain from the B/C pairwise election.

You can go around and around with this. Bottom line, the claim that 
voters will over time gravitate to Approval style voting is based on 
a total misunderstanding of how Range works. It might happen, it 
might not, but it certainly isn't clear that it will, yet Range 
critics often present the matter as if it were obvious that it must 
happen, voters would be "idiots" to not vote that way.

 From my point of view, how to vote Range is pretty simple: rate your 
favorite max, rate the worst candidate min, and then arrange the 
others into two categories: acceptable and not acceptable. Range, if 
it is a single-stage implementation -- which is necessarily inferior 
-- does require that we consider how other voters are likely to vote, 
to have the maximum effectiveness for our vote. Where there is risk 
that lowering the rating of a candidate could cause a candidate to 
lose to a substantially less acceptable candidate, I will push the 
rating of that candidate toward max.

Range, as an election method, is designed to find a candidate who is 
a broad compromise, maximally acceptable, with minimum regret. 
Generally, well-functioning democracies do this with all decisions! 
Otherwise they oscillate, and sometimes they oscillate so severely 
that the structure tears itself apart.

>  This behaviour may make also
>the originally "weak preference" more radical. Voters that
>intentionally want to cast a weak vote (and that are ok with others
>using strong votes) would not be affected.

That's right. The word "weak" is a bit loaded. We need to keep in 
mind that, properly, a correctly weak vote is expressing a weak 
preference, and there is nothing wrong with having weak preferences. 
Indeed, one of the ideas underlying the objection that Range 
allegedly rewards extreme votes is that more intelligent 
understanding often weakens preferences!

> > Look, if there is an election, and I sincerely rank A as 9 and B as
> > 8 (and other candidates lower than that, let's say zero) and B
> > wins, I'm happy! That's an excellent outcome! The danger comes in
> > quite the opposite direction from what Juho proposes. Suppose I
> > rate B as 8 and C wins, with B being the runner-up. Close runner-
> > up. I might regret rating B at 8.
>
>I believe this voting pattern os in line with the Approval style
>strategic voting that I discussed.

But this requires that I cast weak votes in some of the pairwise 
elections, that I give up the idea that my Favorite should win, and 
move toward the idea that the winner should be one of those acceptable to me.


> > If you vote Approval style, you fail to express your true
> > appreciation of the candidates, and this can backfire.
>
>Yes, but typically/statistically Approval strategy improves the outcome.

No. Check out Warren's simulations. Sincere voting (which means 
expressing weak preferences as weak votes) produces the best 
outcomes. Approval style produces acceptable outcomes, relative to 
some other methods.


> > It is just as reasonable to consider that Range elections will move
> > *away* from Approval-style as that they will move toward it.
>
>A simple example of this would be nice.

I've already given the reasons. People are well-accustomed to voting 
for their Favorite, for waving the banner, for very much wanting one 
candidate to win. So they will continue to do this. For a time. 
Gradually, they will relax into a less fanatic mode of voting.... 
that's all. And as they do, outcomes will improve. Voting Approval 
style produces reasonably good outcomes, voting higher-resolution 
Range generally produces better outcomes. As more and more people are 
pleased with election results, and see them as fair, I'd predict that 
partisan fervor will wane. To a degree. I don't know how much.

You know, I've participated in a fair amount of deliberative process. 
You want your motion to win. But, in fact, when the process works 
well and is perceived as fair, losing isn't particularly painful. 
What is painful, really, is never getting a hearing for ideas! And, 
in good deliberative process, what comes out in the end is often 
shifted from what was proposed in the first place. It has been 
shifted to make it more broadly acceptable. As I've pointed out, full 
deliberative process selects the Condorcet winner among alternatives, 
and it does *more* than that, for preference strengths become known 
and are considered.

I've been saying for some time that a good election method should 
simulate deliberative process, and some do so far better than others. 
Any Condorcet method will do fairly well, but is preference-rigid. 
Range considers preference, but can violate the Majority Criterion, 
which is why I'd like to add a ratification step, as a minimum, where 
the Range winner doesn't have a majority preference. Or a runoff 
between the Range winner and a Condorcet winner, in the relatively 
uncommon circumstance that they differ.


> > I expect that they will start out, actually, as close to Approval
> > for many voters. Smith thinks differently, and I really don't know
> > which of us is right. He's got reasons to think his way. We might
> > both be right. I.e., many voters, maybe most, will vote Approval
> > style, and it will be bullet voting. But there will be quite a few,
> > from the start, who do something different.
>
>I think much depends on the media and other discussions before the
>elections.

Sure. And there are always people trying to get others to vote 
against their own interests. So new?

> > I say that we are not going to really know until we see real
> > elections using Range. The alleged devolution to Approval is not a
> > serious harm. It would only mean that some ballot space and a
> > counting effort had been wasted.
>
>Yes, Range could be roughly as good as Approval (with some wasted
>effort, and ability to cast weak votes). The biggest hiccups might
>come in the form of people realizing that their vote was weak
>although they didn't understand that when they voted, or if some
>candidate won as a result of efficient use of strategic voting.

That actually doesn't happen easily under Range (the latter). 
Basically, the most "efficient" strategy for winning is to get as 
many of your supporters as possible to bullet-vote for you. However, 
this can backfire, if you offend those who might otherwise like you 
but consider your recommendation that you vote against your favorite 
to be quite offensive. I know it would offend me!

Can you imagine how it would look of a candidate steps in front of 
the cameras and says: "Don't vote sincerely, it might cause me to 
lose. Vote only for me!"

Political suicide, that's what it would be, if the election were 
Range. Instead, candidates, as now, will simply try to convince 
voters that they are the best, and it is possible, but not certain, 
that they will refrain, a little more, from trying to tear down their 
opponents, for fear of alienating their supporters and thus losing those votes.

> >> Rating the least preferred candidate at 0 reduces the probability of
> >> that candidate getting elected (and doesn't carry any risks with it).
> >
> > But from the conditions of the problem, there was no risk of that.
> > No, I don't buy it. (By the way, none of us involved with Range
> > would recommend giving the "least preferred candidate" any other
> > vote than the minimum. I assumed that PW was being given a 1
> > because voters somewhat liked him, there were *worse* candidates
> > involved.
>
>There were no worse candidates involved. The voter liked PW somewhat.
>But since PW was the least liked candidate and the voter wanted to
>avoid electing him, giving him 0 was a perfect solution. (I thus used
>sincere utility based ratings instead of normalized ones.)

And this is correct voting! Basically, the supposed "sincere" votes 
from which the method devolved into Approval were ignorant votes. I'd 
really suggest that ballot instructions be explicit, suggesting that 
you vote the max for your favorite, the min for your least preferred, 
and whatever you want for the rest.... Range votes are *relative* 
votes. If there were a dozen candidates, and all were quite well 
qualified, we still need to pick one and we will want to pick the 
best. To get good information from the voters, we need them to 
normalize their votes. Otherwise, the necessary resolution is lost. 
If on some absolute scale, all the candidates are 10s, on what basis 
would we choose between them?

No, Range is about *relative* utility. But I prefer to think of voter 
satisfaction. It is about rating candidates as to how satisfied you 
will be if they are elected, with max rating meaning maximally 
satisfied, and min rating meaning maximally dissatisfied. Relatively 
speaking. You might actually be satisfied in an absolute sense with 
any of them, or with none of them.

> > But this contradicts the assumed initial sincere vote! If you want
> > this, why would you vote A=9, B=8 in the first place? By voting
> > this way, you are saying that B winning is almost as satisfactory
> > to you as A winning!
>
>The voter voted originally sincerely since voters were given the
>impression that they should write one's sincere preferences on the
>ballot.

The ballot instructions were, "Write your sincere preferences on the ballot"?

When you vote a ranked ballot, and some systems require full ranking, 
you are putting one candidate at the top and one at the bottom. Some 
allow you to put more than one in each of these positions, or in 
intermediate positions. The method essentially normalizes your vote, 
making it equivalent to a range of 0% to 100% in Range. But ranked 
methods don't consider preferences strength, though some impute it, 
in a way, by considering "defeat strength."

>  Candidate B winning would be quite satisfactory to this
>voter. The voter however wants to make A the winner if he can choose
>between A and B. If A and B were the only candidates, voting A=max,
>B=min would be also risk free.
>
> > I think that people can and will understand that democracy is often
> > about making compromises. It is *not* about crushing the opposition!
>
>I agree, but competitiveness exists despite of this, and that may
>lead to voting with maximum power etc.

Range limits "maximum power" to one vote per voter. And we recommend 
and generally assume that all voters, with rare exceptions, will vote 
with maximum power. That is, they will rank one max and one min and 
they will array the others as they choose. This is maximum power. It 
won't "lead" to this condition, this *is* Range.


> > "Vote -1 to vote against a candidate, vote +1 to vote for the
> > candidate, and vote zero or leave a candidate unrated to have an
> > intermediate effect. The candidate with the greatest sum of votes
> > will win."
>
>Note that negative votes carry some risks.

The issue here is where the default vote is for abstentions. The 
standard in original Range proposals was that it was zero, 
effectively. Average vote disregards abstentions, which is its own 
problem and requires a "quorum rule" to avoid obvious bad outcomes. 
Using negative votes is a means of making the default be other than 
zero, that's all. The range I suggested makes the default be midrange.


>  Let's say there are three
>major parties with one candidate each, and many totally unknown
>candidates. All major parties are afraid of each others and will give
>lots of negative votes to both competing party candidates. The sum of
>all major party candidates may go below 0.

This situation is a setup for a bad outcome. Be careful not to blame 
the voting method for the total disarray and disunity of the 
electorate! Given the setup, it is not clear that there is *any* good outcome!

>  Some unknown candidate is
>mentioned only in very few ballots (let's say his/her family members
>supporting and one neighbour opposing). His score will however be
>positive and he will be elected, not the well known candidates whose
>score was negative.

That's correct. But something was totally neglected in this analysis. 
That candidate is only going to have a couple of votes above zero. 
But the conditions were that there were *many* candidates. Surely 
there is at least one of them who is well-enough known and 
well-enough liked that the candidate gets more than a couple of votes!

Really, if it is true that there are more people in a society opposed 
to a candidate than favor him or her, do you think the candidate 
should be elected! There is a simple solution to the problem given, 
which is a ratification step or runoff. (Not a top-two runoff, but a 
runoff between, say, the votes analyzed as sum and the votes analyzed 
as raw, abstentions zero.)

The problem, if it is a problem -- I'm not sure it is -- could be 
addressed by setting the default lower:

-1: Disliked
0: Acceptable
1: Good
2: Preferred.

Or, alternatively, the simpler Range 3 implementation with blank 
votes defined as -1/2 vote. Or perhaps even some smaller negative 
value, like -1/10. Something to reflect the value that the winner, 
preferably, should be well enough known that the candidate is rated 
by most voters.

This is a question regarding how to treat blank votes. It's an 
unresolved issue among Range advocates.











> > [I suggested that there be a runoff between the Range winner and a
> > Condorcet winner, if they differ]
> >> Let's assume that a Condorcet winner exists. In this case this method
> >> could be said to be a method where the voters are given a second
> >> chance to think again if the Range winner could be seen as a "good
> >> compromise" even though the majority could easily vote as in the
> >> first round and elect the Condorcet winner.
> >
> > Yes. That is, the original ballot analysis showed that this C.
> > winner was rated higher than the Range winner on a majority of
> > ballots.
> >
> >>  I'm not sure this method
> >> would be a very practical method in real life large elections but in
> >> principle the idea of "recommending" the Range winner to the voters
> >> is a positive idea. Some strategies where people would try to
> >> influence who the Range winner will be could take place (i.e. the
> >> Range winner of the second round would not be the sincere range
> >> winner).
> >
> > I think Juho means that the Range winner of the *first* round would
> > not be the sincere Range winner. If there is a second round, it is
> > not held as a Range election. It is a straight which-of-these-two-
> > shall-be-elected vote. Voters will know, this time, if the first
> > election was sincere, which candidate will be most broadly
> > acceptable. Which is more important to them, for their preference
> > to win or for the most broadly acceptable candidate to win?
> > Majority rule.
> >
> > I'd suggest that if their preference was weak, the majority might
> > prefer the Range winner, on reflection. But if their preference was
> > strong, they might insist upon it.
>
>If the first round votes were sincere the Condorcet winner will be
>preferred over the Range winner by majority (since the definition of
>Condorcet winner says so). The Range winner would however be better
>if measured as sum of satisfaction of the voters (if that is what the
>voters marked in the ballots). The opinions could however change
>before the second round as a result of publishing the fact that there
>was a Range winner that was different from the Condorcet winner, and
>the range winner could be supported by a majority at the second round
>(depends on the level of competitiveness etc.).
>
>Juho
>
>
>
>
>
>___________________________________________________________
>The all-new Yahoo! Mail goes wherever you go - free your email 
>address from your Internet provider. http://uk.docs.yahoo.com/nowyoucan.html
>----
>election-methods mailing list - see http://electorama.com/em for list info