Fwd: Voters Choice Task Force Reminder
Jan Kok
From: Matthew Rome <mrro...@holycross.edu>
Date: Jul 24, 2007 11:55 AM
Voters Choice Task Force Members,
This is a reminder that tomorrow, Wednesday July 25th, the second
meeting of the Voters Choice Task Force will take place in House
Committee Room 0112 in the State Capitol Building. The meeting will
begin at 1:00 p.m. and will last until 4:00 p.m. Attached is a copy of
the agenda for tomorrow's meeting.
Abd ul-Rahman Lomax
on one agenda item.
>IV. Decide on Voting Systems Evaluation Criteria
There exists a series of "election criteria," commonly used to
evaluate election methods. But most of them rest on some intuitive
assumption about what makes for good elections.
Unfortunately, these assumptions are often wrong, as we come to
realize when we study elections more closely.
For example, it was clearly an easy assumption that if we ask voters
to vote for their favorite, the one with the most votes should win.
This Plurality Criterion (the candidate preferred by most voters)
actually is a good one, quite frequently choosing the best winner,
but.... it also suffers from some problems, such as discouraging
voters who want to have a say in the outcome from voting for their
preference, if their preference isn't one of the frontrunners. And so
we get "spoiled" elections and the rest.
I'm pointing out a general problem with election criteria: unless
they actually measure election performance, they only give us a
yes/no kind of value to elections. It is well known that common, and
intuitively satisfying, election criteria, are mutually exclusive.
You can't satisfy all of them. (This may not be entirely true, the
mathematical proof did not consider all possible methods, but suffice
it to say that we probably aren't going to find a method which
satisfies all criteria, and, indeed, some very good ones, commonly
accepted, conflict with others that similarly make sense.)
Given that you can't satisfy all the criteria, then, which ones are
most important? Or all they all equally important? Is the best method
the one which satisfies the most of a list of criteria? That,
obviously, might not be true if some criteria are more important than
others. And some are more important than others.
How can we evaluate election criteria? Well, if we have some method
of evaluating actual elections, and we can simulate elections, we can
find out which methods produce the best results by the "result
criterion." What could this be?
There are two basic standards that are proposed: the utility measure
and the plurality measure, the latter of which comes in various forms.
The utility measure assumes that voters have some personal utility
for each candidate, and then, if we look at the election method and
who won it in some case, we sum the personal utilities over all the
voters and determine the utility of the result.
This should not be confused with Range Voting, where voters actually
can vote what look like utilities, but they are really just votes.
That is, if it is, say, Range 10, with votes possible for each
candidate of 0-10, it is as if the voter has 10 votes to cast, and
may cast up to 10 for each candidate. This is quite like each voter
being, really, 10 voters, who can decide among themselves how to
vote. We do not know that Range Votes are utilities, and we do not
know that they can objectively be compared with each other, though
Range advocates would say that it is reasonable to act as if they can.
But in the simulations, we *assume* the utilities, so we know them.
For a good simulation, we try to spread out the utilities in a
realistic way, and there remains a lot of work to do on this.
Nevertheless there are such simulators, most notably IEVS, a
work-in-progress, open-source program, available through the Center
for Range Voting.
The plurality measure considers an election success to be measured by
the percentage of voters who "approve" of the result. "Approve," it
turns out, has no clear definition, so, in the end, it must be a kind
of vote, the voters tell us if they "approve" or not, and a problem
arises when looking at the Approval method: approval admits of no
graduations, there is no difference between slightly approve and
strongly approve. A prominent proponent of the approval plurality
measure says that it properly should minimize the number of voters
who "strongly disapprove" of the candidate, though if we want to
measure this, we need preference strength information.
Underneath the plurality measure is the intuitive appeal of the
principle of majority rule.
The utility measure and majority rule conflict, it is easy to
conclude, because it can happen that a majority has a small
preference, and a minority has a large preference. Consider the case
of a group of people choosing a pizza, they can only buy one. A
majority have a slight preference for pepperoni, a minority actually
can't eat pepperoni, either for religious reasons or perhaps because
of an allergy. If there are only two choices, pepperoni and, say,
mushroom, and, in fact, the majority likes mushroom *almost* as much,
most people would agree that mushroom would be the best choice, even
though it is not the preference of a majority.
Real groups solve this problem by not using election methods to make
decisions! Rather, they use deliberative process, they talk about it,
and they will learn about the problem with pepperoni, and the
majority will no longer consider it the best choice. In the end, in a
healthy society, the best *overall* choice is made, the majority is
willing to give up a little, usually, to help many with greater needs.
There is a compromise that satisfies, reasonably, both utility
maximization and majority rule, and it involves runoff elections.
Real runoffs, under certain situations. Instant runoff is not
adequate for this, I suspect.
As an example, consider an election just like the common ones of
today, but overvoting is allowed. If a majority have voted for one
candidate, and only one candidate receives such a majority, we have a
reasonably clear choice, which would almost always coincides with the
utility maximizing candidate. However, there are two situations where
the candidate with the most votes might *not* be the maximizer. And
they are easy to detect. The situations are that (1) no candidate
gets a majority of votes cast, and (2) more than one candidate gets a majority.
In the first case, many jurisdictions would, under existing rules,
already require a runoff election between the top two. And in the
second, if we make a change to allow voting for more than one, it
would be fairly simple to extend the rules to requiring a runoff with
no majority winner to one with no exclusive majority winner. It might
be a matter of inserting one word to deal with the new possibility
that we have created by allowing overvotes: more than one candidate
gaining a majority.
This combination method has not been tested with simulators yet, but
similar methods have, and there are reasons to believe, as this is
written, that this would be among the very best election methods, as
determined by social utility, and it necessarily would satisfy,
almost always, the Majority Criterion, which some argue Approval
Voting (allowing multiple votes) fails. Yet it is only standard
voting with a common runoff rule, the only difference is that, as Mr.
Kok says, we Count All The Votes. Cheap. Simple. Easy to understand.
And simply a first step, it does not preclude moving, later, to
ranked choice methods or more detailed Range methods. Indeed, there
are very simple and small steps that can be taken in those
directions, it is not necessary to go to full-blown ranked ballots
and complex counting process, or to substantially more complicated
Range ballots of high precision.
You can have the greatest good for the greatest number *and* preserve
Majority Rule. The key is that whenever an action is less than
preferred by a majority, the majority explicitly consents to it. This
is what our pizza group would normally do. If they are bound by some
election methods, they would make the wrong choice. With some other
methods, they'd make the right choice, based on overall satisfaction,
but that choice involves, in the case described, some small sacrifice
by the majority.
In the end, I'm recommending that committee members be aware that
various criteria will be advocated by various groups as being the
most important ones. But all criteria are not created equal, and some
criteria actually will be satisfied at a cost, satisfying the
criterion, no matter how good it sounds in itself, can be worse than
breaking it. There is a temptation, especially for those new to the
study of election methods, to start thinking of this criterion or
that as an absolute necessity; I'd like to warn against that. We need
means of evaluating the evaluation methods!
Preferably these criteria measures would be as objective as possible
and, since nearly all election methods on the table satisfy some
important criteria, they would produce results that can be used to
compare methods, quantitatively. Ultimately, what we need is a way of
judging election *results*, not just methods.
I certainly have my own opinions about that, but my purpose here is
not to advocate my opinions, but to point out and clarify, for the
committee members and others who may be interested, the
considerations, and to suggest that, at the very least, relatively
objective criteria for measuring election performance, with some
accuracy, not just PASS/FAIL, be included in the list of criteria to
be considered.
Terrill Bouricius
1. Mr. Lomax makes a good point that criterion conflict with each other.
However, it appears that the criterion the task force must use for
evaluating voting systems has been delimited by the foundation
legislation, which repeatedly states that promoting majority rule is the
ultimate goal. While some election theorists see majority rule as merely
one of many conflicting goals, nearly all Americans have internalized this
goal (even if merely parroting an un-thinking slogan), and any voting
reform that doesn't promote it has no chance of adoption in our lifetimes.
2. There is another aspect of voting methods that may be beyond this task
force's ability to fully analyze... That is the impact various voting
methods have on the way candidates campaign, and thus the information
voters receive and use to evaluate candidates. Some voting methods may
encourage or discourage negative campaigning, others may promote taking
strong stands on issues, or avoiding taking stands on any controversial
issues (for fear of alienating any voters), etc. This is another place
where "in the lab" computer modeling of hypothetical election results of
voting methods fails.
Voters do not simply HAVE preferences among candidates that are
FIXED...they FORM their preferences in response to how candidates
campaign. How candidates campaign, is influenced by the best strategy
appropriate to the voting method in use. For example...if voting method 1
is in use, candidates present themselves in such a way that a particular
voter prefers candidate A>B>C, but if voting method 2 were in use,
candidates give voters different (perhaps more, perhaps less) information
about policy positions, etc., and that same voter might prefer the
candidates C>B>A. One can use common sense to guess how IRV, Approval,
Borda, etc. would affect campaigns and thus voter information, but
real-world, high-stakes elections are necessary to know for sure.
The fact that these effects are currently unprovable (since many
alternative voting systems have no real-world, high-stakes governmental
election track record) does not mean that these potentially fundamental
impacts should be ignored.
Abd ul-Rahman Lomax
>Very thoughtful email...I would like to add two thoughts...
>
>1. Mr. Lomax makes a good point that criterion conflict with each other.
>However, it appears that the criterion the task force must use for
>evaluating voting systems has been delimited by the foundation
>legislation, which repeatedly states that promoting majority rule is the
>ultimate goal.
There is no necessary conflict with "majority rule." It is well-known
that majority rule properly applies only to Yes/No votes. The proper
comparison here is with a series of conflicting referenda. If we
think of an election as a referendum on a series of questions, Shall
we elect A, Shall we elect B, etc., it is standard that each question
is decided by majority rule; indeed, there is no such thing as a
Plurality distinct from a Majority in a Yes/No question, unless we
count abstentions.
Standard practice when there are such conflicting referenda is that,
if more than one passes, the one with the most Yes votes wins.
However, this can violate the "Majority Criterion," as some
understand it, for it is possible that a majority preferred one of
the referenda with fewer total Yes votes, but because they chose to
vote Yes for more than one of them, this preference is suppressed.
Majority Rule, to reiterate, means that on any single question
presented for vote, the Majority prevails. Elections, in general, are
not single questions. Even if there are only two candidates, in full
deliberative process, which is how majority rule is truly expressed,
a forced choice between the two eliminates a logical possibility that
majority rule would require remain possible: both are rejected.
Now, it is possible to have a question on a ballot: Shall the result
of this election be accepted [according to whatever rules are in
place for it]? If the answer to this question is Yes, that is, more
people vote Yes than No, then, indeed, "majority rule" is fully and
explicitly satisfied if a majority vote Yes. It was a simple
question, no complications!
Having elections without some explicit ratification of the result is
a violation of majority rule, so if we were bound to only pure
majority rule, we wouldn't allow elections to complete without some
process like I described, or, what is more common in small
organizations, without an explicit acceptance of the election result.
I.e., a vote with the question, :Shall the declared result of the
election be accepted?" Or, "Shall Bush be elected?" Yes/No.
It's important to understand that election methods are a shortcut,
sacrificing pure democracy (and pure majority rule) in favor of
efficiency and cost. We are so accustomed to equating democracy with
elections that we often overlook this. In pure democratic process,
election methods may be used, but they never result in a completed
election unless the majority explicitly consents to it.
What I've proposed -- though I'm not asking the task force to
consider it at this time, it's a detail to come up later, albeit what
could be an important one -- is that we use satisfaction-maximizing
methods like Range combined with analysis of majority preference.
Whenever there is a conflict between maximizing overall approval or
satisfaction, however defined, and the preference of a majority (or
even a plurality preference, perhaps), that there be a runoff
election. While this is still not the pure, absolute "majority rule"
I describe above, it gets close. The various election criteria
conflicts occur with elections, generally, where there are more than
two candidates. You can't have a Condorcet cycle, for example, with
two candidates. Range Voting generally reduces to Approval with two
candidates (though there remain a few quibbles about that). IRV would
be irrelevant with two candidates, etc.
> While some election theorists see majority rule as merely
>one of many conflicting goals, nearly all Americans have internalized this
>goal (even if merely parroting an un-thinking slogan), and any voting
>reform that doesn't promote it has no chance of adoption in our lifetimes.
We need a clear definition of "majority rule." If the majority, for
example, chooses to implement a method which can violate it in some
way, is it still majority rule. My own opinion is that the majority
could choose whatever damn fool method it chooses, and as long as it
can change the method election by election (that is, majority rule is
immediate, not bound by the past), it is still "majority rule." The
majority could decide, for example, to delegate a particular election
to an individual or committee, and this would satisfy majority rule.
Most of the methods on the table can generate a result that the
majority would reject in favor of another candidate. So most of the
methods "violate majority rule." However, if, under some conditions,
a single runoff is allowed, you can convert this otherwise-violation
to satisfying, explicitly, majority rule, particularly if it is
guaranteed that the majority preference will make it to the runoff.
The problem with top-two runoff (and thus with IRV, which is top-two
runoff extended) is that the winner can be one who would be rejected
in a pairwise election with the top-two winner. Doesn't this violate
"majority rule?"
The simplest implementation I'd propose that comes much closer to
satisfying true majority rule than what is ordinarily considered is
Count All the Votes, a name for Approval that emphasizes how simple
it really is. (Contrary to what is often claimed, Approval *is* used
in public elections, I gave the example above, conflicting referenda.
And there is in fact another situation where Approval is used, though
it has mostly escaped notice and I don't know how often multiple
voting occurs: it isn't necessarily recorded!)
Now, Approval, it is sometimes asserted, does not satisfy the
Majority Criterion. It is claimed by Approval supporters that when it
violates this criterion, it does so to choose a better winner. The
Majority Criterion is not "Majority Rule," but there certainly is a
relationship. The question of whether or not Approval satisfies the
Majority Criterion is a complex one, and it turns out that experts
can disagree on it. Depends on what it means to "prefer a candidate
over all others." But suffice it to say that, with Count All the
Votes, it is quite possible, even likely, that a voter votes for more
than one even if having a preference between them. Usually one of
these votes is truly moot, but if the voter errs in the strategy, the
voter has, in fact, suppressed preference and thus there could be a
majority with that preference which fails to prevail.
But this situations is detectable, and, it turns out, we already have
standard election practice which gets us there, there is precedent,
and we only need to adapt the precedent to a new possible situation
if we Count All the Votes: what do we do if more than one candidate
receives a majority of votes?
Clearly, with simple approval, as with conflicting referenda, we
would elect the one with the most votes. But this is not necessarily
a majority preference, because the majority may have suppressed that
preference to vote for more than one.
In such a case, though, there will be more than one candidate with a
majority, for we can expect that the majority *will* vote for its
preference as well as another. There is no strategic motivation to
not vote for your preference. So if the majority preference fails to
get the most yes votes, it still must get a majority of votes.
So, presently, in some jurisdictions, the rules require a majority
winner, or there is a runoff between the top two. We simply extend
that rule to cover any situation where there is no *exclusive*
majority winner, and, suddenly, we have detected the situation where
Approval can fail to elect the majority preference, and we then
submit the top two for vote. The majority, now, has no strategic
motivation; if it maintains its preference, that preference will prevail.
It might change the preference, though. The argument of Approval
Voters is that, generally, this more-widely-approved candidate is a
better choice, and the majority *might* agree, there is at least a shot at it.
We can have approval maximization *and* majority rule, or, more
accurately, better implementation of majority rule than we presently
have, far less likely to elect someone who would lose, pairwise, to
another candidate.
So I'm glad Mr. Bouricius brought this up. It is, indeed, a very
important point.
>2. There is another aspect of voting methods that may be beyond this task
>force's ability to fully analyze... That is the impact various voting
>methods have on the way candidates campaign, and thus the information
>voters receive and use to evaluate candidates. Some voting methods may
>encourage or discourage negative campaigning, others may promote taking
>strong stands on issues, or avoiding taking stands on any controversial
>issues (for fear of alienating any voters), etc. This is another place
>where "in the lab" computer modeling of hypothetical election results of
>voting methods fails.
This is completely true; however, we have very little information on
how voting methods affect campaign strategies, and what is often
claimed about it is little more than speculation, and not neutral
speculation but simply another advocacy argument.
Implementing Count All the Votes is not terribly likely to have
immediate effects on campaigning. It does create a motive for
candidates to seek to broaden appeal and thus it *might* have some
effect on reducing negative campaigning; however, reducing negative
campaigning, to my mind, is not a legitimate role of government. It
is arguable that negative campaigning is good for us. I'm not
suggesting that it is, generally, but what if I *do* know that a
certain candidate is a fraud and a cheat? That he is lying? Should I
keep silent on it, because it would be "negative campaigning?"
It's not a simple problem!
Again, using Count All the Votes will probably leave most voters
quite as they have always voted, they will vote for the major party
candidate and that's that. It really only affects, at least
initially, a relatively small number of voters who prefer a third
party candidate, and it is a huge benefit for them, but it also
benefits the major party candidate, *a little*, who can attract
additional approvals from these voters. Usually, though, because of
the constellations of third parties, these votes are already pretty
settled, and, say, a Democrat who wants to win isn't going to gain
too many more votes from Greens that he or she wouldn't already get
if we had Count All the Votes, based on being namby-pamby in
positions. Rather, the motivation for bland -- or centrist is a less
judgemental way of saying it -- is to get Republicans to vote for
you. In other words, the existing situation!
What really enhances negative campaigning is two-party domination....
and we aren't going to solve that, immediately, with our voting
methods. IRV appears to maintain two-party duopoly quite well, where
it is in use. There is theory that Range, and possibly Approval
(which is the simplest Range method), will help incubate third
parties, but I'd expect this to be a slow, long-term effect. Range,
in particular, with higher resolution, allows voters to show
precisely the level of support they could give a candidate, and thus
a party, and thus allows third parties to show their true strength,
which is certainly fair. IRV would also do this, allowing those who
prefer third parties to express that preference.
>Voters do not simply HAVE preferences among candidates that are
>FIXED...they FORM their preferences in response to how candidates
>campaign.
Yes.
> How candidates campaign, is influenced by the best strategy
>appropriate to the voting method in use. For example...if voting method 1
>is in use, candidates present themselves in such a way that a particular
>voter prefers candidate A>B>C, but if voting method 2 were in use,
>candidates give voters different (perhaps more, perhaps less) information
>about policy positions, etc., and that same voter might prefer the
>candidates C>B>A. One can use common sense to guess how IRV, Approval,
>Borda, etc. would affect campaigns and thus voter information, but
>real-world, high-stakes elections are necessary to know for sure.
That's right. I think that the very simple change of tossing the
no-overvoting rule would have no drastic effects, it isn't suddenly
going to elect Ralph Nader as Governor of Colorado. Or as President.
It's just going to make election spoilage by third party candidacies
much less of a problem.
Ideally, there would be trials on a small scale. However, I'd
suggest, the Count All the Votes proposal is so simple, and so
cost-free, that it's worth considering it state-wide immediately. In
the alternative, the state could simply *allow* local jurisdictions
to use it. Thus there would be small-scale tests in jurisdictions
that decided to try it, like towns or cities. (This may already be
possible for small jurisdictions, in which case the work of the Task
Force would party be to help get all the reasonable possibilities on
the table, so that the trials aren't just focused on whatever
proposal has "political momentum," which really means that there is a
focused group of funded activists working for it, for their own purposes.
(Which may indeed be noble purposes, but.... they are not neutral and
they made their decisions long ago, and are not known for
reconsidering them based on new information. They are dedicated
activists and don't spend a lot of time considering alternatives to
what they are funded to do.)
To be more explicit, a jurisdiction could go for IRV. There is little
doubt but that this is some improvement over Plurality. However, it's
relatively expensive, the ballot is complex, the counting likewise.
Precinct votes can't simply be amalgamated and passed on -- if
counting is to take place at a higher level, individual ballot data
must be transmitted.
So, I'd suggest, for a quick and simple change, Just Count All the
Votes. Many experts think this is a better method than IRV anyway!
I'm not trying to resolve that question, and implementing Approval
does not prejudice further change; indeed, my expectation is that if
we get Approval, people will start to want more flexibility in
expression than the Yes/No that is Approval (with Just Count All the
Votes, a blank is a No.)
They will want to be able to simultaneously express their preference,
while being able to participate in other pairwise contests. And there
are then two ways to go: the ranked way and the rating way, i.e.,
ranked methods (like IRV or Condorcet methods) or Range with higher
resolution, starting with Range 2, i.e, the votes are 0, 1, 2, or -, 0, +.
The latter (negative, neutral, positive) is being used for MSNBC
polls and it is fascinating to see how much more information is
collected with this simple increase in options. The Republican polls
showed Ron Paul drastically leading the others, being the *only*
Republican with net positive ratings, and significant ones, on the
order of the best Democrats. If those polls were not biased by armies
of sock puppets (and I think they were not, though this is what the
Republican pundits have claimed, without any proof at all), then
there is one Republican who might be able to beat whoever the
Democrats nominate, at least as things stood when I last looked,
which was some time ago, maybe a month or two. Other polls, generally
plurality polls, vote for one, show nothing like this, Ron Paul is
down in the noise. If armies of sock puppets are deranging the MSNBC
polls, why not the other ones?
The reason is that vote for one requires voters to make strategic
considerations. Further, the MSNBC polls allow both Democrats and
Republicans to vote, and a Democrat in a plurality poll isn't going
to vote, probably, for *any* Republican, under current conditions.
But if the Democrat can express approval of some Democrats *and* a
Republican like Paul (who is really a Libertarian), why not?
On the Democratic side, you could see Edwards and Obama with a
commanding lead, and Clinton struggling with a small net negative
rating. Lots of positives, but slightly more negatives. And that's
pretty much the situation with Clinton, lots of support, but strong
negatives from some that just might, overall, cancel out or more than
cancel out the positives.
Ron Paul, I'd predict, will not win the Republican nomination. And
the main reason is that nobody believes he can win. So people will
vote, under the rules, for someone they think can win, because
otherwise their vote is wasted. It's circular. We need voting systems
that allow us to escape from this trap.
And with Bloomberg coming, if I were a Republican, I'd be looking to
get Count All the Votes in place immediately.
>The fact that these effects are currently unprovable (since many
>alternative voting systems have no real-world, high-stakes governmental
>election track record) does not mean that these potentially fundamental
>impacts should be ignored.
Sure. However, there is no reason to expect a large immediate impact
on campaigning strategy from Count All the Votes. To the large
majority of voters, the ability to cast an additional vote will be
moot. To a few percent, it will help them, but these voters are
generally to the left or right of the major parties, and thus
negative campaigning will be no more effective than it is now. If a
Democrat slanders the Republican, is that going to change a Green
planning to vote only for the Green candidate to vote for the
Democrat? This Green voter is already disposed to see the Democrat,
we can expect, in general, more favorably than the Republican. And a
Libertarian may already be predisposted to vote for the Republican,
the slander isn't going to win votes from Libertarians. No, what it
will do, as it does now, is to bleed some votes away from the
Democrat, from those closer to the middle, or from independents.
And we have that problem already. Count All the Votes won't
*increase* the problem, and it might decrease it *a little*, which is
all we need to know for the short-term. For the long-term, we would
really like to see what happens with various experiments on a local
scale, or elsewhere in the country.
It's not clear that election method will have any drastic effect on
campaigning practice, no matter what method we use. What will change
that, I think I know, but it is an entirely different approach, and
it does not involve governmental action.
When government starts to try to regulate or influence how people
express themselves, within the standard limits of free speech, we
might create a dangerous feedback loop, for expression is an
important part of how government is controlled by the people. Similar
arguments could apply, by the way, to public (i.e, governmental)
campaign funding. But this is, again, an entirely different issue,
and it would be an error to infer my position from the little I've
said about it.