[EM] The path to election reform, was Re: Why the concept of "sincere" votes in Range is flawed.

[Abd ul-Rahman Lomax]

At 11:45 PM 11/25/2008, Kevin Venzke wrote:
>Hello,
>
>--- En date de : Mar 25.11.08, Abd ul-Rahman 
>Lomax <abd at lomaxdesign.com> a écrit :
> > What Approval sincerely represents from a voter is a
> > *decision* as to where to place an Approval cutoff.
>
>But is it not true that what *all* methods sincerely represent from a
>voter are the decisions related to voting under that method?
>
>If a decision makes sense in a given context, then that is a sincere
>decision. Is that not your stance?

No. My stance, or my current point of view, is 
that sincerity is a red herring. Votes are 
generally actions, not sentiments or even 
statements of truth. They represent, ordinarily, decisions.

Under many voting systems, a voter may vote a 
preference that is not the voter's true 
preference, specifically, that *reverses* 
preference. The most common form of this is, of 
course, voting for a frontrunner in Plurality 
when one would prefer a different candidate *if 
that option were considered possible.* In the 
U.S., where write-in votes are generally allowed, 
this means that *almost always,* voters are 
voting *insincerely*, and they will continue to 
do so with Instant Runoff Voting.

There is only one system which truly allows 
"fully sincere voting," where the voter can vote 
for their absolute preference out of all eligible 
candidates, and need not add any other votes, and 
that's Asset Voting, and it does so by a trick: 
it is not a deterministic method, it is really an 
input stage to a deliberative process. 
(Interestingly, this was originally invented by 
Lewis Carroll (Charles Dodgson) as a means of 
fixing the problem of exhausted ballots in STV.)

The best writing on this that I've found, it's 
related to the Dhillon-Mertens work, is that of 
voting on lotteries. Which would you prefer, x% 
chance of A or y% chance of B? If you choose the 
latter, does it mean that you "sincerely" prefer 
B to A. No. It means that you prefer -- choose to 
support -- the outcome that represents the 
highest expected return, i.e., absolute value of 
the outcome times probability of the outcome. We 
vary from this kind of behavior for various reasons, but not too far.

In real elections, the voter makes choices based 
on two factors: personal utilities, which create 
preferences of various strengths, and outcome 
probabilities. With ranked methods, an "insincere 
vote" has a clear meaning, it is one which, on 
its face, supports one candidate over another 
while, in fact, the voter prefers a different 
candidate. Voters vote this way in order, we can 
assume, to the extent that they do, in order to 
improve the outcome; a good method will not 
require or reward this. However, there is a class 
of methods which do not provide any incentive to reverse preference.

But there still is an incentive to equate 
preferences when the voter actually does have a 
preference. This has been called "strategic 
voting," in the basic meaning, confusing the hell 
out of the field, because originally "strategic 
voting" was used with respect to preferential 
voting systems and meant preference reversal, and 
thus strategic voting implied insincere voting. 
But setting an approval cutoff and voting 
according to that isn't *insincere*. It simply 
does not disclose a preference that the voter 
considers less important than voting effectively.

Much of the fluff about Approval Voting has been 
based on the concept that there is some absolute 
Approval cutoff, that it can be assumed that 
voters "approve" one set of candidates and not 
another set, and that if the voter votes 
differently, then the voter is voting 
"insincerely" and the method can encourage this, 
and that therefore "Approval is vulnerable to 
strategic voting." But there is no absolute; what 
outcomes we "approve" is not a quality of the 
outcomes alone, but of our assessment of the probabilities of each.

John offers you $1000. Do you accept it? Doesn't 
it depend on the alternatives? No strings 
attached, sure. But if you have a choice between 
a 10% chance of $10,000 -- with an expected value 
of $1000 --, or a 90% chance of $5,000 which do 
you "approve"? One could imagine that one would 
approve of all gifts, but these aren't purely 
gifts, and they are exclusive. You can't have 
both. By voting for the $10,000 option, one would 
be voting "sincerely," by some definitions. But, 
Burr Dilemma: really one "approves" of both 
outcomes, so by that definition, one is voting 
"insincerely." The whole so-called Burr Dilemma 
is based on an assumption that voters "really" 
approve of two candidates, but only vote for 
their favorite. And the dilemma is that 
supposedly this is a difficult choice, because by 
voting only for their favorite they risk a loss, perhaps, to someone worse.

But this is totally ordinary with choices. And in 
the real situation behind the Burr Dilemma, there 
was no problem, because the bullet votes caused 
majority failure. *That was the design.* Further 
process then ensued. There was no constitutional 
crisis, there was merely some frustration as it 
took some time, and many ballots, before it was 
sorted out. Better, in my view, would have been 
Asset Voting -- or, short of that, top-two runoff, back to the electorate.

*No method can guarantee the best result with a 
single ballot, unless you define "best result" as 
the best that could be obtained by a snapshot of 
accurate voter utility profiles at that moment.*

Preferential voting systems did not allow voters 
to vote with *accuracy*, by which I mean that an 
accurate understanding of voter preferences, 
including preference strengths, could be derived 
from the ballot. Unrestricted Range Voting 
likewise fails, but for a political reason: we 
have decided to equate the preferences of all 
voters *who show up*, thus distorting utilities through normalization.

However, if we accept that restriction, and 
likewise accept distortion through round-off 
error caused by practical limitations, only a 
Range ballot collects the information needed to 
determine an optimized outcome, given "accurate" 
ballots, what others have called "sincere" ones.

However, voters won't vote accurate ballots, for 
a number of reasons. I'd contend that we don't 
even know how to do it. Rather, we are 
*instinctively* programmed to consider 
probabilities. It's relatively easy for me to 
determine, of A and B, that I prefer one to the 
other. But then, introduce C. If A and B remain 
the favorite and least-favored, where do I rate 
C? When we try to think of Range Voting as 
involving "sincere ratings," then, we see the 
purported difficult of Range. There is no 
specific meaning to those intermediate ratings.

But when we simply think about Range Votes as 
fractional votes in an Approval election, that 
they are weights we are tossing in baskets, and 
the heaviest basket will win, we are 
instinctively able to do this kind of analysis. 
We do it all the time with any goal-seeking 
behavior. We don't necessarily put our efforts 
toward the ideal outcome, when we don't think it 
reasonably possible; rather, we devote our 
limited resources to an outcome that is an 
optimal combination of desirability and probability of success.

Because we can conceive of Approval Voting very 
simply, as representing a decision to support a 
set of candidates, all of whom are preferred over 
all non-supported candidates, and because 
Approval never rewards insincerity in this (i.e., 
including a candidate in the Approved set, when 
there is another candidate preferred to that one 
who is left in the unapproved set, or 
vice-versa), Approval is strategy-free, in the 
old sense, and this is why Brams introduced it as 
such. The voter sets the Approval cutoff at will, 
based on election probabilities, presumably, or 
just on pure personal preference (reasons other 
than affecting the outcome), and then will vote, 
we may assume, sincerely with respect to this.

This created chaos in the voting systems world; 
it offended many authors because there was a 
family of sincere votes, not just one. Suddenly 
there was no way to take a preference profile and 
determine from it, alone, a "sincere vote." One 
needed to make some other assumptions. Messy. But real.

Range Voting creates "problems," because it 
allows the expression of critical information, 
preference strength. If voters choose not to 
express this, they may, under realistic 
conditions, find some advantage. But what is the 
alternative? The alternative is much worse: don't 
allow that expression. Range, when voted 
"strategically," "degrades" to Approval. Which 
when voted "strategically," "degrades" to 
Plurality (normally, for some reason we don't see 
criticism of Approval with respect to voters 
"insincerely" approving of candidates when the 
voter doesn't "really" approve of them -- I'll get back to this.)

And Plurality is actually a much better method 
than we've given it credit for, when it is used 
within a generally functional political system. 
Proof? Look at the results of U.S. nonpartisan 
IRV elections. We can assume that if voters voted 
sincerely, with Plurality, the results would 
*always* have been the same as with IRV. This 
phenomenon has not received the attention it 
deserves. There will be exceptions, but they are 
much more rare than I would have expected, 
certainly. It's been known in Australia, though, 
that with optional preferential voting, vote 
transfers heavily favor the leader in the first round.

There are exceptions related to the spoiler 
effect, where vote-splitting hurts only one 
candidate, a phenomenon that arises when a third 
party rises up and gains a few percent of the 
vote, without a counterbalancing third party on the other side of the spectrum.

In any case, isn't it suspicious that voting for 
the favorite is considered "sincere" with Plurality, and not with Approval?

What critics of Range would have us do is to 
continue to forbid the expression of preference 
strength, on the argument that some of us won't 
do it "sincerely," and thereby gain some 
advantage. Simulations show that, as I'd expect, 
Range continues to perform well with various 
levels of "incomplete expression," or "strategic voting."

Because we can't be sure that all voters will 
vote "sincerely," something which is never, 
conveniently, defined, we should prevent all voters from voting accurately?

Because some voters may "get what they want" by 
voting with full strength -- I thought that 
seeking to get what they want is what voters are 
supposed to do! -- we should require all voters 
to accept results that are, overall, inferior to Range Voting?

 From my discussions here and elsewhere over the 
last few years, I've come to see a very clear 
path to election reform. The first step is 
extremely clear, and at one time I was able to 
say that voting systems experts actually were in 
agreement about it, except for the few who had 
fallen into the trap that promoting anything 
other than instant runoff voting was politically 
foolish, because of its "momentum."

(1) Start counting all the votes, and the 
candidate with the most votes wins. I have yet to 
see anyone knowledgeable about voting systems who 
thinks that this would be a step backwards! When 
I started, I, like many, though that IRV was 
better than Plurality; I still think that, but 
only in restricted circumstances, which happens 
to be ones which don't apply for nearly all U.S. 
implementations! IRV is *not* better than top-two 
runoff, it is, quite clearly, worse, and only 
better, if that, from an expense or convenience 
point of view. I.e., let's tolerate worse results 
because voting is too much trouble. Bad idea.

There are, from this point, quite a number of 
alternative paths. I'm not even sure which path to pursue first.

(2) Allow fractional votes. That's Range Voting, 
sum of votes version. Warren, dump Average Range, 
there isn't any solid theoretical foundation for 
it, it is a wild idea that can sound appealing 
until one realizes the complications. 
Sum-of-votes Range is entirely consistent with 
standard Approval practice. Average Range is 
*not* used, and has never been used, for any 
political application, and I don't think that 
natural range voting (the honeybee thing) uses 
Average Range. It is sum of votes, which is 
computationally much simpler for an organism. 
(Actually, decisions are made through a fractal 
hierarchy, so it's more complex than simple sum 
of votes, except for very primitive organisms. 
I'm pretty sure that slime molds decide which 
direction to move in by summing the impact of messenger molecules.)

(3) Require that a majority of voters explicitly 
approve an outcome for that outcome to be 
implemented. Note that we already have this 
reform, in many places. It is standard in 
deliberative process. Even when plurality voting 
is used, it often takes a majority vote to 
implement the result. *This method is Condorcet 
compliant!* (Particularly if we qualify 
"Condorcet winner" to allow another winner when 
preference strength is insufficient to motivate 
voters to change the result, even where they have 
the power to do so. The whole concept of 
preferential analysis needs modification like 
this to allow that there is a difference between 
a weak preference and a strong one.)

(4) Use Asset Voting techniques to generate fully 
democratic results. Asset Voting should really be 
a no-brainer for representation. Only Asset 
allows truly maximized representation; all other 
systems assign representative power without the 
consent of a substantial number of voters.

(5) Shift to a parliamentary system so that 
single-winner elections are handled through 
deliberative process, and so that there are no 
fixed terms; rather, single-winner is only for 
officials or what are essentially "hired 
servants," not democratic representaties; such an 
official should only be able to act when 
supported by, at the minimum, a majority of 
voters or their chosen representatives.

(6) Where single-winner elections are still used, 
use Range, either high-resolution or with means 
of indicating preference within a rating level, 
incorporate an acceptance marker, which might be 
built into the rating system --- i.e., Range 5, a 
rating of 3 or more could be considered 
acceptance (and voters would be motivated, then, 
to vote that way); "acceptance" means, "I prefer 
accepting this result to going through the 
election again." If restricted runoff is used 
(i.e., top two or, better, Range winner vs 
Preferential winner), then use an advanced method 
like Bucklin or Approval in the runoff -- even 
though there are just two candidates on the 
ballot -- and allow write-in votes. This allows 
the electorate to fix a problem with the first 
election, if they care sufficiently. And if they 
don't care sufficiently, then fixing the problem is not important!

(7) Use the electoral college created by Asset 
Voting -- the collection of identified public 
voters, those holding and exercising votes they 
received in the general election by secret 
ballot, which, under reasonably settled 
conditions, means anyone who wants to be such a 
voter and so registers, and gets at least one 
vote -- to create hybrid representative/direct 
democracy. An Assembly seat, then, becomes a 
representative for the purposes of participating 
with deliberative rights, continuing the solution 
to the problem of scale that has, heretofor, 
forced us to switch from direct democracy to 
representative democracy whenever the scale 
became too large, while allowing electors to vote 
directly when they choose to do so.

The result of all this: we'd think of the government as "us." And it would be.

(8) Organize directly outside of the legal, 
political system, to maintain a watch over it. 
Use FA/DP principles and techniques to make this 
easy and efficient and not vulnerable to attack 
at concentrated nodes, because FA/DP is 
independent of such. In short, following the 
excellent book, The Starfish and the Spider, 
create a starfish/spider hybrid, with the 
starfish being the people, with the 
communications structures that FA/DP creates, 
bottom-up, exercising power only through the 
power to advise, and the spider being the 
government, with relatively traditional 
structures, but dependent on the continuous 
consent of the people to maintain power.

"Spider" is used as an example of an organism or 
organization which dies immediately -- becomes 
powerless or ineffective -- when you cut off its 
head. Starfish, some species, cut them up, and 
each piece regenerates a complete organism. It is 
extremely difficult to fight starfish 
organizations; the book makes the point that the 
Navajo survived for hundreds of years fighting 
the Spanish because of their distributed power 
culture. The Spanish would kill the leaders, new 
ones would pop up, quickly and spontaneously. 
Al-Qa'ida is a starfish; fighting starfish as if 
they were spiders is a seriously Bad Idea. Pretty 
much, you have to change the environment to fight 
starfish, you have to motivate the members of 
such to do something else. You have to convince 
them. *Maybe* you can exterminate them, but that 
requires a ruthlessness that isn't generally 
supportable. You have to kill everyone who has 
joined a starfish cell, or even who might join 
one. ("Thinks dangerously.") And I rather doubt 
that this is, in the modern world, successful in the long term.