[EM] Lomax reply 3,/13/12

[Abd ul-Rahman Lomax]

At 04:25 PM 3/13/2012, MIKE OSSIPOFF wrote:


>Abd:
>
>I'm starting this reply today, though I probably won't have enough 
>computer-time
>today to complete this reply.
>
>First, two definitions that you asked about:
>
>AOC:
>
>AOC is Optional Conditional Approval. You can make any of your 
>approvals conditional
>by mutuality. Then, that approval is counted only if it is 
>reciprocated. The best
>way to describe how it actually works is by the pseudocode for the 
>proposed process
>for optional conditionality by mutuality. That pseudocode can be 
>found in my first
>posting that introduced (and mentioned in the subject-line) 
>"MTA-Optional-Conditional (MTAOC)"
>That posting contains the pseudocode for determining whether a 
>conditional MTA middle-rating will
>be counted. AOC uses that same algorithm to determine whether to 
>count a conditional approval.
>Some days later I posted a message with a subject line indicating a 
>fix for a typo, an
>inadvertently-omitted line that needed to be added to the 
>pseudocode. Then later I posted another message in which I
>suggested that all of your top-rated (in MTA) candidates should be 
>regarded as "coalition-suitable"
>(instead of asking the voter to mark some top-rated candidates as 
>"coaliton-suitable") for you.
>In AOC, unconditional approval is, for that purpose, treated as top-rating.
>
>I admit that that is a mess--when my 
>optional-conditionality-by-mutuality algorithm definition
>is in three widely-separated postings. At least I should re-post the 
>corrected pseudocode in
>one posting. Should have already done that before now. Will within a few days.

While there may be value for this in terms of working on improved 
methods, as to theory, as to possible public implementations, not 
method that is so complex to explain has a prayer of seeing 
application outside of specialized societies where they are willing 
to tolerate that.


>u/a election:
>
>u/a stands for unacceptable/acceptable. A u/a election is one in 
>which there is one or more completely unacceptable
>candidates who could win.

If write-ins are allowed, this is theoretically possible for all 
zero-knowledge elections. However, many elections, for many potential 
voters, do not fall into this category. As shown by the fact that 
they do not vote. Obviously, whatever is the result of the election, 
it doesn't make enough difference to them to vote. That's actually 
important. If you somehow entice these voters into participating in 
an election, without shifting something else, you will be adding 
noise, and noise probably biased by such factors as media exposure of 
candidates.

I have been, in general, pointing out that increased turnout can 
actually harm the social utility of results. In runoff elections, the 
common lower turnout doesn't mean that there is anything wrong. It 
means that the potential electorate largely doesn't care about that 
choice, which might be because they are just fine with both, or 
equally diapproving. The ballots have been pre-filtered, we might 
say, for low preference strength, which pushes the results toward 
normalized utility maximization.

>In such an election, avoiding the election of an unacceptable is 
>all-important.

Sure. However, in situations where that election is reasonably 
likely, the voter is teetering on the edge of a falling-out with 
society itself.

>  In a u/a election by ABucklin,
>it's definitely your best strategy to top-rank all the acceptables, 
>and not rank any unacceptables.

This is a black-and-white analysis. "Unacceptable" has been pushed to 
the maximum of unacceptability, i.e., if So-and-So is elected, I 
might as well commit suicide. Given that, if some truly horrible 
candidate is on the ballot, I'll definitely vote, and I'll vote in 
such a way as to defeat the candidate, that's the condition of the 
problem. If there is more than one such "totally unacceptable" 
candidates, each a possible winner, I might really wonder about the 
society itself.

Note that if I don't consider the unacceptable candidate a possible 
winner, the whole argument is false. If I truly believe it's 
impossible, I'll not give it any voting strength; for this not to be 
the case, my horror at possibility of the candidate's election would 
have to be infinite. Otherwise zero times a finite disapproval is still zero.

>  I consider our
>public political elections to be u/a. The Republocrats are the among 
>the unacceptables, though there are probably
>others too. That's just my opinion as a voter, judging by standards 
>such as dishonesty, corruption, bought-ness, etc.

That's your opinion, to which you have a right, but you are rather 
obviously not typical. At all. And how you would vote is not 
particularly relevant, then, to the design of public systems. I would 
argue that public systems should allow you to express your 
preferences fairly. If you want to vote by categorizing candidates 
into two classes, acceptable and unacceptable, voting only for the 
acceptable candidates, with maximum strength, you should be free to 
do so, because you are, by definition, willing to accept the loss of 
choice between the acceptable ones.
I think, Mike, that you may want your cake and to eat it too. You 
want to be able to get the "unacceptables" out of the way, first, 
then you can choose. I only see one way to do that which doesn't 
bring up the various voting paradoxes, and that's a system with at 
least two rounds as a possibility. As long as the first round makes 
the choice you need, you are okay. The problem is that there may be a 
winner of the first round who is still unacceptable to you (but 
obviously is acceptable, or more, for many voters).

Mike, I'll repeat this: most voters will not be thinking like you.

We all support voting systems which will allow you to, at least, act 
clearly to prevent unacceptable results. With Count All the Votes, 
you know how to vote, and it's simple and easy to understand. I'm 
merely suggesting that with some better knowledge of the 
probabilities, you may be able to maximize your expected return. This 
black-and-white understanding of what is acceptable leads you to the 
black-and-white voting. It reduces your real effectiveness in the 
world. But it's only one vote, after all. Your participation in the 
other aspects of the political process can be much more important.

>The Republocrats are a set, effectively a party, consisting of two 
>nearly identical subsets called
>Democrats and Republicans. Gore Vidal said that we don't have a 
>two-party system--We have one party with two right wings.

Must be true if Gore Vidal said it.

No, if he was actually making sense, we have one party with a right 
and a left wing. Sort of. So we have a one-party system. We are 
accustomed to thinking of that as a bad thing. Is it? I'm not sure at 
all. We have one government, and that is really "one party." Imagine 
the communist regimes, where the Party nominates the candidates, for 
unopposed elections. Is this undemocratic? Not necessarily. It 
depends on the Party's nomination process, and whether or not the 
Party truly represents the people, as it claims. The problem was that 
it didn't. Or doesn't, as the case may be.

If the people, voting directly in secret ballot elections, were able 
to actually reject the election of the Party's nominee, that would be 
democratic! At least in part!

>End of definitions of AOC and u/a election.
>
>At 04:04 PM 3/5/2012, MIKE OSSIPOFF wrote:
>
>You said that achievability, getting from here to there, is more 
>important than optimality, for
>a voting system proposal. Yes, and that's why strongly suggest that 
>Approval should be the first
>proposal. In later proposals, vote-management options could be 
>added. They include:

Count All the Votes. Terminally simple, clear improvement, only bogus 
arguments against it.

Not Perfect. So?

>AOC, MTAOC, MCAOC, AOCBucklin (Bucklin with the MTAOC kind of 
>optional conditionality),
>the delegation option of SODA, and fractional voting (RV). All of 
>these options could be offered
>for the same Approval election. Of course they needn't all be 
>offered in the same enhancement
>proposal. I'd start with AOC. But, first, Approval should be proposed.

Count All the Votes. When it's is presented as Approval, hosts of 
distracting questions are raised, such as "Do I really approve of 
that frontrunner?"

Voting is better understood as adding weight to a balance, rather 
than as a "sincere expression of an opinion," though we can design 
systems that make the adding of the weight relatively easy to deduce 
from the opinion. But, always, it's an action, not a sentiment. And, 
my view, we are responsible for the anticipable results of our actions.

>Approval, though entirely adequate as a destination method, is also 
>the best route to ABucklin or
>SODA or maybe RV.
>
>Or, alternatively, RV could be the first proposal:

In NGO elections, quite possibly. In public elections, I'm obviously 
not opposed, but ... understand that Range can badly fail, because of 
utility normalization problems. I'm a whole lot happier with Range as 
an element in a runoff system.

Range polling of course, is drastically superior to any other kind. A 
Range ballot is maximally expressive, with one exception. It doesn't 
express approval, which is a separate decision. We need either a 
voter-determined approval cutoff, which is complicated, or we need to 
predefine the ratings, classifying them into approved and 
not-approved. We may not need so many unapproved ratings as we do 
approved ones, so the proposal of using mid-range with a linear range 
ballot is certainly not the only one.

Approval, in a runoff system, means "I prefer this candidate to a 
runoff being held." That pushes off compromise to the runoff, if it's 
a difficult compromise. That's quite proper. If the stage is 
deterministic, then, approval means something a little different, it 
means "i prefer this candidate to the expected outcome of the election."

>Of course it goes without saying that Approval voting is 
>automatically a way of voting in RV (you
>top-rate your approved candidates and bottom-rate everyone else). 
>Then, all of the abovementioned
>Approval vote-management options could be proposed too.

Which is how you would vote in the u/a scenario, right?

>Approval, then, is naturally, easily, expandable to AOC, AOCBucklin, 
>ABucklin, RV, SODA, etc.,
>as vote-management options in the Approval election.

Yes. It all starts with counting all the votes.


>You spoke of the "one-person-one-vote" (opov) objection to Approval.
>
>Yes, that objection amounts to a complete misunderstand of what opov 
>was originally intended to mean.

It reduces it to technical compliance with a particular and shallow 
understanding of the intention.

>Opov meant that each person should have the same voting power. One 
>person shouldn't be able to outvote
>two or three people. Approval doesn't violate opov.

At all.

>If someone wants more elaboration about that, point out to them 
>that, in an Approval election, my ballot
>can cancel your ballot, no matter how many candidates you vote 
>for--I could do that by voting for all
>the candidates you don't vote for, and not for any that you do vote for.

This is true, but doesn't seem directly convincing to me. I prefer to 
simply point out that my vote has either the ultimate effect of one 
vote, or has no effect. (there is one exception, in a runoff system, 
and it brings up some interesting possibilities.) This argument is 
true, in basic Approval.

>For example, suppose that you vote for all of the candidates except 
>for one. I can cancel your ballot
>by voting only for the one candidate that you didn't vote for.

Of course.

>So, obviously, you don't have more voting power by voting for more 
>candidates. In a 10-candidate election,
>in the above example, you've voted for 9 times as many candidates 
>I've voted for, but I've canceled you
>out. Voting for more candidates didn't give you more voting power. 
>Any ballot can be cancelled out by
>an oppositely-voted ballot.

It's true.

>Approval is a point system. Like the 0-10 point system, or the 0-100 
>point system, Approval is a point
>system. Approval is the 0-1 point system.

Of course.

>Approval asks you to rate each candidate as acceptable or 
>unacceptable. Your vote on a candidate's
>acceptability is no stronger than anyone else's.

You can interpret it that way. I prefer, simply, that you are allowed 
to vote for any candidate, equally. In basic Approval, same ballot as 
plurality (the place to start), you are adding a vote to the total of 
each candidate you "approve." This is equivalent to voting against 
any candidate you do not approve. If it is truly take that way, then 
awaring the election to a candidate without a majority is quite 
equivalen to awarding it to someone when the majority voted against 
that candidate. That's why, in direct deliberative process, a mere 
plurality is never enough.

"Approval" can be understood in two ways: as an action allowing 
something to happen, in this case, the election of the candidate, or 
as a sentiment. A lot of the confusion over approval is due to 
understanding it the second way.

>Approval elects the candidate who is acceptable to the most voters.

Not exactly, but close. It elects the candidates who were given the 
most votes. The winning total is the same as the number of voters who 
"approved" the candidate. Any given voter has only contributed one 
vote to that total, which is why it is still one-person, one-vote.


>I've probably mentioned this in my other reply, but all of the 
>Approval strategies that we've discussed at EM
>amount to voting for all the candidates who are better than your 
>expectation for the election. In other words,
>vote for a candidate only if you'd rather appoint hir to office 
>instead of holding the election.

Yes, assuming that this is the only choice you have. "The election" 
includes all the other candidates and probabilities you assess. It is 
inherently a "strategic" decision, in a meaning of the word.

>So, Approval maximizes the number of voters for whom the winner is 
>better than their expectation was. The candidate
>who was better than what they expected. The candidate who maximizes 
>the number of pleasantly-surprised voters.

Yup.




> >(First, when I speak of ABucklin as an Approval option, that's only
> >for brevity and simplicity,
> >because, starting from Approval, I'd offer AOC as an option or an
> >alternative method, before ABucklin.
> >AOCBucklin, for me, would be next, after AOC. Still, as Abd
> >suggested, ABucklin is the natural
> >1-balloting implementation of a reasonable and obvious collective
> >deliberating process, and, thereby,
> >might be the first alternative voting system that people would
> >accept--Then AOCBucklin could be reserved
> >for a later enhancement.)
>
>Unfortunately, I don't know what "AOC" is, and I searched the subject
>headers here. It was used in many posts, none of which explained what
>the method was, some contained long sentences with unexplained
>alphabet soup..... Sigh.
>
>You continued:
>
>Bucklin is a simululation of a *common* collective
>deliberating process, not just a reasonable one. Bucklin/runoff is a
>more sophisticated simulation that splits up the compromising process
>into two ballots.
>
>[endquote]
>
>...But the runoff would complicate the proposal. You could propose 
>it later, though.
>But when you complicate or elaborate a method, FBC-failure tends to sneak in.

Maybe. But if the existing method is top-two runoff, you could be 
making things worse by, say, going to raw Approval, which often 
reduces to plurality.

>In your posting you suggest a runoff between some two members of 
>{CW, Bucklin-winner, RV-winner}.
>
>That adds complication, and so, as you probably know, it shouldn't 
>be proposed other than as a
>later enhancement. And deciding which two members of that 3-element 
>set should go into the runoff adds another complication.

I agree it is a later enhancement, very likely. The earlier step I 
would propose is using a Range ballot for a Bucklin method. (Since I 
already think Bucklin ballots,  i.e., 3-rank Bucklin, are range 
ballots, I mean, by this, that the ballot has more than one 
unapproved rank. If the method was traditional Bucklin, where we 
might assign ratings of 0 (default, no vote), 2, 3, 4, to the ranks, 
we would add an unapproved rank, meaning "not approved -- not a part 
of a majority if that is required" but better than the worst. In the 
runoff, this might be used, under some conditions. And then it might 
be used in a primary, to determine a Range winner, and test this 
against the Bucklin winner. And to determine a Condorcet winner, the same.
These are actually easy to express and understand.

Part of the goal is simply to collect better data on real preferences 
and expressions, in real elections, for the advancement of election science.

If you want to know what I'd really propose, as to voting system 
reform, I propose the creation of committees to study voting systems, 
deeply, with the responsibility of making recommendations to those 
that appoint them. These committees should take whatever time it 
takes to thoroughly study the issues. I saw such a committee in 
Colorado largely fail because it really wanted a quick decision, and 
was then easily vulnerable to the sound-bite arguments of UnFairVote. 
Such a committee should be encouraged to be careful, above everything 
else, and to propose, first, reforms that will do little or no harm, 
if they fail to realize the promises that the theory predicts.

Count All the votes, regardless of whether or not the existing system 
is Plurality or Top Two Runoff (and maybe IRV, I don't know if people 
have noticed it, but Count All the Votes would have some salutary 
effect on IRV, and would do no harm), is an ideal first reform. Any 
reform process should include study and followup. We can expect from 
Approval Voting, that it will not be a vast advance over Plurality. 
Most elections it will make no difference. But elections where there 
is a significant spoiler effect from a minor candidate, it can help. 
It doesn't need to be a large percentage of the vote.

And whatever reform is then considered, unless it can be seen as 
actually harmful, will keep Count All the Votes and will improve on it.

>Adding that runoff would mean making a new method, instead of just 
>having Approval with vote-management options.
>I'm not saying that that rules it out, but an enhancement other than 
>an option would, of course, require more
>justification and more convincing.

Consider, instead, that top two runoff already exists. Proposing 
Bucklin, say, as a way to avoid some runoffs, then, makes complete 
sense. Then the methods for choosing the runoff candidates can be 
improved. (It could start top two, perhaps).

Or simply Counting All the Votes will make some improvement, and 
might avoid a few runoffs. Because Bucklin does allow multiple 
approvals while preserving a level of Later No Harm *protection* -- 
not avoidance of all possibility of "harm" -- Bucklin will encourage 
the addition of additional approvals over raw Approval in a top two 
runoff context, avoiding more runoffs.

>Besides, anything using the CW almost surely violates FBC.

Technical FBC compliance is not an essential characteristic of voting 
systems under all conditions.

Condorcet *failure* is far more serious. It means that the electorate 
has rejected a winner, compared to another candidate. That, to me, 
means that the election should not complete, for the other candidate.

Consider a Range ballot. As we know, Range is not Condorcet 
compliant. We sometimes imagine that Range is theoretically optimal. 
It is not, because of the normalization problem. Suppose we assume 
that voters vote Borda style in Range, because it fits their real 
preferences. One voter's A>B>C is not generally equivalent in real 
preference strength (i.e, utility distance) to that of another. To be 
optimal, Range ballots must allow absolute utilities, which is not 
possible except with devices like a Clarke tax. So even setting aside 
strategic voting, which cannot be avoided, normalization can cause 
error. Yet awarding an election to a Condorcet loser ignores a basic 
principle of democracy, majority rule (which is best understood with 
narrow Yes/No votes, but, next to that, simple choice between two 
alternatives).

In a runoff system, though, the runoff tests the reality of the 
apparent preference as well as testing absolute utilities of the 
voters. If voters absolutely have high preference, they will turn out 
to vote. And if it is low, they will not. Thus the results are pushed 
in the direction of absolute utility maximization.

Usually the Range winner will be the Condorcet winner. When ballot 
analysis shows, instead, that by raw preference, some candidate beats 
the Range winner, it makes sense, then, to present the two in a runoff.

In this discussion, I've come across, from prior comments of others, 
the issue of clones. Voters voting to place two clones in a runoff is 
a possible phenomenon that bears further examination. I know of no 
real-world examples, except, of course, that someone with preferences 
far from normal may think that runoffs are always doing this!

(Indeed, that was the express view of Nader in 2000, that Gore and 
Bush were Tweedledum and Tweedledee, and seems to be your view, Mike. 
So I really wonder, then, would Nader supporters have turned out in a 
runoff between Gore and Bush? However, Gore and Bush making it into a 
runoff would not be the result of voters voting for both, the 
strategy suggested with real clones! If Nader could make it into the 
runoff, he might win. The trick for a Nader>Gore>Bush voter might be 
making sure that a runoff was between Gore and Nader, if possible. 
And that could require voting for both in the primary. If the primary 
is Bucklin, that would mean voting Nader>Gore or maybe Nader=Gore. 
This maximizes the possibility of both candidates being in the 
runoff, unless it causes both to gain a majority.)