[EM] [CES #8848] Re: MAV on electowiki

[Abd ul-Rahman Lomax]

At 12:22 AM 6/21/2013, Jameson Quinn wrote:

>  Classic approval strategy suggests approving all candidates above 
> the expected election value. We've seen that advised again and again.
>
>
>Right. Which is not necessarily the same as the average of the best 
>and worst candidates, even in the zero-knowledge case.

Of course it is not necessarily the average. However, that's a start, 
a ready approximation. I'll disagree, though, about the 
zero-knowledge case, if it is actually zero-knowledge. If the 
election of any candidate is equally likely, then the election 
expectation will be mean utility. It's correct that this is not 
necessarily midrange, but there is an assumption that Jameson is 
making about where the extremes are. The reasonable *first setting* 
of the voting range with the internal utility range would be to set 
midrange, the approval cutoff, at the expected election utility. Then 
one categorized candidates into above-expectation and 
below-expectation. And spreads them out accordingly.

I have suggested that the "meaning" of an approval in a runoff system 
is "I prefer the election of this candidate to the decision being 
postponed." That is a comparison of absolute utilities. That sensibly 
is made in consideration of election expectation.

The system I've proposed for considering Bucklin utilities, 
interpreted from votes, is balanced, with one exception: the midrange 
vote is considered approval. More technically, it is a stand-aside, 
an acceptance of that result.

>That's a flaw in your logic.

Logic smodjic.

>  But it's a minor one. The bigger problem is that you are stuck 
> inside your logic, and simply cannot see things another way.

Jameson, you have only a little clue as to what I see. I see far more 
than I express.

>Ask a normal person if grade D should be allowed; they'll say "why not?".

Now ask them if they want to choose a physician who got a median 
rating of D from those who reviewed his record.

>  Ask them what numbers ABCDF should correspond to; they'll say "43210".

Got that. Hence C is midrange. And C is supposed to mean "average." B 
above average, and A is excellent. Electing an "average" candidate is 
the *worst* we should do.

>  Ask them if there should be special rules to make D an unapproved 
> rank; they'll say "huh? You're confusing me."

No, I'm claiming that it naturally *is* an unapproved rank, with what 
'approval' means in approval voting. But voters don't think of grades 
as ranks. They don't think of grades as weighted votes, except toward GPA.

>If you really want to analyze this as a Score ballot, I agree, the 
>approved-unapproved gap between D and F should be larger than other 
>gaps. Fine; for that analysis, which most people will not care 
>about, make F equal -2.

And then you just expanded the system, making it more complex. No, D 
is an *unapproved rating*. That is *simple.* It represents something 
a "little better than failure." That's how people think about Ds.

Other things being equal, and with multiple majority-approved 
candidates -- D not being approval -- a voter having voted D for a 
candidate could elect the candidate over the voter having voted F. 
That's all. It's a slight improvement in social utility, as measured 
by the votes. That voter is *not* pleased, but is not as displeased 
as the voter might have been if another candidate the voter rated F 
was elected.

D means "below election expectation." That's a common standard for 
approval, my point.

>  Do not put extra rules into the system, making it seem more 
> complex. People will rightly suspect that if you have to make 
> things complex, you're either hiding something, or dealing with a 
> delicate and finicky system that needs careful fine-tuning, or both.

There isn't an extra rule. I'm suggesting that an initial system not 
even have a D rating. I don't like the use of the grade levels, in 
general, because of confusion introduced that could readily lead 
voters to disempower themselves.

Or *if* the D rating is used, use it intelligently. Use it for 
pairwise analysis or score summing. Do *not* pretend that it is an 
approval. It is not.


>Jameson
>
>It is *offensive* to disregard the additional approvals after counting them.
>
>
>But you're not disregarding them, any more than you are disregarding 
>the first round results when you run a runoff.

First round result in true deliberative process *are* disregarded, 
the become completely moot. In political runoff elections, first 
round results nominate candidates to appear on the runoff ballot, but 
in many systems, do not prevent other candidates from winning as 
write-ins. The first round results, beyond ballot placement, become moot.

Here we have round results showing that more voters approve of A than 
of B. But because there is a multiple majority, the method backs up 
and disregards that round. There is only an anomaly if we assume that 
the B voters didn't wish for A to win, but mistakenly added approvals 
for A, perhaps thinking that C could win. That's possible, and is why 
I'd suggest results like these might be appropriate for runoff. But 
it's also *unlikely*, in general. If significantly more voters 
approve of A, than of B, then A is very likely to be the SU winner.

Backing up, those votes for A are not considered, if B is allowed to 
win because of earlier round results.

>Getting a majority is necessary in order to be eligible to win.

True, but A *got more votes than B* but it's proposed that B win, 
because of first round results. That A got a majority is then *moot*. B wins.

That is *possibly* a better result, but *likely* not.

>  And as to *offensive*: you're stating an idiosyncratic judgment as 
> a universal fact.

There are lots of voting system characteristics that I propose as 
offensive to this or that sensibility. "Offensive," is, by 
definition, not a fact, not a universal. It's a judgment, an 
intepretation. I was stating my opinion, and that of those who agree 
with the problem of disregarding cast and canvassed votes. I find 
this aspect of Bucklin *offensive,* that there are cast votes that 
may not be counted. *At least,* I suggest, they should be counted. If 
people go to the trouble of casting them, they should always be counted.

What's specifically offensive here is a superior majority being 
distregarded for an inferior one, based on ... what?

I *agree* that this creates a certain value for how people might 
vote, it may encourage additional approvals, but I do *not* consider 
that a necessary good. If it's necessary to get the system 
implemented, okay, I could accept that. But we are not at that stage, 
we are considering, to remind everyone, a reasonably simple next step 
in voting system reform beyond Approval Voting.

Bucklin is an Approval system. It *is* Approval Voting and commonly 
reduces to Approval in the canvassing. It clearly simulates a series 
of repeated approval elections with descending approval cutoff, thus 
simulating a normal compromise process. To do this, it uses a limited 
Range ballot. It was never thought of that way, but ... they *did*, 
in Oklahoma, attempt to set up a fractional vote system.

Basic Bucklin is simple to vote, to understand, to canvass, and is 
probably a bit better than Warren's simulations showed, because I 
think he assumed that Bucklin was a ranked method and did not treat 
the Bucklin ballot as a range ballot, thus losing the subtlety of 
expression. So he artificially elevated poor second and third choices.

>  No voting system is perfect, and if you start getting *offended* 
> by any perceived flaw, you're probably forgetting how bad plurality is.

I'm not offended, I stated that something was "offensive," which is 
to a principle I commonly assert and stand for, Count All the Votes.

That principle does not absolutely require that all votes actually 
affect the result, but we should realize that we already have 
approval voting in public elections, with multiple ballot questions. 
The tradition is clear: if there are multiple conflicting ballot 
questions, and both pass, the one with the most votes prevails. Now, 
that's *simple.* I'll agree that it's not necessarily *optimal.* But 
that is probably the best that could be done with the data available.

Voters, on a Bucklin ballot, have expressed preferences. They *did* 
express a preference between A and B, some of them. Some did not. 
Some may have voted for both A and B in the multiple-majority round. 
Some may have indicated a preference at a lower round, not yet canvassed.

MAV takes a simplistic approach that neglects preference strength 
considerations (doesn't use the expressed utilities) and that 
disregards votes in the current round, only looking above. I'm seeing 
a need to examine this much more specifically. *How much* of a 
compromise are we making with optimizing results, in order to avoid 
the so-called "chicken dilemma."