[EM] DYN

[Abd ul-Rahman Lomax]

At 07:51 PM 4/4/2010, fsimmons at pcc.edu wrote:

>I waould like to advertise Delegable Yes No (DYN) voting again.  It 
>overcomes the main difficulty with Approval, which is that the 
>common voter will will not feel sure about approving or disapproving 
>some of the candidates.

Well, that's not the main difficulty with approval, that's the main 
difficulty with *elections*, as Dodgson noted in abou 1884. The Asset 
Voting solution, which this is a restricted version of, is the only 
solution I've seen that doesn't involve going back to the voters.

>Under DYN you check Y next to all of the candidates that you are 
>sure that you want to approve (like your favorite), you check N next 
>to all of the candidates that you are sure that you disapprove (like 
>the candidate you most detest), you leave blank the choices for the 
>candidates you are not sure about, and you circle the name of the 
>approved candidate that you want to delegate the remaining Y/N decisions to.

Or there is a separate check box for that.

Can we imagine someone who wants to delegate the remaining decisions 
to, but who doesn't approve the candidate for the office? Interesting 
conditions that would take. "I trust this person to choose who will 
make the decisions, but not to make the decisions." In the office, 
the person can get advice. In a sensible system, they could name a 
proxy to actually make decisions, and, in any case, many legislators 
and office-holders do functionally delegate much or most work, even 
though they still have to sign off on it.

>After the sure Y/N votes are counted, the candidates make their 
>proxy votes for their supporters on the delegated choices.  The 
>partial talleys already available to them help to counteract 
>disinformation from the media.

The candidates, generally, will be far more informed about the other 
candidates than the media. Problem is, the media can influence those 
No votes. In other words, the voters, under media influence, may 
restrict their trusted candidate, not allowing him or her to make the 
best decision in their interest. I prefer pure Asset Voting, and I 
prefer that it be used for accurate proportional representation of 
the voters' trust, and that elections, then, be done as parliamentary 
process, for terms at the pleasure of the assembly. Majority rule.

Asset Voting can, in fact, function similarly to Yes/No voting, if we 
understand that any voter who does not want to fully delegate voting 
power can become a "candidate" and vote for himself or herself, thus 
obtaining the right to vote in the final processes.

I call the set of those who hold votes to distribute or use in a 
determinative asset election the "electoral college," and, because 
they are public voters, they can be enabled to actually vote in the 
resulting assembly. Asset can become far more than a voting system, a 
method of election. It can become a hybrid representative/direct 
democracy. The restriction necessary because of the scale is that 
participation in *deliberation* in the assembly is restritcted to 
those who hold seats. In a sense, then, for most electors the issue 
will become the choice of an effective proxy, with the ability 
remaining to effectively cancel the vote of the proxy by voting 
directly. My sense is that direct voting would be relatively rare, 
once the enormous freedom of being able to choose representation 
rather than elect it is realized.

>This is one way to solve the problem
>
>49 C
>
>26 A>B
>
>25 B>A
>
>Where the 25 threaten to bullet B.

And why would they do that? And, indeed, what is the best outcome of 
this election? Without some idea of preference strengths, it's not 
really visible. It is obvious that we have a 51:49 majority, very 
thin, who prefer either of A or B to C. The Condorcet winner is A. 
But this could easily be a situation where the Condorcet Criterion 
violates maximized social utility.

Suppose these are the absolute utilities, but normalized to the 
candidate set over all voters. (Not individually). Define utility as 
the *actual* pleasure/pain response of each voter to the election of 
a candidate, as viewed by our omniscient observer. If election had a 
dollar value to voters, and voters had equal ability to pay, the 
difference between the utility of two candidates would be what the 
voter would consider break-even (any lower cost would be a bargain, 
any higher cost would be a loss).

49: A 00, B 40, C 100 (These utilities for C and A define the scale)
26: A 70, B 60, C 50
25: A 60, B 70, C 50

Just for fun, consider that 50 is, in this case, indifference, the 
candidate causes neither pain nor pleasure.

mean: A 33.2, B 52.7, C 74.5.

Does this predict the votes shown? Yes. The C voters do not approve 
of either A or B. In Bucklin, this is how they might vote. Except 
that it's a naive vote, if this is single-round deterministic. I'll 
get to that. They would also circle C.

The A and B voters have the preferences indicated, and, because of 
the restricted choices, are not likely to vote for C, even though 
they approve of C (minimally, I include indifference under absolute approval).

The A and B voters do not have strong preference. They, in fact, 
would accept the election of C. It is clear from the utilities 
assumed that the best winner is the candidate who would win under 
Plurality. This is why I suggest, strongly, that voting systems 
theorists start to work with utility profiles, because preference 
order tells us way too little to make sense of voter behavior, and we 
cannot understand the value of an election without that information. 
The Condorcet and Majority criteria are deficient, understood naively.

In a real election, if the profile is of the overall eligible 
electorate, and if this is a special election, not mixed with other 
elections, the C voters would turn out to vote preferentially, thus 
electing C straightaway. Thus we can also see how comupulsory voting 
prevents an important function of voting process, the test of 
preference strength that allows simple voting systems to perform much 
better than preference analysis would indicate. In real deliberative 
process, when voters don't have significant preference, they abstain, 
as long as either outcome is acceptable to them.

With DYN, it is possible that the A and B voters will not vote No on 
A. After all, they don't have strong feelings *against* this 
candidate. The C voters, however, will definitely vote No on A, and 
might vote No on B.

Suppose they vote No on A, but the A and B voters do not vote No on 
C. This leaves us with 49 % for C, 26% for A, and 25% for B. B will 
win, because at this point there are three voters (A, B, and C), and 
they each know the voting power, and C is constrained to vote either 
for self or for B. A runoff would be between C and A. C would win a 
runoff, I'd predict from the utilities, because of differential turnout.

Unstated was how the proxies would vote, and using what method.

Let me look at this election as Bucklin. I'd suggest that this would 
be the likely Bucklin ballots:

49: C
26: A>B>?
25: B>A>?

If this is a majority-required election, i.e., Bucklin is used as a 
primary with a runoff if needed, some percentage of the A and B 
voters will rationally add a third preference for C, because they 
would prefer or be indifferent to the election of C to the cost of a 
runoff. If more than 1% of them do this, C wins. Otherwise there is a 
runoff between C and A, I'd assume. A is the Condorcet winner, so, by 
the runoff rules I've proposed, there is no exception there. The 
runoff tests preference strength; that's why C wins. And that is the 
utility-maximizing outcome, minimizing regret, and, in fact, given 
the utility profile, nobody regrets the election at all.