[EM] National Popular Vote & Condorcet

[Raph Frank]

On Thu, Jul 2, 2009 at 10:03 PM, Dave Ketchum<davek at clarityconnect.com> wrote:
> Each state controls how it interacts with its voters - so let them choose
> their own way, such that their voters' desires get properly added into the
> national X*X array.

This depends, there needs to be rules on what states can do.

For example, they could create  the array so that it is

State winner>(others)

for all voters in the state.

In effect, this returns things to the current, pre-NPV, system.

>> Plurality:
>> A vote for candidate A is considered
>> A>(others)
>
> This reads as giving the same power as if ranking ONE candidate in Condorcet
> - simple and declarably accurate.

It also means that only the expected national top-2 can get votes from
this state.

Ofc, this isn't as extreme as currently, and if other states support
more open methods, at least candidates can gain publicity in one
election for a challenge on the next.

>> Condorcet:
>> Matrix is provided directly
>>
>> IRV:
>
> Here the voters could have ranked exactly as in Condorcet, but standard IRV
> counting does not extract all that the voters say.  I would leave it to the
> state - perhaps they will do an X*X matrix.  I do not like what I read below
> - better for such states to avoid such as IRV when  they do not fit with
> what is reasonably the standard.

Right, if they collect ranked ballots, it would be best if they
publish the full results.

However you need to somehow handle the case where states use IRV.

For example, the State might have a rule that all EC votes for the
State are assigned to the IRV winner in the state, so they don't
publish ballot info.

>> Approval/Range
>
> For approval my first thought is that they are presumably doing approval and
> my first choice for them is whatever Condorcet states do when their voters
> vote with approval thinking.

Again, the States may not publish the date required.

I would agree that a voter who approved candidate A and B should
ideally be considered

A=B>(others).

However, you can't extract that info from the approval results.

> For Range the thinking is much as I do above for IRV.

Again, if the full ballot info is released, it would be worth
converting the range votes into condorcet, but there is a need to
handle things if the State doesn't provide the info.

It occurs to me that you could just include rules for comparisons
rather than trying to work out the votes for approval.

Approval

A: 800
B: 400
C: 300
Total 1500

Comparing A and B
800-400

This means that the votes are assumed to be
200: A=B
600: A>B
200: B

A>B: 600-200

Comparing B and C
400-300

B>C: 400-300

Comparing A and C
100: A=C
700: A
200: C

A>C: 700-200

Thus the rule when determining the pairwise comparisons is to assume
that the number of equality votes are minimum.

This may not even effect the result depending on completion method used.

For example:

A: 800
C: 300

must be
A=C: X
A: 800-X
B: 300-X

Thus the win margin is (800-X) - (300-X) = 500.  As long as the win
margin is all that matters we can determine the exact national result
without knowing exactly the number of ballots where there is a tie.

Thus, approval, condorcet (if matrix is provided) and plurality
results can be converted to an exact matrix (or equivalent).

IRV cannot be fully supported, and ballot info is lost.