[EM] Conditional YES votes

[Richard Moore]

DEMOREP1 at aol.com wrote:

> A later place YES vote might be deemed conditional -- i.e. so as NOT to 
> injure an earlier place YES vote that is not paired equally.    
> 
> Example-
> 
>    (* Yes conditional vote)
> 
>  2  A B*
> 24 A B
> 24 B A
> 49 C


A similar system could be obtained by modifying IRV to allow voting
of tied preferences. The above ballots could then be cast as:

2 
A > B > C
24 
A=B > C
24 
B=A > C
49 
C > A=B

In each case, the conditional B voters are expressing their desire for
their vote for B to count only if A does not win. Counting is like
"conventional" IRV, except that the elimination doesn't stop when one
candidate gets more than 50% of the ballots; it must continue until
all candidates but one are eliminated. That is because, with tied
preferences, a 50%+1 quota is not sufficient; a block of voters might
have their votes transferred to a tied preference on the next round.

For three candidates, I think this IRV variant is equivalent to Demo's
idea, but not for four or more. The ranking in IRV is really a statement
of multi-level conditionals, while Demo's method is restricted to one level
of conditional ("If A loses, then make mine a B vote").


> The above is why simple Approval is defective -- i.e. it does not rank the 
> choices.


What Demo sees as a defect, I see as a reasonable tradeoff. When methods try
to achieve more expressivity than Approval, it always seems to be at the
cost of bringing in worse defects, such as nonmonotonicity/inconsistency.

I haven't looked into how this method, or the IRV variant, will do along
those lines, though I suspect they will fail as IRV does. Also, as in IRV,
we would be dealing in manufactured majorities. One thing that is obtained
in the trade, besides greater expressivity, is a simplification of strategy.
For example, in a replay of Gore v. Bush v. Nader, a Green voter (whose
preferences are probably "Nader >> Gore > Bush") could vote "Nader > Gore
 > Bush" in the modified IRV (and have his/her votes treated just as they
would be in conventional IRV), or "Nader Gore*" in Demo's scheme, which
would receive the same treatment. There's no agonizing over whether to
vote Nader+Gore or just Nader in ordinary Approval. That doesn't mean
I like this method better than ordinary Approval -- it needs more
study.

One more potential advantage, from the standpoint of EM politics (or
polemics?), is that the IRV folks would be free to always vote an IRV-style
ballot, and the Approval folks would be free to always vote an 
Approval-style
ballot. Unfortunately, there's no pairwise treatment, so the Condorcet
folks are left out. And to echo Mike O., the IRV folks never seem to be
willing to accept a compromise method anyway.

Just based on gut feelings, I think that, for four or more candidates, 
Demo's
idea is better than the modified IRV. Since the latter attempts to achieve
a high level of expressivity, I suspect it will fail monotonicity and
consistency more frequently and/or with worse consequences than the former
method, particularly when the number of candidates is large. I think either
method would be better than conventional IRV.


 -- Richard