[EM] Forest--Your method proposal

[Michael Ossipoff]

Forest--

The use of those checkmarks to inform automatic strategy is a shortcut to 
DSV's way of determining each voter's best strategy.

It was you who first suggested that it's in a voter's best interest to 
approve every candidate who is better than that voter's expectation in the 
election. Several of us agree, including me.

Automatically having each ballot approve every candidate that it ranks above 
the most likely winner is a rough approximation to that, when we let the 
merit of the most likely winner stand in for your expectation.

But then wouldn't it be better to use a simplified DSV that automatically 
has everyone approve everyone they like better than the winner of the 
previous count? Wouldn't the previous count's winner be a better estimate of 
the most likely winner?

That would be easier to explain to people than would Hoffmann's or Crannor's 
calculation of the pairwise tie probabilities based on previous elections. I 
don't know how much merit difference there'd be between these different 
automatic strategy methods for DSV.

One suggestion for your method and for DSV: For public acceptance, it's got 
to be optional. The voter must have the option to manually place his/her 
Approval votes, instead of letting DSV or any other automatic process do so 
for him/her.

Some years ago I looked at a method that, based on everyone's rankings, 
elects the candidate who'd win at Nash equilibrium in (say) an Approval 
vote. From the examples I looked at, it seemed as if that method had 
properties very similiar to those of wv Condorcet. And, though DSV was more 
difficult to examine in that way, my impression was that DSV is similiar in 
that regard. If these attempts at idealness--DSV and Nash Equlibrium 
Selection--have the properties of wv, then that gives strong support to wv, 
as the best that one can do, for strategy-freeness.

I suspect that wv can't be beat for strategy freeness by any 
nonprobabilistic method. Of course wv, and other methods with similar 
properties, such as MDDA and MAMPO, can have their strategy-freeness greatly 
enhanced by add-on options such as ARLO.

Jobst suggested a method, or a class of methods, that matches wv, by trading 
Condorcet's Criterion for FBC. With some electorates that would be better, 
but the important thing is that both kinds of methods are equally good; it 
simply depends on which of those two criteria you choose. Timid, 
overcompromise-prone voters need FBC more than Condorcet’s Criterion.

One strategy improvement over wv would be to disqualify majority-beaten 
candidates, even if it means disqualifying everyone. That would guarantee 
that if a majority prefer X to Y and vote sincerely, Y can't win--without 
all the stipulations of SFC and GSFC. But of course it would be 
prohibitively indecisive. So that method demonstrates a limit on how 
strategy-free a method can be.

The subject of how good a nonprobabilistic method can be, in terms of 
strategy-freeness is an interesting one. Those are my impressions about it.

Mike Ossipoff