[EM] "STAR voting..." paper by Wolk, Quinn & Ogren

[Kristofer Munsterhjelm]

On 2024-01-08 09:25, C.Benham wrote:
> Complete garbage.
> 
> With STAR  Push-over strategy is more tempting and much less risky than 
> it is with IRV.   Suppose you are sure that your favourite (or at least 
> a candidate you find acceptable) will make
> the final but are fearful that the other finalist will be able to 
> pairwise beat that candidate.
> 
> In that case you obviously have incentive to give at least 4 stars to 
> all the candidates you think would pairwise lose to the candidate you like.
> 
> And what is supposed to be the excuse for failing Condorcet?
> 
> Let me suggest some good methods that use the same ballot information as 
> STAR.
> 
> Smith//Approval (more than 1 star)
> 
> Condorcet//Approval (more than 1 star)
> 
> Smith//Approval (above average rating of Smith-set members)
> 
> Approval (above average rating of remaining candidates) Elimination
> 
> The last two might be too complicated and the last one probably fails 
> mono-raise, but they are all vastly better than the nightmare festival 
> of Push-over and Compromise
> that is STAR.

Presumably Smith-Range (eliminate non-Smith then renormalize everybody's 
ballots and do Range) could work, although it may not be quite as 
chicken-proof.

Or this idea I've occasionally mentioned on list, "taking vNM utilities 
more seriously":

Let A >= B if in every three-candidate normalized L2-cumulative 
sub-election (A, x, B) of this election, A is not the loser. 
(Alternatively "A is the winner")
Include pseudo-sets (A, A, B) and (A, B, B) where the given candidate is 
cloned, to not fail clone independence outright.

Determine the Smith set based on the relation. Then do some Condorcet 
method restricted to this set, or this set comma some Condorcet method.

-km