[EM] Hare (aka IRV) versus STAR

[robert bristow-johnson]

> On 04/13/2024 3:37 PM EDT Closed Limelike Curves <closed.limelike.curves at gmail.com> wrote:
> 
> 
> Are the ballots in your example supposed to be honest or strategic?

Should it make any difference?

 ---- “My system is only intended for honest men.” - Jean-Charles de Borda

In a political election with high stakes, why wouldn't voters mark their ballots in such a way to maximize their political interest?  The purpose of these election reforms to FPTP is to disincentivize tactical or strategic voting.  A method should *not* depend on voters voting honestly in the secrecy of the voting booth if they believe that insincere voting will further their political interests more than the sincere vote.  The method should expect voters to vote in such a way that these voters would expect to promote their own political interest.  That's why Borda Count is so vulnerable to Burial and Borda (and his critics, including Condorcet) knew it.
____________

The ballots are derived from the Burlington 2009 Mayoral election which had a Condorcet winner that was not elected using IRV.  It was scaled from circa 8900 ballots to 100.  The preferences between the three most significant candidates are:

 - 34 Left voters:
   --- 23 ballots: L>C>R
   ---  4 ballots: L>R>C
   ---  7 ballots: L only
   
   - 29 Center voters:
   --- 15 ballots: C>L>R
   ---  9 ballots: C>R>L
   ---  5 ballots: C only
   
   - 37 Right voters:
   --- 17 ballots: R>C>L
   ---  5 ballots: R>L>C
   --- 15 ballots: R only


The rationale to assigning scores from this preference data is based on sane voters that understand how STAR works.  These voters want their 1st-choice candidate to win and they want their 2nd-choice candidate (if they have one) to defeat their least-favorite candidate in the contingency that their 1st-choice candidate does not win.

So, given any partisan voter with a first-choice, A, and second-choice, B, where D is the candidate they do not want to win at all.  Their ranking is A>B>D.  So how would that voter naively best serve their political interests under STAR?

A gets 5 stars and D gets 0 stars, correct?  It's up to you to show why A and D would be scored any differently.  I think there's nothing to dispute here.

Then what does this voter do with B, their 2nd-choice candidate or, equivalently, their "lesser evil" candidate?  They don't want their 2nd-choice candidate to beat their 1st-choice, but if their 1st-choice is not going to win, that is that their 1st-choice does not make it into STAR's Automatic Runoff, then they want their 2nd-choice to beat the candidate they hate.

Now, in that case, all they need to do is score their 2nd-choice candidate incrementally higher than the score for the candidate they hate.  Scoring their 2nd-choice any higher does *nothing* to help that candidate beat their hated candidate in the final runoff.  Scoring their 2nd-choice any higher does *nothing* to help their 1st-choice beat their hated candidate in the final runoff.  But scoring their 2nd-choice higher *does* harm their 1st-choice get into the final runoff.  There is nothing naively to be gained by scoring their 2nd-choice any higher than 1 star more than their hated candidate.

____________

Indeed, for the 15 Center voters in the example above that ranked C>L>R, if six or more of them score their 2nd-choice candidate lower (that is with 0 stars), they get their candidate elected.  This burial strategy serves their interests better than if they "honestly" score their 2nd-choice higher than than their hated candidate.  But if exactly four of them do that, they get their hated candidate elected and the burial strategy backfires.

But if some of the 17 Right voters that ranked R>C>L *had* bumped their score for C higher (enough for a 6 star increase in C's score), they could prevent L from winning.  This failure mode is exactly comparable to the 2009 IRV election in Burlington.  There is no similar parallel to the Center voters strategically burying their 2nd-choice candidate with IRV because of Later-No-Harm (but the R voters *did* harm their interests by ranking R highest, since R lost).

So, with IRV, Burial doesn't serve any voter's interest but Favorite-Betrayal does.  But with STAR, *both* Burial (for the Center voters) or Favorite-Betrayal (for the Right voters) *can* serve the political interest of these voters' interest by preventing the election of the Left candidate.

--

r b-j . _ . _ . _ . _ rbj at audioimagination.com

"Imagination is more important than knowledge."

.
.
.