[EM] Clone-proof STAR (Score+Automatic Runoff)

[Ted Stern]

[Related to a topic that was discussed earlier on the Election-Methods list
--
http://election-methods.electorama.narkive.com/Vlwq75Zy/em-top-two-approval-pairwise-runoff-ttapr
]

STAR is described here: http://www.equal.vote/

It is vulnerable to crowding -- too many clones could prevent the top-two
score winners from capturing the variance-minimizing candidate in the
pairwise runoff.

One relatively simple way to deal with clones is to reweight the ballots of
voters, and then choose one of the remaining candidates in some way.

Following that approach, let me propose the following, with no claims to
efficiency. I'm just interested in whether it would be closer to Condorcet
in a Yee plot.


- Find traditional STAR winner: pick the top two scoring candidates,
then take the pairwise-preferred of the two. Call that candidate STAR1.
- Assuming a score of 0 to 5, reweight ballots by a factor of (1 - score
/ 5). In other words, if a voter scores STAR1 at a 4, their ballot is
reweighted to (1 - 4/5) = 20%. This is effectively a simplified form of
multiwinner PR reweighting to find two representatives.
- Tabulate the weighted scores on the remaining ballots, picking the top
two weighted score winners for the second round. Also tabulate the
weighted pairwise preferences. In other words, if a 60% ballot rates
candidate Herbert over candidate Franklin, Herbert receives 0.6 preference
against Franklin.
- STAR2 is the winner using the sum of new pairwise preferences.
- Finally, the winner would be the most preferred of STAR1 versus STAR2,
using the original non-reweighted preferences.

In terms of efficiency, one could skip the SW2 round if there are no other
candidates preferred to SW1. In other words, if regular STAR finds the
Condorcet winner, you're done. But if not, ballots lose as much power as
they gave toward choosing STAR1, leaving behind the non-cloned candidates.
Choosing the second round finalist using STAR again would deal with cloning
at that level.

Is this method vulnerable to pushover?

[Kristofer Munsterhjelm]

Post by Ted Stern
[Related to a topic that was discussed earlier on the Election-Methods
list --
http://election-methods.electorama.narkive.com/Vlwq75Zy/em-top-two-approval-pairwise-runoff-ttapr
]
STAR is described here: http://www.equal.vote/
It is vulnerable to crowding -- too many clones could prevent the
top-two score winners from capturing the variance-minimizing candidate
in the pairwise runoff.
One relatively simple way to deal with clones is to reweight the ballots
of voters, and then choose one of the remaining candidates in some way.

Some DSV thoughts along those lines:

The most obvious way of making crowding go away is to simply remove the
ballots of the voters who voted for the winner. But that may cause
regret because of a vote-management situation, e.g. X has total rating
100, Z has total rating 91, and Y has total rating 90; and someone who
voted for both X and Y gets his vote removed, whereas he could have
helped Y get more votes if he hadn't voted for X. So he would have
preferred to only vote for Y unless that would have made X lose.

So suppose X is the Range winner and so is chosen as one of the top two.
Then we could do a (relatively simple) DSV this way:

For each other candidate Y, maximize Y's score by turning "X and Y"
votes into "only Y" votes, subject to:
- X has to be the Range winner (i.e. top scoring candidate) after the
transformation too - or Y has to be the Range winner and X second,
- You have to remove all remaining untransformed "X and Y" votes when
counting Y's support.

The transformation would only be temporary, i.e. when considering Y's
score, use the unmodified ballots + transformation to get Y's score,
then when considering Z's score, use the unmodified ballots +
transformation to get Z's score. The candidate with the greatest
modified score is the second candidate for the runoff.

Unfortunately, this is rather complex, and simplicity is supposed to be
STAR's thing, so I don't know how useful it is. There may be some second
order/pushover strategy incentive as well, e.g. someone who wants Z to
win, where Z is not very well liked by X-voters and Y is in second place
by unmodified ratings, votes Z and Y so as to decrease the margin X has
down to Y, in an attempt to deny "X and Other" votes from going to Other.

(Using this kind of DSV is much easier with threshold-based methods
because then the threshold is fixed, so the second order strategy is
rendered moot.)
----
Election-Methods mailing list - see http://electorama.com/em for list info

[Ted Stern]

My current thinking on cloneproofing STAR is a simpler version of my
earlier post

Pick Top Two scoring finalists.

Select pairwise winner between those two. This candidate is STAR1.

Deweight ballots that voted for STAR1 by the fraction of their vote that
contributed to the score:

Score 5 (100%) is fully deweighted (X factor of (1 - 1.00 ) = 0)
Score 4 (80%) is deweighted by 80% (X factor of (1 - 0.80) = 0.2)
Score 3 (60%) is deweighted by 60% (X factor of (1 - 0.60) = 0.4)
etc.

Top scoring candidate among remaining candidates is STAR2.

Winner is pairwise winner between STAR1 and STAR2, using original pairwise
array.

Note that deweighted scores can be accumulated summably during first pass,
along with pairwise array, using similar storage requirements and number of
operations.

Post by Kristofer Munsterhjelm
The most obvious way of making crowding go away is to simply remove the

ballots of the voters who voted for the winner. But that may cause regret
because of a vote-management situation, e.g. X has total rating 100, Z has
total rating 91, and Y has total rating 90; and someone who voted for both
X and Y gets his vote removed, whereas he could have helped Y get more
votes if he hadn't voted for X. So he would have preferred to only vote for
Y unless that would have made X lose.

Case 1: X and Y are the initial top two scoring candidates.
The voter for Y shouldn't have regret if X and Y are members of the same
party, because the pairwise winner between them will become STAR1.

Case 2: X and Y share voters, but those voters do not vote for STAR1.
X & Y voters won't lose votes after STAR1 ballots are deweighted, so X and
Y scores will remain unchanged. The higher scoring of the two becomes
STAR2. This has the same voter management as in traditional Score voting.


On Tue, Jul 25, 2017 at 1:46 PM, Kristofer Munsterhjelm <

Post by Kristofer Munsterhjelm

Post by Ted Stern
[Related to a topic that was discussed earlier on the Election-Methods
list --
http://election-methods.electorama.narkive.com/Vlwq75Zy/em-
top-two-approval-pairwise-runoff-ttapr
]
STAR is described here: http://www.equal.vote/
It is vulnerable to crowding -- too many clones could prevent the
top-two score winners from capturing the variance-minimizing candidate
in the pairwise runoff.
One relatively simple way to deal with clones is to reweight the ballots
of voters, and then choose one of the remaining candidates in some way.

The most obvious way of making crowding go away is to simply remove the
ballots of the voters who voted for the winner. But that may cause regret
because of a vote-management situation, e.g. X has total rating 100, Z has
total rating 91, and Y has total rating 90; and someone who voted for both
X and Y gets his vote removed, whereas he could have helped Y get more
votes if he hadn't voted for X. So he would have preferred to only vote for
Y unless that would have made X lose.
So suppose X is the Range winner and so is chosen as one of the top two.
For each other candidate Y, maximize Y's score by turning "X and Y" votes
- X has to be the Range winner (i.e. top scoring candidate) after the
transformation too - or Y has to be the Range winner and X second,
- You have to remove all remaining untransformed "X and Y" votes when
counting Y's support.
The transformation would only be temporary, i.e. when considering Y's
score, use the unmodified ballots + transformation to get Y's score, then
when considering Z's score, use the unmodified ballots + transformation to
get Z's score. The candidate with the greatest modified score is the second
candidate for the runoff.
Unfortunately, this is rather complex, and simplicity is supposed to be
STAR's thing, so I don't know how useful it is. There may be some second
order/pushover strategy incentive as well, e.g. someone who wants Z to win,
where Z is not very well liked by X-voters and Y is in second place by
unmodified ratings, votes Z and Y so as to decrease the margin X has down
to Y, in an attempt to deny "X and Other" votes from going to Other.
(Using this kind of DSV is much easier with threshold-based methods
because then the threshold is fixed, so the second order strategy is
rendered moot.)

[robert bristow-johnson]

 
a few weeks ago, someone posted here a reference to a story about RCV in Sante Fe.  the story was in an Albuquerque newspaper but it's about Santa Fe.  i  found a current op-ed about it and wrote a response.  Rob Richie is not happy with me.  i've initially
replied to him, but need to write something else, because i actually am on board with the mission of FairVote, but *not* their efforts on IRV.  (and i was specifically unhappy when i discovered that they repackaged IRV with the label "RCV" which compounds their continuing effort to
bring election reform by implying that any form of RCV must be decided with IRV rules and hiding the very real flaws of IRV.)  i understand that this might be ill-timed because the city has voted to adopt "RCV" (in the form of IRV) 9 years ago and have not implemented it, ostensibly
due to issues of modifying the voting machines.
anyway, here is the op-ed:
http://www.santafenewmexican.com/opinion/my_view/looking-in-the-differences-in-ranked-choice-voting/article_77d7e472-6876-529b-acc9-ca7fffdcc896.html
please be gentle. :-)

--
r b-j                  ***@audioimagination.com
"Imagination is more important than knowledge."