Re: [EM] STAR Voting Variations
... continuing ...
Here's another small tweak to create a similar, but much improved method ... let's call this tweak of STAR, "CHAMP."
STAR elects the pairwise winner between the score winner X and the candidate Y who is the score winner among the other candidates (besides X).
CHAMP elects the pairwise winner between the score winner X and the candidate Z who is CHAMP winner among the other candidates (besides X).
Both methods pit the score winner against a runner-up from among the other candidates (besides the score winner).
In STAR the method for choosing the runnerup is score (restricted to the other candidates) which simply yields Y, the candidate with the second highest score over-all.
In CHAMP the method for choosing the runner-up is CHAMP itself (restricted to the other candidates).
[At first glance this looks like a case of circular definition, but in mathematics definitions of this form can (under certain conditions satisfied in this context) be considered as logically convergent spirals called "recursions," rather than circles.]
It turns out that this description of CHAMP is just a particularly compact [and elegant, in the eyes of software engineers] reformulation of Sequential Pairwise Elimination (SPE) applied to the "agenda" of candidates listed from least to greatest scores.
SPE is a highly respected method that has been used in many kinds of deliberative bodies for centuries, and recommended by "Robert's Rules of Order" for choosing from among more than two options ... two-at-a time from an agenda with the least promising agenda items listed first. In the CHAMP context "least promising" means lowest score.
CHAMP is clone proof, monotonic, and Condorcet compliant ... what simpler method has this much going for it?
... to be continued ...
Sent from my MetroPCS 4G LTE Android Device
-------- Original message --------
From: Susan Simmons suzerainsimmons@outlook.com
Date: 7/10/21 5:55 PM (GMT-08:00)
Cc: election-methods@lists.electorama.com
Subject: STAR Voting Variations
STAR is a great gateway method because there are so many different directions to tweak/elaborate/improve/generalize it to transform it into other simple methods based on similar ideas.
For example, Majority Judgement Top Two Runoff, that uses the exact same score ballots, voted exactly the same.
While STAR pits against each other the two candidates with the greatest average ("mean") scores, MJTTR compairs the two with the greatest median scores.
Why median instead of mean? Because use of the median gives less strategic incentive for voting closer to the extremes, and the un-exagerated ratings of MJ voters makes the pairwise comparisons more accurate due to fewer strategy induced tied ballot ratings.
Majority Judgement is a score based version of Bucklin that has an ingenious way of breaking the median ties that are extremely likely, as anyone used to dealing with medians will understand.
Why do medians discourage exageration? Suppose the median score for candidate X is three, then changing any ballot rating above three to another one above three would make no difference ... same for ratings below three: the median is less rewarding to exagerating manipulators.
Experiments by the MJ inventors seem to support the idea that voters under MJ experience little to no temptation for exagerating their best honest estimates for the deserved ratings of the candidates.
That said, there is another metric called "chiastic approval" that is a very nice compromise between mean and median ballot ratings ... which we will come back to in the future, after considering more of the low hanging fruit in another direction ... (to be continued)
Sent from my MetroPCS 4G LTE Android Device
-------- Original message --------
From: Susan Simmons suzerainsimmons@outlook.com
Date: 7/8/21 9:04 PM (GMT-08:00)
To: Daniel Carrera dcarrera@gmail.com
Cc: election-methods@lists.electorama.com
Subject: Re: [EM] STAR Voting
BTW its not just the clone thing that makes me think that STAR is strategically equivalent to Approval, rather it is the combination of these two facts:
(1) Score Voting is strategically equivalent to Approval ... optimal strategy entails voting at the extremes (mostly if not exclusively).
(2) If voters vote only at the extremes, the Score winner will be the pairwise winner of every head-to-head contest.
So in the case of sophisticated voters the top two runoff is redundant.
For naive voters it is harmless and may serve some valuable psychological purpose.
On Mon, Jul 12, 2021 at 8:41 PM Susan Simmons suzerainsimmons@outlook.com
wrote:
... continuing ...
Here's another small tweak to create a similar, but much improved method
... let's call this tweak of STAR, "CHAMP."
STAR elects the pairwise winner between the score winner X and the
candidate Y who is the score winner among the other candidates (besides X).
CHAMP elects the pairwise winner between the score winner X and the
candidate Z who is CHAMP winner among the other candidates (besides X).
It took me a while to figure out what you were saying. So... a recursive
definition of CHAMP... Btw, what does CHAMP stand for (if anything)? Let me
see if I understand. If I wanted to compute the CHAMP winner I would:
- Sort the candidates by score.
- Pick the pairwise winner "X" between the bottom two candidates.
- Pick the pairwise winner between "X" and the next candidate up.
- Repeat.
So CHAMP Is kind of like BTR-STV but you are sorting by score instead of
first-choice votes? CHAMP = BTR-Score ? Or have I completely misunderstood?
It turns out that this description of CHAMP is just a particularly compact
[and elegant, in the eyes of software engineers] reformulation of
Sequential Pairwise Elimination (SPE) applied to the "agenda" of candidates
listed from least to greatest scores.
SPE is a highly respected method that has been used in many kinds of
deliberative bodies for centuries, and recommended by "Robert's Rules of
Order" for choosing from among more than two options ... two-at-a time from
an agenda with the least promising agenda items listed first. In the CHAMP
context "least promising" means lowest score.
CHAMP is clone proof, monotonic, and Condorcet compliant ... what simpler
method has this much going for it?
A recursive method will be tough for most people. I studied Lisp and it was
a bit of a curve ball for me. I think an SPE definition might be easier to
follow even if it is mathematically inelegant.
If I was trying to sell this to the STAR team, I would tell them that they
can even keep the same name they have.
STAR = Score Then Automatic Runoff
There is nothing in that name that requires only one runoff. You could
repurpose the name to mean that you Score candidates and then run N runoffs
from the bottom up. Boom! Now you've turned STAR into a clone proof,
monotonic, Condorcet method and hopefully created a method that fans of
Score and fans of Condorcet can both get behind.
Am I right to think that your method, and any other BTR-* method is also
Smith-efficient?
Cheers,
Daniel
SPE, not BTR is Ihe correct (and customary) terminology for this kind of elimination, because the agenda (score order) is set before any of the eliminations take place.
BTR, as in IRV-BTR, is a procedure where there is no "agenda" list .. the order (after the first elimination) is determined on the fly after the eliminations have already begun (which is why BTR methods fail monotonicity), not before the start of the run-off.
The recursive definition was never intended for the public, but only for EM list subscribers and lurkers, to show them that a change of one word ... score to CHAMP (or STAR if you prefer that name) ... could radically improve STAR, making it ISDA compliant while retaining the monotonicity of Score. It just happens that the one word change makes the definition recursive ... a concept that any EM List reader (or any other educated person) should not be afraid to learn about before their brain fossilizes ... and what better context for EM list readers than a message on that list?
To be clear, the intended audience for my messages to the EM list is the subscribers and curious lurkers for the list, not the general public. Like other older EM contributors, I'm trying to help interested new-comers get a better understanding of the various election methods (both good and bad) and the useful principles for comparing them, by walking them through some of the most salient comparisons. Like the oldest EM members, I continue to learn ... including from young people with their fresh perspectives and interesting questions.
Sent from my MetroPCS 4G LTE Android Device
-------- Original message --------
From: Daniel Carrera dcarrera@gmail.com
Date: 7/12/21 7:49 PM (GMT-08:00)
To: Susan Simmons suzerainsimmons@outlook.com
Cc: election-methods@lists.electorama.com
Subject: Re: [EM] STAR Voting Variations
On Mon, Jul 12, 2021 at 8:41 PM Susan Simmons <suzerainsimmons@outlook.commailto:suzerainsimmons@outlook.com> wrote:
... continuing ...
Here's another small tweak to create a similar, but much improved method ... let's call this tweak of STAR, "CHAMP."
STAR elects the pairwise winner between the score winner X and the candidate Y who is the score winner among the other candidates (besides X).
CHAMP elects the pairwise winner between the score winner X and the candidate Z who is CHAMP winner among the other candidates (besides X).
It took me a while to figure out what you were saying. So... a recursive definition of CHAMP... Btw, what does CHAMP stand for (if anything)? Let me see if I understand. If I wanted to compute the CHAMP winner I would:
- Sort the candidates by score.
- Pick the pairwise winner "X" between the bottom two candidates.
- Pick the pairwise winner between "X" and the next candidate up.
- Repeat.
So CHAMP Is kind of like BTR-STV but you are sorting by score instead of first-choice votes? CHAMP = BTR-Score ? Or have I completely misunderstood?
It turns out that this description of CHAMP is just a particularly compact [and elegant, in the eyes of software engineers] reformulation of Sequential Pairwise Elimination (SPE) applied to the "agenda" of candidates listed from least to greatest scores.
SPE is a highly respected method that has been used in many kinds of deliberative bodies for centuries, and recommended by "Robert's Rules of Order" for choosing from among more than two options ... two-at-a time from an agenda with the least promising agenda items listed first. In the CHAMP context "least promising" means lowest score.
CHAMP is clone proof, monotonic, and Condorcet compliant ... what simpler method has this much going for it?
A recursive method will be tough for most people. I studied Lisp and it was a bit of a curve ball for me. I think an SPE definition might be easier to follow even if it is mathematically inelegant.
If I was trying to sell this to the STAR team, I would tell them that they can even keep the same name they have.
STAR = Score Then Automatic Runoff
There is nothing in that name that requires only one runoff. You could repurpose the name to mean that you Score candidates and then run N runoffs from the bottom up. Boom! Now you've turned STAR into a clone proof, monotonic, Condorcet method and hopefully created a method that fans of Score and fans of Condorcet can both get behind.
Am I right to think that your method, and any other BTR-* method is also Smith-efficient?
Cheers,
Daniel