What should STAR-PR be? SRV-PR is horribly broken (it's not even proportional) but STAR-PR doesn't have to be.

[parker friedland]

Why is SRV-PR broken? SRV-PR is not even proportional. In this forum discussion on CES, Warren included an example of when SRV-PR can fail to elect a single Democrat even when 49% of voters are Democrats and give max stars to the Democrat and min stars to the Republicans: https://groups.google.com/forum/#!searchin/electionscience/proportional$20srv$3F%7Csort:date/electionscience/aOMyxkoey4Q/KepDTrY7AwAJ

What could STAR-PR be? Well, here are some possible definitions of a STAR-PR method:

1. Harmonic voting ( http://scorevoting.net/QualityMulti.html ), except that between the outcomes with the highest quality according to harmonic voting's quality function, hold an automatic runoff between those two outcomes where everybody's vote automatically goes to the outcome out of the two that had candidates that the voter gave a higher scores on average too.

2. Normal SRV-PR, exept that each round, the number of ballots the candidate with the second highest score needs to be preferred on to beat the candidate with the highest score gets greater and greater. In the 1st round, the candidate with the 2nd highest score would only need to be preferred on 1/2 ballots to win, on the the 2nd round, they would need to be preferred on 2/3rds of ballots, on the 3rd round, 3/4th of ballots, on the forth round, 4/5th's of ballots, on the 5th round, 5/6th of ballots, etc.

3. RRV, except that the first winner is elected via STAR instead of score.

4. Duplicate each vote so you have two versions of everybody's ballot (you don't have to actually duplicate each person's ballot, but it is easier to explain it this way). We shall refer to these two parts as two different types of votes: score votes and a preference votes. One of the two ballots will be used for scores and the other will be used for the runoffs. Each round, use the score votes to pick the top 2 candidates and the preference votes to chose which of the two candidates gets elected that round. Each round, the score votes are re-weighted using the RRV re-weighting equation. However, the preference votes will not be re-weighted. Instead, a droop quota % of the preference votes that preferred the runoff winner will be exhausted (and the % of those votes chosen for exaustion will be the % that gave the highest difference in scores to each of those two candidates that made it to the runoff).

All four of these methods would be proportional (I think) while still reducing to STAR voting in the single winner case.

[parker friedland]

Which of these methods should STAR-PR be defined as?

I also posted this discussion on the EndFPTP reddit because STAR voting is becoming a popular topic there: https://www.reddit.com/r/EndFPTP/comments/8at51y/what_should_starpr_be_srvpr_is_horribly_broken/

[parker friedland]

> 3. RRV, except that the first winner is elected via STAR instead of score.

I much prefer proportional score voting (that's what I'm now calling proportional approval voting + the KP transformation) because it is more proportional, so instead of defining STAR-PR as RRV but electing the first winner through STAR, another proposal could be defining STAR-PR as PSV but electing the first winner through STAR. However, the downside to this would be that PSV is a bit more complicated then RRV. But the extra proportionality of PSV might be worth it.

On Sunday, April 8, 2018 at 2:18:20 PM UTC-7, parker friedland wrote:

[Nevin Brackett-Rozinsky]

One straightforward answer is to modify RRV so that in every round the two candidates with highest score undergo a “virtual runoff”. In the virtual runoff, consider all ballots that score one of the runoff candidate higher than the other, and count each such ballot as contributing its current weight toward the one it scores higher. Whoever wins this “weighted runoff” is awarded the next seat, and the ballots are reweighted accordingly.

Thus, every round acts like star voting with weighted ballots, and the standard RRV method for reweighting the ballots is used between rounds.

Nevin

[Phil Uhrich]

Why not just apply the weighting to the runoff round too. It wouldn't fix it if we let people score with 3 decimal places but since we stick to integers and usually max out at five that is less of a problem. Unless my math sucks (very possible) the GOP would need to have 54.54% of the voters and rate them all 1 to the rest all voting 5 before they would get the first two wins.

Also, there is a problem with the weighting formula:
" 1 / (1 + sum/max), where sum is the total of the scores you've given to winning candidates. So if you gave 5 out of 5 points to the winner of the first seat, your subsequent ballot influence would be one half (1 / (1 + 5/5)"

If the GOP gave everyone 1's and wins the first two seats after round one they get weighed down by 1/(1+(1/5)) and after round two 1/(1+(2/10)) which = 1/(1+(1/5)) So they don't lose any power for future rounds. You would probably need to change the formula to: 'weight last round / (1 + score given round winner/max score)'

With the updated weighting formula the GOP would need 59% of votes to win the first 3 spots. Beyond that it is totally implausible to imagine democrats only running 1 candidate for a 4 winner election. Not to mention highly unlikely GOP voters would be that well disciplined to only vote 1's.

[Phil Uhrich]

Actually it's pretty implausible to imagine any strong party in a reasonably favorable district running only 1 candidate for a multi winner election. I would almost say they deserve to lose if they can't find more than 1 person to run in that election.

[parker friedland]

> One straightforward answer is to modify RRV so that in every round the two candidates with highest score undergo a “virtual runoff”. In the virtual runoff, consider all ballots that score one of the runoff
> candidate higher than the other, and count each such ballot as contributing its current weight toward the one it scores higher. Whoever wins this “weighted runoff” is awarded the next seat, and the ballots
> are reweighted accordingly.
>
> Thus, every round acts like star voting with weighted ballots, and the standard RRV method for reweighting the ballots is used between rounds.

That is exactly how the broken SRV-PR method worked: https://groups.google.com/forum/#!searchin/electionscience/proportional$20srv$3F%7Csort:date/electionscience/aOMyxkoey4Q/KepDTrY7AwAJ

But, it was a naive way of making STAR voting proportional and it turned out that it didn't make STAR voting proportional.

> Actually it's pretty implausible to imagine any strong party in a reasonably favorable district running only 1 candidate for a multi winner election.  I would almost say they deserve to lose if they can't find
> more than 1 person to run in that election.

You can try to justify those non-proportional SRV-PR examples any way you like, but just the fact that they can happen makes it difficult for many people to be able to categorize SRV-PR as a proportional voting method.

[Nevin Brackett-Rozinsky]

The post you have linked (twice now) does not appear to mention the thing you claim it does.

Nevin

[parker friedland]

OK, so among the 4 options,

method 1 has the biggest connection to (and shares the most properties with) normal STAR voting.

method 4 resembles normal STAR voting the most.

and methods 2 and 3 are the simplest but they basically function like RRV.

Before we break down the different method's properties, let me clarify the different types of monotonicty that I will refer to:

The normal monotonicity criterion - A winner cannot be changed to a loser by experiencing an increase in support and loser cannot be changed to a winner by experiencing a decrease in support. Or in other words, giving candidates higher rankings or ratings while not changing anything else should never hurt them and giving candidates higher rankings or ratings while not changing anything else should never help them. This is the type of monotonicity described on CES's montonicity page: https://www.electology.org/monotonicity

Warren's strong monotonicity criterion (this is my best guess as to the type of monotonically that Warren was refering to on this form about STAR voting: https://groups.google.com/forum/#!searchin/electionscience/star%7Csort:date/electionscience/Gj1UPgmL5B0/KtkDVvphAwAJ) - Take a given ballot then make a list of the ratings/rankings provided on that ballot ordered from highest rating/ranking to lowest rating/ranking. Then make a ranking of all the candidates from highest scored to lowest scored on that ballot. Then move a candidate (we will call her candidate x) up the rank without changing anything else. Then pair the scores with the rankings of candidates so that the n highest score corresponds to the n highest ranking. Replacing the ballot with a new ballot in which each candidate gets the score that is paired with them should not hurt candidate x. Similarly, if candidate x was moved down the rank, then replacing the original ballot with the new one should not help candidate.

Both definitions of monotonicity are equivalent for ranking methods but different for rating methods.

Let's break this down by individual properties:

                                              Normal monotonicity criterion              Warren's strong monotonicity criterion

Normal STAR voting                                 ✓                                                                X

Method 1                                                   ✓                                                                X

Method 2                                                   X                                                                X

Method 3                                                   X                                                                X

Method 4                                                   X                                                                X

Broken SRV-PR                                        X                                                                X

Normal score voting                                 ✓                                                                ✓

RRV                                                          ✓                                                                X

Harmonic voting                                       ✓                                                                 ✓

Others might want to double check these but I'm pretty sure that these are all correct.

So in short, no matter which notion of monotonicity is used, methods 2 - 4 can not be considered monotonic. The only proportional STAR voting method that is monotonic according to the most common notion of monotonicity () would be the first one.

[parker friedland]

> The post you have linked (twice now) does not appear to mention the thing you claim it does.

Yea it does. Just scroll to the top and go to Warren's first comment in that discussion.

[parker friedland]

> Let's break this down by individual properties:
>                              

>                 Normal monotonicity criterion              Warren's strong monotonicity criterion
>
> Normal STAR voting                                 ✓                                                                X
>
> Method 1                                                   ✓                                                                X
>
> Method 2                                                   X                                                                X
>
> Method 3                                                   X                                                                X
>
> Method 4                                                   X                                                                X
>
> Broken SRV-PR                                        X                                                                X
>
> Normal score voting                                 ✓                                                                ✓
>
> RRV                                                          ✓                                                                X
>
> Harmonic voting                                       ✓                                                                 ✓

Correction:

                             

                Normal monotonicity criterion              Warren's strong monotonicity criterion

Normal STAR voting                                 ✓                                                                X

Method 1                                                   X                                                                X

Method 2                                                   X                                                                X

Method 3                                                   X                                                                X

Method 4                                                   X                                                                X

Broken SRV-PR                                        X                                                                X

Normal score voting                                 ✓                                                                ✓

RRV                                                          ✓                                                                X

Harmonic voting                                       ✓                                                                 ✓

[Mark Frohnmayer]

The example Warren comes up with does not at all represent a serious critique of STAR-PR. We'll start at the end. Warren writes:

"This scenario sort of reflects a 1-party domination scenario where the Repubs have taken over the universe and almost squashed all opposition.  The Dems are so weak and oppressed they are only able to run a single candidate, their Champion, whom they highly revere.  Even the Repub voters themselves are so disgusted with the situation they only give their party's candidates epsilon score.  But still "proportional score runoff" refuses to elect a Dem, even in this pushed-to-the-edge end of the world scenario."

The reality is that his last sentence should have been the very first: "In a pushed-to-the-edge end of the world scenario" where N candidates get elected and the 49% party only runs one, that one might not win if the 51% party votes 100% strategically.

Is this a reasonable failure mode? Hmm.

As Phil writes above, "it's pretty implausible to imagine any strong party in a reasonably favorable district running only 1 candidate for a multi winner election."

As long as each party runs several candidates (in reality, any major party would likely run as many candidates as there are seats), this scenario won't happen, and that strategy posed would severely punish voting blocs that use it because none of their candidates would make the top two for any seat.

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.
To unsubscribe from this group and stop receiving emails from it, send an email to electionscien...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[parker friedland]

Mark, I agree that there is unlikely to ever be a scenario in which a democrat can't win a single seat when 49% of voters give a max score to the democrat and a min score to everyone else. But this example demonstrates that RRV's re-weighting equation isn't very comparable with STAR voting. The re-weighting equation in RRV re-weights ballots based off of how much of a voter's voting power they spent on electing winners. (Kind of. It's a bit more complicated then that because RRV doesn't use strict quotas like STV so this is obviously an oversimplification. If you want to understand what RRV is actually doing, you will want to understand harmonic voting). However in STAR, the amount ballots contribute to electing a winner is very different then in score. For example, if out of the two candidates with the highest scores, you gave one of the candidates 1 star and the other 0 stars, under score voting, your vote wouldn't have much voting power, but under STAR, it would have a LOT more voting power. So in RRV, the sum of the voting power you used is equal to the sum of the scores you contributed, in STAR voting, the total voting power a voter has used is very different number all together, however dispute this, SRV-PR still uses the score voting definition of voting power instead of a star voting definition of voting power (which it is probably impossible to come up with a good value for the voting power a voter has spent in STAR voting) even though that is not the actual voting power a voter has used in STAR voting. Because of this error, voters can and will have voting weights that don't reflect how much voting power they have spent, and this relationship between voting power and ballot weights is what makes RRV proportional. Sure, STAR voting can be just as proportional when there are a lot of candidate cloning as a result of parties running multiple similar candidates, but that candidate cloning basically converts SRV-PR into RRV because when there are a lot of clones, a candidates will inter runoffs with their clones making it so that at least one of the two almost identical candidates wins the runoff. So if your reasoning of why SRV-PR is proportional is that parties will run many similar candidates, thus parties will inter runoffs with themselves, you are basically argueing that SRV-PR will be proportional because the candidate cloning effect will basically convert it into RRV, so if that is going to be your reasoning behind SRV-PR's proportionality, you might as well just push for RRV instead of SRV-PR.

[parker friedland]

While methods 2 and 3 are the simplest and their flaws will not be as severe as SRV-PR's flaws, they still share many of the same flaws and I'm very doubtful about their proportionality.

So, if equal vote wants a multi-winner voting method that reduces to STAR in the single winner case, they should define STAR as either the 1st or the 4th method. And while the 4th method looks like it should be proportional, I'm not sure about that one either.

[Mark Frohnmayer]

Parker, what would be the implication if all ballots were rescaled to the maximum range of all candidates still in the race before computing each round? Would that mitigate the strategic concern?

[parker friedland]

Then any voters attempting the "dominate the runoff" strategy would just give max stars to write-in candidates to avoid their votes being normalized.

[Warren D Smith]

The RRV process can provide proportionality even if if it is subjected to
"insults." By an insult, I mean, elect somebody at some moment during the
process, who is not the person RRV wanted to elect at that point,
but rather is selected in some other way (I do not terribly care how).

After the insult, we just continue on as usual, updating the ballot weights
based on whoever actually did get elected.

The point is that the reweighting process automatically "self corrects" (or
tries) to restore proportionality later. Whoever is elected at some moment
essentially always destroys/hurts proportionality, and the subsequent
reweighting tries to bring it back. It is a continual
mathematical fedback-correction process.

So anyhow, since I was a fan of range voting, I not surprisingly was
electing the
individual winners using range voting at each moment
(with weighted ballots). But if you used whatever other election
method you prefer,
such as STAR, then the resulting process would still yield PR, PROVIDED that
(1) it still had enough "gas left in the tank" to be able still reach PR --
if the alternate process elects 20 Reds while PR calls for only 10,
then the feedback-correction process it will be unable after that
point to regain PR.
If you elect only 10, even if rather
"ahead of schedule," it should be able to readjust.
(2) at some point you need to revert to usual insult-free
RRV process to fill the remaining seats, and at that
point PR must not yet have been irrevocably
destroyed.


--
Warren D. Smith
http://RangeVoting.org <-- add your endorsement (by clicking
"endorse" as 1st step)

[parker friedland]

The RRV process can provide proportionality even if if it is subjected to
> "insults." By an insult, I mean, elect somebody at some moment during the
> process, who is not the person RRV wanted to elect at that point,
> but rather is selected in some other way (I do not terribly care how).
>
> After the insult, we just continue on as usual, updating the ballot weights
> based on whoever actually did get elected.
>
> The point is that the reweighting process automatically "self corrects" (or
> tries) to restore proportionality later. Whoever is elected at some moment
> essentially always destroys/hurts proportionality, and the subsequent
> reweighting tries to bring it back. It is a continual
> mathematical fedback-correction process.

The auto-correcting process works when it has an accurate account of the amount of "voting power" a voter has spent to help elect candidates. It doesn't work when voters are using a lot more "voting power" then the re-weighting process thinks they're using.

> (with weighted ballots). But if you used whatever other election
> method you prefer,
> such as STAR, then the resulting process would still yield PR, PROVIDED that
> (1) it still had enough "gas left in the tank" to be able still reach PR --
> if the alternate process elects 20 Reds while PR calls for only 10,
> then the feedback-correction process it will be unable after that
> point to regain PR.
> If you elect only 10, even if rather
> "ahead of schedule," it should be able to readjust.
>
> (2) at some point you need to revert to usual insult-free
> RRV process to fill the remaining seats, and at that
> point PR must not yet have been irrevocably
> destroyed.

Not necessarily. You are right that the auto-correcting process will work when voters vote honestly, but when voters strategically take advantage the insult adding seats, the RRV process can't correct for this.

RRV's auto-correcting capabilities only work in voting methods that when the voting power they use is equivalent to the utility they indicate on their ballot (which is tgeor honest utility when voters are voting honestly). RRV's auto correcting nature depends on this relation between voting power and utility. In voting methods where giving a candidate a score of x does not equate to spending x amount of voting power when that candidate is elected, then RRV's auto correcting feature will auto-correct the results assuming that certain voters got a lot less bang for their buck then they actually got. Sure, if you elect enough seats with the usual insult-free election process afterwards, you will eventually drown out the disproportional by electing a whole lot more proportional seats, but that's just the same thing as achieving proportional representation by electing 100 seats proportionally to drown out the 10 seats that were elected under FPTP. It will drown those FPTP seats, but unless your using MMP, the proportional method will not correct for those seats by giving voters of that ideology less voting power in the proportional seat election. It's the same thing with when using RRV with a voting method that doesn't share a relationship between voting power and utility: after you elect the first seats with the other voting method, if when electing those seats, voters cheated the system by finding a way to use a lot of vote power without that vote power being measured by the RRV re-weighting system, then the only thing RRV can do to still achieve somewhat proportional results, is to drowned out those earlier elected candidates by electing a lot more people proportionally.

[Toby Pereira]

An alternative to your number 4 (quoted below) is to reweight all preference votes that elected the winner, rather than completely eliminating some of them. So, after having elected 1 candidate, they would be at 1/2 weight, after two it would be 1/3 and so on - assuming we're using D'Hondt divisors.

I see this as probably the most natural analogue to RRV for a proportional form of STAR voting.

On Sunday, 8 April 2018 22:18:20 UTC+1, parker friedland wrote:

[Nevin Brackett-Rozinsky]

On Sunday, April 8, 2018 at 10:47:49 PM UTC-4, parker friedland wrote:

> The post you have linked (twice now) does not appear to mention the thing you claim it does.

Yea it does. Just scroll to the top and go to Warren's first comment in that discussion.

In the future, please link directly to the post that contains the relevant information.

• • •

Back to the discussion at hand, it not clear to me why you keep talking about “voting power” (which I do not see a definition for).

If the goal is to achieve proportional representation, then it makes sense to measure the amount of *representation* each voter achieves. That is exactly what RRV does: it reweights the ballots by how well-represented each voter is. The precise mechanics behind *how* they came be represented that amount may be interesting, but it does not affect the fact that they *are* represented a certain amount.

[Warren D Smith]

> Not necessarily. You are right that the auto-correcting process will work
> when voters vote honestly, but when voters strategically take advantage the
> insult adding seats, the RRV process can't correct for this.

--Umm. Sorry, I'm feeling confused.

OK, first of all, I think due to a brain-malfunction
that I meant not RRV, but rather
MSPAV
( http://rangevoting.org/HolyGrailPR.html#multPSV )
as the mathematical fedback readjustment process
which does not care about "insults." :(

Second, returning to RRV,
RRV's proportionality guarantee applies
when the voters vote in "racist" color-based style,
i.e. give MAX score to same-color-as-them candidates,
MIN to all others.
In that case they effectively are using approval voting,
not range voting. Now as it happens , STAR
voting is the same thing as range voting if
voters vote in approval-style (all MAX & MIN scores,
none intermediate).

Therefore, if the RRV process were run but at each stage electing,
not the range-voting winner with the current weighted-ballot set, but
rather the STAR winner, were elected, then
it ought to obey the same proportionality theorem
as for RRV. (You would have to use
STAR as redefined to work with possibly-unequally
weighted ballots, though: weighted-sum the scores,
find the top 2 candidates, then weighted-sum the
signs of their score-differences to find winner.)

And in fact this method, and RRV, will be equivalent methods in
"racist color-based voting" elections.

In other kinds of elections, they will not be the same thing.
But that is ok since the proportionality theorem only speaks
about what happens with "racist voting."

[parker friedland]

> --Umm.  Sorry, I'm feeling confused.
>
> OK, first of all, I think due to a brain-malfunction
> that I meant not RRV, but rather
> MSPAV
> ( http://rangevoting.org/HolyGrailPR.html#multPSV )

> as the mathematical feedback readjustment process

> which does not care about "insults."  :(

I believe that harmonic + runoff is the best optimal quality function multi-winner adaptation to STAR voting, however quality function methods can often be impractical without eliminating a set amount of candidates first through an SNTV to limit the number of possible election outcomes, so perhaps your STAR voting version of the MSPAV method could be the best sequential multi-winner adaption to STAR voting.

> In other kinds of elections, they will not be the same thing.
> But that is ok since the proportionality theorem only speaks
> about what happens with "racist voting."

Which is why I prefer proportional score voting (PAV + KP) over RRV, because RRV doesn't do as good of a job at producing proportional results as proportional score voting when voters are not maximally racist.

[Jameson Quinn]

I've missed out on this thread so far, but I just wanted to point out the similarity between method 4 and E Pluribus Hugo voting. Very roughly speaking, method 4 and EPH will elect a similar set of candidates, but EPH goes by eliminating bottom-up while method 4 goes by electing top-down. 

Personally, I believe that it's not a good idea to look for equivalence in single-winner and multi-winner methods. In my opinion, the best way to get a good multi-winner system is to go for breadth of choice, and given voters' finite tolerance for ballot complexity, that inevitably means sacrificing depth of choice. If you take this idea to the ultimate extreme, you get some form of asset voting; but if you merely take it to the limits of what is possible with equal-weighted legislators, you get something like PLACE voting. On the other hand, for a single-winner method, it's not good to sacrifice depth of choice. Even if you do use delegation for single-winner, it's crucial that it's optional, as in SODA. So I think that the best single-winner options will be something like 3-2-1 or STAR.

But. If you are looking for a STAR-like PR method, I think that any of methods 2, 3, or 4 would be good.

[parker friedland]

> But. If you are looking for a STAR-like PR method, I think that any of methods 2, 3, or 4 would be good.

Why not method number uno. Because it is too impractical? The first method would work if you used a top ≈10 non-partisan SNTV primary to limit the number of possible election results. Method would be the theoretical best multi-winner adaption of STAR, but 4 might be more practical in some scenarios.

As for methods 2 and 3, they still exhibit the same error that exists in SRV-"PR": the re-weighting equation can re-weight a voter's weight less then it actually should be re-weighted. If in method 3, a voter gives the winning candidate 1 star but still decided a very close runoff, RRV's re-weighting equation then reduces their weight a lot less then it should be reduced. Because the re-weighting equation is the culprit of the errors you get when you mix STAR and RRV, perhaps methods 2 and 3 should work a lot better in reverse:

5. (fixed version of 2): Normal SRV-"PR", except that each round, the number of ballots the candidate with the second highest score needs to be preferred on to beat the candidate with the highest score gets greater and greater. In the last round, the candidate with the 2nd highest score would only need to be preferred on 1/2 of ballots to win (The last time I forgot to mention that ballots that are re-weighted would still count less to whether one candidate is preferred to another), on the the 2nd to last round, they would need to be preferred on 2/3rds of ballots, on the 3rd to last round, 3/4th of ballots, on the forth to last round, 4/5th's of ballots, on the 5th to last round, 5/6th of ballots, etc. I'm not 100% sure if this method is proportional. Could someone please look into how proportional this voting method would be?

6. (fixed version of 3): RRV, except that the last (instead of the first winner) winner is elected via STAR instead of score.

So far, I think methods 1, 4, and 6, are good multi-winner adaptions of STAR voting, and methods 2 along with SRV-"PR" are broken and should not be used. However, I'm not sure on how proportional method 5 is and while I'm pretty sure that methods 1, 4, and 6, pass Warren's Strong PR criterion, I would like a proof of whether or not method 5 passes his criterion.

Warren's Strong PR criterion:

Suppose the electorate consists of both colored and uncolored voters and candidate, where the colored voters give the max possible scores/rankings to every candidate of their color, and where the uncolored voters whose ratings do not depend on each candidate's color. A voting method passes this criterion if for each colored subset of the electorate, the portion of candidates of that candidate that win is guaranteed to be at least as much as the portion of the electorate that color makes up minus 1 seat, provided that for each color, there are at least enough candidates of that color running to represent that portion of the electorate (e.g. we cannot elect 5 Greens if only 3 run).

Any voting method that actually is proportional should be able to pass this criterion.

[Ciaran Dougherty]

I still think that the method I came up with a while ago (q.v. here) is a good multi-seat, Score-based method.  But then, that's because it's an Iterative approximation of Monroe's method (which Parker was kind enough to explain to me, recently).

Under STAR, it would be simple, and somewhat related to #4 above:

STAR for seat N

Set aside the ballots that contributed most to that candidate being seated.

STAR for seat N+1, considering only ballots not yet set aside.

--

You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.

To unsubscribe from this group and stop receiving emails from it, send an email to electionscience+unsubscribe@googlegroups.com.

[Jameson Quinn]

I like this idea a lot. Easy to explain. Here's my version:

STAR for seat N: Considering only ballots not yet set aside,

Find 2 highest scores
Find pairwise winner between those two

Set aside:

Order ballots by winner's absolute score

As a tiebreaker, lexically by lowest score of each remaining candidate, with candidates' lexical priority in order of highest to lowest average score on remaining ballots (so loser of top two is first to be checked).

Set aside the first Hare quota of ballots in that ordering.

I wouldn't use "lexical" in describing it to a layperson, but it's actually quite simple, and could be done with a "book club" of ballots.

I still think that we should not link single-winner and multi-winner proposals. Summability, breadth of choice, and ballot simplicity are all bigger considerations in multiwinner than single-winner. But the above is is a "Star PR" that I would not be ashamed to describe to a layperson.

[Ciaran Dougherty]

I would just like to point out the reason that I went with "difference from average" is that a 4/2/1 ballot is more discriminating (+2,+3) than a 5/4/3 ballot is (+1,+2).

[Jameson Quinn]

I suggested ordering by highest for the winner, and then by lowest for the runner-up, and then by lowest for the second runner-up, and so on. An alternative would be to order by difference between winner and runner-up, then by difference between winner and second runner-up, and so on. These both prioritize high scores for winners and low scores for runners-up (and both look at runners-up in order and stop when they see a difference); so in essence, they are similar.

I have no strong preference between these two alternatives. I think either one is better than going by "difference from average", because both correctly ignore irrelevant candidates, and require less math.

Actually, wait a minute, I like my original idea better than the alternative I described above. Consider the following ballots, in an election to choose 4 winners:

49: A10 B10 C10 D0 E0

25: A0 B0 C4 D10 E0

26: A2 B0 C0 D0 E10

ABC is one party with ~2 Hare quotas, D is a party with ~1, and E is a party with ~1. Best result is ACDE. My original version chooses C first and eliminates 25 of the ABC voters, then chooses D, then E, then A and B are tied (broken towards A by original score). The alternative method also chooses C but eliminates 25 of the D voters for doing so, leaving both B and A to also be selected; final winners would be CABE. In other words, the incentive is to rank all members of a party you don't like as equal.

So I'm going to say I strongly prefer my original STAR-PR proposal. Your vote will only get used up on electing a compromise when you are one of the ones who compromised least; if there is anybody for whom that candidate was an ideal choice, they'll get used up first.

[Ciaran Dougherty]

Requires less math?  A ballot's average can be calculated once at the beginning.  Indeed, the difference between Candidate X and BallotAverage can be calculated once at the beginning.  Further, I suspect that it would be mathematically equivalent to treat those calculated differences as the voter's ballot.

> The alternative method also chooses C but eliminates 25 of the D voters for doing so

Do you mean your alternative methods?  Because the "difference from ballot average" doesn't select D>C ballots for C:

• The ballot average for the D>C voters is 2.8 (14/5), and their vote for C is 4, for a difference of 1.2
• The ballot average for ABC voters is 6 (30/5), and their vote for C is 10, for a difference of 4
• 4>1.2, so C's quotient would comprise 25 voters from the ABC faction.
• Final Result
• C(25ABC)
• A(24ABC, 1D>A)
• C(25D>C) or D(25E>A)
• D(25E>A) or C(25D>C)
  

But I think you were assuming Score rather than STAR.  Under STAR, A should be selected first:  Top Two are A vs C, and of the ballots that express preference, there are 26 A2>C0 and 25 C4>A0.  In that case, however, the quotient still selects exclusively from the ABC faction:

• The ballot average for the E>A voters is 2.4 (12/5), and their vote for A is 2, for a difference of -0.4
• The ballot average for ABC voters is 6 (30/5), and their vote for A is 10, for a difference of 4
• 4>-0.4, so A'a quotient would comprise 25 voters from the ABC faction.
• Final Result
• A(25ABC)
• C(24ABC,1D>C)
• E(25E>A)
• D (24D>C,1E>A)
  

Perhaps you were thinking that I meant that ballots were removed in ascending order of Difference from Ballot Average?

But I think the biggest takeaway I have for this is that you can't merely compare the difference of the Top Two; no matter who wins the first seat, C (score winner) or A (STAR winner), the difference between the top and runner up for the ABC faction is 0, and would therefore be selected last.

[Toby Pereira]

On Thursday, 12 April 2018 00:31:44 UTC+1, parker friedland wrote:

> But. If you are looking for a STAR-like PR method, I think that any of methods 2, 3, or 4 would be good.

Why not method number uno. Because it is too impractical? The first method would work if you used a top ≈10 non-partisan SNTV primary to limit the number of possible election results.

As you say, number 1 only works if you limit the election results. If you hold an automatic run-off between the top two under harmonic voting, then they will probably be very similar only differ in the least popular of the candidates to be elected. I don't think it would be worth the extra complication for what it might achieve (not very much).

 

So far, I think methods 1, 4, and 6, are good multi-winner adaptions of STAR voting, and methods 2 along with SRV-"PR" are broken and should not be used. However, I'm not sure on how proportional method 5 is and while I'm pretty sure that methods 1, 4, and 6, pass Warren's Strong PR criterion, I would like a proof of whether or not method 5 passes his criterion.

Warren's Strong PR criterion:

Suppose the electorate consists of both colored and uncolored voters and candidate, where the colored voters give the max possible scores/rankings to every candidate of their color, and where the uncolored voters whose ratings do not depend on each candidate's color. A voting method passes this criterion if for each colored subset of the electorate, the portion of candidates of that candidate that win is guaranteed to be at least as much as the portion of the electorate that color makes up minus 1 seat, provided that for each color, there are at least enough candidates of that color running to represent that portion of the electorate (e.g. we cannot elect 5 Greens if only 3 run).

Any voting method that actually is proportional should be able to pass this criterion.

I'm not sure I follow or recognise that particular definition, but if there are two parties that each have supporters that approve (or top rate etc.) all their candidates, but that there are also other candidates that are approved by supporters of both parties (so are universally approved by these voters), then for a method to obey strong PR, the party candidates should be elected in the same proportion to each other regardless of these other candidates. For example, take the following approval ballots with three to elect and parties A and B.

2 voters: A1, A2, A3

1 voters: B1, B2, B3

Here, the proportional result is two from party A and two from party B. Then we have this example, with six to elect:

2 voters: U1, U2, U3, A1, A2, A3

1 voter: U1, U2, U3, B1, B2, B3

A method that passes strong PR would elect U1, U2, U3 and two candidates from party A and one from party B. A method that doesn't obey strong PR might elect U1, U2, U3, A1, A2, A3. And in fact that's what RRV and that general family (the Thiele family) of methods does. So I don't think any of 1, 4 or 6 would pass.

[Jameson Quinn]

Using difference from ballot averages encourages voters to max-rank any candidate they think can't win (or at least, can't win before the ballot is eliminated), in order to last as long as possible. That has obvious problems.

The more I think about it, the more certain I am that ballots should be selected on highest score for winner, with ties broken by lowest score for runner-up, then for next runner-up, etc; where runners-up are, of course, defined by average score. As with any proportional method, this does have some free-rider-strategy incentives; but I think they're as weak and self-limiting as possible.

[parker friedland]

https://www.starvoting.us/pr

This recently added page shows that the equal vote coalition is still promoting the broken version of STAR-PR that just combines STAR voting with RRV's re-weighting algorithm.

I've tried to communicate the problems with SRV-PR and suggested solutions at the real choice voting forums, which is where I can find most of the STAR voting supporters:
https://www.loomio.org/d/XhDQ5piK/glossary-for-voting-enthusiasts
https://www.loomio.org/d/xcfqVm45/multi-winner-adaption-of-star-voting

Mark? Can you comment on equal vote's position on a proportional version of STAR voting? Are you still promoting the broken SRV-PR method? Because nobody who is an actual voting theorist is claiming that that method is proportional.

To unsubscribe from this group and stop receiving emails from it, send an email to electionscien...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.

To unsubscribe from this group and stop receiving emails from it, send an email to electionscien...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.

To unsubscribe from this group and stop receiving emails from it, send an email to electionscien...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.

To unsubscribe from this group and stop receiving emails from it, send an email to electionscien...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.

To unsubscribe from this group and stop receiving emails from it, send an email to electionscien...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.

To unsubscribe from this group and stop receiving emails from it, send an email to electionscien...@googlegroups.com.

[Brian Olson]

A while ago I proposed a ratings-ballot PR method I called "Instant Runoff Normalized Ratings (Proportional)" or IRNRP
https://bolson.org/voting/IRNRP.html

I've implemented in this in code a couple times and run simulations and I think it works okay.

Copied from above page, a concise statement of the algorithm:

   1. Maintain a 'weight' for each choice, staring at 1.0.
   2. Sum up the normalized de-weighted preferences from the voters
      For each vote:
      - multiply each voted preference by the current weight for that choice. (deweight)
      - divide these by the square root of their summed squares (normalize)
      - add those results to the per-choice sums
   3. The total vote is the sum of the vote summed up across choices. The quota is (total vote)/(seats + 1)
   4. For each choice with a sum greater than the quota, that choice's weight becomes ((old weight) * (quota / sum))
      If negative ratings are allowed changes may cause a choice's sum to fall, in which case if it was above the quota it may fall below the quota.
      It's weight should be adjusted upwards to be (weight * (quota/sum)), but it's weight shall be limited to 1.0.)
   5. If no choices are greater than quota then the choice with the lowest sum has its weight set to zero.
   6. Repeat steps 2-5 until the right number of candidates have a sum greater than or equal to quota.

*Parker*, I looked at https://www.starvoting.us/pr and don't see a clear statement of a proposed election algorithm. Lots of talking *about* it and its parts, but no one statement of what the whole system is end to end.

To unsubscribe from this group and stop receiving emails from it, send an email to electionscience+unsubscribe@googlegroups.com.

[parker friedland]

> *Parker*, I looked at https://www.starvoting.us/pr and don't see a clear statement of a proposed election algorithm. Lots of talking *about* it and its parts, but no one statement of what the whole system is end
> to end.

Here is their definition of the "proportional" algorithm:

> What is the STAR-PR Algorithm?
> With STAR-PR voters fill out their ballots as usual, giving the highest score to their favorite(s) and so on. The scores are totaled and the two candidates with the highest scores advance to an automatic
> runoff. Just like in single-winner STAR Voting, the candidate you scored higher gets your full vote. The first winner has been elected!
>
> In each subsequent round, your ballot influence is 1 / (1 + sum/5), where "sum" is the total of the scores you've given to winning candidates.
>
> Let's explain that again without the math. To find the next winner, the remaining candidate's scores are counted again with one change. If a candidate you had supported won a previous round, your ballot's
> influence is reduced in proportion to how much you've "won" so far. If you gave them a zero your ballot still has it's full strength, but if, for example, you gave them a full support, your ballot would be half as > powerful going forward. So, in the second round if you had given your next favorite a score of 4 it would be counted as half that, 2.

> As in the first round, the the candidates with the highest scores advance to a runoff. Whichever you scored higher gets your full vote, multiplied by your ballot influence, in the runoff. This process continues

> until enough candidates are elected to fill all the available seats.

If they want the method to be proportional, they should either only use star voting in the last round and use normal score voting in the rest of the rounds, or define STAR-PR as one of the other multi-winner adaptations of star voting (that are actually proportional) discussed on this thread.

[Brian Olson]

Yeah, I'm pretty sure that's not proportional.

60% of the people vote for the same favorite, they got one so they get de-prioritized, but they get de-prioritized too much. That 60% bloc should probably get to elect 60% of the seats. My IRNRP system accounts for this by adjusting votes based on how much over quota a candidate won by. A win by more over quota should result in a smaller adjustment so that voters still have voting power to distribute to other choices.

To unsubscribe from this group and stop receiving emails from it, send an email to electionscience+unsubscribe@googlegroups.com.

[sa...@equal.vote]

Hi All, Sara here. Chief petitioner for the Portland STAR Voting initiative. I wanted to weigh in because my post on PR was mentioned here: starvoting.us/pr

The situation: In Portland (PDX) there is huge pressure from a certain branch of the voting reform movement to implement PR for PDX City Council, which is currently at large plurality. There is also huge pressure to do the same thing in BC, Canada. Sightline and FairVote are the big forces behind that, and as you all know they are in some cases specifically willing to accept a questionable single winner system IRV because it pairs with STV, their endgame. We also like PR and want it to be a part of the picture, though I'll admit that most of us aren't head over heels in love with any of the options. We think STV is somewhat better than IRV because multi-winner corrects for some inaccuracy, but still. We don't want STV presented as the only option. 

Our PDX Equal Vote chapter (I'm not speaking for Mark or Clay, though they agree with some of this at least,) believe that our best chance at getting an actually accurate voting system is to offer up a package deal, with a great single winner system as well as a PR version. Both should use the same ballot (0-5 score/star) and voter education. We also are open to and supportive of those that would rather reform multi-winner election to single winner districts. Our current position on that is that we would like to host that conversation so that each locality can make an informed and inclusive decision. Including and listening to community groups, especially people of color, is critical for the eventual success of the initiative. I'm committed to democratic process in the pursuit of democracy! Here in Portland many support districts because they want "proportional" geographical representation. 

Currently our County Commissioners are elected by single-winner districts. That council is all women with 3/5 women of color. In contrast PDX City Council is at large and is all white and majority male with almost all hailing from the rich side of town. Same voters. 

Equal Vote PDX's position: So, we are advocating STAR Voting for single winner and STAR-PR (0-5) for multi-winner where PR is desired. (To clarify: STAR-PR and SRV-PR are the same thing, we just rebranded. It includes reweighing for each seat and runoff.) Right now we have 2 county wide ballot initiatives for single winner STAR Voting. We don't have a finalized, ballot measure ready proposal for multi-winner and our current STAR-PR proposal is still open to input, vetting, study, and data. We'd like more science on the topic to make a final determination. We'd love to see multi-winner VSE/bayesian Regret and more data/conversation around types of proportionality and what our actual goals are. We are open to revising the algorithm if there is another version that we find does a better job at meeting/balancing the 5 pillars of a just voting system: Equality (or Equity in the case of PR), Accuracy, Honesty, Simplicity, and Expressiveness. Viability would be great too! 

In the meantime we plan to get STAR-PR tried and used more in smaller scale elections for events, boards, neighborhoods, etc. STAR-PR is supported by our star.vote app so try it yourself! I believe we can all agree that STAR-PR as proposed would offer much better results than plurality at large.

I'm happy to define STAR-PR as a multi-winner voting using a 5 star ballot which provides fair representation. Think of our current proposal as a working draft. I invite discussion and study on what the final should be, though I would love it if that were done in private threads where the general public wont get freaked out by discussion or disagreements. I know you all are excited by the science here but please be aware of the politics and obstacles of actually passing any reforms that offer any level of improvement. A private Loomio thread or similar would be good. Anyone willing to engage open-mindedly and respectfully is invited. Please try and keep in mind that what you say and write publicly can and will be used against us all. Politics, algorithms, and change are scary to lay people. Clarification that Plurality is horrible is always a good safeguard to include in discussions as a disclaimer. 

Here are some questions that I think are worth exploring: 
1. Will the final proposal have reweighing on every round and runoff (as is)? Some (last round only?) None (aka is RRV the best STAR-PR?)? 
2. What quotas are best? 
3. Do we want to meet Warren's perfect proportionality criterion? 
4. Is it desirable to have an algorithm that would put Nazis or other super antagonizing and polarizing candidates at some level of disadvantage while still allowing non-hateful minorities to get elected? 
5. Should we round up or down for quotas? 
6. Is there a geographic scale that's too big for non-precinct summable systems or too big because geographical representation is important too?
7. Is there a nice hybrid of single winner and PR that can offer both kinds of proportionality? (ie. STAR-PR House and STAR Senate.)
8. Is it possible to have a 0-5 ballot, proportional results, and an algorithm that can be explained to lay people? 
9. Is there a way to explain or demo STAR-PR using a board game? A computer game? 

My philosophy is this: Question everything and the truth will come out stronger for it. 

[Jameson Quinn]

Hi, Sara. Your post brings up a lot of interesting questions.

First off, I'd like to look at the democratic values you espouse:

> Equality (or Equity in the case of PR), Accuracy, Honesty, Simplicity, and Expressiveness. Viability would be great too!

First, you have equality/equity. In my mind, if you're electing more than one similar position (ie, a council, board, or legislature), proportional representation is the truest form of equity. Yes, a good single-winner system is still better than FPTP, but I side with Ernest Naville: "In a democratic government the right of decision belongs to the majority, but the right of representation belongs to all." If some people aren't (proportionally) represented in the legislature, their interests will be systematically excluded from decision after decision. If they are represented, their interests will be equitably taken into account; that's not a guarantee that the final decisions will be equitable, but it is pretty close to being a precondition. 

Second, accuracy. I don't have a lot to say about that; it's obviously a worthy and achievable goal.

Then, the last three: honesty, simplicity, and expressiveness. To a certain degree, expressiveness is in tension with the other two. Imagine two rated voting methods, one of which uses a 3-level ballot and the other of which uses a 100-level ballot. It's pretty obvious that the 100-level one is more expressive, while the 3-level one is simpler. Perhaps less obvious is that the 3-level one is likely to be more honest, too. With fewer levels, there's simply fewer options for strategic voting.

Of course, there's a limit to that logic. A 2-level approval ballot is probably more, not less, strategic than a 3-level one. That's because voters often have at least 3 different kinds of candidates they want to distinguish — often that is a favorite, a "lesser evil" frontrunner, and a greater-evil frontrunner; though other configurations are possible — but it's much rarer to need to make 4 or more levels of distinctions. Since a 2-level ballot can't make those 3 distinctions, voters need strategy to shoehorn their preferences into approvals.

A purist would of course object that by even mentioning frontrunners, I'm already assuming strategic voting. And yes, that's technically true; but remember, the base value we're talking about is "honesty". Strategy is only really a problem in my book if it means lying about or hiding the distinctions you consider most important. It's much less of a problem if it's just a matter of collapsing distinctions that are relatively minor to begin with.

So, on the basis of both simplicity and honesty, I think that a 3-level ballot is better than a 6-level one. And I realize that's probably not a popular opinion here, in a forum for ultra-engaged political geeks. Personally I'd be happy to rank all candidates 0-5, if I felt that I could afford to do so honestly without ending up a strategic chump. But there are a lot of voters who'd rather choose a favorite, perhaps vote weakly for or strongly against a few others, and go home, without worrying about those fine distinctions. And people like that don't tend to be on this mailing list.

As you might guess, when I argue for 3-level ballots, I have specific methods in mind: 3-2-1 for single-winner, and PAD for proportional representation. Both of these allow voters to rate candidates at 3 explicit levels — good, OK, or bad — plus "I don't know" — filled in as OK or bad using predeclared ratings from the candidates the voter rated "good". So a ballot in either case would look something like this:

Candidate	Good	OK	Don't know (delegated)	Bad
Alice	▢	▢	▢	▢
Bob	▢	▢	▢	▢
Carol	▢	▢	▢	▢
Eve	▢	▢	▢	▢

Why do I think these are better than STAR and (the version I describe below of) proportional STAR? Because they lack safe, effective strategies, so that a voter's naive ballot is more likely to be the same as their strategically-optimal ballot.

For a single-winner example, consider the chicken dilemma. Voters' true preferences are:

35: A>B>>C

25: B>A>>C

40: C>>A=B 

So in STAR, the honest ballot would be 5/4/0 for the first two groups and 5/0/0 for the third; and A wins, which is arguably the best outcome. But the B voters can, with no risk, change their vote to 5/1/0, and thus win unless the A voters respond defensively.

Of course, the B voters could in theory go still further than that, and vote 5/0/0. But while this would improve their chances of strategic success, it's risky, and involves a substantial sacrifice of their own expressiveness. So in this discussion, I'll assume they don't go that far.

In 3-2-1, the honest ballot would be good/OK/bad or good/bad/bad. The B voters have no risk-free strategy, and in fact it would take a full 20 of the 25 B voters before even a risky strategy would work. I believe that because of that, in practice they'd just vote honestly. So voting would be simpler not just in terms of the number of options of how to cast a ballot, and not just because they'd have the option of delegating, but also because the cognitive burden of strategy (and of worrying about being a "chump") is lifted.

Now, none of this is to say that STAR isn't a good voting method. It's great; much much better than FPTP, and significantly better than other reforms such as IRV, approval, or score. It's just that it's not quite as good as 3-2-1 in my eyes.

What about multi-winner? Well, to discuss this, I'm going to first have to define what I mean by "proportional STAR":

Voters rate candidates 0-5

Define a quota as a Droop quota: ceiling([V/(S+1)]+ε)

Repeat until all seats are filled:

    Find the STAR winner of the remaining ballots and give them a seat

    Remove 1 quota worth of ballots. To do so, sort ballots first from highest to lowest score for the winner, then by lowest to highest score for the runner-up, then by lowest to highest score of the candidate who's 3rd in overall scores, etc.

Unlike the "reweighted" version of proportional star, this is actually proportional. In my opinion, it's reasonably easy to explain and intuitive as well.

Why is PAD better than the above? As above, because PAD lacks the "safe strategy" options that P-STAR has. In this case, the scenario is "free rider", not "chicken dilemma". With 3 seats:

31: A1>A2>A3>B1=B2

30: A1>A3>A2>B1=B2

39: B1>B2>A1=A2=A3

In P-STAR, if the second group strategically downrates A1 from 5 to 4, then A1's quota will be removed from the first group, so A3 will beat A2. This is a safe strategy, very unlikely to backfire. (Tweaking the rules of P-STAR would change the details of how free-riding worked, but no voting method lacks a free-riding incentive entirely.)

In PAD, if the second group strategically downrates A1 from "good" to "OK", they are risking A1 not getting a seat at all. Free riding is still a possibility, but it's inevitably a risky one.

Another advantage of PAD is that it works just fine as an "Open MMP"-like system, with some seats allocated by district and others at large. I'm currently trying to get both the plaintiffs and the defendants in a lawsuit against the at-large voting in the city of Lowell, MA to appreciate the benefits of PAD so they can settle the case. (Because of how the specific law involved works, this is a grassroots strategy outside the courtroom, not something that would be argued in court.)

....

So, I understand that all of the above (except for the definition of P-STAR) may be something you end up disagreeing with. If there's already a strong STAR movement in Portland, then switching to 3-2-1 might not be the best tactic. But I do encourage you to at least consider it. 

And, on a meta level, what I'd really like would be to run a focus group/deliberative poll/citizens' assembly with ordinary voters, where advocates for different voting methods could each make their cases and we could settle the viability question more empirically. Of course, that would take a fair amount of funding, because you'd want to do various things to help ensure the people involved were taking it seriously, and to make sure they truly held the reins in terms of the ultimate decision.

So, as to your questions:

2018-05-15 22:06 GMT-04:00 <sa...@equal.vote>:

Here are some questions that I think are worth exploring:

1. Will the final proposal have reweighing on every round and runoff? Some? None (aka is the best STAR-PR RRV?)?

STAR-PR-RRV is badly broken. The P-STAR I describe above is not. It uses ballot removal, not reweighting. 

 

2. What quotas are best?

Meh... I don't think this question is super-important, but I'd go with Droop.

 

3. Do we want to meet Warren's perfect proportionality criterion?

I'm sorry, I forget what that was now.

 

4. Is it desirable to have an algorithm that would put Nazis or other super antagonizing and polarizing candidates at some level of disadvantage while still allowing non-hateful minorities to get elected?

P-STAR does this, and to a lesser extent PAD does. If you wanted to strengthen this aspect in PAD, you could add a rule that no candidate could win without some threshold amount of approval-or-above votes; that threshold could be anywhere from a Hare quota to 2 or more Droop quotas.

 

5. Should we round up or down for quotas?

Even less important, but up.

 

6. Is there a geographic scale that's too big for non-precinct summable systems or too big because geographical representation is important too?

I think that the real limit isn't geography, but total number of candidates on the ballot that a voter has to rate. The delegation aspect of 3-2-1 or PAD mitigates that issue substantially, so that bigger elections are feasible.

As for summability, there is a n² summable approximation to PAD that's quite feasible. This will never be true for any proportional version of STAR.

 

7. Is there a nice hybrid of single winner and PR that can offer both kinds of proportionality? (ie. STAR-PR House and STAR Senate.)

Yes, PAD offers MMP-style hybrid option, with almost no additional complexity.

 

8. Is it possible to have a 0-5 ballot, proportional results, and an algorithm that can be explained to lay people? 

I think that P-STAR is reasonably good at that. I doubt there's anything better. There may be other tweaks that are approximately as good.

 

9. Is there a way to explain or demo STAR-PR using a board game? A computer game? 

Good question, I don't have an answer right now.

[Jameson Quinn]

Note regarding the O(n²) summability of PAD: insofar as voters delegate, it's actually O(n) summable. In other words, it may be feasible to keep only 3 tallies per candidate (fully delegated "good", all "good", and undelegated "OK"); and then, as long as non-delegated votes are rare enough, to keep track of upper and lower bounds on the possible values of those tallies such that it's clear who won without even the full matrix of X-good-and-Y-OK tallies.  

[sa...@equal.vote]

Thanks Jameson, 
I do like 3-2-1 and I'll look more into PAD. I'm glad you're working on it and I hope it gets a real world trial. I don't think Portland or OR will be that place. I do mention 3-2-1 in Q and A's a lot when VSE comes up or if I say that there are lots of systems out there but in general we're well past the "comparing systems" stage. It was an insanely monumental task to get people to unite around STAR to the extent that they have, and I don't think a total course change would be possible any time soon, even if it was desirable.

Personally I think 3-2-1's biggest obstacle is in the explanation. Not that it's that complex or anything, but the non-numerical and proxy components are barriers to getting people on board in the first place. Obviously if it was implemented I think everyone could figure it out, but still. I think your best bet would be rebranding it to somehow use a "ranking" ballot type explanation, but allowing ties, but because of IRV people would get confused by that. Tricky. For your sake I hope I'm wrong. It's an innovative approach and we need that kind of thinking to solve these questions. 

As for the product testing/polling idea to measure systems usability I also think that has somewhat limited value. In order for people to compare systems they need to understand all the options in some detail. Explaining more than 1 system at a time to lay people adds an exponential magnitude of confusion so I think the results would be kinda meaningless. Doing the same experiment with non-lay people (like this forum) also is of limited value in measuring real world viability because you guys/we think really differently than most people. Maybe the best way to do that would be to get a huge sample and divide it into groups? Each group gets pitched 1 idea only, uses it for a vote, then has to re-explain how to do it and how its counted back to the pollster. Then they rate (0-5 ;) ) how much they like it as a voting system. Then you look at the ratings for their comprehension, explanation, and satisfaction. To do it right would require a huge polling size with a lot of controls and the amount of money required to do it well could do a lot more elsewhere. If anyone can afford it I highly recommend they donate here instead: starvoting.us/donate

Viability is a whole other question than usability. What I mean by viability is: can it get passed and used in the current political climate and would it get repealed? If important political decisions we're actually made on the merits of the ideas themselves we'd live in a very different world. Ugh. 

Re: honesty, simplicity, and expressiveness.
Simplicity and Expressiveness are definitely inversely correlated. To me in order to have full honesty and accuracy you need a minimum degree of expressiveness. I think a minimum there is Favorite, Pretty good, better than worst case scenario, worst. Neutral would be good too. 

For a minority voter in a red state there's a huge difference between a pretty good second choice and a lesser evil who is only worth considering in order to prevent a worst case scenario. 

The fact is the 5 star rating is already a thing. In an alternate universe 0-3 or 0-6 or 0-9 could all be fine, but in our universe going with something already used around the world is huge. STAR Voting ties neatly into our own design as humans. 0-5 scoring is encoded in our 5 fingers, and the runoff can be represented by a "raise your hand if you prefer.."

Proportional STAR v. STAR-PR and algorithms in general:
I confess that I'm not a numbers person. I'm glad you all are. I love a good thought experiment and am a visual thinker. To convince me, someone would have to explain the algorithms and such in words, (as I did in starvoting.us/pr). I don't get why STAR-PR as proposed "is broken" or "fails" proportionality while RRV would "pass". I'm not even totally clear on what the definition of  "proportionality" here is. I thought someone mentioned a Warren PR criterion but looking back I didn't see the comment. As I recall the link for that was incomprehensible anyways. 

STAR-PR story problem: The analogy that makes some sense is that voters are hiring their candidates. Each voter starts out with an equal budget to spend. Voters fill out their ballots showing via scores how much they are willing to pay each based on how represented they would feel by each candidate. During the tabulation you only spend some of your vote if a candidate you liked gets hired. How much you spend depends proportionally on how much you liked them. Once all your "money" is spent you are fairly represented. The result is perfect Equity. Is that wrong? Is that still the concept for your P-STAR suggestion?

Proportion of what? Honestly, I also am a bit concerned about what exactly is getting proportioned out and if that actually leads to the kind of representation many find so attractive. With geographical districts, each local area is equally represented. Since the goal of most local governments is providing services to their constituents that makes a lot of sense. Party list systems proportion out seats based on ideology with each faction getting fairly represented. (If our parties were less broken this might be more attractive.) Many seem to want demographical/racial/gender equity. We just had an election yesterday here and many candidates ran on a platform based around who they plan to represent (seniors, poor people, young people, Latinos, Black people.) Others ran platform based campaigns (gun rights, housing crisis, tax reform, election reform for city council.) RRV/STV systems seem to do some combination of all or none of those so I wonder if different voters focus on being represented by different metrics if that works out in the end? Does perfect mathematical proportionality = good representation across the board? 

Real life is messy:

Recently we were approached by a neighborhood that wanted to adopt STAR-PR. Their last two at large plurality elections (with a majority requirement) had run into major issues and people were effectively gaming the system using coordinated strategic block voting. Their stated goals and reasoning for choosing STAR-PR was to help all people, including people of color have a voice and be fairly represented. They wanted to encourage better and more candidates to run and have and more open elections. This was a part of a package deal that included everything from paper ballots and early voting, to better voter verification, to better bylaws for a more inclusive process. 

Unfortunately the reality of the problem was that they have 1 super-toxic board member who has effectively dominated the board for years, using petty bylaws and threats of litigation (he's a lawyer) to coerce more and more power for him and his voting block, using regressive rules and policy to exclude people and prevent change. They had a supermajority of people in support of STAR-PR but were unable to get to vote it in while everyone was there due to delay tactics, so it didn't pass. This guy is opposed by a supermajority on the board but the neighborhood isn't engaged and is turned off by the stagnation so they don't know the 1/2 of it. Utility would dictate that this guy not get re-elected, but because of the toxic climate they are unable to get many candidates to consider being on this board. As such he is effectively guaranteed a win. The only hope is that they could use the majority requirement against him, but that would require a massive negative education campaign against him and they are afraid of retaliation. 

My assessment is that STAR-PR or any PR would make this scenario worse. Right? With PR he would only need a 1/8 sliver of people to support him to get reelected, even if everyone else hates him. What they really need (for an effective happy board) is a system that can keep hateful, divisive, and toxic players out, while letting non-hateful cultural minorities or other demographics and members on. They'd get a better board with plain old multi-winner STAR Voting where the top 8 win the 8 seats and where that asshole would lose in all/most runoffs. This real world scenario has me stumped as it kind of goes against the equity ethos that seems good. Add this to allegations that PR in Europe helped the Nazis rise to power and I'm concerned. What am I missing here?

Is PR a good idea for actual elected positions of power? Should it be paired with rules to allow voting people off the council if they are deemed abusive/destructive by a supermajority or something? PR certainly is great for a council or advisory board... Knights of the round table. When is PR good and when is it bad? 

[Ted Stern]

Hi Sara,

Watching Portland attempt to get PR and voting method reform is very interesting from the viewpoint of Seattle, which shares some features with Portland.

A couple of years ago, Seattle moved to a 7 district + 2 at-large member city council, after having been a 9 member at-large member council for the previous 25 years or so.  The 7 district members are elected for 4-year terms in the same off-year election as the mayor, while the 2 at-large members are elected for 4-year terms in the other odd-year election cycle, giving voters at most 2 city races to consider in any off-year election.

The 7 district model is fairly well suited to the city's geography and population pattern, but I could see PR being useful also.

Here in Seattle, I would propose a hybrid of PR and districts as follows:

• 4 districts, adjusted every 10 years (using some "fair" relatively compact method such as shortest splitline, Brian Olson's method, etc.).
• 3 PR members per district, with the 12 total PR members chosen in the same odd-year election as the mayor.
• 3 city-at-large PR members, chosen in the alternate odd-year election.

Seattle is now large enough that I think a 15 member council would give reasonable representation.

With Droop quotas, this means that a council member would need 25% of a district or 25% city-wide to get elected.  This is small enough to enable minorities and diverse political stripes to be represented, while still being a high enough threshold to avoid problems.

Any thoughts on whether this would work in Portland?

[Sara Wolf]

Interesting! Are they talking about doing something like that with STV? With a recent change like that it's hard to imagine Seattle going in for a whole new redesign, plus Democracy is illegal in WA. No wonder Sightline is looking at Portland. I think Portland could do something similar with a couple districts and PR reps within each. A hybrid of districts and PR is imo the only PR option that has any chance. I also think that having less seats in each PR election could be a good thing. 1/3 of the vote is a reasonable minimum quota. Less than that and concerns around polarizing candidates become more valid. 

--

-Sara Wolk

Chief Petitioner for "STAR Voting for Multnomah County"

Portland Equal Vote

971-222-9364

sa...@equal.vote

​www​

.

​starv​

o

​ting​

​

.

​

​us​

www.equal.vote

“The fact is that FPTP, the voting method we use in most of the English-speaking world, is absolutely horrible, and there is reason to believe that reforming it would substantially (though not of course completely) alleviate much political dysfunction and suffering.”

-Jameson Quinn, The Center For Election Science

[Jameson Quinn]

Responses inline

2018-05-16 21:56 GMT-04:00 <sa...@equal.vote>:

Thanks Jameson, 
I do like 3-2-1 and I'll look more into PAD. I'm glad you're working on it and I hope it gets a real world trial. I don't think Portland or OR will be that place. I do mention 3-2-1 in Q and A's a lot when VSE comes up or if I say that there are lots of systems out there but in general we're well past the "comparing systems" stage. It was an insanely monumental task to get people to unite around STAR to the extent that they have, and I don't think a total course change would be possible any time soon, even if it was desirable.

Understood. That's pretty much what I expected. 

Personally I think 3-2-1's biggest obstacle is in the explanation. Not that it's that complex or anything, but the non-numerical and proxy components are barriers to getting people on board in the first place. Obviously if it was implemented I think everyone could figure it out, but still. I think your best bet would be rebranding it to somehow use a "ranking" ballot type explanation, but allowing ties, but because of IRV people would get confused by that. Tricky. For your sake I hope I'm wrong. It's an innovative approach and we need that kind of thinking to solve these questions. 

For me, the non-numerical aspect of 3-2-1 (heh, despite the name) is actually helpful in understanding it. With numerical scores, people will inevitably think in terms of totals, and that leads to exaggeration strategy. With verbal categories, intent is clear, and it's easy to think of tallies for one category at a time — "3 semifinalists with the most 'good' ratings, 2 finalists with the fewest 'bad' ratings".

 

As for the product testing/polling idea to measure systems usability I also think that has somewhat limited value. In order for people to compare systems they need to understand all the options in some detail. Explaining more than 1 system at a time to lay people adds an exponential magnitude of confusion so I think the results would be kinda meaningless. Doing the same experiment with non-lay people (like this forum) also is of limited value in measuring real world viability because you guys/we think really differently than most people. Maybe the best way to do that would be to get a huge sample and divide it into groups? Each group gets pitched 1 idea only, uses it for a vote, then has to re-explain how to do it and how its counted back to the pollster. Then they rate (0-5 ;) ) how much they like it as a voting system. Then you look at the ratings for their comprehension, explanation, and satisfaction. To do it right would require a huge polling size with a lot of controls and the amount of money required to do it well could do a lot more elsewhere. If anyone can afford it I highly recommend they donate here instead: starvoting.us/donate

I was thinking in terms like the 2004 British Columbia Citizens' Assembly, where over 100 people randomly selected from around the province took almost a year of meetings, and eventually came close to consensus. Obviously something like that is a huge undertaking. Even if you scaled it back to, say, 25 people and a few days of meetings, it would be a lot. But I think that something at that latter scale would actually be able to compare voting methods and appreciate their relative pros and cons, if they had access to a good set of experts. (Of course, I'd want to be one of those experts.)

 

Viability is a whole other question than usability. What I mean by viability is: can it get passed and used in the current political climate and would it get repealed? If important political decisions we're actually made on the merits of the ideas themselves we'd live in a very different world. Ugh. 

I think that STAR, 3-2-1, PLACE, and PAD are all viable, or at least, will be if and when they are better-known. I'd also put majority judgement in that category, as well as the P-STAR I described in my previous message.

I think that IRV and approval are probably viable, but I have reason to doubt that, especially for IRV. Even if these methods were in use, I think that there would be people with legitimate and fixable complaints about them, and I fear that this could end up regressing to plurality.

I think that STV is not viable at the state or federal level in the US, but is at the municipal level. Unfortunately, that makes it a bit of a dead end.

I think that STAR-PR is a mess and not viable at any level.

 

Re: honesty, simplicity, and expressiveness.
Simplicity and Expressiveness are definitely inversely correlated. To me in order to have full honesty and accuracy you need a minimum degree of expressiveness. I think a minimum there is Favorite, Pretty good, better than worst case scenario, worst. Neutral would be good too. 

For a minority voter in a red state there's a huge difference between a pretty good second choice and a lesser evil who is only worth considering in order to prevent a worst case scenario. 

The further out of the local mainstream a voter is, the bigger the difference between a purely-expressive and a (perfect-knowledge) purely-strategic ballot is for them. 3-2-1 already covers all voters whose first or second choice is either of the two true frontrunners. Say you have a highly divided election where the percentage which favors each candidate first goes 30%, 25%, 16%, 12%, 8%, etc. In 3-2-1, the voters who will be able to be both expressive and strategic will be at least 30% + 25% + (16% + 12%)/2 = 69%, a solid majority. Meanwhile, the 8% voting for candidate E might be in the situation you suggest, so that they end up strategically voting something like E=D>C=B>A (that is, not expressing their true E>D preference), but that is still a lot more expressivity than they'd get under plurality or even approval. Or, they could vote E>D=B=C>A, and the chances that this would actually be strategically weaker are low.

It is possible to modify 3-2-1 or PAD to have a 4-level ballot, while still keeping honest and strategic ballots the same for most voters? Well, kinda, but it adds complexity and compromises honesty. I can imagine what those 4-level methods would look like, and they're not too bad, but still not as good in my book. And going to 5-level ballots, within the 3-2-1 or PAD framework, is quite definitely a bad idea in my book; strategically speaking, the middle rating would basically never be a good idea.

 

The fact is the 5 star rating is already a thing. In an alternate universe 0-3 or 0-6 or 0-9 could all be fine, but in our universe going with something already used around the world is huge. STAR Voting ties neatly into our own design as humans. 0-5 scoring is encoded in our 5 fingers, and the runoff can be represented by a "raise your hand if you prefer.."

Thumbs up, sideways, or down?

 

Proportional STAR v. STAR-PR and algorithms in general:
I confess that I'm not a numbers person. I'm glad you all are. I love a good thought experiment and am a visual thinker. To convince me, someone would have to explain the algorithms and such in words, (as I did in starvoting.us/pr). I don't get why STAR-PR as proposed "is broken" or "fails" proportionality while RRV would "pass". I'm not even totally clear on what the definition of  "proportionality" here is. I thought someone mentioned a Warren PR criterion but looking back I didn't see the comment. As I recall the link for that was incomprehensible anyways. 

I'll try to do this later, but I don't have time to do a good job right now. 

STAR-PR story problem: The analogy that makes some sense is that voters are hiring their candidates. Each voter starts out with an equal budget to spend. Voters fill out their ballots showing via scores how much they are willing to pay each based on how represented they would feel by each candidate. During the tabulation you only spend some of your vote if a candidate you liked gets hired. How much you spend depends proportionally on how much you liked them. Once all your "money" is spent you are fairly represented. The result is perfect Equity. Is that wrong? Is that still the concept for your P-STAR suggestion?

In the P-STAR suggestion, the whole group together decides which candidate to "hire" next, and then the people who are happiest with that decision spend all their money on that hire and leave the group. At the end, each winner represents the same number of voters.

  

Proportion of what? Honestly, I also am a bit concerned about what exactly is getting proportioned out and if that actually leads to the kind of representation many find so attractive. With geographical districts, each local area is equally represented. Since the goal of most local governments is providing services to their constituents that makes a lot of sense. Party list systems proportion out seats based on ideology with each faction getting fairly represented. (If our parties were less broken this might be more attractive.) Many seem to want demographical/racial/gender equity. We just had an election yesterday here and many candidates ran on a platform based around who they plan to represent (seniors, poor people, young people, Latinos, Black people.) Others ran platform based campaigns (gun rights, housing crisis, tax reform, election reform for city council.) RRV/STV systems seem to do some combination of all or none of those so I wonder if different voters focus on being represented by different metrics if that works out in the end? Does perfect mathematical proportionality = good representation across the board? 

Again, responding fully to this takes more time than I have right now.

 

Real life is messy:

Recently we were approached by a neighborhood that wanted to adopt STAR-PR. Their last two at large plurality elections (with a majority requirement) had run into major issues and people were effectively gaming the system using coordinated strategic block voting. Their stated goals and reasoning for choosing STAR-PR was to help all people, including people of color have a voice and be fairly represented. They wanted to encourage better and more candidates to run and have and more open elections. This was a part of a package deal that included everything from paper ballots and early voting, to better voter verification, to better bylaws for a more inclusive process. 

Unfortunately the reality of the problem was that they have 1 super-toxic board member who has effectively dominated the board for years, using petty bylaws and threats of litigation (he's a lawyer) to coerce more and more power for him and his voting block, using regressive rules and policy to exclude people and prevent change. They had a supermajority of people in support of STAR-PR but were unable to get to vote it in while everyone was there due to delay tactics, so it didn't pass. This guy is opposed by a supermajority on the board but the neighborhood isn't engaged and is turned off by the stagnation so they don't know the 1/2 of it. Utility would dictate that this guy not get re-elected, but because of the toxic climate they are unable to get many candidates to consider being on this board. As such he is effectively guaranteed a win. The only hope is that they could use the majority requirement against him, but that would require a massive negative education campaign against him and they are afraid of retaliation. 

My assessment is that STAR-PR or any PR would make this scenario worse. Right? With PR he would only need a 1/8 sliver of people to support him to get reelected, even if everyone else hates him. What they really need (for an effective happy board) is a system that can keep hateful, divisive, and toxic players out, while letting non-hateful cultural minorities or other demographics and members on. They'd get a better board with plain old multi-winner STAR Voting where the top 8 win the 8 seats and where that asshole would lose in all/most runoffs. This real world scenario has me stumped as it kind of goes against the equity ethos that seems good. Add this to allegations that PR in Europe helped the Nazis rise to power and I'm concerned. What am I missing here?

This is an interesting example. You're right that under almost any PR method, as long as there are 1/8 of voters who prefer that guy over all others, he will get a seat. Any mechanism that would allow the non-toxic (super)majority to veto his election, would also allow a racist (super)majority to veto a minority candidate. One could imagine a mechanism that would let a supermajority veto just one candidate, so that racists could knock down only one of (presumably) several minority candidates... but even though this is feasible, it mostly amounts to adding epicycles that would leave the voting method impossible complex.

But! I actually can seriously propose a method that would solve this: PLACE (or LPR, or any other strictly biproportional method). In these systems, seats are tied to districts, so in order to get this guy off the board, voters would only have to make sure one of his same-district opponents wins.

  

Is PR a good idea for actual elected positions of power? Should it be paired with rules to allow voting people off the council if they are deemed abusive/destructive by a supermajority or something? PR certainly is great for a council or advisory board... Knights of the round table. When is PR good and when is it bad? 

I think it's pretty much always good, but explaining all my reasons would take pages.

[Ciaran Dougherty]

> Ernest Naville: "In a democratic government the right of decision belongs to the majority, but the right of representation belongs to all."

This is an excellent quote, and actually highlights some of what concerns me about STAR as a single winner method: in cases where the Runoff step selects differently than Score would have, it lessens the representation of the people who preferred the Consensus candidate.

> With fewer levels, there's simply fewer options for strategic voting.

That's one way of looking at it.  The other way of looking at it is that it forces strategic voting, because there are fewer options for honest voting.  
I would even go so far as to argue that saying it allows fewer options for honest voting would be more accurate, given that according to the VSE Simulation, shifting from an 11 point range to a 3 point range for either Score or STAR  has little to no effect on the 100% Strategic vote results, but does significantly cut down on the benefit of 100% honest voting.

Indeed, given the fact (when normalized) a smaller range approaches mathematical equivalency with the accepted definition of strategic voting, saying that decreasing the range helps prevent strategy is like saying that if you put down your dog, it won't get run over by a car: yes, it's true, but I don't understand how causing the problem that you are trying to prevent actually improves the situation.

> It's much less of a problem if it's just a matter of collapsing distinctions that are relatively minor to begin with.

I disagree that changing the distinctions makes them minor changes.  What is the difference between changing a 5 to a 10 (on 0-10) vs changing a 1 to a 2 (on a 0-2)?  In both cases, you're changing 50% of maximum possible support to 100% of maximum possible support; all you're doing is making the numbers smaller, at the expense of a voter having the option of giving a candidate more accurate levels of support.

> But there are a lot of voters who'd rather choose a favorite, perhaps vote weakly for or strongly against a few others, and go home, without worrying about those fine distinctions

I'll agree with that for many elections, but that becomes problematic with several real world examples.  Seattle recently had a mayoral race with 21 candidates. It's currently filing week, here, and there are no fewer than 28 candidates running for the WA Senate Seat this year.  I'm not certain that a 3 way distinction could accurately reflect the voters' intent.

The questions I don't answer below are those I agree with Jameson's assessment.

> 1. Will the final proposal have reweighing on every round and runoff? Some? None (aka is the best STAR-PR RRV?)?

I largely agree with Jameson on this, though I'm not convinced that the Removal Priority he uses in his version of P-STAR is better than the "Difference from Ballot-Mean" that I originally proposed.

> 2. What quotas are best?

I have to argue in favor of Hare, because of the Naville quote above: if you use Droop, slightly less than one Quota doesn't really have representation (or, put another way, the last seat corresponds to nearly twice as many voters as the others).

> 4. Is it desirable to have an algorithm that would put Nazis or other super antagonizing and polarizing candidates at some level of disadvantage while still allowing non-hateful minorities to get elected?

What happened to "let's not use triggering scenarios?"

But no, even if such an algorithm were possible (prioritizing minorities based on your analysis of their values rather than their community's), which I don't believe it is, I strongly question whether such actions are a good idea.

Steven Pinker makes a decent argument that attempting to silence such people actually helps them gain traction.  Let the snotwaffles have their <5% of a legislature, let their ideas be neatly shattered by the >35% of the population who hate such ideas and will speak out against them.  Find some way to exclude them, and they will leverage that (reasonable, justified) oppression to recruit.

> 5. Should we round up or down for quotas?

I agree that it's largely irrelevant.  If you're using Droop, it's predefined: ceiling(Votes/(Seats+1)).  If you're using Hare, it might be best to simply Round (privileging neither ceiling nor floor), as such bias would increase the size distortion of the last seat.

>  6. Is there a geographic scale that's too big for non-precinct summable systems or too big because geographical representation is important too?

I would modify Jameson's response to say that for "non-precinct summable" voting methods, it's not actually the number of candidates, but the number of possible ballot types (which is, in turn, a function of candidates).  Ordinal methods would have to return a number for every possible ballot type (on the order of n!), and star would have to STAR would have to return n+(n choose 2) values.

> 7. Is there a nice hybrid of single winner and PR that can offer both kinds of proportionality? (ie. STAR-PR House and STAR Senate.)

What do you mean by "kinds of proportionality"?  I don't see what the question is; with sensible multi-seat voting methods, the single seat version is nothing more than the multi-seat version with a seat count of 1.  That's (perhaps the only) good thing about IRV/STV: once people understand the (slightly more complicated) multi-seat algorithm, the single seat one falls out.

[Ciaran Dougherty]

> I don't get why STAR-PR as proposed "is broken" or "fails" proportionality while RRV would "pass"

I think you're making assumptions, there; RRV doesn't actually pass proportionality to any greater degree than Reweighted-STAR does. 

Indeed, I originally designed and offered (a slightly different) version of the algorithm Jameson has been calling P-STAR for Score/Range voting because I observed that RRV was not properly proportional, especially with slates of candidates or as Party List elections (such as would be the case with proportional allocation of Presidential Electors).  . 

The numbers I ran based on California's 2016 Presidential vote showed that under RRV, Clinton and Trump would have gotten 100% of the California Electors between them, despite the fact that Johnson and Stein voters numbered enough to warrant 1.85 and 1.08 electors, respectively.

> With PR he would only need a 1/8 sliver of people to support him to get reelected

He only needs approximately that number currently, doesn't he?  If he represents one of eight districts, he actually only needs support of roughly 1/16th of the electorate at large, because he has a simple majority of his district.  On the other hand, if his district is an "At Large" one, he's already got (at least) a plurality of the voters supporting him, so they'd be no worse off.

> What they really need (for an effective happy board) is a system that can keep hateful, divisive, and toxic players out, while letting non-hateful cultural minorities or other demographics and members on

And how do you imagine such a system would work, precisely?  If such a snotwaffle has as many or more votes than a reasonable candidate, how could you justify excluding them without "making democracy illegal"?

> Is PR a good idea for actual elected positions of power?

That depends: Do you consider it better to leave more, or less, of the electorate without representation?  If you agree with Ernest Naville, Jameson, and myself, then yes, giving a greater percentage of the populace a voice is better, and that's what PR does.

> Should it be paired with rules to allow voting people off the council if they are deemed abusive/destructive by a supermajority or something?

If the definitions of "abusive" and "destructive" are objective, then yes, that's a good idea regardless of whether you're using single-seat elections or actually proportional systems.

> When is PR good and when is it bad? 

Again, I agree with Jameson: the only scenarios in which (I believe) that someone could reasonably conclude that PR is bad (inherently single seat positions, such as Executive, notwithstanding) would be in scenarios where you did not actually believe in democracy; the inherent definition of PR is that it more accurately reflects the will of the people.

On Wed, May 16, 2018 at 6:56 PM, <sa...@equal.vote> wrote:

[sa...@equal.vote]

Re: Sara: STAR-PR story problem: Jameson: This is an interesting example. You're right that under almost any PR method, as long as there are 1/8 of voters who prefer that guy over all others, he will get a seat. Any mechanism that would allow the non-toxic (super)majority to veto his election, would also allow a racist (super)majority to veto a minority candidate. One could imagine a mechanism that would let a supermajority veto just one candidate, so that racists could knock down only one of (presumably) several minority candidates... but even though this is feasible, it mostly amounts to adding epicycles that would leave the voting method impossible complex.

But! I actually can seriously propose a method that would solve this: PLACE (or LPR, or any other strictly biproportional method). In these systems, seats are tied to districts, so in order to get this guy off the board, voters would only have to make sure one of his same-district opponents wins.

  

It's not a big neighborhood so subdividing it is not an option. I asked. Turnout is low enough as it is. Part of the problem is it's hard to even find 8 good candidates in the first place to do a no pay, boring volunteer job. This toxic member thing is actually a big enough problem that there's talk of cutting all funding for all Neighborhood Orgs. This guy is notorious and there are others like him that have taken over and derailed everything on other boards. It's a pattern. 

Here in PDX racists aren't a problem on these boards. The odds of a (super) majority of racists getting elected and blocking people of color is zero. On the other hand the asshole bureaucracy-enthusiast problem is real and current. 

Re: "When is PR good and when is it bad?"

I'm starting to think that PR is only good if the seat up for election is highly desirable? If there is low incentive to run then you risk having not many more candidates than seats. This means that almost anyone who runs can win even if they're barely supported by anyone. How many "extra" candidates are needed in a given race in order for there to be a culling where only the best win?

Re: Ciaran: I would even go so far as to argue that saying it allows fewer options for honest voting would be more accurate, given that according to the VSE Simulation, shifting from an 11 point range to a 3 point range for either Score or STAR  has little to no effect on the 100% Strategic vote results, but does significantly cut down on the benefit of 100% honest voting.

Exactly!

RE:  Ciaran "What do you mean by "kinds of proportionality"?

This is referencing my "what are we proportioning out" question. Geographical areas? Focus on specific issues or platforms? Cultural, ethnic gender or Identity based representation? A candidate recently said something like this: "There are three types of groups working in politics: Geographical, Idealogical, and Identity based groups. If you can get all three working together then you can accomplish anything." 

My concern is that by trying to proportion out all three at once you end up not proportioning any of them out fairly. Even if "perfect" proportionality is met in a mathematical sense. In some ways we are dividing apples and oranges. I recently read about the Kurdish clans in the N Syrian mountains. They're experimenting with Democracy and have a novel type of council in each town. Each town has a council and each (4 person?) council is required to have 1/2 women. Each is also required to have one rep from each ethnic group. (Kurd, Turk, Armenian) or something like that. Their population is roughly evenly split between the 3 identity groups. I'm sure I've messed up those details, but you get my point. It's interesting and I'll try and find the link. A system like that would fail Warren's PR criteria because it has nothing to do with math and everything to do with guaranteeing proportional outcomes. (Disclaimer: I'm not advocating anything here, I'm not convinced of anything yet.)

 

[sa...@equal.vote]

Hi Ciaran, thanks for the clarification around RRV. I thought you guys thought it was PR. I'm looking forward to a "why not" answer in laymens terms because it sure seems proportional to me. I believe you, and that demo example helps there, but I want to get why. 

> With PR he would only need a 1/8 sliver of people to support him to get reelected Ciaran: "He only needs approximately that number currently, doesn't he?"

No, it's currently "At Large" and you have to be voted in by 50% of voters. So any candidate who doesn't meet that threshold doesn't get elected. If 8 people don't make the threshold they actually just end up with a smaller board and they have to have another election to add more people. This actually seems kind of ideal for someday voting the asshole off so I'm secretly a bit glad PR didn't pass and they weren't able to get rid of that majority criteria. The problem is that anyone with name recognition at all can get the "majority" of voters, and most voters have no clue about the actual politics beyond being turned off in general. They aren't paying enough attention to figure out who the problem actually is and take sides. 

> "What they really need (for an effective happy board) is a system that can keep hateful, divisive, and toxic players out, while letting non-hateful cultural minorities or other demographics and members on..." Ciaran: And how do you imagine such a system would work, precisely?  

I don't know, that's why I was asking. I'm not even saying it's desirable necessarily... But here's a stab at it: Polarizing hateful minorities (racists or assholes) are more likely to be given 0s by anyone outside of their block. Non-polarizing minorities (other identity groups like hippies, people of color, rural people, etc. and non assholes) are likely to get some support from people outside of their block. One of my volunteers, Dan Adler, described it as negative vs positive bullet voting. A negative bullet vote in a 4 candidate race is: 5, 5, 5, 0, this is a strong vote against a highly polarizing candidate. A positive bullet vote is 0, 0, 0, 5, is a strong vote for your favorite. A system that gives advantage to negative bullet voters and disadvantages positive bullet voters would prevent polarizing candidates. 

Again, I'm not advocating for anything, just theorizing... but wouldn't doing RRV and then adding a non-reweighed runoff for each seat give that kind of outcome? The good guy off that neighborhood I mentioned, Matt, had that idea. I don't think this passes "Equality" criterion, but it would give PRish results with an twist to give coalition builders an edge over divisive candidates. If it worked as we're thinking it might I think that would lead to better legislation and outcomes and a more constructive council or board, even though it sacrifices mathematical proportionality in the process. It's kind of a compromise between PR and At Large. 

> Is PR a good idea for actual elected positions of power? Ciaran: "That depends: Do you consider it better to leave more, or less, of the electorate without representation? If you agree with Ernest Naville, Jameson, and myself, then yes, giving a greater percentage of the populace a voice is better, and that's what PR does."

This is getting good! So, single winner STAR Voting already gives everyone a voice, which = representation and the ability to hold candidates accountable. I think we are all entitled to a voice, to not have our vote wasted, and have our vote actually make a difference. Single winner STAR doesn't give everyone an elected rep, but it gives them a voice that is never wasted and always makes a difference. This is actually why I love it so much as a single winner system. Even if none of my favorites can win, my vote in the runoff goes to the finalist I preferred if I had a preference. This is why there's a huge strategic incentive for minority voters to actually show their full honest preference order. 

So let's take a minority voter for example: Even if none of my favorites can win, my runoff vote is still 100% powerful and helps prevent my worst case scenario. This allows a black woman in the rural south to vote her conscience and also vote against the KKK with all her might because she can give a less hateful republican 1 star. Similarly a right wing extremest in Portland can vote his conscience and still give 1 star to a candidate who isn't anti-guns and vote against a candidate that is working to repeal the 2nd amendment. Both of these votes matter and help the voter have a better outcome and better representation, even if nobody they love gets elected. 

We are all entitled to a voice, but we don't all deserve power over others. Getting elected = power over others. How much support does a group need before they have earned that power? That's the question. In math a Droop or Hare quota makes sense. In real life I'm not sure it's that simple. What quota actually offers the best utility in outcomes? I don't know. Is is STAR or proportional STAR or something in between or a combination? With fair districts single winner STAR offers perfect geographical proportionality and local governance. Bigger districts = less local representation and less accountability. 

Question: Do you want 100% direct democracy where every person gets to vote on every issue, even if they have no background and don't have an educated opinion? Or are we better served by having less people, but those people do their homework and actually figure out the best option? An example is the anecdote about the recession (or climate change) where the introverted wonks we're warning the board for years that a train-wreck was eminent. The non-wonks (who were profiting from things as they were) didn't listen and didn't care because they didn't understand the issue at stake. 

Off the cuff it seems like having as many voices at the table and heard is essential in drafting good policy and developing a broad perspective in the first place, but perhaps it's best to have a team that's non-biased and well qualified voting that legislation in. 

In conclusion: Sorry to get esoteric here. "Where and when do we want PR?" is a totally different question from "what is the simplest algorithm for perfect STAR-PR."

[Jameson Quinn]

Focusing on one part of the debate:

 

> With fewer levels, there's simply fewer options for strategic voting.

That's one way of looking at it.  The other way of looking at it is that it forces strategic voting, because there are fewer options for honest voting.  
I would even go so far as to argue that saying it allows fewer options for honest voting would be more accurate, given that according to the VSE Simulation, shifting from an 11 point range to a 3 point range for either Score or STAR  has little to no effect on the 100% Strategic vote results, but does significantly cut down on the benefit of 100% honest voting.

Indeed, given the fact (when normalized) a smaller range approaches mathematical equivalency with the accepted definition of strategic voting, saying that decreasing the range helps prevent strategy is like saying that if you put down your dog, it won't get run over by a car: yes, it's true, but I don't understand how causing the problem that you are trying to prevent actually improves the situation.

My model is that there are basically 4 kinds of voters, in terms of strategic heuristics:

Type A: Always be as honest as possible.

Type B: Rule 1, don't be a chump. Rule 2, be honest/expressive if it doesn't conflict with rule 1. "Don't be a chump" means, be defensively strategic if it's necessary; that is, if there's a plausible risk that you could lose due to offensive strategy. "Plausible risk" usually means "assume the worst about the number of voters who disagree with me who are type C"; that is, make the highest plausible guess about how many there will be, which will tend to mean assuming odds of type-C-ness that are at least twice as high as the true odds.

Type C: Always be as strategic as possible.  

Type D: Be strategic if you think most other people like you are being strategic; be honest if you think that most other people like you are being honest.

I think that there will generally be more voters of types B and D than of types A and C, and that the overall fraction of type C voters could systematically vary across groups from barely above 0% to around 20%, and that it's likely to be higher in more extremist groups (who are a priori likelier to be very wrong, though almost as likely to be about right).

Fewer ratings levels prevents a "strategic slippery slope", because it's comforting to type B voters. They'll see fewer type C's on the other side (because there will be less opportunity to take honest votes as strategic votes), so they'll feel safer being honest themselves.

> It's much less of a problem if it's just a matter of collapsing distinctions that are relatively minor to begin with.

I disagree that changing the distinctions makes them minor changes.  What is the difference between changing a 5 to a 10 (on 0-10) vs changing a 1 to a 2 (on a 0-2)?  In both cases, you're changing 50% of maximum possible support to 100% of maximum possible support; all you're doing is making the numbers smaller, at the expense of a voter having the option of giving a candidate more accurate levels of support.

I'm saying that if your true "realistically-normalized" (that is, normalized on a scale that leaves the worst plausible outcome at 0, even if there are unserious candidates who are worse than that) rating for a candidate is 7 on 0-10, then rounding that to 1 on 0-2 is "minor", and if it's 8 then rounding that to 2 is "minor". I actually think that people's prospective utility judgements are noisy enough that VSE overestimates the quality of any method with more than about 4 or 5 rating levels.

> But there are a lot of voters who'd rather choose a favorite, perhaps vote weakly for or strongly against a few others, and go home, without worrying about those fine distinctions

I'll agree with that for many elections, but that becomes problematic with several real world examples.  Seattle recently had a mayoral race with 21 candidates. It's currently filing week, here, and there are no fewer than 28 candidates running for the WA Senate Seat this year.  I'm not certain that a 3 way distinction could accurately reflect the voters' intent.

Of those 21 or 28 candidates, by the time election day arrives, I'd wager that no more than 3-6 are seriously "in the race". 

[Jameson Quinn]

> 2. What quotas are best?

I have to argue in favor of Hare, because of the Naville quote above: if you use Droop, slightly less than one Quota doesn't really have representation (or, put another way, the last seat corresponds to nearly twice as many voters as the others).

In practice, around half of the "remainder" in the last seat will be voting against the winner. So with Hare, you get a last seat that represents about half a Hare quota, and half a Hare quota of wasted votes; with Droop, you get a last seat that represents about 1 Droop quota, and 1 Droop quota of wasted votes. So Hare is "better" in terms of having fewer wasted votes, but Droop is "better" in terms of having winners each representing an equal proportional share of voters — that is, not over-representing one arbitrarily-chosen minority view. I think that the harm of 1 actively-harmful representative is worse than the harm of 1 missing representative, so I prefer Droop.

[Jameson Quinn]

2018-05-18 21:07 GMT-04:00 <sa...@equal.vote>:

Re: Sara: STAR-PR story problem: Jameson: This is an interesting example. You're right that under almost any PR method, as long as there are 1/8 of voters who prefer that guy over all others, he will get a seat. Any mechanism that would allow the non-toxic (super)majority to veto his election, would also allow a racist (super)majority to veto a minority candidate. One could imagine a mechanism that would let a supermajority veto just one candidate, so that racists could knock down only one of (presumably) several minority candidates... but even though this is feasible, it mostly amounts to adding epicycles that would leave the voting method impossible complex.

But! I actually can seriously propose a method that would solve this: PLACE (or LPR, or any other strictly biproportional method). In these systems, seats are tied to districts, so in order to get this guy off the board, voters would only have to make sure one of his same-district opponents wins.

  

It's not a big neighborhood so subdividing it is not an option. I asked. Turnout is low enough as it is. Part of the problem is it's hard to even find 8 good candidates in the first place to do a no pay, boring volunteer job. This toxic member thing is actually a big enough problem that there's talk of cutting all funding for all Neighborhood Orgs. This guy is notorious and there are others like him that have taken over and derailed everything on other boards. It's a pattern. 

In this case, "PLACE by astrological sign of voter" would be a good solution, with non-incumbent candidates able to choose which sign they ran under. The asshole in (say) the Libra seat would draw a Libra challenger; the other incumbents could run unopposed.

Of course, that's not viable as a serious proposal. So... um... limited vote? Is the only answer I can think of which might work, and doesn't end up systematically disenfranchising a minority if most voters are racists. But it's a tough problem.

I'm starting to think that PR is only good if the seat up for election is highly desirable?

I agree that it's bad to have too many things up for election. If it's one of many boards of something-or-other, just have signature-driven nominations and then let the main elected council make appointments off of that list.

This is referencing my "what are we proportioning out" question. Geographical areas? Focus on specific issues or platforms? Cultural, ethnic gender or Identity based representation? A candidate recently said something like this: "There are three types of groups working in politics: Geographical, Idealogical, and Identity based groups. If you can get all three working together then you can accomplish anything." 

My concern is that by trying to proportion out all three at once you end up not proportioning any of them out fairly. Even if "perfect" proportionality is met in a mathematical sense. In some ways we are dividing apples and oranges. I recently read about the Kurdish clans in the N Syrian mountains. They're experimenting with Democracy and have a novel type of council in each town. Each town has a council and each (4 person?) council is required to have 1/2 women. Each is also required to have one rep from each ethnic group. (Kurd, Turk, Armenian) or something like that. Their population is roughly evenly split between the 3 identity groups. I'm sure I've messed up those details, but you get my point. It's interesting and I'll try and find the link. A system like that would fail Warren's PR criteria because it has nothing to do with math and everything to do with guaranteeing proportional outcomes. (Disclaimer: I'm not advocating anything here, I'm not convinced of anything yet.)

There is actually math on how to ensure constraints and yet be voting. Tommy Ratliff is the name I know in that subfield. But this is probably more theoretical than practical.  

[Sara Wolf]

RE: Jameson: So with Hare, you get a last seat that represents about half a Hare quota, and half a Hare quota of wasted votes; with Droop, you get a last seat that represents about 1 Droop quota, and 1 Droop quota of wasted votes. So Hare is "better" in terms of having fewer wasted votes, but Droop is "better" in terms of having winners each representing an equal proportional share of voters — that is, not over-representing one arbitrarily-chosen minority view. I think that the harm of 1 actively-harmful representative is worse than the harm of 1 missing representative, so I prefer Droop.

​So if we're willing to accept this kind of margin of error or inaccuracy due to wasted votes (aka ratio of voters not matching number of candidates?) why not specifically aim that that extra margin never goes ​to the polarizing candidate? Sounds like that would be more like Hare, but neither of the above, am I right? 

​I started 2 new threads. One for PR theory/goals and also one for PR ​algorithms. It seems like this thread is trying to cover a lot all at once. Feel free to paste stuff from here to refocus the conversation. I'm still looking for the 0-5 PR system we could come together around. 

​Maybe the Neighborhood example ​conversation should stay here in this thread? I originally had hoped that the neighborhood could be a model for PDX city council but I now don't think that's possible as the situation is really different. I'm leaning towards plain old STAR Voting multi-winner with a minimum average of 3 stars. The board would have as many members as received a passing rating and it would be up to board members to recruit and vote in more members if the minimum wasn't reached between elections each year. The thing is the vast majority in SE PDX is totally happy to elect people of color so we don't need pr to get those outcomes in this specific scenario. We do need to be able to hold elected officials accountable and vote them off. That's the real need. 

Eek, are Equity and Accountability negatively correlated? It sure seems like the ability to vote expressively and honestly is negatively correlated with strategic resilience. We picked the 0-5 scale for star specifically based on Jameson's calculations on VSE and strategy resilience. We thought 0-5 was the best balance of honesty vs accuracy. In other news FV has agreed to take down the three worst articles against STAR, Score, and Approval!!! Yay!!  It's in a thread in our Loomio group "Cross Team Dialogue." Their attack article on Score sucks, but it had some valid points about different voters using the scale differently, it's a lot like what Jameson is saying above actually. I think 0-5 as opposed to 0-10 does a lot to address that concern. 

What is limited vote? 

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.
To unsubscribe from this group and stop receiving emails from it, send an email to electionscience+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ciaran Dougherty]

If there is low incentive to run then you risk having not many more candidates than seats. This means that almost anyone who runs can win even if they're barely supported by anyone. How many "extra" candidates are needed in a given race in order for there to be a culling where only the best win?

How many "extra" candidates does a multi-seat method need to return a better result?  One.

If you have an 8 seat body, elected in single-seat elections, and you have 9 candidates, there is a pretty decent chance (I believe it's upwards of 85%, and I'm sure others can correct my statistics) that you're going to end up with the 9th best candidate elected instead of someone who is better than them, simply because your elections were something like 1|2|3|4|5v8|6|7|9.  Indeed, without multiseat (PR) elections, it's possible that the 2nd best candidate overall would lose because they happened to run for the same position as the best overall candidate.

This is referencing my "what are we proportioning out" question. Geographical areas? Focus on specific issues or platforms? Cultural, ethnic gender or Identity based representation?

The only question that someone who believes in Democracy should truly care about at all in that case is how significant a role Geography should play in the question.  It is clear, I would think, that the people of NYC, or Chicago, or LA, would have vastly different concerns (gun control, crime, public transportation) than someone in grain belt (water rights, farm bills, etc).  Given that, the question is how to ensure that those different areas, with different regional concerns, have their voices heard, rather than being outnumbered by the huge cities (as NYC, Chicago, and LA/SF do to NY, IL, and CA, respectively, every presidential election).

Political Issues?  Identity Politics?  That is not for the voting method, nor voting activists, nor anyone else to decide a priori.  If the voting method is good, then you could have a candidate running a campaign on purely ideological questions, and another running a campaign purely on identity politics, and the voters would decide who wins.  Because that's what democracy is.

[Ciaran Dougherty]

 Single winner STAR doesn't give everyone an elected rep, but it gives them a voice that is never wasted and always makes a difference

You do understand, I trust, that the same (empty) argument could be made of FPTP, right?  Indeed, I'm having a hard time seeing how the Runoff step "always makes a difference" any more than the Top Two primary that CA and WA have currently.

If it doesn't give everyone an elected rep, that means that the people who aren't represented have no voice, by your own "voice = representation" assertion.  Under STAR, if you aren't one of the 50%+1 of those who expressed a relative preference, then your representative has nothing to lose from upsetting you, just as a Republican has nothing to lose from upsetting a Democrat voter that didn't vote for them under the current voting method.

--

[Jameson Quinn]

2018-05-19 18:45 GMT-04:00 Sara Wolf <sa...@equal.vote>:

RE: Jameson: So with Hare, you get a last seat that represents about half a Hare quota, and half a Hare quota of wasted votes; with Droop, you get a last seat that represents about 1 Droop quota, and 1 Droop quota of wasted votes. So Hare is "better" in terms of having fewer wasted votes, but Droop is "better" in terms of having winners each representing an equal proportional share of voters — that is, not over-representing one arbitrarily-chosen minority view. I think that the harm of 1 actively-harmful representative is worse than the harm of 1 missing representative, so I prefer Droop.

​So if we're willing to accept this kind of margin of error or inaccuracy due to wasted votes (aka ratio of voters not matching number of candidates?) why not specifically aim that that extra margin never goes ​to the polarizing candidate? Sounds like that would be more like Hare, but neither of the above, am I right? 

This is the (or at least, a) right question, but wrong answer. You can, indeed, have a method that looks for a "compromise" rather than "polarizing" candidate at each step — the basic STAR rule is a pretty good example of that. But then, when you turn STAR into a proportional version, that means that you have a smaller pool of ballots for the later seats. If you want more compromise, you want that smaller pool to be as big as possible. Thus, Droop has more compromise than Hare. 

In fact, consider the possibility of quotas even lower than Droop. If the quota were 0, you'd just have at-large STAR voting, successively electing N candidates using the same set of ballots — that's maximum "majoritarian" compromise, but no proportionality. So if you had a quota that was less than Droop, you'd have some middle point between majoritarian and proportional. The only problem is, there's not really a "natural" way to set the quota aside from Droop and Hare. 

Separate from the Droop/Hare question is the question of how you decide which ballots to eliminate. I continue to argue that the RRV idea in STAR-PR is broken at a level that's not worth trying to rescue. The right thing to do is put the ballots in some order of "how much they support the winner over alternatives", and remove the first quota of votes. That's what my P-STAR does. I'm not super-attached to the specific order I proposed in P-STAR (highest score for winner, then lowest score for 2nd, then lowest score for 3rd, etc.), but it is at least something I've looked at for a while and not found bad pathologies with. I could imagine another order (perhaps: highest score for winner, then lowest average score for other remaining candidates?) that would also work well. This is in my view a detail with several right answers, though there are still definitely some answers that would be wrong in creating unreasonable levels of free-rider strategy. 

 

​I started 2 new threads. One for PR theory/goals and also one for PR ​algorithms. It seems like this thread is trying to cover a lot all at once. Feel free to paste stuff from here to refocus the conversation. I'm still looking for the 0-5 PR system we could come together around. 

​Maybe the Neighborhood example ​conversation should stay here in this thread? I originally had hoped that the neighborhood could be a model for PDX city council but I now don't think that's possible as the situation is really different. I'm leaning towards plain old STAR Voting multi-winner with a minimum average of 3 stars. The board would have as many members as received a passing rating and it would be up to board members to recruit and vote in more members if the minimum wasn't reached between elections each year. The thing is the vast majority in SE PDX is totally happy to elect people of color so we don't need pr to get those outcomes in this specific scenario. We do need to be able to hold elected officials accountable and vote them off. That's the real need. 

I get that PDX doesn't have a huge problem with a majority knee-jerk voting against a minority. But from the point of view of math, there's really no way to tell the difference between a minority animated by a hateful ideology, and an oppressed racial minority who absolutely deserves representation. So I'm really reluctant to propose anti-minority mechanisms that would prevent a Nazi from winning, because they could in another context just as well prevent a Jew from winning.

But. The problem of "that guy" isn't one of a coherent minority of voters. There isn't 20% of the voters who really want to make sure that the board has at least one of "that guy". There's just a few candidates who are "that guy", and if they were somehow disqualified, the voters that support them could perfectly well find somebody else to vote for who wouldn't be a toxic personality. So it is definitely possible to imagine voting mechanisms that would eliminate "that guy".

I can think of basically two categories of mechanisms that would do that. First, explicit votes against. It's easy to think of dozens of ways this could work, but the basic problem is that it's all pretty overtly about "naming and shaming". If voters like to maintain a comfortable fiction that things are amicable, they won't like any of these mechanisms.

The second category is cage matches — biproportionality. That's what PLACE does, and it's possible to imagine variants. Numbered seats, where you have to choose which seat you're running for. As Ciaran points out, these systems are unlikely to get the optimal result, but as long as there are people willing to run specifically against "that guy", and as long as voters can tell that those alternatives are better, then "that guy" will lose.

Neither of these solutions is perfect. That's why I suggested limited voting: each voter gets X votes, which they can use for one candidate or spread out over several. This is typically known as a "semi-proportional" method, because a coordinated minority can elect their preferred candidate if they want to. But "that guy"'s voters probably won't act as a coordinated minority like that — they'll give him a vote on name recognition, but also spread their other votes among the other names they recognize. So this ends up being kinda like a way to have explicit votes against without admitting that's what you're doing. But it might not work.

[Ciaran Dougherty]

7 on 0-10, then rounding that to 1 on 0-2 is "minor", and if it's 8 then rounding that to 2 is "minor".

Sure, that might be minor, but it is still forcing strategy, and actively distorts the results.  Again, look at the VSE results, which show that 100% Honest is 0.011 better than 100% Strategic for Score0-10, but 0.019 worse for Score0-2.  According to your own simulation, Score0-11 is 0.035 better than Score0-2 for 100% honest voting, and 0.010 better for 50% Strategic, and 0.005 points better for 100% Strategic voting.  To my thinking, that makes it hard to argue that shrinking the voting space doesn't actually improve anything.

 

Of those 21 or 28 candidates, by the time election day arrives, I'd wager that no more than 3-6 are seriously "in the race". 

Turns out, that's a pretty safe bet.... with FPTP, or other voting methods that involve vote splitting.  The Seattle Mayoral race had 5 candidates who won 5/6 of the vote.

I have to question whether or not such a paring down is meaningful with methods that don't limit voter behavior so significantly.  After all, one of the classic scenarios of why I think Range/Score is better is when you have 95% of the populace thinking A or B is ideal, and the other is horrible, and thus the centrist (whom both major factions agree is decent) ends up winning.  Under that scenario only 2 of the 3 are "seriously in the race," but both of them lose to the candidate that isn't "seriously in the race."

But even if you are correct, even if only about a quarter of candidates that file are "seriously in the race," that would mean as many as 5-7 candidates would still be seen as viable, and a 3 way distinction would not allow the voters a meaningful distinction between all candidates perceived as viable, and, again, actively force the electorate into strategic voting. 

At that point, the only of your voter categories (which I largely agree with) that won't be voting strategically would be A: B (IMO the largest group) would feel forced into strategy because honest voting would be less likely to "work out" for them or the community at large, which would then tip the scales for D towards strategy.

[Ciaran Dougherty]

you get a last seat that represents about half a Hare quota, and half a Hare quota of wasted votes

It seems to me that you're still considering the way the last seat will fall based on the premises of our current method; the great advantage of cardinal methods is that the trend to find consensus among the electorate, rather than dominance; Score in particular tends to actively avoid extremists.

 

I think that the harm of 1 actively-harmful representative is worse than the harm of 1 missing representative, so I prefer Droop.

In that statement you are both contradicting your previous assertion of concurrence with Naville, and preuming that an actively harmful individual is more likely to come from a smaller party than not, which is not necessarily a safe assumption.  
Indeed, I assert that an antagonistic individual is more likely to win if they have major backing than if they do not >coughTrumpcough<. 

What's more, I would argue that in many cases, the earlier seats might be more prone to extremism than the later ones, because they will correspond to a subsection of the electorate that is large enough that it doesn't have to compromise with other factions. 

--

[Sara Wolf]

>Re Ciaran: "How many "extra" candidates..?" 

So you're saying if the good candidates run against each other in a district then one of the loses, even if at large they would be in the top x. That makes total sense for districts that are small enough that that local representation isn't needed. My favorite candidate just lost her race because of that. She went up against the Incumbent. I was actually thinking of that question in relation to non-pr STAR multi-winner compared to STAR-PR (ideal version) though. 

>RE: Importance of geographical representation. 

I see the cons of what you are saying, but local representation is also important for accountability, direct outreach and connection between communities and their reps, and because a regional rep has no clue about mundane local stuff like that problematic intersection. In general for economy of scale if quality is a factor I tend to favor mis sized. 

> RE: Me: "Single winner STAR doesn't give everyone an elected rep, but it gives them a voice that is never wasted and always makes a difference"

Ciaran: "You do understand, I trust, that the same (empty) argument could be made of FPTP, right?"

Totally different. For many voters in FPTP there are tons of people who's votes literally and predictably will make no difference to help them get a better candidate. 3rd party voters. Those who's party always wins, those who's party always loses... the list goes on. In IRV your vote is wasted if your first choice is eliminated and your next choice isn't counted. 

With STAR (and score) your vote and scores help your candidate advance, but  even if they can't, if there's a top 2 runoff (instant or not) your vote still goes against your worst case scenario in the runoff. That does make sure your vote isn't wasted. 

Yes, even plurality Top-2 helps prevent wasted votes and makes sure that every vote in the general makes a difference. Unfortunately it also excludes voters from the election that matters most and exacerbates the spoiler effect in the process. I don't support Top 2 because I think there are better ways to accomplish the same things without the problems, but if you're stuck with it for now (like WA), combining it with Score mitigates most of those concerns as much as possible. 

>Re: "If it doesn't give everyone an elected rep, that means that the people who aren't represented have no voice" 

A vote is a voice. The ability to communicate and hold your rep accountable is a stronger voice. An actual rep you like is an even stronger voice. To be clear I absolutely support PR as part of the package. I think we all can agree that there are populations with a small enough demographic that they don't necessarily deserve a rep. Let's say a hate group with 2% of the population. I'm debating how big a group should have to be to deserve a rep and if size is the only important factor or if degree of polarization and coalition building should be factors too.  

[Sara Wolf]

Thanks Jameson, those responses around quotas and everything make sense. I could see Droop quota being good with a cop on number of seats per district. 8 seats on one PR board would mean you need only a sliver of support. ~4 seems much more reasonable. If you want more seats than that your could elect the rest at large or even in a second PR election so the quota ends up being 25%. 

I get your concerns around targeting polarization, but I also see the opposite argument. I expect that the truth is that you're right to an extent. To what extent I'm not sure. It would be nice if there were ways to compare efficacy in action across boards elected with different systems. STAR-PR vs STAR at large vs STAR with a midpoint quota.

 I'm definitely curious to find out more and compare P-STAR and STAR-PR in that new thread. 

>Re: "from the point of view of math, there's really no way to tell the difference between a minority animated by a hateful ideology, and an oppressed racial minority who absolutely deserves representation." 

I hope good minorities would be, on average, better at coalition building than hate groups. Is that a naive assumption? It seems like that could be studied and proven somehow.