Voting in the Senate

[CmdrWilkens]

Without getting into names, unless our PM chat continues and I get the go ahead, one of the longer discussions I've been having recently has been in relationship to HOW we actually implement voting within the Senate. The point which brought it up was the most recent nomination thread and the brief brouhaha it caused amongst folks who realized that with the system of "first past the post" for voting on member nominations there is a huge distortion between votes with one candidate and votes with multiple candidates.

While most of the US based members (though not a majority themselves a significant plurality) are most familair and comfortable with the system it does have weaknesses and defects in multi-candidate races and mroeover in multi-option questions such as the recent Schuyler Colfax vote. What I attempted to do was offer multiple options allowing for second preferences in the event primary preference does not win. The best way of putting it is that I structured a very limited version of an instant runoff vote whereby the given options allow someone to select a first and a second preference. The obvious error, and I freely admit this, is that the way I structured the options you could vote for a less "harsh" punishment as a secondary option but not a "harsher."
In other words I didn't offer the option to say:
1st Preference: TempBan
2nd Preference PermaBan
This obviously could have distorted the voting, though we wont know due to the annonymity of votes, as some of the TempBan voters might have preffered that option if available.

The problem is threefold: If this had been a straight first past the post style vote despite "None" only having 2 votes the remaining three punishments all would have lacked sufficient impetus to pass (PermaBan was 8 votes and TempBan 2 votes shy of the neccessarry majorities) so none would win by default even though there was a clear impetus for SOME punishment. The second is that if we get into multiple run-off votes (say had we eliminated "Title" as an option and re-voted) we would be taking up close to 2 weeks worth of voting plus the time I requested for the discussion on alternatives to emerge. Latly going with the system I used the current board software (there are efforts underway to update this) limits the number of poll options to 10 of which "Abstain" needs to be one.

Current Senate rules give me essentially unlimited discretion in how to structure the votes but given that I think it is worthwhile that we take this thread and cover two key points:

A) Should we continue to use "first past the post?"
- If not what system should we use (including a hybrid with some first past the post votes)
- If so what chages should we make to avoid the twofold problem above (or even if you see that as a problem)

B) If the vote limits/types/categories where it needs to be? In other words is 7 days too long? too short? Is PM notification the best way to go, etc? Are the majorities and quorums reasonable for the votes they represent?

In other words i think we need to discuss this and I'm not neccessarily suggest we actual change the rules (we may need to but we may not) but rather at the least we need to be clear as to how vores will be conducted going forward.

[The Duchess of Zeon]

I'm very in favour of some sort of a Single Transferable Vote implementation, Greg.

[CmdrWilkens]

Thanks Marina,

Guys I'm serious please I'm looking for ideas here, thoughts, motions, short speeches tinged with greatness looking to roll of the tongue...anything. I don't want to just say: This is how we shall vote and have it be law.

[Thanas]

Qui tacet, consentire videtur.

[Stuart]

I think that the existing first-past-the-post system still has much to commend it. It has the virtue of giving unambiguous answers most of the time. Looking back over past votes, it seems to have worked pretty well.

[CmdrWilkens]

Qui tacet, consentire videtur.

Which may be true but its not so much I'm looking for consent as ideas. I have plenty as to how to do this but I'm just one brain and things would probably be better if there were other contributions. I likely, likely, make the final decision but I don't want my thoughts and those I've been exchanging with Starglider via PM to be the only ones.

Beyond that we've had what I think is a rather productive back and forth on how to conduct a better form of preferrential voting I still would like to add some extra voices to the mix. If its just me making a decision with methodological insight from him then that's a poor way to construct a system in my opinion. Sure it could come out sparkly and clean but I'd rather work on the presupposition that there is not uniform agreement on voting system we should use, and moreover even if there is it may not be uniformly in agreement with what I would come up with alone.

[RedImperator]

Well, as I mentioned in the HoC, I like the idea of having the option for preferential voting for when we have more than two options. Not just in the Senate, either; I'm sure people could find creative ways to use it on the rest of the board. I also like the sheer nerdity of it. "Bitch, on our forum, we Condorcet." As long as the code can be implemented without breaking the board, I don't see any reason not to do it.

In the Senate itself, I'm willing to defer to Wilkens's judgment on how any given vote should be conducted. There are circumstances where first-past-the-post will do just fine, and others, like the SC poll, where it produces absurd results. We're not bound to use any one system for every given vote. For votes with multiple options like the SC vote, I'd stick with the system Wilkens just used, warts and all. The only other way that would work is to dispense with the poll and just have people list their choices in their posts.

[Stark]

Preference voting is so familiar to me that it actually strikes me as absurd that anyone would think the current 'one vote, first to xyz wins' makes sense. The Colfax example really highlights the weakness of this system (it worked fine as Wilkins set it up, but it was a vote-option hackjob to get around the limitations and apparently confused people).

I understand proper preference voting may require code changes, however, and I'm not sure if Mike has time to mess about with that.

[Kuroneko]

What I attempted to do was offer multiple options allowing for second preferences in the event primary preference does not win. The best way of putting it is that I structured a very limited version of an instant runoff vote whereby the given options allow someone to select a first and a second preference.

An implementation of various voting rules would be an interesting option to have, but in the meantime, let's explore how the current limitations may be put to maximum use.

Latly going with the system I used the current board software (there are efforts underway to update this) limits the number of poll options to 10 of which "Abstain" needs to be one.

Noted.

The obvious error, and I freely admit this, is that the way I structured the options you could vote for a less "harsh" punishment as a secondary option but not a "harsher."

For four punitive choices, that's eight possibilities: four in the form of "X only", four more ones in the form "X or less" and "X or more" (some of them are equivalent, which is why it's not eight more). We might as well ask if we can use all ten options in an efficient manner.

Implicit in the structure of your previous poll was that there is a general consensus of the punitive rankings. I'm not sure if that's really the case (e.g., one may consider a derogatory custom title either a minor inconvenience or a stigma that follows the person more than a temp-ban would), but if it is, then it's natural to allow a vote for both lower and upper bounds of acceptable punishment.

In other words, out of a set of ranked options {1,...,n}, the voter selects an interval [a,b]. There are C(n+1,2) ways to do that, so for four options (none:title:tempban:permban), there are exactly C(5,2) = 10 possible voting choices, which makes full use of the current limitations. Examples:
-- [X,X] votes for only option X, e.g., [none,none] for no punishment whatever.
-- [title,permban] asserts that some punishment is appropriate, but involves no preference as to what it might be.
-- [none,permban] corresponds is the "abstain" option; it adds to all categories equally, and thus does not affect the final outcome at all.

Under this system, tally counts are unambiguous and straightforward. There is also another possible tiebreaker rule: if X and Y are tied, then X wins over Y iff [X,X]>[Y,Y]. That would be completely optional, since it doesn't resolve all possible ties, but then, we've a rule for that already.

Edit: Since someone still needs to count the tally, a specific order might be best. For example, in the following:

Code: Select all

0[N,P] 1[C,P] 2[T,P] 3[P,P]
4[N,T] 5[C,T] 6[T,T]
7[N,C] 8[C,C]
9[N,N]

the tally for option X is [X,X] and all elements upward or to the left of it (including other rows and columns).

[CmdrWilkens]

Except its not fully representative. With 4 options (along with abstain) lets call them:

A: PermaBan
B: TempBan
C: Title
D: None

The full set of options are:

1. A, B, C, D
2. A, B, D, C
3. A, C, B, D
4. A, C, D, B
5. A, D, B, C
6. A, D, C, B
7. B, A, C, D
8. B, A, D, C
9. B, C, A, D
10. B, C, D, A
11. B, D, A, C
12. B, D, C, A
13. C, A, B, D
14. C, A, D, B
15. C, B, A, D
16. C, B, D, A
17. C, D, A, B
18. C, D, B, A
19. D, A, B, C
20. D, A, C, B
21. D, B, A, C
22. D, B, C, A
23. D, C, A, B
24. D, C, B, A
25. Abstain

That's 25 total possibilities UNLESS we make assumptions about voter preference based upon prior selection. Moreover "Abstain" as a seperate line item is indicative of "no preference" and thus is not included in any of the line chance but must be represented. The only way to shorten this down would be to make a set of assumptions and frankly that opens up the very can of worms I'm talking about.

[Kuroneko]

Except its not fully representative. With 4 options (along with abstain) lets call them ...

Yes, with abstain, it's n!+1. However, if we're voting on the level of punishment, some of them fail to satisfy some natural criteria, e.g., that tempban < permban in terms of punitive value. Those are assumptions, yes, but they don't seem to be a unduly restrictive.

That's 25 total possibilities UNLESS we make assumptions about voter preference based upon prior selection.

Correct.

Implicit in the structure of your previous poll was that there is a general consensus of the punitive rankings.

In the previous poll, you've assumed that title is less punitive than a tempban. I don't necessarily disagree with that. My point is that if the ranking of the (seriousness of the) punishments are agreed upon, one of them being "none", then four choices gives exactly 10 options, including abstention.

Moreover "Abstain" as a seperate line item is indicative of "no preference" and thus is not included in any of the line chance but must be represented.

Incorrect. Abstention is represented as one of the above 10 choices.

The only way to shorten this down would be to make a set of assumptions and frankly that opens up the very can of worms I'm talking about.

The only assumption is that the choices monotonically escalate in punishment, A < B < C < D. You've already made this assumption; this is just an improvement on it.

---Addendum: it somehow got lost in an edit, but in the previous post, my labeling scheme was N = none, C = (derogatory) custom title, T = tempban, P = permban.

[CmdrWilkens]

Except its not fully representative. With 4 options (along with abstain) lets call them ...

Yes, with abstain, it's n!+1. However, if we're voting on the level of punishment, some of them fail to satisfy some natural criteria, e.g., that tempban < permban in terms of punitive value. Those are assumptions, yes, but they don't seem to be a unduly restrictive.

They wouldn't except that I've been hit up via PM for making assumptions of just this type in the previous vote. So its not so much that I disagree that we could narrow it down (Starglider actually sent me this:

3 nominations (A, B, C): limited runoff, a full selection would require 25 choices, but I'd suggest these 10 as the most useful for capturing preferences (order choices by preference). People can express first and second candidate preferences, or chose none and a 'least worst' candidate if none loses.
1 : A, B, (abstain)
2 : A, C, (abstain)
3 : B, A, (abstain)
4 : B, C, (abstain)
5 : C, A, (abstain)
6 : C, B, (abstain)
7 : none, A, (abstain)
8 : none, B, (abstain)
9 : none, C, (abstain)
10 : abstain
sum 1+2, 3+4, 5+6, 7+8+9, if none is not highest eliminate and go to 1+2+7, 3+4+8, 5+6+9, eliminate lowest and apply votes to remaining two, highest wins.

That would probably be the closest match to the situaiton we had (if we take Perma=A, Temp=B and Title=C)


So anyway before I stray further my point isn't so much that I have an issue with making assumptions (which I admit it may have come across as) but rather I think the critical point is that any system we devise needs to acknowledge those assumptions and work within them. Moreover if we ackowledge the assumptions then it serves the purpose I want with this thread which is to make sure the Senate as a whole is aware of the assumptions and the "how the sausage is getting made" side of how I conduct the votes. Moving forward with some sort of ranked approval list and instant runoff voting is how I want to run things but once we get past 2 options (and "none") there have to be assumptions. So long as the Senate is aware and understands what those assumptions are then Im fine. This is sort of a "speak now or forever hold your peace" thread for folks to question the assumptions made.



(And yes this post was sponsored by the word "assume" and its family)

[The Yosemite Bear]

code in that Fgalkin and Marina will always vote for the harshest penalties, and that I will always vote for liency (not true but close enough)

[Kuroneko]

They wouldn't except that I've been hit up via PM for making assumptions of just this type in the previous vote.

You've expressed that the main concern was the lack of choice for a harsher punishment. This deals directly with the problem. If one takes your original scheme, adds the option to do harsher punishments, and also adds the ability to select "middles" (say, "Title or Tempban, but not None or Permban"), one finds that there are exactly ten unique choices, and they can be represented in an easy manner.

So anyway before I stray further my point isn't so much that I have an issue with making assumptions (which I admit it may have come across as) but rather I think the critical point is that any system we devise needs to acknowledge those assumptions and work within them.

Then what's the problem? For the punitive case, the only assumption is that voting choices are restricted to be consecutive. That's very natural when the options escalate monotonically, although it is inappropriate for candidate votes. (For example, one advantage is that allows one to say, "yes, some punishment is deserved, but I'm not sure what it should be", which is not available in the instant runoff version.)

Edit:
I should add that Starglider's scheme is quite excellent under the current restrictions for candidate votes, but it's not quite appropriate for punishment votes because, e.g., it makes very little sense to select "none" and then "permban" as the "least worst" option. We could use some extra slots to say things like "definitely not a permban" or "definitely at least some punishment" instead, which the "vote for an interval" scheme allows.

[fgalkin]

code in that Fgalkin and Marina will always vote for the harshest penalties, and that I will always vote for liency (not true but close enough)

Huh? Whatever gave you that impression?

Have a very nice day.
-fgalkin

[The Duchess of Zeon]

code in that Fgalkin and Marina will always vote for the harshest penalties, and that I will always vote for liency (not true but close enough)

Huh? Whatever gave you that impression?

Have a very nice day.
-fgalkin

Seriously, I didn't even vote to permban SC despite the fact I could have and still have my vote counted for a tempban if it didn't go through, and I can't recall pushing for a banning in some time.

[Coyote]

As SC actually started to draw my attention, and I started paying more notice of his shenanigans, I actually voted for total perma-ban. Usually I see myself as pretty damn lenient, myself...

[CmdrWilkens]

Then what's the problem? For the punitive case, the only assumption is that voting choices are restricted to be consecutive. That's very natural when the options escalate monotonically, although it is inappropriate for candidate votes. (For example, one advantage is that allows one to say, "yes, some punishment is deserved, but I'm not sure what it should be", which is not available in the instant runoff version.)

Edit:
I should add that Starglider's scheme is quite excellent under the current restrictions for candidate votes, but it's not quite appropriate for punishment votes because, e.g., it makes very little sense to select "none" and then "permban" as the "least worst" option. We could use some extra slots to say things like "definitely not a permban" or "definitely at least some punishment" instead, which the "vote for an interval" scheme allows.

No there isn't a problem, and the was my point in saying I may have come across poorly in my initial response. I don't disagree with anything you've offered but rahter I just want to make sure that everyone is aware of the assumptions being made so that if we use that system everyone understands the compromise to wrok within the software.

That said Starglider and I continue to exchange PMs and we are both of the opinion that his system for a 3 candidate race works best for nomination votes while your system probably would work best for a graded punishment vote. I have no issue with using alternate system for different votes my main concern is that this thread be clear as to the benefits and assumptions of each system.

[Kuroneko]

I might critique the voting rules somewhat. Assume that each senator votes honestly in the sense that the first choice is always the senator's first pick. This naturally requires that sometimes the voted-for second choice is actually the senator's third pick, rather than second. Thus, a preference of ANBC is a vote for AB due to limited selection of valid choices.

SG-0

[ 1. AB | 2. AC | 3. BA | 4. BC | 5. CA | 6. CB | 7. NA | 8. NB | 9. NC | 10. Abstain ]
sum 1+2, 3+4, 5+6, 7+8+9, if none is not highest eliminate and go to 1+2+7, 3+4+8, 5+6+9, eliminate lowest and apply votes to remaining two, highest wins.

Some scenarios (100 voters, except 4):
(1) 99 ANBC and 01 CBNA, with overwhelming support for A. Then A is both the majority and the Condorcet winner, but B wins the election despite only one voter finding B deserving.
(2) 15 ABC, 05 BAC, 40 ACB, 40 BCA. Here, C would lose each individual pair of elections (A:C = B:C = 60:40), and hence is the Condorcet loser, but nevertheless wins the election. (One can distribute N's somewhat abitrarily, but mostly in the second place.)
(3) 15 ABC, 10 BAC, 30 ACB, 30 BCA, 10 CAB, 05 CBA. Here, there is no majority winner. A is the Condorcet winner, but C, the Condorcet loser, wins instead.

SG-1
----
As SG-0, but alter the rules a bit:
1) Sum A = 1+2, B = 3+4, C = 5+6, N = 7+8+9, which are first preferences. If any has majority, declare it winner.
2) Keep only the top two choices and add secondary preferences to the tally. If tied, pick the one with greatest number of first preferences.

If N is the Condorcet winner but not the majority winner, one of {A,B,C} is still elected. Hence SG-1 is not Condorcet. Even if the condition is relaxed to apply to {A,B,C} only, SG-1 is still not condorcet: consider 35 ABC 34 CAB 31 BCA. Additionally, for, say, 40AB 27BC 33CB, B drops out in step 2 and C wins. However, if 14 people change their preference from ABC to CAB, increasing support for C, then A would drop out instead, and C loses. Hence SG-1 is not monotonic.

Approval
--------
This system selects a subset out of three candidates {A,B,C} that the voter finds approves of, including the empty ("none"). This gives nine options when "abstain" is included. In other words the choices would be:
1. None 2. {A} 3. {B} 4. {C} 5. {A,B}, 6. {A,C} 7. {B,C} 8. {A,B,C} 9. Abstain.
The goal here is slightly different. Instead of trying to find the "most favored" candidate, it selects the "most tolerable" candidate, unless a plurality votes "none".

SG-2
----
As SG-0, but with:
1) Sum A = 1+2, B = 3+4, C = 5+6, N = 7+8+9, which are first preferences. If any has majority, declare it winner.
3) Redefine A = 1+2+7, B = 3+4+8, C = 5+6+9, and pick highest.
It's essentially SG-0 with an enforced majority rule criterion.

Code: Select all

+-------------+------+------+------+------+
| Vote Method | SG-0 | SG-1 | Appr | SG-2 |
+-------------+------+------+------+------+
| Monotonic   | YES  |  NO  | YES  | YES  | Increasing support for X cannot make X lose.
| Condorcet   |  NO  |  NO  |  NO  |  NO  | If X would win every XvsY duel, X is the winner.
| CondorcetL  |  NO  | YES  |  NO  |  NO  | If X would lose every XvsY duel, X is not the winner.
| Majority    |  NO  | YES  | n/a  | YES  | If X is the majority's first choice, then X wins.
| MajorityL   | YES? | YES  | n/a  | YES? | If X is least favored by the majority, then X does not win.
+-------------+------+------+------+------+

Comments/suggestions/corrections?

[Kuroneko]

I would like to put on the table that approval voting makes a lot of sense for candidate votes.

Approval
--------
For three candidates {A,B,C}, the ballot is, in no particular order,
1. None 2. {A} 3. {B} 4. {C} 5. {A,B}, 6. {A,C} 7. {B,C} 8. {A,B,C} 9. Abstain.
Rules:
(1) Let A = 2+5+6+8, B = 3+5+7+8, C = 4+6+7+8, N = 1, if the order is as above. If N has the most votes, no one wins.
(2) The candidate with the most approval votes wins, as long as it is more than half of non-abstentions.

This has the following reasonable properties:
-- If "None" has more votes than any candidate, then no one wins.
-- The winner is always the candidate deemed qualified by the most voters.
-- If X wins, then X would pass a single-cadidate X vs None vote (due to condition (2)).

A while ago, we had an issue regarding the peculiarity of single-candidate election. The above approval voting directly addresses this concern. For example, for 44% None, 11% A, 20% {A,B}, 10% {B,C}, 15% {A,B,C}, A has the leading approval of 47%, but no one is elected due to the clause in rule (2). This is a reasonable, because in the corresponding single-candidates races, if the voters stick with their judgments, A would lose 47%:53%, B would lose 46%:54%, and C would lose 26%:74%.

It is true that strictly speaking, approval voting does not satisfy that Majority and Majority Loser criteria. Thus, it is possible for the candidate considered worst by the majority to win the election. However, for our purposes that is irrelevant, because in all such cases, this candidate must be considered qualified by the majority of voters. It's still impossible to elect a candidate disapproved by the majority, so the Majority Loser holds in this modified form, and similarly for Majority (i.e., the winner must have majority approval). And really, what's the harm in electing the qualified candidates out of their true order of popularity, so long as they are still qualified? The more popular ones can have another go a month later.

Code: Select all

+-------------+------+------+------+------+
| Vote Method | SG-0 | SG-1 | SG-2 | Appr |
+-------------+------+------+------+------+
| Homogenous  |  NO  |  NO  |  NO  | YES  | Always consistent with single-candidate (XvsN) races.
| Monotonic   | YES  |  NO  | YES  | YES  | Increasing support for X cannot make X lose.
| Condorcet   |  NO  |  NO  |  NO  |  NO  | If X would win every XvsY duel, X is the winner.
| CondorcetL  |  NO  | YES  |  NO  |  NO  | If X would lose every XvsY duel, X is not the winner.
| Majority    |  NO  | YES  | YES  | yes* | If X is the majority's first choice, then X wins.
| MajorityL   | YES  | YES  | YES  | yes* | If X is the majority's last choice, then X does not win.
+-------------+------+------+------+------+ * with a modified criterion; see above

In the case of SG, consider the following distribution of preferences:
-- 20ANBC, 20CNAB, 15BNAC, 15NA, 15NB, 15NC
Then A wins the election, but would lose an AvsN race 20:80. A's winning is purely an artifact of being forced to select a second candidate, even if the voter judges only one to be qualified.

Based on the above, I believe that approval voting is the most suitable for candidate votes for our purposes.

Comments/suggestions/corrections?

[The Yosemite Bear]

I don't know that's giving me a headache, but I've only been awake for half an hour...

[Kuroneko]

No comments at all?

I'll transplant this one from a different thread:

You might as well get a definitive suitable/unsuitable answer on a single candidate (i.e. Steve), rather than deal with the whole split voting mess again.

Approval voting with the winner being the most approved candidate doesn't have that problem--a single vote can handle up to three candidates as things stand now, and more than three can be split arbitrarily across multiple votes without introducing any detrimental artifacts. It behaves the same way for both low (single-candidate) votes and high (four-or-more) votes.

[CmdrWilkens]

Clarifying somewhat the general sense has been that some form of instant-runoff/approval voting is both acceptable and desireable. So we will be using such votes in the future but the exact system is the one question remaining. Based on what we've worked through it will likely be some variant of either Starglider or Kuroneko's contribution.