Clarify the voting process

[jeffjaffe]

The last line in Section 7.3 (about votes) says

'In the case of Advisory Board and TAG elections, "one vote" means "one vote per available seat".'

I think this line is a holdover from previous voting procedures. We now use STV. A literal interpretation of this line is that (e.g.) in an AB election with 4 open seats, each AC rep would have 4 votes: i.e. 4 opportunities to use STV. This is absurd.

I recommend dropping this line.

[michaelchampion]

I very strongly disagree. I would argue instead that the current implementation of STV is incorrect because it violates this point in the Process Document. One might argue, I suppose, that the "single" in STV means that the AC's intent in approving STV voting implied that we were moving from "one vote per available seat to "one vote", but to my shame I didn't realize this until recently. But STV was sold to us as an alternative tabulation system using ranked rather than binary preferences, not an alternative voting system where we only got one vote no matter how many seats were open.

I don't know enough about STV mechanics to suggest an alternative ranked preference tabulation system that allows one vote per available seat, but we should investigate.

[dwsinger]

I agree with both Jeff and Mike. We undoubtedly have an inconsistency. Mike, I cannot think of any STV system in which one gets one vote per seat; equal-rank voting sort of comes close, but isn't really this, and anyway, we can't implement it.

We either delete this sentence, or we re-think STV. I fear the latter is too soon and too large.

[michaelchampion]

My inclination is to take this to the AC. Clearly there is a logical inconsistency in the Process. I thought we were approving a one vote per seat ranked preference system, but apparently there is no such thing. It would be informative to know what other people on the AC expected... Other than an end to the debate few understood or cared about!

I think the most responsible thing to do is tell the AC "we screwed up... help us figure out how to rectify our error." They'll hate us for making them think about voting systems yet again, I suppose, but the alternative of running elections that are in violation of the process is worse.

[dwsinger]

I think it's way too early to go back to the well...

[jeffjaffe]

It is clear that the intent of the pieces of the process document (the portion on STV and the last line of Section 7.3) are talking about different voting systems. There are four ways I can think of to deal with the situation.

1. We simply delete the line in Section 7.3. That is what I proposed above, but Mike didn't like that idea.

2. We can harmonize STV with the line in Section 7.3, by giving everyone 4 votes and then everyone will deliver 4 identical ballots so we just create 4x the number of votes with the same result. That is what I referred to as absurd, above - but we could do it.

3. We can simply ignore the line in Section 7.3, which I believe was the intent of Process 2017. While we could do that, it is not the neatest thing to do.

4. We can propose a broader fix to election issues. That seems to be what Mike is proposing. I'm not opposed to that idea, but I don't have any ideas at the moment - so I'll leave it to Mike or others to propose something.

[michaelchampion]

I guess I have a taste for the absurd: I thought we were doing something like #2 when we agreed to STV voting in W3C elections. I'd prefer #4 but don't have a concrete proposal. We have to live with #3 until the process can be updated.

[dwsinger]

I agree #2 is absurd; we just multiply all the numbers by 4. Equally absurd would be asking people for 4 rankings, potentially different, for the 4 seats. "For the first seat I prefer Mike over Chris, for the second, Chris over Mike..." huh?. In the next process revision, we either need to do #4, or (as seems likely in the time available) #1. Leaving #3 is poor form, but I suppose...

[michaelchampion]

I'm not sure #2 is so absurd if it is described as:

• All AC members get one vote per seat, but are asked to do only one ranking.
• The first open seat is filled according to the current STV algorithm
• The winner is removed from the candidate pool
• The next open seat is filled by according to the STV algorithm with the remaining set of candidates
• Repeat until the n seats are filled.

As best I can recollect, that's how I THOUGHT the new voting system worked when we voted to add STV to the Process. I admit to not thinking through exactly how I thought STV would work in a system where each voter gets multiple votes, but I sortof assumed that since we weren't talking about removing the "one vote per available seat" rule, I thought someone smarter than me had figured out how the two constraints would work in harmony. I suspect I'm not the only AC rep who was under that delusion.

That's consistent with the current process document that specifies both an STV system and gives AC members one vote per open seat.

That mitigates the concerns about one's 2nd thru n rankings being more or less meaningless. Once your top-ranked candidate wins a seat, your 2nd-ranked candidate "gets your vote".

It may not meet the original goal of tilting the system toward diversity, I'm not sure.

It adds some complexity for the team I suppose

I think it would be more robust than the current STV system in the face of extremist candidates strongly supported by slightly over 1/n of the electorate but low ranked by (n-1)/n of the electorate. We can argue whether that's a bug or a feature. I believe it's a feature in a consensus-based decision culture even if it's plausibly a bug in a voting-based decision culture.

Am I missing some absurdity here?

[jeffjaffe]

Mike, I'm not sure I understand your proposal.

As David said above, if we give each AC rep 4 opportunities to do an STV vote and then add up all of the results - presumably most AC reps fill out their 4 ballots identically. This has the effect that everyone gets 4x the number of votes, but does not change any of the outcomes compared to single vote STV. The fact that it does not change outcomes is the reason that David and I called it absurd.

Is that indeed what you are proposing we should do, or did you have something else in mind?

[dwsinger]

OK, I think I get what Mike is proposing. Today, we tabulate the votes; imagine that the top vote-getter gets just the quota. The ballots that placed that candidate first are never considered again.

In contrast, Mike is suggesting that after the first seat is filled, we re-run the tabulation using all the ballots, but marking the elected candidate as having withdrawn, so all votes cascade past that candidate.

It's an interesting thought experiment, but I want someone who understands voting and tabulation to weigh in.

[chaals]

A very rough answer is no. In more detail, I think it is a terrible idea that makes bad outcomes very likely. Mike asserted that second preference votes didn't count but all the information I have seen suggests that is simply false. The people elected were preferred over those not elected by the tabulated rankings. Mike's proposal seems to be a return to the old system that tended to a winner takes all outcome. An example from us history is to assert that Walter Mondale had less than 2% support since he only won a single state - and then translate that into a Senate with 2 democrats.While this might make agreements easier to reach it seems unlikely to help ensure they reflect a real consensus. More when I'm back working. Cheers17:50, 28 July 2017, David Singer notifications@github.com:OK, I think I get what Mike is proposing. Today, we tabulate the votes; imagine that the top vote-getter gets just the quota. The ballots that placed that candidate first are never considered again. In contrast, Mike is suggesting that after the first seat is filled, we re-run the tabulation using all the ballots, but marking the elected candidate as having withdrawn, so all votes cascade past that candidate. It's an interesting thought experiment, but I want someone who understands voting and tabulation to weigh in. —You are receiving this because you are subscribed to this thread.Reply to this email directly, view it on GitHub, or mute the thread. Private mail - Chaals is Charles McCathie Nevile - Standards Declaimer

[michaelchampion]

@jeffjaffe here's why I don't agree with your summary of my proposal on 7/27/2017

if we give each AC rep 4 opportunities to do an STV vote and then add up all of the results - presumably most AC reps fill out their 4 ballots identically. This has the effect that everyone gets 4x the number of votes, but does not change any of the outcomes compared to single vote STV.

Consider the example I've used before:

• Granger and Weasley are mainstream candidates with broad but somewhat lukewarm support; few strongly prefer one over the other.
• Malfoy is an extreme candidate, passionately supported by a substantial minority but fervently opposed by a majority.

In the old system, Granger and Weasley both get 65% of the votes, and Malfoy gets 35%, thus Granger and Weasley are elected. In the current system, Granger gets ranked first by 33% and second by 32%; Weasley gets ranked first by 32% and second by 33%; and Malfoy gets ranked first by 35%. Thus Malfoy and Granger are elected – Malfoy in the first round since he got over the threshold, Granger in the second round after Weasley was eliminated and his voters’ second place votes went to her.

In my proposal, which as best I recall is how I thought this would work when I voted for the STV system, there are n STV elections for the n open seats. The threshold is 50% since each is a single-winner election.

In the example first round of the election for the first seat, nobody meets the threshold. Weasley is eliminated, and his first place votes are given to his voters' second choice, Granger. Thus Granger is elected in the second round.

In the STV election for the second seat, since Granger has already been elected, first place votes for her are allocated to her voters' second ranked candidate, Weasley in this contrived example. Weasley is over the threshold so gets elected in the first round.

So this isn't just the current STV system multiplied by n; having n elections for the n open seats is different because the threshold is different with n elections for a single seat as opposed to 1 election for multiple seats. It is consistent with the current process document because it gives AC reps 1 vote per open seat and runs an STV election for each seat.

It this a more desirable election system than the one we have used in practice (in violation of the literal wording of the process doc)? I'm pretty sure there will be much disagreement. I biased my contrived example by giving the 35% candidate the name of an evil character, what if it were "Lovegood" [1], i.e. someone with an outsiders perspective but with the best interests of all at heart? So, the real question is whether we -- as an organization that operates by consensus rather than voting -- should we fear the consequences of a non-mainstream "bad" candidate more or less than we hope for the benefits of a non-mainstream "good" candidate? Living in a kakistocracy created by the quirks in my country's election system :-) I'm feeling more paranoid than hopeful about voting systems at the moment myself.

I do submit that if there is no agreement to change the process document to eliminate the sentence about 1 vote per open seat, the Team is obligated to implement the process as it stands, and my proposal provides one mechanism to do so.

[1] http://harrypotter.wikia.com/wiki/Luna_Lovegood

[jeffjaffe]

@michaelchampion Thanks for providing a detailed example. Now I at least understand what you are proposing.

Indeed, what you are proposing is an approach that one could use for elections and indeed it would give a different result that how we have implemented STV. I see your point how this can be viewed as both STV (since it has n instances of STV voting with one winner) and also has n votes per 7.3 in the Process Document.

I don't personally agree with your conclusion, however, that this is the correct interpretation of the process document.

1. Section 2.5.2 talks about using an STV system (singular). It does not talk about multiple STV systems.
2. STV as commonly defined is a single vote system for multiple seats; not multiple instances of votes for each for a single seat [1].
3. (Most importantly): when we ran several trial elections and spent numerous hours discussing it - the method that we used in the trials was the method that we used in the actual election last spring. Not once did someone suggest the novel interpretation you have provided. Hence I believe the membership voted on that one when they made the change in Process 2017.

[1] https://en.wikipedia.org/wiki/Single_transferable_vote

[michaelchampion]

@jeffjaffe writes

(Most importantly): when we ran several trial elections and spent numerous hours discussing it - the method that we used in the trials was the method that we used in the actual election last spring. Not once did someone suggest the novel interpretation you have provided. Hence I believe the membership voted on that one when they made the change in Process 2017.

I doubt very many of the AC members voting understood that they would henceforth have only one vote no matter how many seats are open in AC and AB elections, and that the contradictory words to in the process were an obvious bug. I certainly didn't. And there was no example in the test runs that illustrated the anomalies found by analyzing the first really competitive election.

My point is that having multiple votes is not logically incompatible with ranked preference voting, even if this is a "novel interpretation" or one believes that the term STV suggests a single vote. (I assumed it meant a single vote for each seat that could be transferred across candidates, not a single vote for all seats).

It seems like the time has come to take this question to the AC: Given the incompatibility between the formal process and the actual procedure, do they want to make the process align with the procedure by eliminating the "one vote per open seat" language, or do they want the team to implement a ranked preference procedure that gives them one vote per open seat?

[dwsinger]

On Aug 20, 2017, at 10:48 , Michael Champion notifications@github.com wrote:

@jeffjaffe writes

(Most importantly): when we ran several trial elections and spent numerous hours discussing it - the method that we used in the trials was the method that we used in the actual election last spring. Not once did someone suggest the novel interpretation you have provided. Hence I believe the membership voted on that one when they made the change in Process 2017.

I doubt very many of the AC members voting understood that they would henceforth have only one vote no matter how many seats are open in AC and AB elections, and that the contradictory words to in the process were an obvious bug. I certainly didn't. And there was no example in the test runs that illustrated the anomalies found by analyzing the first really competitive election.

My point is that having multiple votes is not logically incompatible with ranked preference voting, even if this is a "novel interpretation" or one believes that the term STV suggests a single vote. (I assumed it meant a single vote for each seat that could be transferred across candidates, not a single vote for all seats).

It seems like the time has come to take this question to the AC: Given the incompatibility between the formal process and the actual procedure, do they want to make the process align with the procedure by eliminating the "one vote per open seat" language, or do they want the team to implement a ranked preference procedure that gives them one vote per open seat?

I asked David Baron, and he doesn’t feel enough of an expert.

As I understand it, we want to compare two ways of using STV; and I would need analysis on the second before going to the AC. Does it work? Has it been used? What consequences does it have?

1. Classic STV. One preference list, and so on, with votes cascading on a n-seat quota until all seats filled. What we did.
2. STV seat-by-seat. The first seat is filled as if it were a one-seat election (50% threshold?). Then that candidate is marked ‘withdrawn’ and the same set of ballots used to fill the second seat. And so on until all seats are filled.

I don’t see #2 as being any more or less “one vote per seat”. It might be different, but I do not wish to experiment on live elections, and I am unwilling to ask the AC until we have done or read some analysis.

David Singer Manager, Software Standards, Apple Inc.

[chaals]

+1 to Jeff's original proposal. -lots to Mike's suggestion, which would combine the complexity of STV ("You have to rank the candidates in your order of preference") with the unrepresentative outcomes of the old system ("the largest bloc can win everything, regardless of whether their support is 90%, or 19% with 75% preferring "anyone else or nobody at all").

I'm also not keen to replace the live elections with an apparently untried system. I can't find evidence of anyone using this approach anywhere, and it makes Schulze_STV seem tried and true by comparison.

It also appears that the rationale for suggesting it is false. The election results are not an anomaly caused by strangeness in Meek vote tabulation, they are a result of people winning more votes at higher rankings than the other candidates.

However surprising the outcome may be (and I have to confess I didn't find it very shocking, and even less so on reflection), it shows why we have secret ballots - because people may not be prepared to state their perceived interest publicly.

I also find Jeff's third argument reasonably compelling: what we are using is what we trialled and the AC worked with.

[dwsinger]

I find myself between Mike and Chaals.

Yes, STV will have the effect of (pick your euphemism here) [electing more diverse candidates | electing more fringe candidates]. Yes, STV also means that first preferences have more impact than second, they more than third, and so on (to the extent that I don't think any 4th preferences factored into the last election which was 4-seat), whereas in the past if we voted for N candidates, those N votes had equal impact. No, STV doesn't eliminate strategic voting, it just changes what that means (e.g. to try to get someone NOT elected, you vote strategically to raise the levels of those who might otherwise have trouble beating them). Yes, this is all a change, and yes, I am not sure we understood all this.

But equally, treating an N-seat STV election as N one-seat elections is something I can (in a limited search) find no analysis, literature or usage information on. I am, with Chaals, completely unwilling to run a live experiment without at least some intellectual understanding of it in theory.

I suppose the AB could request we run that tabulation on an election (possibly even the last one), but that doesn't substitute for analysis by voting-systems experts, IMHO. We could even re-tabulate, for AB and team eyes only, the last election. But I think I would only make that request if we understood an analysis.

So at the moment, I support only removing the mis-leading "one vote per seat" sentence.

[michaelchampion]

I'm going to oppose removing the "one vote per seat" sentence until we have a frank AC discussion along the lines "We messed up and need your guidance on how to fix it." I have no idea whether others on the AC supported the new system in the belief there was no contradiction between ranked preference voting and having one vote per open seat.

Given that -- for practical purposes -- there is a contradiction, we should re-ask the AC whether they prefer the old one vote per open seat un-ranked selection system to the one vote per election STV system. Some reasons for preferring the old system include:

• It allows one to vote for a slate of candidates, not force one to impose artificial rankings on equal choices.
• It makes it hard to elect fringe candidates who are opposed by a majority of the electors

Of course, each has a flip side:

• The old system doesn't allow one to express a preference order
• It makes it hard to elect fringe candidates whose virtues are not know to a majority of voters

[jeffjaffe]

@michaelchampion @dwsinger I think the best way to stimulate an AC discussion is to propose removing it and let the discussion take place.

If we don't remove it, the AC won't notice anything and there will be no discussion.

[dwsinger]

we re-open this issue and re-insert this contradiction to give us an actionable issue as a hook for further debate

[dwsinger]

We agree that this sentence should be removed to reflect reality that we have STV. A separate issue will be filed on whether we should revert from STV to the previous voting system, or something else.

[frivoal]

I agree with @michaelchampion that it is likely that a significant portion of the AC did not realize that the introduction of STV-as-we-did it was in contradiction with "one vote per available seat". I think there are 3 possibilities:

1. AC Reps meant STV in the form that we implemented, and did not notice the lingering sentence about "one vote means one vote per available seat"
2. AC Reps understood the combination of the two to mean what @michaelchampion said (or some other thing that combined STV and "one vote per available seat", and just failed to notice that the STV variant we implemented was not what they had in mind.
3. AC reps did not notice that STV and "one vote per available seat" were contradictory.

Voting systems are complex and subtle, the raw numbers were not public, and between "one vote per available seat" and "using a Single Transferable Vote system" the first of the two sentences is the more obvious one. Based on that, I don't think we can easily assume that most AC-Reps who voted for this meant (1).

Resolving this is only editorial if all/most people meant (1), and I don't think we can know this for a fact. I think we have to go back to the AC.

Whether we should stick with meek-STV, go back to the old method, do @michaelchampion 's suggestion or something else altogether is a significantly more difficult question, given the big consequences that subtle differences in voting systems can have.

The feature with STV is that it gives a better chance than the old system to minority-favored candidates.

I think* the downside is that it:

• does not strongly distinguish between candidates that have the favor of a large minority but are strongly disliked by the rest, and candidates that have the favor of a large minority but are relatively unknown to the rest.
• if anything, it favors the strongly polarizing ones, as for the others, it is quite possible that those who have the support of large minority but are relatively unknown to the rest would be ranked high but not first by that minority, who still knows incumbents and may approves of (some of) them equally.

I think* @michaelchampion's suggestion does not have that downside, that it rejects loved-by-some-hated-by-most candidates, and that it does favor liked-by-many-but-not-always-as-number-1 candidates. It does have the downside that chaals mentions: the largest coherent bloc can win everything without an actual majority.

I would be interested in seeing the results of past elections with @michaelchampion 's method applied.

I think* that meek STV with equal ranking mitigates both the "large coherent block" downside and the "liked-by-many-but-not-always-as-number-1" downside.

---

 * IANAVSS (I Am Not A Voting System Scientist), so what I think is true about their various properties might not be.

[chaals]

If a candidate has sufficient support they will be elected in either system. STV slightly favours the candidates who are not polarising over those who are.

Candidates who are disliked by many, under STV, can be voted against by the many, by ranking them last, rather than the limit provide by the old system of one unranked preference for every available seat.

The fact that STV encourages, instead of discourages, a broader pool of candidates reinforces this effect.

Under the old system, it made more sense to only vote for a single candidate - and we know that this occurred often - in the order of 25%-35% of votes. This strategy strongly favours polarising candidates. It also strongly favours coordination in advance of the election to minimise the pool of candidates (again, something that we saw occur regularly under the old system).

Under STV, where a voter can rank all candidates without prejudice to the value of their vote, there is an incentive to do so. In comparison to the old system, this favours "compromise", and "somewhat unknown" candidates, because they will get preferred above "polarising" candidates by those who don't consider them top choices.

[chaals]

Equal ranking merely saves time for voters who don't want to choose between two candidates.

Equal ranking has no real impact on the "large minority block wins everything" issue. That is a feature of the old system and something Meek actively works against.

It has a minimal impact* on polarising vs unknown candidates - Meek (and other STV systems) already address this issue as noted above.

It seems intuitively strange, but it is mathematically true, that tossing a coin is a thoughtful rational way to rank candidates where you don't actually have a real preference. Equal ranking stops people having to think about why that is so, and thus removes apparent complexity.

Equal ranking does not make it more likely that if you cannot decide then both your candidates will be elected. Nor does it penalise indecision: your candidates do get the full value of your vote.

*minimal impact: In principle it would have none, but since a "polarising" candidate is almost never going to be absolutely first or last choice for all voters, some will likely rank them equally when that is available who would have otherwise explicitly put them last. In practice, this reality of behaviour can slightly attenuate the system's built-in favouring of compromise/unknown over polarising candidates. In practice, it is unlikely to make a lot of difference to real outcomes.

[jeffjaffe]

@frivoal While the Team prefers to stay neutral on how the AC decides to do voting, I feel obligated to bring you up to speed on the previous discussions about this issue - since I infer that you are not happy with the current method.

Unless and until we change the process again, the Team (which has been the one interpreting the process document in practice for elections) has felt that the proper voting mechanism to use is the one that we used during the couple years of experimentation. That is what we have been using until now.

Looking forward, the Team will implement whatever changes are agreed to.

[frivoal]

My assumption is that in case of equal preference without the ability to equal rank, a fair amount of people will be tempted to rank the incumbent first, "just in case", because it makes them feel that since they cannot express their preference, they should be voting strategically. This works against candidates that are perceived to have lower chances. I believe that having the ability to rank equally limits this effect.

Regardless, my main point was to agree that the current process is self contradictory, that given that contradiction and how non obvious voting system are it is not self-evident what the AC thought it was agreeing to, and that resolving it is non editorial and needs to go back to the AC, with an explanation of the contradiction, and at least a choice between the old system and the one implemented, and possibly some variants that fall within the range of what people might have thought they were agreeing to. This is without prejudice for or against the systems. -- Sent from my phone. Please excuse my brevity.

[frivoal]

@jeffjaffe I understand that, and have no expectation that the team will change anything for the ongoing election. I am mostly commenting on this now to inform the AC voters of what my position regarding this open issue would be if elected, as requested in the AC forum.

I don't think I am in strong opposition against the current voting method. It seems to me that meek-STV with equal ranking would be better (even though @chaals seems to think that my intuitions on this are wrong), but I can live with either.

What doesn't seem reasonable is to stick with a system that a non trivial part of the AC feels we picked it by accident or by misunderstanding. We should either make sure that we are picking it intentionally, or revert. Living with the ambiguity is not appeasing.

[jeffjaffe]

@frivoal I think the entire AB agrees that it is desirable to add equal ranking. The operational problem is that we have not found software that tabulates this, and it is too expensive to implement ourselves.

[dwsinger]

equal-ranking: Since I have wanted to use equal-ranking, I have not opposed it. But I am concerned that people understand it. @chaals "Equal ranking merely saves time for voters who don't want to choose between two candidates." -- I don't think it's merely that. If I equal-rank 4 candidates first, and you rank order them, they get 1/4 of a vote from me in the first round, and one of them gets a full vote from you in the first round (with the rest getting no vote from you). That's a difference of impact. If we introduce equal-ranking, I want to be sure we understand it.

separate-stv-per-seat: I remain deeply concerned about adopting such a system without any live examples around the world, and without any analysis by voting system scientists.

[chaals]

Equal ranking: Why it is not trivial to implement (it was described by Meek decades ago) is that it effectively needs to expand the apparent number of votes. So in your example, your equal ranking effectively turns into one vote for each possible ranking of the 4 candidates, and mine turns into an equivalent number of votes, all of which rank the candidates in the same order.

Clearly, if these were the only two votes, my vote would be decisive.

That makes sense - you don't mind which of the four are chosen, I have expressed a preference, we both got what we asked for.

Let's expand the example, assuming there are in fact only 4 candidates, and there there are two other votes. Both rank my last choice and my third choice equal-first, my second choice third, and my first choice last.

At that point, the number of seats matters. It's close to two against one, and the first elected is going to be one of those top-ranked by the two. In practice, My preference for 3 over 4 will be what combines with the votes from those 2 to elect my third choice first.

Update In this precise case, I think you might get a tiebreak situation. In a real case where there are 20 votes not 4, a very slight variation amongst them will resolve it. Although I still haven't done the precise maths.

The point is that when you equal-rank two or more candidates, no candidate is penalised in terms of overall votes. It is only in the choice between those candidates where your vote no longer has an impact. Because you expicitly asked for that.

[dwsinger]

Yes, with single seats I might agree. I think the confusing effects might come up in a multi-seat election, where those who rank in strict order (no equal preference) repeatedly get more weight behind a candidate in each round, over the population who gave equal ranking. example: even in one seat, if A,B,C,D each equal rank X,Y and D,E,F rank Z first and W second, the the second group wins the seat because Z has 3 votes while X=Y=2 votes (4 half votes); but the population supporting X,Y is 4 voters while only 3 support Z,W. Or something like...

[chaals]

@dwsinger wrote

those who ranked repeatedly get more weight behind a candidate in each round.

I don't understand what you mean.

[dwsinger]

(updated)

[chaals]

First, this assumes nobody ranked all four candidates, and second, I don't think so, running it in my head.

In your scenario, Z doesn't have a quota for a single seat, so W gets eliminated, with nothing to distribute. One of X or Y are then distrbuted by tiebreak, whose votes transfer to the other at full value, so the other of X or Y gets elected.

In the two-seat scenario, Z has about a quota so either: Z gets elected, a bit gets distributed to W, but W is then eliminated, and one of X or Y is elected on a tiebreak. Or: W gets eliminated with nothing to transfer, then X or Y get eliminated with votes are transferred to the other at full value who is elected followed by Z. (This latter is an example of votes being extinguished. Because the X/Y voters didn't rank all the way down the line, the leftover bit gets lost. Not that it makes much difference in this tiny example, but with more votes and more seats it might).

In 3 seats, its Z, X and Y.

[chaals]

@frivoal wrote:

My assumption is that in case of equal preference without the ability to equal rank, a fair amount of people will be tempted to rank the incumbent first, "just in case", because it makes them feel that since they cannot express their preference, they should be voting strategically.

They make that assumption. But it would be based on a fundamental misunderstanding. The reason for changing the voting system was precisely to make sure that such an assumption went from being pretty much guaranteed correct to wrong and irrelevant.

The whole point of STV is that if you ranked someone first who turned out to be eliminated, your vote transfers to your second-ranked candidate. Instead of not counting for anything - which is what happened under the old system.

In addition, if you happened to vote for someone who would have won without your vote, then your vote still gets transferred - but as a fraction. For example, if a candidate got 20 votes, but only needed 10, then each paper which ranked them first only uses half their vote. So for each paper, the second-ranked candidate gets an additional half a vote to add to whatever they already had.

[chaals]

@frivoal We ran three votes with both systems - the old one, and an experiment that would be analysed but not change the outcomes.

I would expect the AC to be surprised if we implemented anything other than what was put in front of them repeatedly as the experimental version, in which most who voted actually participated.

By comparison, that people noticed what the editor and others considered an obvious error wildly inconsistent with what had been presented in practice and what had been explained repeatedly, and assumed it meant either that we would use the old system, or that we would use a different system that as far as anyone know has not existed in practice, seems very unlikely to me.

[dwsinger]

I agree with @chaals; we used what had been explained and trialed, and simply didn't notice this sentence.

[michaelchampion]

And I agree with @frivoal "What doesn't seem reasonable is to stick with a system that a non trivial part of the AC feels we picked it by accident or by misunderstanding." I am an AC member who had no idea that this change would eliminate the one-vote-per-open seat provision in the process. It not only was not removed from the draft voted on, it was NEVER MENTIONED in the years of discussion, as far as I can find.

The experiment did not generate enough data to illustrate the different outcomes that different voting systems could create, nor were enough data released for anyone not familiar with the arcana of voting systems to infer that the vote-counting process the team used didn't actually follow the letter of the process document.

It is LOGICALLY possible to have an STV ballot for each open seat. I now understand that is not what "STV" means to voting theorists, but it was not at all apparent to me when I voted for this process change, and I'm sure I'm not the only AC member in that position.

[dwsinger]

@michaelchampion "it was NEVER MENTIONED in the years of discussion". Every single experiment and every single explanation clearly explained we were using STV voting, and how it would work. We studied STV modes in detail. It was blindingly obvious that STV voting works with a Single Transferable Vote. That's what STV means.

[michaelchampion]

If it was blindingly obvious then why didn't the contradiction with "one vote per open seat" leap out at everyone who read the draft document for several years?

The easiest way forward would be to ask the AC how they want to run elections now that the misunderstanding has been cleared up.

[dwsinger]

We missed that sentence when adjusting the process document.

[fantasai]

I wanted to drop a link here to @bkardell’s post from last December where he expresses some very serious reservations about the current system: https://bkardell.com/blog/W3CTAGElectionChange.html

Having talked to both @bkardell and @cwilso about that post, I know that Chris spent a lot of time trying to explain through a lot of confusion how the system actually works and what its implications are. To say that the AC didn't understand the implications of the changes seems much more likely than not.

The system @michaelchampion describes in https://github.com/w3c/w3process/issues/60#issuecomment-318534082 makes sense to me, and is probably the most consistent with the current wording in the Process document. If I had to interpret the Process as it stands now, this is what I would do. (And, fwiw, is also the interpretation I came up with before reading @michaelchampion’s comment.)

https://www.w3.org/2018/Process-20180201/#AB-TAG-elections

The Advisory Board and a portion of the Technical Architecture Group are elected by the Advisory Committee, using a Single Transferable Vote system...

If, after the deadline for nominations, the number of nominees is ... Greater than the number of available seats, the Team issues a Call for Votes that includes the names of all candidates, the number of available seats, the deadline for votes, details about the vote tabulation system selected by the Team for the election, and operational information. ... When there is a vote, each Member (or group of related Members) may submit one ballot that ranks candidates in the Member's preferred order.

https://www.w3.org/2018/Process-20180201/#ACVotes

In the case of Advisory Board and TAG elections, "one vote" means "one vote per available seat".

https://www.w3.org/2002/10/election-howto#votes

STV Meeks used as Tabulation system; OpenSTV 1.7 used to perform the computation.

The interesting situation is that, regardless of what the AB decides to do in the future, we are now operating under the 2018 Process document. I would argue that enacting @michaelchampion’s proposal is what we must do, per Process 2018, unless someone else has a different proposal which is also consistent with the Process as it stands now.

[bkardell]

As just a 'concerned individual', I remain very confident in my previous assertions that ACs do not understand the new voting system. Aside from a few people on the AB who I have discussed with at great length, I have yet to meet someone in the W3C (AC or just member) who seemed they could actually explain it or use it to accurately represent what they thought they were representing, or that what they thought they were representing wasn't actually even possible to represent. To be entirely honest, every time I go through and explain it again, I find something new perplexing about it.

If you combine the low participation numbers with the actual expressive ability of the votes and then the fact that people don't even understand it to begin with... Well, I'm not sure how to best verbalize what I am trying to say here, but... I'm not sure what the results even "mean" at that point. If that makes sense.

Sorry, I realize this is somewhat of an aside on the issue and ultimately not very helpful.

[chaals]

Separate STV per seat:

I'm highly skeptical about this, because I don't clearly understand the proposal itself. From a voter perspective, do voters provide 4 complete rankings? If so, how does that work - do we have 4 sequential elections? And if not, how does it match the "one vote per seat" motivation of the proposal, rather than just being a novel proposal for an alternative method of counting STV votes?

[chaals]

@michaelchampion asked:

If it was blindingly obvious then why didn't the contradiction with "one vote per open seat" leap out at everyone who read the draft document for several years?

As far as I recall - and I was the editor at the time so this is clearly my fault - I simply failed to notice, or forgot to remove the sentence when making the particular edit. (It may be one of the cases where I was given a particular text change by this CG, and didn't carefully enough check whether it was complete).

I believe it was the clear intention of Process 2017 to implement exactly the voting system which had been trialled by the AC on 3 live elections, and that this was a failure of the editor. I have no doubt that had I looked carefully I would have reached the conclusion is the first comment in this issue.

As I recall, when it was brought to my attention that was my reaction to my mistake.

@michaelchampion objected to fixing the error, but agreed that we should be following the intention of Process 2017. I believe the rationale was so the anomaly would continue until it raised enough heat to reopen the whole voting systems discussion.

...10 months later... It is unclear to me as a participant in this group whether that has now occurred, and is the focus of the discussion, or whether the issue is that STV and Meek's counting algorithm are somewhat complex and people would like to understand them better.

I note Meek's algorithm was explicitly designed in the late 1960s to prioritise using careful mathematics, to ensure quality data, and rely on computers for the processing, in order to get a result that reflected what voters were asking for, over being easy to run through in your head. Probably for that reason, when I try to run through a scenario like the examples above in my head, I sometimes discover I was not careful enough and got it wrong. Luckily, computers are more widely available than when Meek was a working computer scientist and mathematician in the 1960s and 70s.

If you read this far down, you may be interested in the actual proposal. I found it in French, and it's 15 pages without complicated mathematics, but if anyone has a link in english, or to something more accessible than a scanned PDF, I would be grateful. part 1: please mind the copyright statement) and part 2 (which justifies equal ranking - same copyright).

[michaelchampion]

@chaals said no

I believe it was the clear intention of Process 2017 to implement exactly the voting system which had been trialled by the AC on 3 live elections, and that this was a failure of the editor."

We all failed; you didn't catch what you clearly would have understood to be an inconsistency, I (and I suspect others on the AC) failed to understand the details of STV well enough to know that it wasn't just adding a ranked-voting option to the one-vote-per-open-seat system, and the team failed by implementing what they thought the Process was supposed to say rather than what it actually said.

As @bkardell notes, nobody really knows what this all means. Given there is an inconsistency in the "axioms" expressed in the process document, obviously there is no logical answer (or rather , all answers are equally logical ). Since there's no consensus among those of us who maintain the process document whether the inconsistency should be corrected by removing the STV language or the one vote per open seat language, I don't see any way forward other than to poll the AC on which way they want to run elections, now that we've had some experience with both systems.

[frivoal]

On May 9, 2018, at 7:57, fantasai notifications@github.com wrote:

I would argue that enacting @michaelchampion https://github.com/michaelchampion’s proposal is what we must do, per Process 2018,

Technically I agree. At the same time, while the current method is not widely understood (as @bkardell https://github.com/bkardell and @fantasai detailed), it was indeed trialed for a while, and explanations were given (even if the consequences were not fully understood by many).

So, violating the letter of the process (what we do now) is not good, but doing something else than what was explained and trialed does not seem great either.

Hence, we should go back to the AC, explain again, and vote again (and this time make sure that the text of the Process matches the intent).

Also, since I don't expect that people want to have voting system discussions all the time, if there are proposals for different voting systems we should include that in the discussion as well rather than just resolve the ambiguity/contradiction now and have a separate discussion later.

As for the current election, short of throwing out the votes that have already been cast and possibly redoing the nomination phase as well, I don't see how we can switch voting systems half way through. This kind of scramble does not sound desirable, especially given that we have used the present system already in the past.

[dwsinger]

On May 8, 2018, at 20:37 , Michael Champion notifications@github.com wrote:

As @bkardell notes, nobody really knows what this all means.

I am quite sure that that is not true (for example, there are people in the AC who study voting systems, as we well know). I am also aghast at the suggestion that some of the brightest people I know can’t understand STV, as applied to multi-seat elections. I would expect erudite discussion of the quota, or the details of the algorithms. Not basic “I don’t know what this all means”.

David Singer Manager, Software Standards, Apple Inc.

[dwsinger]

On May 9, 2018, at 2:40 , Florian Rivoal notifications@github.com wrote:

On May 9, 2018, at 7:57, fantasai notifications@github.com wrote:

I would argue that enacting @michaelchampion https://github.com/michaelchampion’s proposal is what we must do, per Process 2018,

Technically I agree.

I do not. You have a vote in every seat that’s open, today — a vote per seat. Your preference list is used for every seat until either every seat is filled, or your preference list is exhausted (which is roughly similar to voting for fewer people than the open seats, under the old system). I could make at least as plausible a case that we are not in violation of the process as Michael’s case that we are. I do think that the sentence adds nothing and could cause confusion, but it’s an editorial drop.

Honestly, the complaint is not over this sentence: it’s that we no longer support block voting. And that was the intention: STV makes it more likely we’ll elect a diverse group, that a heavy block vote won’t sweep the day and exclude all the diverse candidates. That is what we wanted — and, I think, still want. If anyone wants to go back to the system which reduced diversity, please say so (and why).

It’s mildly concerning that we can’t express equal preference, and the AB has wanted to allow it as soon as we find tabulation software that supports the documented and analyzed ways to do it. But the absence of it is hardly a show-stopper. If you really don’t mind whether Amalasuentha or Gaiseric are elected, then it doesn’t matter which order you rank them; flip a coin.

David Singer Manager, Software Standards, Apple Inc.

[fantasai]

@dwsinger,

You have a vote in every seat that’s open, today — a vote per seat. Your preference list is used for every seat until either every seat is filled, or your preference list is exhausted (which is roughly similar to voting for fewer people than the open seats, under the old system).

This is absolutely not true if you implement STV under Meek’s as one vote per election rather than one vote per seat (as @chaals is explaining was intended). It is true if we follow @michaelchampion’s suggestion of applying the given ballot per seat.

To give a simple proof-by-example that it is not true: if each AC provided a fully-ranked ballot, and 5 of the candidates each received exactly 20% of the first place votes (and 2 of the candidates received none, only 2nd place or later), then at no point in Meek’s algorithm do we look at any ranked position other than the first-place vote. The entire remainder of every ballot is perfectly discarded.

@chaals,

I'm highly skeptical about this, because I don't clearly understand the proposal itself.

Reread https://github.com/w3c/w3process/issues/60#issuecomment-318534082 , which I believe has been edited for clarity.

From a voter perspective, do voters provide 4 complete rankings?

No, they provide one ranking. There is a single ballot containing one ranking. This is very clearly specified in the Process document, and is not a condition we can violate. (Nor does it make any sense to provide 4 different rankings, obviously.)