LWN: Comments on "Debian's election results"

Debian's election results

emorrp1 — Sat, 24 Apr 2021 21:14:01 +0000

> People tend to underestimate how much other people's preferences and thought processes differ from their own

That's a really good way of putting it, and (meta) even does well at assuming good faith despite disagreeing with said People. I'm reminded of a conversational technique of putting the opponents position into your own words and checking they're happy with this interpretation before continuing. It's a lot of effort that is often missed in short form prose, but inevitably raises the quality of conversation by e.g. realising there is no underlying problem just disagreement on definitions of terms.

So please continue to wade into controversial topics with your opinions as much as possible, because whenever I see your name I know I'm about to read a measured response which I may or may not agree with.

Debian's election results

itvirta — Thu, 22 Apr 2021 09:47:32 +0000

It wouldn't be Debian if they weren't discussing. ("They" since I'm also only a user.) The mailing list archives are public, you can go and have a look. Actually, it would probably be a good idea to go closer to the source before drawing conclusions about other people's mistakes and misunderstandings. After all, the comment section here is a completely unrelated site, and not all the Debian voters are here to discuss their votes.

Debian's election results

rra — Wed, 21 Apr 2021 17:39:22 +0000

And yet, people have repeatedly confirmed that they understood just fine and voted exactly the way that they intended to vote.

I'm not sure arguments from personal incredulity are all that persuasive when it comes to voting. People tend to underestimate how much other people's preferences and thought processes differ from their own, and think that mixes of opinions that other people sincerely hold are nonsensical or obvious errors. This comes up all the time in politics, and I think argues for a lot of humility and deference to other people's ballots as cast.

Yes, people possibly made errors; that certainly happens. But it feels more productive to me to focus on the errors that people identify themselves. If a voter says "ugh, I meant to express preference A but got confused by the ballot and expressed preference B instead," that's a valuable data point. But saying "this voter who I have never talked to voted something that I find nonsensical and therefore obviously made a mistake" is not helpful and usually isn't even true.

Debian's election results

geuder — Wed, 21 Apr 2021 17:22:08 +0000

Thanks for the correction! Of course I had not read the fine manual (Debian constitution)... It seems I (as an outsider) am not the only one who has misunderstandings about how the system really works if they are discussing this on their list now. Those who voted should have probably made sure before they understand how it really works. As noted in a sibling thread some votes looked really weird, like the voter in question contradicting themselves.

Debian's election results

guillemj — Wed, 21 Apr 2021 14:29:42 +0000

When the project voted on whether the GFDL was a DFSG-compliant license, there was a substantial amount of voters that stated that it was never a free license, regardless of presence of invariant sections. <https://www.debian.org/vote/2006/vote_001>

So while the project decided it is a free license when no invariant sections are present, and that means one can upload such works into the Debian archive, that does not mean that people that didn't or don't agree with that position will be willing to maintain packages containing such licensed works, and might thus remove or not install them as would be done with other non-free files.

I don't think at the time, the GFDL issue was an excuse to distance from GNU or the FSF though, more a disagreement with the desire for it to be solved. To me it resembled more the situation when Debian dropped KDE due to licensing concerns, which ended up with those concerns being eventually addressed. <https://www.debian.org/News/1998/19981008.en.html>

Debian's election results

itvirta — Wed, 21 Apr 2021 10:04:31 +0000

No, unranked options just rank lowest, and quorum/majority is based on votes against the "default option" FD.

So, if a ballot puts e.g. 8 for FD, and leaves option X unranked, that does count against X for majority and means
X is "completely unacceptable" for the voter. But the same could be achieved by putting 7 for FD and 8 for X,
or any other vote that puts FD above X.

The email call for votes (https://lists.debian.org/debian-vote/2021/04/msg00218.html) says that:

> You may skip numbers, leave some choices unranked, and rank options
> equally. Unranked choices are considered equally the least desired
> choices, and ranked below all ranked choices.

And e.g. the irrelevance of an all-equal ballot was just confirmed by the Project Secretary
(https://lists.debian.org/debian-vote/2021/04/msg00429.html)

> Ranking all options the same has no effect on the result. It does
> not have an effect on the quorum or majority. The only effect it has
> is that more people voted.

Also see the formal description in the Debian Constitution A.6. Vote Counting (https://www.debian.org/devel/constitution).
Point 1 says the same as the call for votes, points 2 and 3 clearly say quorum and majority are based on defeating FD.

Debian's election results

anton — Wed, 21 Apr 2021 08:51:37 +0000

Or maybe a gfdl section. Not everyone wants to open the door to proprietary software just because they want documentation for free software.

But they don't stop at putting documentation in non-free. For Gforth, Debian excised the documentation completely (it has no invariant sections).

My impression is that a portion of DDs has wanted to distance Debian from GNU and the FSF for many years, and the documentation issue is one area where they won. The present issue is one where they lost, but it was close.

Debian's election results

geuder — Wed, 21 Apr 2021 06:54:06 +0000

I don't think so. The first 2 are equivalent, I agree.

The one with the hyphens is different I believe. Isn't it so that a dash means "completely unacceptable" while the highest number means "least acceptable". In practice the dashes are used to calculate quorum. If an option has too many dash votes it does not reach quorum and cannot be accepted even it would win enough one on one races.

That leads to the question of tactical voting: Not many voters use dashes, but high numbers (relative for their vote). So if you cast a dash for a "bad" option, but the option reaches quorum your vote is "lost". Had you voted a high number (relative within your own one again), your vote would still be counted.

That's at least my understanding. It could be wrong. In all ballots I can participate in real life I have only 1 or 2 votes for better or worse.

Debian's election results

smurf — Tue, 20 Apr 2021 20:44:13 +0000

Well, I think it's irrational, but whoever voted that way didn't ask me, so …

Debian's election results

hummassa — Tue, 20 Apr 2021 20:06:38 +0000

To me, this vote convey "I ONLY want Option 1 to win. If they don't remove the entire board, I think it's inconsequential if RMS resigns or if he's laureated Emperor of Rome." I don't see the irrationality on that.

Debian's election results

bpearlmutter — Tue, 20 Apr 2021 19:22:53 +0000

1-------. It is impossible for someone strongly supporting Option 1 to be indifferent to Option 2 vs Option 5.

Debian's election results

itvirta — Tue, 20 Apr 2021 13:09:02 +0000

Showing only the differences can give the impression that they're all that matters for resolving the winner.
But as said, in the case of loops, it's not so.

Debian's election results

smurf — Tue, 20 Apr 2021 12:36:33 +0000

What's confusing about that?

The full weight is irrelevant IMHO. We want to know how tight the vote was. You can easily replace the single number with an explicit difference if you want, but what would that help?

Debian's election results

bpearlmutter — Tue, 20 Apr 2021 09:28:52 +0000

Not a 4-cycle. Changing one ballot to reverse that "1" arrow would have caused a 5-cycle among all five options that passed majority.

Debian's election results

itvirta — Tue, 20 Apr 2021 09:05:13 +0000

It's also confusing that the graphs they post on the vote results page have the difference, not the full weight.

Debian's election results

dgm — Tue, 20 Apr 2021 06:21:46 +0000

I want to read your comment as "tongue in cheek", and I understand that the FSF is open to criticism, as any other organization.

But, taking into account the running disprestigiation campaign against them, I think that it would be far better to put humour asside and make complaints explicit, and verifiable. Otherwise it's just throwing dirt on their name.

An interesting contrast between two organisations

rodgerd — Tue, 20 Apr 2021 04:30:40 +0000

Results aside, it's interesting to contrast the two organisations: Debian has clear criteria for who gets to vote, and a robust process; any given individual may not be happy with the outcomes of a vote, but it's transparent and open.

The FSF, on the other hand, has a completely opaque governance structure, with the voting power being a closely-held secret.

Debian's election results

pabs — Tue, 20 Apr 2021 00:20:44 +0000

I have following GNU documentation packages from Debian non-free installed, several of them definitely still have back-cover texts:

cpp-doc gcc-doc gdb-doc make-doc tar-doc

It would be great to see these packages moved to Debian main.

Debian's election results

IanKelling — Mon, 19 Apr 2021 23:48:59 +0000

> And for GNU documentation, they are never met

I think that is wrong. Afaik, most GNU documentation does not use invariant sections and is included in debian free section https://www.gnu.org/licenses/fdl-howto-opt.en.html

Debian's election results

mattheww — Mon, 19 Apr 2021 21:42:24 +0000

As I understand it, the 1.0 constitution says that if A has a 2:1 supermajority requirement and B doesn't, then for the pairwise comparison between A and B you'd compare half the number of ballots voting A ahead of B with the number of ballots voting B ahead of A.

So I suppose when they decided they didn't like that happening, it was a natural idea to start to treat FD specially and apply the supermajority requirement only there.

Debian's election results

Sesse — Mon, 19 Apr 2021 21:33:39 +0000

Hm, where's the built-in advantage? Are you saying there would be a problem if there were no Condorcet winner, and then the A-FD link would be too strong?

Debian's election results

mattheww — Mon, 19 Apr 2021 21:29:42 +0000

That's what the v1.0 constitution did, before changing to the current system in 2003.

I think the main difference is that if you have three choices

A: try to solve some problem by a means requiring a 2:1 supermajority
B: try to solve the problem by a means that doesn't require a supermajority
FD: further discussion

then the new system allows a third of the voters to veto A (by voting FD over it), but as long as that doesn't happen the choice between A and B will effectively be by simple majority.

While with the old system B would have had a built-in advantage over A, even if everyone involved was happy in principle to do the thing that requires the supermajority (overrule someone, or whatever) if A won.

Debian's election results

NYKevin — Mon, 19 Apr 2021 19:20:27 +0000

Ah, I misread https://www.debian.org/devel/constitution#item-A to interpret the weights as the net margin by which A defeats B rather than the absolute number of votes favoring A over B. This is why you have to read the rules carefully.

Debian's election results

Sesse — Mon, 19 Apr 2021 19:18:23 +0000

But couldn't one just double or triple all the beat strengths for that?

Debian's election results

mattheww — Mon, 19 Apr 2021 18:38:00 +0000

The way I remember it, the special handling for Further Discussion was added as a way to get a reasonable definition for "2:1 majority" and "3:1 majority".

I think the effect on elections without a supermajority requirement was largely a side effect.

Debian's election results

itvirta — Mon, 19 Apr 2021 18:20:23 +0000

There's a post from the Debian Project Secretary stating that the winner of that hypothetical vote-flip
would actually be #4 "Call on the FSF to further its governance processes", since it's the total weight
that counts there, not the difference, see: https://lists.debian.org/debian-vote/2021/04/msg00384.html

Debian's election results

chris_se — Mon, 19 Apr 2021 17:41:29 +0000

> Switching the one vote margin in the other direction would not change much, as this would be the weakest link in the created cycles.

Playing around a bit with the numbers (disclaimer: I'm not sure I've understood the algorithm well enough to guarantee this is correct): if 9 additional people had voted 2 > 7 > FD (and the rest somewhere in that chain, it doesn't really matter where), then option 2 would have won. That means that if 5 existing people had voted 2 > 7 instead of 7 > 2, then a completely different option would have won. While this isn't a margin of 1, this is still very close.

Debian's election results

matthias — Mon, 19 Apr 2021 17:14:34 +0000

> Had either margin been switched, the Schwartz set would contain five (!) options.

This is true for the "introspection" option. But if the "RMS resignation" option would have defeated the "no statement" option, the Schwartz set would be trivial. The "RMS resignation" option defeats options except the "no statement" option.

Switching the one vote margin in the other direction would not change much, as this would be the weakest link in the created cycles.

Debian's election results

calumapplepie — Mon, 19 Apr 2021 16:17:01 +0000

It is easily the closest ever. Just checked through all the archives since the current method was adopted. There has never been a non-trivial Schwartz set (and there wasn't this time, either, thanks to a one-vote margin).

Only 2 votes have had the final winner defeat another contender by a single-digit margin: the 2006 and 2007 DPL elections of Anthony Towns and Sam Hocevar (respectively), both of whom nearly lost to Steve McIntyre. However, even then the margins were 6 and 8, as opposed to a single vote. Further, in both cases the Schwartz set would STILL have been trivial, since Steve defeated all other candidates.

This vote, by contrast, had the winning option nearly lose to TWO candidates: the call for RMS's resignation and the call for some introspection by the FSF. In the latter case, that was a one-vote margin. Had either margin been switched, the Schwartz set would contain five (!) options.

So yeah. Pretty controversial stuff. It is definitely true that the "support RMS" options were VERY unpopular, but a few votes could have changed the result from "Debian issues no statement" to anything from "Debian signs the letter demanding the entire board of the FSF resign" to "Debian asks the FSF to have a good hard think about these things".

Debian's election results

NYKevin — Mon, 19 Apr 2021 15:56:59 +0000

Yes, there probably was some tactical voting here. But because the FD option is a form of approval voting, that may have been *sincere* tactical voting (i.e. tactical voting which is not inconsistent with a voter's true preferences). In other words, instead of thinking of FD as an outcome, think of FD as a line: Items above it are approved and items below it are not. Then, under that characterization, some voters may have sincerely decided that they preferred no statement over an anti-RMS statement, but could still live with the latter as long as it meant no pro-RMS statement. The reverse probably also happened for some voters.

This is a form of tactical voting, but crucially, it does not require you to reverse the order of any of your true preferences (provided you *don't* think of FD as a "real" outcome).

Debian's election results

hmh — Mon, 19 Apr 2021 15:39:06 +0000

They mean exactly the same thing to the vote counting algorithm.

The exact numbers used are NOT compared across votes, just used *within that specific vote* to order the choices.

So, these vote examples actually mean the same: I (the voter) prefer *this* higher ranked option above all others, and I (the voter) am not disclosing how I feel about the other options (either because I don't want to, or because they're actually all the same as far as I am concerned).

Debian's election results

rgmoore — Mon, 19 Apr 2021 14:04:07 +0000

Personally, I don't see the difference between this and FD

Both options start from the idea that none of the statements are good, but they go in opposite directions after that. Making no statement is saying there's never going to be a statement that encapsulated Debian's position, so they should shut up about it and go back to making a distribution. Further discussion is saying there's hope of coming up with a statement that encapsulates Debian's position, so they should talk about it more. So people who are tired of arguing about it would prefer no statement to further discussion.

Debian's election results

zdzichu — Mon, 19 Apr 2021 10:53:34 +0000

Could you provide an example of such irrational votes?

Debian's election results

bpearlmutter — Mon, 19 Apr 2021 10:23:11 +0000

Yes: if you actually scan the individual ballots (they're online) it is clear that many of the voters did not know how to express their preferences correctly. I'm not saying I disagreed with them: I'm saying that there are ballots that cannot possibly correspond to a rational vote (either honest or strategic) cast by someone with any position.

It's really interesting, because Debian is an ideal setting for this kind of vote: voters who actually care about the issue, technically highly sophisticated, etc. Yet even so, they're not very good at filling out a rank ballot.

Debian's election results

Sesse — Mon, 19 Apr 2021 09:57:25 +0000

I know (I voted in the election). But save for the special handling in devotee, they would sound functionally equivalent, so the logical thing to do would be to vote for it immediately above FD (or equal to FD). It seems a large amount of people had other alternatives between this alternative and FD.

I never understood exactly why the voting system handles FD in a way that breaks with the Condorcet criterion (it is possible to use FD to vote tactically, and I've had votes among a very small number of people where it mattered), but I've been told it was discussed on -vote back in the days.

Debian's election results

bpearlmutter — Mon, 19 Apr 2021 09:54:17 +0000

It would actually be a 5-cycle!

Debian's election results

NYKevin — Mon, 19 Apr 2021 08:38:11 +0000

Further discussion is special. Ranking something below FD means "I oppose this option." If more people support FD than a given option, then that option is considered defeated, and it is ineligible to win the election.

Usually, this doesn't matter, because options which lose to FD tend to lose to everything else, too. But if there is a Condorcet cycle which involves FD (which should be quite rare in practice), then any option which loses to FD is eliminated before the cycle-breaking machinery even gets involved. This ensures that an option which is actively opposed by a simple majority cannot win under any circumstances. It also means that, if you want a "do nothing" option, you need to nominate it separately.

Debian's election results

cortana — Mon, 19 Apr 2021 08:36:59 +0000

Debian's vote graphs were totally mysterious to me until I read somewhere that the number on each edge represents the number of votes by which the option at the tail end beat the option at the head end.

Debian's election results

Sesse — Mon, 19 Apr 2021 08:11:41 +0000

I agree this is also a very reasonable interpretation. My point was more that there's no way I'm reading the outcome as “Debian secretly agrees with Stallman”.

Personally, I don't see the difference between this and FD, so I'm surprised they ranked so dramatically different.

Debian's election results

NYKevin — Mon, 19 Apr 2021 07:46:43 +0000

I just sat down and worked out how the Schulze method handles a cycle like that, and it appears to be more or less the following:

0. For the sake of the hypothetical, imagine that "Call on the FSF to further its governance processes" had defeated "No statement" by 1 vote (instead of the other way around). Now we have a four-outcome cycle as you describe.
1. Delete all outcomes other than the four involved in the cycle.
2. Delete the edge from "...further its governance processes" to "No statement" (because it has the smallest weight, at 1 vote).
3. Now there's no cycle anymore, and you still get "No statement" as the final winner.

So it would have taken a swing of more than one vote to actually change the outcome. Nevertheless, I agree with your conclusion: Not only did Debian choose not to explicitly support Stallman, those options were overwhelmingly defeated. "No statement" won a much more narrow victory over the options which explicitly opposed Stallman (yes, it was more than one vote, but "more than one vote" is not what most people would think of as a wide margin). Arguably, they don't need to issue a statement; those numbers speak for themselves. Everyone will interpret the outcome slightly differently, but it is very difficult to interpret it as substantively supportive of RMS (procedurally, maybe, but not substantively).