Re: Evolving away from source package realms
I myself am *very* happy to have other Debian people (DDs, DMs) git push and dput fixes to any of "my" packages. No need for an MNU or delay or permission: just do it. Zero friction. In the unlikely event you do something I'm uncomfortable with I'll just revert it and discuss. This has nothing to do with a mono repo. It's a social convention, and can be done with per-package repos. In fact, I believe the salsa.debian.org "debian" group is intended for this, with packages having their packaging repos there treated in roughly the above fashion. That's where I put my own packages, unless they belong in some team group. People interested in this communal maintenance idea should be aware of the Low Threshold NMU list https://wiki.debian.org/LowThresholdNmu which is basically the same idea, and I think may have led to a bit of confusion about what a repo being in salsa.debian.org/debian/ means. --Barak Pearlmutter
Re: Range Voting - the simpler better alternative to Condorcet voting
Sam Hocevar [EMAIL PROTECTED] wrote: On Fri, Jun 15, 2007, Barak A. Pearlmutter wrote: You make another point, which is interesting, but which actually when carried to its logical conclusion ends up being in support of Range Voting over Condorcet. If you continue with the logic asking what happens when Range Voting voters vote strategically, you find that in Range Voting where all voters are well informed of the opinions of the electorate in general, and all voters cast optimal strategic ballots, you get approval voting, which results in ... the winner being what would have been the winner of a 100% honest Condorcet election. Unless you can tell us how to be well informed of the opinions of the Debian electorate in general, especially with ballots that go beyond yes/no questions, yet still want to make a point about the Debian voting system, you should not use that assumption in your reasoning. I don't really understand your point here. Under any particular set of symmetric assumptions people have made, Range Voting seems to perform better than Condorcet. These are practical assumptions, rather than theoretical ones. Like: everyone tries to vote honestly. Or: some people try to vote honestly others try to vote strategically using whatever information they know. Etc. One limiting case of this is where people think they know who the front runners are, and many people vote to try to get their preferred candidate elected. This is behavior you might expect in practice, right? Under those conditions, Range Voting performs well, and Condorcet runs into horrible problems. In fact, when people think they know who the front runners are and are correct, RV with highly strategic voters leads to electing the honest Condorcet winner! That shows how very robust RV is to strategic voting. Unlike Condorcet, which basically falls apart when people act like ... well, like people. Certain situations are more mathematically tractable and easier to explain than others. That is why I talked about that particular limiting situation in the message you responded to: because it is easy to see what's going on, and it is interesting. Not because only under *those* assumptions does RV beat Condorcet. In fact, under all realistic conditions I've seem discussed, RV seems to beat Condorcet. But if you have some other actual specific situation in mind where you think Condorcet would perform well relative to RV, then let's hear it! (The only situation in which Condorcet beats RV proposed in the discussion so far is: RV voters voting strategically vs Condorcet voters voting honestly. When pressed as to why this startling asymmetry would arise, it's because someone official tells the voters---incorrectly---that they'd hurt themselves by voting strategically with Condorcet. That is a pretty outlandish assumption, and basically goes against human nature: it assumes people won't often try their best to get their favored candidate elected. And it sort of contradicts itself, in that the official instructions to voters says something untruthful in order to obtain a better global outcome. In other words the official instructions to voters are themselves strategic rather than honest! And for some reason, the same strategic instructions to be honest aren't given with RV. And even under this very silly assumption, Condorcet only beats RV by a tiny bit.) -- Barak A. Pearlmutter Hamilton Institute Dept Comp Sci, NUI Maynooth, Co. Kildare, Ireland http://www.bcl.hamilton.ie/~barak/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Range Voting - the simpler better alternative to Condorcet voting
The example you give is a perfect instance of the DH3 problem. You have a population of voters whose true preferences are 31x ACBXD 32x BCAXD 37x CBAXD Assuming the B supporters know that C is the front runner, some of them might notice that if a handful of the B supporters do not cast the honest ballot BCAXD but instead cast the ballot BACXD then B could win. Especially if some of the A supporters do the analogous thing. Woo hoo! If a few of the C supporters are smart, they'll notice that, since C is the front runner but B is the runner-up, they'd be wise to increase C's slim margin of victory --- or perhaps preserve C's deserved victory and prevent the unjust election of B by a few clever B supporters --- by not casting the honest ballot CBAXD but instead casting CABXD As you have doubtless noticed, if enough of the C and B supporters exhibit this cleverness, then A will win. The existence of an X option does not eliminate this problem. As you note, if everyone ranks X second, the winner is again not any of the top three honest options, but is instead X, even though every single voter believes X a worse option than any of the three options A, B, C. So having an X on the ballot doesn't help. -- Barak A. Pearlmutter [EMAIL PROTECTED] Hamilton Institute Dept Comp Sci, NUI Maynooth, Co. Kildare, Ireland http://www.bcl.hamilton.ie/~barak/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Range Voting - the simpler better alternative to Condorcet voting
It always amuses me how people pull out these examples close to Condorcet cycles as examples of strategy in Condorcet methods while ignoring the strategy issues in even simpler Range Voting elections that push it towards Approval-style voting. Well (a) that wasn't *my* example, and (b) it wasn't cyclic! (It looks like it was based on the idea that everyone thought CBAXD except that the A supporters pulled A to the front of that list and the B supporters pulled B to the front. That seems about as non-cyclic as you can get.) You make another point, which is interesting, but which actually when carried to its logical conclusion ends up being in support of Range Voting over Condorcet. If you continue with the logic asking what happens when Range Voting voters vote strategically, you find that in Range Voting where all voters are well informed of the opinions of the electorate in general, and all voters cast optimal strategic ballots, you get approval voting, which results in ... the winner being what would have been the winner of a 100% honest Condorcet election. In other words, if you want to see the winner be whichever candidate *would* have won an honest Condorcet election, you'd be better off having a Range Voting election and encouraging people to vote strategically. That is a strength of Range Voting, not a weakness. -- Barak A. Pearlmutter Hamilton Institute Dept Comp Sci, NUI Maynooth, Co. Kildare, Ireland http://www.bcl.hamilton.ie/~barak/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Range Voting - the simpler better alternative to Condorcet voting
You also make the point 60x A=60,B=40 40x B=60,A=40 It only takes six of the second group to vote B=99,A=0 to change the outcome, which is a major victory for the extreme supporters but a loss for honest moderates. I think your point is that, with Range Voting, if *some* of the voters vote strategically while others don't, the ones who vote strategically carry extra weight in the election. (a) This is ***EXACTLY RIGHT***!!! In fact, that is the *definition* of strategic voting. If voting strategically didn't give a voter extra power to influence the election in their favour, we wouldn't call it strategic. (b) This is also true of Condorcet. Unfortunately (Arrows theorem etc) there is no voting system for 2 candidates which avoids the possibility of strategic voting. And Condorcet is, unfortunately, in fact, particularly susceptible to strategic voting. In Condorcet, if all the voters vote strategically, you often get a really bad candidate winning; this is the DH3 pathology, see http://rangevoting.org/DH3.html for details. But with Range Voting, if everyone votes strategically, you get ... the honest Condorcet winner. Which isn't really so bad. As an addendum: in Range Voting voters are told to rate the candidates, with min/max for their least/most favoured candidates. So in your particular example with only two candidates, the actual votes would have been 60x A=99,B=0 40x A=0,B=99 which would leave no opportunity for strategic voting. But that's no great trick, since there are only two candidates. With 2 candidates, it is easy to make examples where strategic voting by Range Voting voters would make sense. -- Barak A. Pearlmutter Hamilton Institute Dept Comp Sci, NUI Maynooth, Co. Kildare, Ireland http://www.bcl.hamilton.ie/~barak/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Range Voting - the simpler better alternative to Condorcetvoting
... simulations seem to show that they are much less harmful ... than analogous circumstances with Condorcet. Usually the best statistics are those which were faked by yourself, so please point us to some scientific paper(s) which support(s) your claim. The claim was based on simulations described at http://rangevoting.org/vsi.html A more academic-style presentation of similar results is available at http://math.temple.edu/~wds/homepage/rangevote.pdf The code used is also made available, so you can re-run the simulations, or modify them to fix any perceived deficiencies. -- Barak A. Pearlmutter Hamilton Institute Dept Comp Sci, NUI Maynooth, Co. Kildare, Ireland http://www.bcl.hamilton.ie/~barak/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Range Voting - the simpler better alternative to Condorcetvoting
You make some interesting points. I'm curious here: a.) Can you give any example of any election we've had so far that has resulted in an outcome not expected by the voters (that is based on the cast votes, not based on predictions) I'm not sure if this fits your criterion, but according to Warren Smith's analysis http://rangevoting.org/Debian2003.html there were questionable results in a number of DPL elections. The analysis of 2003 is particularly relevant. For instance: ... this also means the Robinson-supporters could have prevented Michlmayr from being a Condorcet Winner, by agreeing among themselves to rate GarbeeMichlmayr a few percent more often. In that case Garbee would have been the Condorcet Winner. It thus is conceivable that Garbee really should have won, but was prevented from doing so as a result of misfired strategic voting by the Robinson (or Zadka) supporters aiming to prevent one or the other from being a Condorcet Winner. b.) Can you provide a rewritten devotee that makes use of range voting? Free software communities generally work on a Codes speaks louder than words-policy, so working code is the best first step towards adaption; such a contribution would also be useful even if Debian sticks to Condorcet. The method is quite trivial. But that is a good idea; I'll try to figure out dvt well enough to add the option. Right now I'm getting the feeling you're just pushing a personal agenda. It might well be that there is substance behind your words, and that range voting really is better than Condorcet, but Condorcet has the advantage of already being in use for several years within the project, and thus familiar to almost everyone in the project. I cannot vouch for others, but I had no troubles whatsoever to figure out how our current voting system worked the first time I voted, so I find it hard to believe that Condorcet is too complicated for DD's. In the above analysis of the 2003 DPL election, Smith notes: 200 votes were rejected as invalid versus 510 accepted as valid, even after 5 years of experience and the best software developers in the world voting and programming He takes this as evidence that the voting system in use is not so easy for DDs as you speculate. (Disclaimer: my own ballots have had no trouble, and this criticism would not have occurred to me.) But as to your more global point: actually I agree, Condorcet is doing okay for Debian. But I would like to see Range Voting adopted in other more critical places, and Debian seems particularly ripe for adopting this improvement, minor though it might be compared to Debian's current rather good voting sytem. -- Barak A. Pearlmutter Hamilton Institute Dept Comp Sci, NUI Maynooth, Co. Kildare, Ireland http://www.bcl.hamilton.ie/~barak/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Range Voting - the simpler better alternative to Condorcetvoting
Given that almost all of the publicly reported problems with ballot validity I've seen have been due to getting the PGP signing wrong ... I find this conclusion highly dubious. Good point; you're probably right. I did modify Smith's presentation to textually separate the factual information he mentioned from the conclusion he drew from them, because the logic really didn't seem watertight. -- Barak A. Pearlmutter Hamilton Institute Dept Comp Sci, NUI Maynooth, Co. Kildare, Ireland http://www.bcl.hamilton.ie/~barak/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Range Voting - the simpler better alternative to Condorcet voting
election outcomes, like candidate C winning even though all voters prefer A to C and all voters also prefer B to C. That cannot happen with Range Voting. With range voting, each individual voter's interests are best served if they only use the extremes of the scoring range. That isn't always true - see above. The only respect in which this is different from approval voting is that the naive voter is given the false impression that they might usefully give some candidates middling scores. So approval voting is strictly superior to range voting - why aren't you plugging it ? For two reasons. One is the above: optimal Range Voting ballots aren't always pegged at min/max, due to a number of factors: uncertainty about the other voters, intrinsic voter honesty, and also a desire to express one's own uncertainty. The other is that it is good to provide voters with more expressive power, so they can more accurately express their opinions should they so desire. As to why they might so desire, here are a few reasons. The DPL elections are iterated, so if we regard saturating all number to min/max as defecting then the iteration may discourage defecting. The ballots are made public, and intermediate values (and veridical ballots) will probably be seen as looking good for Debian, which would also lead a rational Debian voter to cast a more veridical ballot. There is also a natural human tendency towards honesty and generosity, which in this case would manifest as not casting zeros for people who don't really deserve a zero. All of this ability to cast a nuanced ballot---in other words, the expressive power of a range voting ballot---would be thrown away with approval voting. And some of it is thrown away with Condorcet. -- Barak A. Pearlmutter Hamilton Institute Dept Comp Sci, NUI Maynooth, Co. Kildare, Ireland http://www.bcl.hamilton.ie/~barak/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Range Voting - the simpler better alternative to Condorcet voting
Oops: what I said here: there are no circumstances in which a rational (in the game theoretic sense of that term) Range Voting voter will cast an anti-veridical ballot. is not true when there are N3 candidates. There do exist circumstances etc. However, computer simulations seem to show that they are much less harmful (less likely, and when they occur less likely to result in a really poor candidate being elected) than analogous circumstances with Condorcet. -- Barak A. Pearlmutter Hamilton Institute Dept Comp Sci, NUI Maynooth, Co. Kildare, Ireland http://www.bcl.hamilton.ie/~barak/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Range Voting - the simpler better alternative to Condorcet voting
I still fail to understand how Condorcet with the default option suffers from the DH3 pathology (I did understand how Condorcet without the default option does suffer from DH3). Could you enlighten me? It is pretty straightforward to add some extra candidates whose existence causes voters to use up their default option elsewhere on the ballot, below the dark horse candidate. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Range Voting - the simpler better alternative to Condorcet voting
Let me see if I have this straight. We would both agree that: - Range Voting is more expressive than Condorcet. (In the sense that it is possible for a voter who so desires to more precisely express their true opinion.) - Range Voting is simpler (easier to understand) than Condorcet. - With honest voting, Range Voting beats Condorcet. - With strategic voting, Range Voting beats Condorcet. - When voters switch from being honest to being dishonest, the performance of Range Voting degrades less than does the performance of Condorcet. So (aside from calling me names) it seems like your argument boils down to the idea that somehow Range Voting causes voters to be less honest than they are with Condorcet? And not just a little more dishonest, but a *lot* more dishonest! And the reason you hypothesize that this will occur is that Condorcet performs so horribly when voters are highly strategic that they will recoil in shock at the prospect and in reaction take a vow of utter honesty. That's what I take it you mean by this: With the difference that with Condorcet, you can confidently say to voters: if you lie about your opinions of the candidates when you vote, you are much more likely to hurt yourself than to help yourself, regardless of how you think others will vote. although the statement itself would be a lie. (You could say it confidently if you want, but it would remain false. Complete honesty is not a stable strategy---in the Nash equilibrium sense---with Condorcet. Or with Range Voting, for that matter. In fact as Arrow's Theorem shows, there is no voting system which etc.) You also write this: the fact is that with Range Voting, the *best* real-world outcome one can reasonably expect is total strategic voting on the part of the electorate which is also false. In practice, when actual experiments have been conducted, even when they intend to vote strategically in Range Voting, voters do not generally peg all candidates to min/max. Instead, they tend to give intermediate values to intermediate candidates, but with a push towards the extremes. This corresponds to being slightly strategic, rather than fully strategic as assumed in the table I quoted earlier. Such behavior would put Range Voting's performance somewhere near that of Condorcet without strategic voting in that table. There are several other forces that would tend to further discourage extreme strategic voting with Range Voting in Debian, even aside from whatever aspect of human nature seems to cause people in general to exaggerate less than game-theoretically-optimal with Range Voting. One such force is the Debian culture, which encourages honesty, albeit at times perhaps brutal honesty. Another is that most discussion is on open forums, so secret collusion would be difficult to arrange. Yet others are that many of the top candidates are actually pretty similar, and voters are not very rigid in their preference structures and typically do not care so very strongly about who wins the DPL election. Most voters would also probably not wish to publicly trash (by rating with zero) candidates who are good and valuable Debian contributors but happen to not be their favorite. If you really want to resolve the question of what the best voting system might be, instead of the current combination of ad homenim attacks and falsehoods your posts on this topic have devolved to, I'd suggest we try an experiment. One proposal: it would be easy to give people the option of including a Range Voting ballot along with their Condorcet ranking in the next DPL election. It would be very interesting to see how the two correlate, both on the individual level and on the ultimate decision level. -- Barak A. Pearlmutter [EMAIL PROTECTED] Hamilton Institute Dept Comp Sci, NUI Maynooth, Co. Kildare, Ireland http://www.bcl.hamilton.ie/~barak/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Range Voting - the simpler better alternative to Condorcet voting
Ireland uses STV, which is not a Condorcet voting system. Although true, from a voter's perspective they're pretty much identical: rank the candidates, any left unranked are implicitly at the bottom. Furthermore, I would dare to venture that even our sophisticated Debian Developer voters by-and-large do not understand the minutia of our particular Condorcet resolution mechanism. I will however admit that I've been told by a number of Irishmen that one reason for the use of STV here is that it makes for a lot of fun in the post-voting and postmortem parts of the elections. Nice try though, it was a very good troll otherwise. If by trolling you mean pointing out when Steve Langasek makes an invalid argument, then I plead guilty as charged. To quote the relevant lines from http://rangevoting.org/vsi.html where VSI is the Voter Satisfaction Index, Voting system VSI AVSI B Magically elect optimum winner100.00% 100.00% Range (honest voters) 96.71% 94.66% Condorcet-LR (honest voters) 85.19% 85.43% Range (strategic exaggerating voters) 78.99% 77.01% Condorcet-LR (strategic exaggerating voters) 42.56% 41.31% ... These experimental results also strongly suggest that range voting is the least susceptible to strategic voting, of these common methods. It would appear that in this table, when voters merely exaggerate their preferences, which is a natural human tendency, Range Voting degrades from ~95% VSI to ~78% VSI while Condorcet-LR degrades from 85% to 42%. OUCH! It seems like even a small amount of strategic voting in Condorcet would push its performance below that of Range voting with fully strategic voting. -- Barak A. Pearlmutter Hamilton Institute Dept Comp Sci, NUI Maynooth, Co. Kildare, Ireland http://www.bcl.hamilton.ie/~barak/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: Range Voting - the simpler better alternative to Condorcet voting
http://rangevoting.org/vsi.html itself makes it clear that honest voting in Condorcet performs better than strategic Range Voting. Right, there are a variety of asymmetric conditions under which Condorcet may perform better than Range Voting. You came up with one such condition: Condorcet with honest voting vs Range Voting with strategic voting although to be fair Condorcet would only win there with some distributions of preference structures. I came up with some others: Condorcet with honest voting vs Range Voting with voters replaced by lobotomized monkeys Condorcet with honest voting vs Range Voting by with all desirable candidates removed from the ballot Condorcet with honest voting vs Range Voting with ballots cast by hostile aliens Seriously, it would appear that when people have attempted to make fair comparisons, Range Voting has won. Perhaps there were methodological flaws in those studies, but if so you need to be a little more precise about it. For instance, perhaps as you argue people would tend to attempt to vote strategically more with Range Voting and less with Condorcet, although I'm not convinced of that. But even if so, the question is what the magnitude of that effect would be, and whether it would swamp the apparent intrinsic advantage of Range Voting. If you want to make a convincing argument, you'll need to get a quantitative handle on these effects. Regarding this Condorcet discourages strategic voting proposition, that is---if you'll forgive me---hooey. I'm living in Ireland and there was a major election here just a few days ago using a Condorcet system. There was rampant strategic voting, and the newspapers had a great deal of fun discussing it in the aftermath. And Debian Developers certainly do vote strategically in DPL elections, although they wouldn't call it that; they'd call it ranking people they really don't want to see elected below wacko cranks and people they've never heard of before. Moreover, the lack of expressiveness of Condorcet makes is impossible to cast a ballot saying: I think A and B are really good, with A just slightly better than B, while I think C and D are both pretty bad, and E and F are both terrible. In attempts to shoehorn this into Condorcet people do things like putting X between B and C or between D and E, but that only increases the expressive power slightly and also does not actually have the desired effect. Even worse, Condorcet makes it impossible to say dunno about a candidate. Bottom line: I don't think your flip canard actually outweighs serious scientific attempts to figure out whether Range Voting is better than Condorcet under realistic conditions. I'm not saying the jury is in, but what you said isn't much of an argument. As far as I can see, at this point there is pretty much zero serious argument that Range Voting would be inferior to Condorcet for Debian elections, and some pretty strong arguments to the contrary. --Barak. -- Barak A. Pearlmutter Hamilton Institute Dept Comp Sci, NUI Maynooth, Co. Kildare, Ireland http://www.bcl.hamilton.ie/~barak/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]