alien@cpoint.UUCP (Alien Wells) (11/22/89)
The previous example on STV voting shows clearly how STV works. Let me point out what I consider a number of deficiencies in STV. For the following, consider the recent .aquaria vote with a ballot: A) sci.aquaria B) rec.aquaria C) NO - There is no way to vote NO for a particular name. The best you can do is put it last (or omit it) on your list. Take the example of *.aquaria where the group was uncontroversial (and thus there would be few NOs in an STV vote) but the name was controversial. If most of the sci.aquaria votes came in ABC and the opposition came in CB - it is definitely conceivable that sci.aquaria could have still won even with the large opposition. - This brings me to my main objection to STV - and this is a fundamental one. STV is an attempt to find the choice with the strongest support. MAUVE (and similar concensus schemes) are an attempt to find the choice with the widest concensus. This is a subtle but significant difference. STV might stop votes like sci.aquaria - but it might not. The problem with sci.aquaria wasn't a weakness of the support, but a lack of concensus. If the goal is to stop power plays and minimize flaming, a concensus voting scheme is preferable over a ranking voting scheme. - Supporters of STV have complained (correctly) that in a concensus scheme you can only list what you are willing to accept, you can't give a weighting to them. I would turn that around and say that in STV you have no way to say 'I'll take any of these, but this other group is unacceptable'. You are forced to rank your preferences, even if you really don't care. In addition, the ranking is very primitive. There is no indication in the vote of whether there is a large or a small difference in preference between two ranked choices. This leads to a case where mild preferences have the same strength as vehement objections. - If you ARE going to use a ranking voting scheme, I would much rather see a weighted parallel vote (WEIP) where you have something like: Vote for the following using this scale: 10 Strongly approve 7.5 approve 5 don't care 2.5 disapprove 0 strongly disapprove The votes for each get averaged, the highest average vote (over a certain threshold with a minimum number of votes) wins. Note: The previous 8 lines qulify the scheme under the 'explainable in 25 lines' criteria ... Consider the following example of a vote under WEIP, STV, and MAUVE: NAME WEIP STV MAUVE Ballot 1: (67%) sci.aquaria 10 1 YES rec.aquaria 9 2 YES No Group Ballot 2: (33%) sci.aquaria 0 NO rec.aquaria 10 1 YES No Group 2 Imagine a situation where 2/3 of the voters sent in the first ballot and 1/3 of the people sent in the second. Under STV, sci.aquaria wins. Under either MAUVE or WEIP rec.aquaria wins. - Finally, STV can even fail to do what it claims to, finding the group with the most support. Imagine a situation where people have widely varying first choices, but everyone has the same second choice. That second choice could be eliminated in the first round, leading the vote to settle on a choice that had much less overall support and much more vehement opposition. Both MAUVE and WEIP would correctly find that support. Consider the following example: NAME WEIP STV MAUVE Ballot 1: (20%) sci.aquaria 10 1 YES sci.bio.marine 6 3 YES rec.pets.fish 2 5 NO rec.pets.aquatic 2 4 NO rec.aquarium 1 6 NO rec.aquaria 9 2 YES Ballot 2: (20%) sci.aquaria 6 3 YES sci.bio.marine 10 1 YES rec.pets.fish 2 5 NO rec.pets.aquatic 2 4 NO rec.aquarium 1 6 NO rec.aquaria 9 2 YES Ballot 3: (20%) sci.aquaria 2 5 NO sci.bio.marine 1 6 NO rec.pets.fish 10 1 YES rec.pets.aquatic 8 3 YES rec.aquarium 6 4 YES rec.aquaria 9 2 YES Ballot 4: (20%) sci.aquaria 2 5 NO sci.bio.marine 1 6 NO rec.pets.fish 8 3 YES rec.pets.aquatic 10 1 YES rec.aquarium 6 4 YES rec.aquaria 9 2 YES Ballot 5: (20%) sci.aquaria 2 5 NO sci.bio.marine 1 6 NO rec.pets.fish 4 4 NO rec.pets.aquatic 3 3 NO rec.aquarium 10 1 YES rec.aquaria 9 2 YES In this case, you get the following vote results: MAUVE: sci.aquaria YES 40% NO 60% sci.bio.marine YES 40% NO 60% rec.pets.fish YES 40% NO 60% rec.pets.aquatic YES 40% NO 60% rec.aquarium YES 60% NO 40% rec.aquaria YES 100% NO 0% *** wins *** WEIP: sci.aquaria 4.4 sci.bio.marine 3.8 rec.pets.fish 5.2 rec.pets.aquatic 5.0 rec.aquarium 4.8 rec.aquaria 9.0 *** wins *** STV: Well, in this particular example STV elimates rec.aquaria and runs into deadlock. If the votes were slightly assymetric (not all exactly 20%), you can construct scenarios where almost ANY of the others can win, depending on what order the 'dominos' fall. I didn't try to artfully construct a paradox (I tried to use 'common sense' votes based on what people might want to do), but paradox shows its ugly head anyway. CONCLUSION: I feel that STV is not only complicated to explain and adjudicate, but it is fundamentally flawed as well. It does make an attempt to choose based on preference, but playing some simple games with non-trivial votes shows how truly non-intuitive the percolation of votes can be. Its ranking is also very primitive, and doesn't come any closer to showing the strength of a vote than the simple concensus MAUVE scheme. I think that the MAUVE scheme, as simple as it is, accomplishes the main objective of newgroup voting, finding the proposal with the widest approval and the least objection. It is very easy to explain and adjudicate, and it is intuitively obvious what the effect of your vote will be. If MAUVE is deemed unacceptable because of its lack of preference expression, I think that the WEIP scheme implements preference voting in a fundamentally sound way. It not only has much more flexibility than STV (it is able to show not only the preferences, but the magnitude of preferences - and even ties in preference), but it also incorporates the basic algorithmic nature that makes MAUVE work - the mutual INDEPENDENCE of the votes. Keeping the votes independent removes the cross-dependencies that lead to paradox and non-intuitive vote results under STV. I thank you for having the patience to read this far ... -- --------| If you need help, call the FBI. Alien | - Deteriorata --------| decvax!frog!cpoint!alien bu-cs!mirror!frog!cpoint!alien
brad@looking.on.ca (Brad Templeton) (11/23/89)
STV was explained (the voting process only) in 25 lines in news.groups. But the fact that we're seing dozens of 100 line clarifications and questions does not bode well. We're going to get this with each newcomer, each new self apointed net.police officer. -- Brad Templeton, ClariNet Communications Corp. -- Waterloo, Ontario 519/884-7473
edhew@xenitec.on.ca (Ed Hew) (11/26/89)
In article <51370@looking.on.ca> brad@looking.on.ca (Brad Templeton) writes: >STV was explained (the voting process only) in 25 lines in news.groups. > >But the fact that we're seing dozens of 100 line clarifications and questions >does not bode well. We're going to get this with each newcomer, each new >self apointed net.police officer. ....and all this without even so much as a contested vote! Considering the net.flameage we see with the current simple scheme, I can just imagine the results the very first time someone doesn't like the outcome of an STV vote. Sure, it may be more flexible, but it also appears to be open to so many interpretations that it will settle nothing. >Brad Templeton, ClariNet Communications Corp. -- Waterloo, Ontario 519/884-7473 Ed. A. Hew, SCO Authorized Technical Trainer, XeniTec Consulting Services edhew@xenitec.on.ca -or- ..!{uunet!}watmath!xenitec!edhew