Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!rice!brazos.rice.edu!bbc From: bbc@titan.rice.edu (Benjamin Chase) Newsgroups: news.groups Subject: Re: Proposal for changes to the newsgroup creation guidlines. Summary: Long example of preferential voting Message-ID: Date: 20 Oct 89 06:21:35 GMT References: <14718.2538b6f4@kuhub.cc.ukans.edu> <14980.253c557f@kuhub.cc.ukans.edu> <15094@kuhub.cc.ukans.edu> Sender: root@rice.edu Reply-To: Benjamin Chase Distribution: na Organization: CPRC, Rice University Lines: 178 In-reply-to: SLOANE@kuhub.cc.ukans.edu's message of 19 Oct 89 16:16:06 GMT SLOANE@kuhub.cc.ukans.edu (Bob Sloane) writes: First off, Peter da Silva has informed me that the "preferential voting" that I proposed is the same as STV, which is the same as the Australian System. > As I understand STV, the people that just want the group created would > have to vote for ALL of the possible names for the group to express > what they want. This means that there is a large body of voters that > are voting for EVERY name. Yes. Every name (except write-ins), should appear on the ballot submitted with the call for votes. Voting is done by reordering these names. This shouldn't be too arduous, if the voter's mail software cooperates with a text editor. > Suppose I am the vote taker for a new group to discuss lithuanian > sheepdogs. I want it to be in the sci hierarchy. I will just count > the vote for sci.lith-dogs first, and it will probably pass because of > the large vote from those who don't care about the name. No, you can't do that. As the vote taker, you have no control over the outcome, once you get down to tallying votes. The voting is deterministic at that point. Clearly, an example is needed (please forgive any mistakes): ------------------------------------------------------------ Sample ballot: sci.lith-dogs rec.sheep-dogs.lith rec.pets.dogs.sheep.lith none-of-the-below Victory conditions: A single candidate receives twice as many votes as any of the other candidates not yet thrown out. (Fairly harsh, but it will improve the instructive qualities of my example.) Voting community (it is important to inject humor into such dry postings): A lazy voter who wants the group B rec group proponent, deep hierarchy advocate C Richard Sexton :-) D Someone trying and failing to be cute, but who wants the group E rec group proponent, who doesn't think sheep-dogs are pets F Person who wants the group, but is in Lithuania, which only gets rec G Person who wants the group, but thinks lith-dog might make some people think of dogs made of lithium, especially if it's in sci, where people talk about lithium a lot. (cf sci.physics.fusion) H Me Votes: A B sci.lith-dogs rec.pets.dogs.sheep.lith rec.sheep-dogs.lith rec.sheep-dogs.lith rec.pets.dogs.sheep.lith none-of-the-below none-of-the-below sci.lith-dogs C D sci.lith-dogs misc.sheep-dogs.lith ; A write-in none-of-the-below rec.pets.dogs.sheep.lith rec.pets.dogs.sheep.lith rec.sheep-dogs.lith rec.sheep-dogs.lith sci.lith-dogs none-of-the-below E F rec.sheep-dogs.lith rec.sheep-dogs.lith none-of-the-below rec.pets.dogs.sheep.lith rec.pets.dogs.sheep.lith none-of-the-below sci.lith-dogs sci.lith-dogs G H rec.pets.dogs.sheep.lith sci.lith-dogs rec.sheep-dogs.lith rec.pets.dogs.sheep.lith none-of-the-below rec.sheep-dogs.lith sci.lith-dogs none-of-the-below Tallying: First, tally all first-choice votes (the one listed first by each voter). First Round sci.lith-dogs 3 rec.pets.dogs.sheep.lith 2 rec.sheep-dogs.lith 2 misc.sheep-dogs.lith 1 Predictably, the write-in gleans the lowest number of (first-choice) votes, so it is chosen as the loser. Taking all the ballots cast for the losing candidate (ie, Voter D's ballot only), the vote taker recasts those ballots according to their next most (ie. second-choice in this case) desired candidate (ie. rec.pets.dogs.sheep.lith). Second Round sci.lith-dogs 3 rec.pets.dogs.sheep.lith 3 rec.sheep-dogs.lith 2 This time, rec.sheep-dogs.lith is on the bottom of the heap. The vote taker redistributes the votes of the loser, according to their next most desired candidates. (ie. E's vote is recast for "none-of-the-below" and F's vote is recast for rec.pets.dog.sheep.lith) Third Round rec.pets.dogs.sheep.lith 4 sci.lith-dogs 3 none-of-the-below 1 Next, none-of-the-below would be thrown out, except it represents the collection of unhappy voters whose voice must be heard. Instead, sci.lith-dogs gets the axe. C's vote is recast for "none-of-the-below". A's vote would be recast for rec.sheep-dogs.lith, but that candidate has already lost. So, A's next choice is used, and A's vote is recast for rec.pets.dogs.sheep.lith. H's vote is recast for his next choice, rec.pets.dogs.sheep.lith. Final Result rec.pets.dogs.sheep.lith 6 none-of-the-below 2 ------------------------------------------------------------ First things first. I don't know what is the right thing to do when two candidates tie for last place in a given round. I cooked the previous example so that wouldn't happen. Choosing one at random, or allowing the vote taker to pick, would of course not be deterministic. Perhaps Peter can supply the correct Australian behaviour when this happens? The handling of A's ballot may seem odd to some. A's second choice was passed over because not enough voters preferred it enough to keep it in the running. But if A's ballot were thrown out earlier, and recast for A's second choice, then that candidate might have survived longer, or perhaps even won the election. If A had only known that sci.lith-dogs was going to eventually lose, perhaps A would have ranked the candidates differently, bumping other candidates up in the list of preferences. This is peculiar, and I don't understand all the ramifications of this right now. Perhaps Peter will comment on this phenomenon? Does it occur in his STV scheme also? My example shows that the vote taker has no influence on the outcome, _once the votes are received_. However, if the vote taker is allowed to prepare the ballot, the results can be influenced. In particular, lazy voters (or voters that have no preference!), such as voter A, may not modify the list substantially, thus supporting the vote taker's preferences. I suppose different methods of tallying the votes are possible. Perhaps point values could be assigned to each rank, and the points for all the occurrences of all the candidates are totalled and the highest wins. This is less (conceptually) complicated than the scheme I showed above. I'm not sure how the results turn out. The vote could still be influenced by careful crafting of the sample ballot combined with lazy voters. Also, something would have to be done with write-in votes. However, this voting scheme does allow the simultaneous selection of a name from a list of candidate newsgroups and a vote on its creation. ------------------------------------------------------------ Actually, I don't give a rat's ass whether or not this proposal is accepted as the new voting procedure or not. I bothered to generate this overly long example in the hopes that it would help people understand this method of voting. But I think that while I've been frank about the pitfalls of preferential voting, Bob Sloane has failed to mention any detriments that his proposal might have. I feel like I'm being sold a used car. For instance, was it mentioned that people using wildcard votes might be unwittingly voting for write-in candidates? Are there ways to make the ballot simpler for the voter to understand? Have you considered using "all" or "any" instead of "*" to signify a lack of preference? I've seen "all" used on the net with that meaning. Surely you've thought of some of these points? Blah, blah, blah, -- Ben Chase , Rice University, Houston, Texas