Forest wrote: > 2 % approve A only. > 25 % approve both A and B > 23 % approve C only. > > But note that (in the sample) every ballot that approves B also > approves A. If this is a real feature of the population, it will be > impossible for B to get more approval than A. > > I think I would put my approval cutoff on the C side of A.
I would too, if I were convinced that the correlation was a real feature of the population. It depends on what your model of the electorate is like. If you have a model based solely on distributing the candidates' expected vote totals, you'll want to use something in the style of strategy A. If you create a distribution based solely on the ballots themselves (so that the numbers of different ballots are independent of one another), you'll get something quite different. Here's a more dramatic example: 30% approve A and B 20% approve C and D Say you rank the candidates A>B>C>D. Strategy A would recommend voting only for A. But the ballot-distribution approach would recommend voting for both A and C, skipping B. The only approval strategy I know of that takes as input only vote totals and ever skips is Warren Smith's moving-average strategy. > If nothing else, it is easy to approximate the winning (and tie for > first place) probabilities by repeated Montecarlo simulations based on > the distribution of the approval ballot profiles in the poll. I'm working on using a similar kind of idea to develop an objective way to evaluate plurality, approval and Borda strategies for use with DSV. -- Rob LeGrand, psephologist [EMAIL PROTECTED] Citizens for Approval Voting http://www.approvalvoting.org/ __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com ---- election-methods mailing list - see http://electorama.com/em for list info