Oops! And a good thing for double checks!
The good thing about having implemented all these things 4-5 times is that I had another set of code to check myself against, and that agreed with Mr. LeGrand's calculations. I found the bug in my new code which erroneously reported that IRV and VRR disagreed on this one. Now all is agreement, so we must even be counting tie-rank-votes the same. Fixed results of a bunch of systems at these spots: http://two.bolson.org/a.html http://two.bolson.org/2b.html On Mar 10, 2006, at 11:25 PM, Rob LeGrand wrote: > I took the raw data from the site Brian Olson posted and ended up > with a > slightly different pairwise table: > > (0) (1) (2) (3) (4) (5) > (0) - 5545 6747 3991 6790 7336 > (1) 3556 - 5165 3397 5136 5875 > (2) 1161 1289 - 804 2028 3290 > (3) 4763 5730 6961 - 7027 7351 > (4) 987 1318 1869 603 - 3094 > (5) 323 311 431 194 510 - > > (0) Hinda Miller > (1) Kevin J. Curley > (2) Louie The Cowman Beaudin > (3) Bob Kiss > (4) Loyal Ploof > (5) Write-ins > > It's identical to the one at http://two.bolson.org/a.html except > for the > first column: I have Hinda Miller with many more votes against > her. I've > looked for errors in my results but haven't found the problem. If I'm > right, the Condorcet winner changes from Miller to Bob Kiss, the IRV > winner. I think finding a real-world election in which IRV failed to > elect the voted Condorcet candidate would be quite significant, so I'm > hoping my results are wrong, but they do seem more consistent with the > final round of the real IRV election (although they're still off by a > handful of ballots). It'd be great if someone else could take the raw > data from the election page and triple-check the results > independently. > Regardless of the result, thanks to Brian for pointing this out. > > -- > Rob LeGrand, psephologist > [EMAIL PROTECTED] > Citizens for Approval Voting > http://www.approvalvoting.org/ ---- election-methods mailing list - see http://electorama.com/em for list info