I read somewhere that for most legislators, their voting record can be defined using 2 dimensions.
(Found something that might be it: http://saweis.net/svd/ ) Anyway, the basic point is that you can place each member of a legislature on a 2-d plane based solely on their voting records and then use that to predict future votes. You don't need to actually know what the votes are about. I wonder if a similar trick could be used with voters who submit a ranked ballot. You could try each possible ordering of the candidates. The result that gives the fewest contradictions could be the considered the actual ordering of the candidates. You could then go through all the ballots and place them in a position that minimises the number of contradictions. (Alternatively, you could just place them beside their favourite on the same side as their 2nd preference). Finally, you could then pick the percentile candidates. Ofc, there may be some serious strategy issues with people's rankings. ---- Election-Methods mailing list - see http://electorama.com/em for list info
