Reply #13: A little off topic, about IRV

[redeye]

IRV positively sucks. It prevents the spoiler effect, but it becomes terrible once there are more than two viable candidates. If there are three candidates - a leftist, a centrist, and a rightist - each of whom has roughly 1/3 of the total first-place vote, with supporters of the centrist being equally divided among the other two candidates fro second-place and supporters of the two other candidates all voting for the centrist for second-place, then we can have some bizarre results.

If the leftist in total has the same number of second-place votes than the rightist, and has 1000 more first-place votes than the centrists who has 100 more first-place votes than the rightist, then the rightist will drop and thus let the centrist win by a 2-to-1 margin. However, if 200 leftists break ranks and vote R, then L, then C, instead of L, then C, then R, then the centrist will drop, transferring votes for the leftist to win with a bare majority of 600 votes.

In other words: voting against a candidate can make him win.

Moreover, counting IRV is a nightmare. It's impossible to proceed to round two until all of the votes in round one have been counted, and thus all ballots have to be sent to a central counting location before counting can begin (or else there might be tampering with ballots in the interim). It is theoretically possible to tally all permutations, but the number of such permutations is n! whereas n is the number of candidates, which means that the number of permutations is non-polynomial (Condorcet, by the way, also has a non-polynomial counting time but only if there are cyclical ties, while it can be reduced to a matrix of n² entries, which is polynomial).