After the first round of RCV, there were 17808 votes in the pool. By Round 19 that had fallen to 8200. So, yes, only 46% of the ballots made it that far, and Supervisor Cohen's 4321 first, second, and third place combined votes were enough to beat Tony Kelly's 3879. In the first round, Kelly had 2102 and Cohen 2097, while Lynette Sweet had 2150, topping both of the eventual final round candidates. The point here is that Kelly and Cohen were more 'palatable' than Sweet, across the three choices. Had we had a simple runoff style election based solely on first round voting, the choice would have been between Sweet and Kelly, not Cohen and Kelly.
You've got to ask, too, how many people come out one month after having just voted for another election? Would 46% of the people who were interested enough the first time around vote again? Barely. In the last couple of elections in which we had run-off elections, voter turnout dropped roughly 50% (http://www.sfrcv.com/) between November and December.
So, in 2010, if we'd had a run-off election, it would have been between Kelly and Sweet, and perhaps 8900 people (50% of the November election) would have cast ballots. It would have been a close election, and whoever won would have done it with somewhere slightly over 4450 votes. While we would have ended up with someone different, the vote tally that got them there would have been nearly the same as using RCV.
RCV isn't perfect, it's made out to be more confusing than it really is, but does it lead to voter suppression, or your vote not counting in the end? Maybe. If we didn't have RCV during the 2010 election, people who voted for any one of the candidates who didn't end up in the run-off would be given the chance to choose again in December between the remaining highest vote-getters. With RCV, people who voted for a losing candidate the first time and whose ballots were exhausted were not given the chance to select their preference from the remaining candidates. In the end, 9600 voters for whom neither Kelly nor Cohen was their choice are left with a supervisor they didn't put in their top three. So did 4300 voters override the 9600 to put Cohen in office? Seems that way to me, but I could be wrong.
I suspect the bottom line is that 3-choice RCV works for a small number of candidates, but for the numbers we saw in D10 last year and in the Mayor's race this year, it falls apart. Perhaps RCV should be changed so that when the number of candidates is below 10, you get three choices; below 20, you get four; above 20, you get five choices.
