It's more nuanced than this. RCV helps the median in the way you suggest when the median is more popular, by allowing the more extreme voters to avoid splitting

It's more nuanced than this. RCV helps the median in the way you suggest when the median is more popular, by allowing the more extreme voters to avoid splitting the vote. In the example, by allowing Nader voters to fall back to Gore once Nader is eliminated. But, when the race is closer, it flip flops to have the opposite direction. Say Nader is marginally more popular among liberals. Now Gore is eliminated first, and Nader beats Bush. But hold on, what about the conservatives, who voted Bush > Gore > Nader? They've been screwed, because Gore was eliminated before Bush, so the system never considered their preference for Gore over Nader. And actually, if enough of them had flipped their vote and gone Gore > Bush > Nader, then Bush would be eliminated first, resulting in Gore's victory -- the lesser evil. If you are liberal and have a hard time identifying this as a problem, consider the reverse: an election in the south, where voting D > R > Tea party results in the tea party candidate winning, but voting R > D > Y elects the mainstream Republican. So the short summary is that, under RCV, it's only safe for you to put your favorite candidate first if: A) Your favorite is likely to win, or B) The election is not close -- you know your favorite(s) will be eliminated before your next choice, who is likely to win. It is still an improvement over FPTP, where you always have to vote for a likely-to-win candidate, but not as good as any form of score voting (e.g. approval, STAR).