http://en.wikipedia.org/wiki/Instant-runoff_voting#Voting_sy...
80 people: A, C, B
50 people: B, C, A
35 people: C, B, A
IRV eliminates C (as it has the fewest first-place votes) and elects B. But voters on the whole prefer C over B (115 to 50). This is the failure that Pinckney refers to.
Under ranked pairs, each pair of candidates is considered individually, largest majorities first.
In bradbeattie's example we have:
C > B: 115 votes
C > A: 85 votes
B > A: 85 votes
So the group preference is C > B > A and C wins.Ranked pairs fails the participation criterion[2], but it is still possible to vote tactically in a way that guarantees you will not harm your first-choice candidate by participating (since if your first choice is A you can rank A first and all the others joint second. Similarly if your goal is to prevent C from winning, you can rank C second and all others joint first.)
[1]:http://en.wikipedia.org/wiki/Ranked_pairs [2]:http://en.wikipedia.org/wiki/Participation_criterion
http://en.wikipedia.org/wiki/Preferential_voting
However, AFAIK, IRV is the system which most closely fits what gradstudent described.