tl;dr: Instead of eliminating the candidate in last place, eliminate the loser of a head-to-head matchup between the bottom two candidates. This ensures the Condorcet winner (if one exists) is never eliminated.
In instant-runoff voting (IRV), each voter ranks the candidates in order of preference. The winner is then determined by repeated elimination, using the following procedure:
A major drawback of IRV is that it is not a Condorcet method. That is, IRV can elect a candidate A even when another candidate B would beat every other candidate (including A) in a head-to-head matchup.
We can tweak IRV to make it a Condorcet method by changing what happens in step 1 of the above procedure. Rather than automatically eliminating the last-place candidate, we hold a head-to-head matchup between the bottom two candidates and eliminate whichever one loses. The tweaked procedure is as follows:
Why does this make it a Condorcet method? A Condorcet winner, by definition, beats every other candidate head-to-head. So the Condorcet winner (if one exists) can never lose one of these bottom-two matchups, and therefore can never be eliminated, and therefore must be the last candidate standing.
Ties: In both standard IRV and tweaked IRV, there may be ties when two candidates appear as first preference (among remaining candidates) on the same number of ballots. These ties can be resolved in tweaked IRV using the same methods as for standard IRV.
| Percentage of voters | Ballot ranking |
|---|---|
| 38% | (1) Alice, (2) Bob, (3) Charles |
| 15% | (1) Bob, (2) Alice, (3) Charles |
| 15% | (1) Bob, (2) Charles, (3) Alice |
| 32% | (1) Charles, (2) Bob, (3) Alice |
Bob has the fewest first-place votes (15% + 15% = 30%) but is the Condorcet winner, so the two methods diverge immediately: