On removing Condorcet effects from pairwise election tallies