I am using this year’s NCAA tournament as a testing ground for my cumulative connections ranking method (explained here) compared to other ranking methods (both computer and human). The evaluation is simple. I am just going to keep track of the percentage of games in which the higher ranked team in each scheme actually won the game. Keep in mind that the rankings are all based on pre-tournament data only and will not change throughout the tournament.
After the first 4 games (a very small sample), the cumulative connections ranking scheme has predicted the winner four out of four times. In other words, the winner of each game was ranked higher by the cumulative connections ranking method. This result seems rather impressive because each of the 1st 4 games were between identically seeded teams meaning that predicting the outcome should have been relatively difficult.
Here are the overall results so far…
If a ranking method correctly predicted the outcome it received 1 point, if it did not, it received 0 points. If it did not predict the game (both teams were unranked or had the same ranking) then the cell is left blank.