Pairwise Comparison Ratings

elo chess

inv func theorem

31

MLE Issues

nNon-uniqueness

nIdeally, wins help and losses hurt

nUndefeated / Winless Teams

n

n

n

nUpdate ratings based on linearization

(time dependent, n large)

n

n

• Use a linear approximation to f

• Don’t use the WIF.

• Use prior distribution (Bayesian)