Fig. 3. Efron and Morris’ baseball data (The left panel shows sum of squared prediction error for several methods, plotted against number of players reversed. The line at the top of the plot is for the player-wise maximum likelihood estimator; the black dots and lines are for an empirical Bayesian shrinkage estimator; the red dots and lines are for a mixture of Dirichlet processes model. The lines extend from the 25th to 75th percentiles of sum of squares, accounting for different reversals. The gray line at the bottom of the plot is an estimate of non-eliminable prediction error. The right panel provides an estimate of the reduction in non-eliminable risk when compared to the player-wise maximum likelihood estimator. Blue for the maximum likelihood estimator, black for the empirical Bayes estimator, and red for the nonparametric Bayes method.).
© 2016 Communications for Statistical Applications and Methods