Stretching the Effectiveness of MLE from Accuracy to Bias for Pairwise Comparisons