The method for rating teams in Example 7.8 is based on actual and predicted point spreads. This method can be biased if some teams run up the score in a few games. An alternative possibility is to base the ratings only on wins and losses. For each game, you observe whether the home team wins. Then from the proposed ratings, you predict whether the home team will win. You want the ratings such that the number of predictions that match the actual outcomes is maximized. Try modeling this. Do you run into difficulties?