EXPLANATION OF DATA AND
Probability of Winning
Since late in the 2004 season, I have been using my NGE analytical model to forecast the scores for both teams and the winning or losing margin. The question often arises, “How accurate is this model for predicting the results of the game. The chart above shows the probability of winning a game as a function of the NGE Model's predicted margin.
The ability of this model to predict the winner of a game is really a function of the predicted difference. If the predicted difference averages 6 points per game, then the model should predict the winner correctly for about70 to 75 percent of the time. When the average predicted margin falls to 3 points, then the probability of correctly predicting the winners falls to about 60 to 65 percent.
Since I use the model primarily to predict the scores for one team, the success of the model will rise and fall with the success of that team. When Kentucky is performing with a NGE of 0.140, as it had last year, the average predicted margin was about 7 points and the model was correct about 90 percent of the games with respect to the predicted winner. This season, the team has a NGE of about 0.05, and the average predicted margin has been about 3 points. The Model has predicted about 2/3 of the winners correctly.
The more powerful the team, the higher their NGE and the higher the average predicted margin for the games.
These same relationships also illustrate how NGE relates to a team's winning percentage.
Copyright 2006 Richard Cheeks