Once I have assigned pre-season strength ranks and ratings to teams, I combine those with teams' schedules to "predict" where teams will end up at the end of the season following completion of the conference tournaments. The process to do this requires background work:
1. For each game, I calculate the game-location-adjusted rating difference between the teams, using their assigned pre-season strength ratings as the base. Since the assigned strength ratings are based on what the average historic NCAA RPI ratings are for those teams' ranks, my game location adjustments increase the home team's rating by 0.0085 and decrease the away team's rating by 0.0085, for an overall adjustment of 0..0170. This is the value of home field advantage for the current version of the NCAA RPI with the "no overtime" rule in effect.
2. For the location-adjusted rating difference between the teams, I calculate each team's expected win, loss, and tie likelihoods. These likelihoods are based on a study of the location-adjusted rating differences and results of all games played since 2010 (excluding 2020).
[NOTE: For a detailed explanation of how I determine the game location adjustment and the win, loss, and tie likelihoods, go to the RPI for Division I Women's Soccer website's page RPI: Measuring the Correlation Between Teams' Ratings and Their Performance.
3. Rather than assigning the opponents in a game either a win, a loss, or a tie result, I assign each team its win, loss, and tie likelihoods since these will give a better picture of what a team's overall record will be given its entire schedule. As an example, Colorado and Michigan State will play on August 14 at Colorado. Their location-adjusted rating difference is 0.0333 in favor of Colorado. For that rating difference, referring to a result probability table for the current NCAA RPI in a "no overtime" world, Colorado's result likelihoods are 53.6% win, 19.9% loss, 26.4% tie (which don't quite add up to 100% due to rounding). Those numbers go into Colorado's win-loss-tie columns for NCAA RPI computation purposes. Michigan State's win-loss percentages are the converse. I assign these likelihoods for all teams' games, add up each team's percentages, and convert them from percentages to numbers . Thus Colorado ends the season with 12.2 wins, 4.9 losses, and 4.9 ties. These are the numbers I use for Colorado as its wins, losses, and ties when computing its NCAA RPI.
4. I then compute all teams' NCAA RPI ratings and ranks. It is important to understand that these are different than the teams' assigned pre-season strength ratings and ranks. This is because the NCAA RPI does not measure team strength. Rather, it measures team performance based on a combination of teams' winning percentages and their strengths of schedule (as measured by the NCAA RPI formula). Thus two teams with identical strength ratings and ranks will end up with different NCAA RPI ratings and ranks if they have different winning percentages and/or different strengths of schedule. Below are my computed end-of-season (including predicted conference tournaments) ratings and ranks for teams. You can compare the ranks to the ones in the preceding post to see the differences between teams' assigned pre-season strength ranks and team' predicted end-of-season NCAA RPI ranks. (NOTE: I have corrected these rankings since their initial publication to fix a programming error. The changes are relatively minor.)
No comments:
Post a Comment