Each year, my computer applies a program to the full season schedule to predict where teams will end up in the RPI rankings at the end of the season (including conference tournaments). It also predicts where teams will stand in relation to automatic qualification for and at large selections (and also seeds) for the NCAA Tournament. This article shows the program's pre-season predictions.
But first some background on the program, which I will put in this article and not repeat in my weekly updates. You can use the background to do your own assessment of the reliability of the program's predictions.
Background
The program predicts the outcome of each game by comparing a pre-season NCAA RPI rating I have assigned to each team involved in the game and taking into account the value of home field advantage. (Home field is worth 0.0166 in relation to the rating difference between the two opponents.) In predicting the outcome of each game, for purposes of the rating computations, the program does not simply award the higher rated team a win. Rather, for each game it assigns a win, tie, and loss probability based on historic probabilities in relation to opponents' location-adjusted rating differences. This avoids, for example, having a team with a 51.0% win probability in each of 10 games being awarded a win in each game. Instead, it creates a record for the team of 5.1 wins, 2.8 ties, and 2.1 losses, reflecting that the win probability in each game is 51.0%, the tie probability is 27.7%, and the loss probability 21.3%. From a statistical perspective, this is the record one would expect for the 10 games (of course, rounded off to whole numbers for actual game outcomes) rather than expecting the team to win all 10.
Teams' predicted RPI ratings are based on teams' average Balanced RPI ranks over the last 7 years. I use the last 7 years because, of the possible measures I have considered, using that average produces the best match with the next year's ratings. The measures I have considered include average NCAA RPI ranks over the last 1, 2, 3, ... 15 years and average Balanced RPI ranks likewise over those numbers of years. Once I have the 7 year average Balanced RPI ranks of teams, I put them in order from best to worst, assign them rank positions, and then assign each team the average historic NCAA RPI rating associated with its rank.
Using this method, the program does best at predicting where teams' ranks will end up at the top and bottom ends of the rankings and the poorest in the middle. This is because the ratings are most spread out at the top and bottom of the rankings and most compressed in the middle. Thus at the top and bottom of the rankings, a predicted rating "error" of X may not be sufficient to change teams' ranks, in other words will be inconsequential. In the middle of the rankings, however, the same rating "error" may be equivalent to a significant number of rank position changes.
From an NCAA Tournament perspective, the teams that matter are the NCAA RPI Top 57 (see below). Based on a review I did applying the program to the 2022 season, a reasonable expectation for teams ending up in the NCAA RPI Top 100 is that teams in the following rank groups will have actual end-of-season ranks, on average, within the indicated number of positions of their pre-season predicted ranks:
Teams ranked 1 through 10: 2 positions
Teams ranked 11 through 20: 3 positions
Teams ranked 21 through 50: 8 positions
Teams ranked 51 through 100: 12 positions
The other thing to bear in mind is the NCAA RPI rank groups that, based on past history, are candidate groups for NCAA Tournament seeds and at large selections at the end of the season. These are as follows:
#1 Seeds: teams ranked #7 or better
#2 Seeds: teams ranked #14 or better
#3 Seeds: teams ranked #23 or better
#4 Seeds: teams ranked #26 or better
#5-6 Seeds: teams ranked #30 or better
#7-8 Seeds: teams ranked #49 or better
At Large: teams ranked #57 or better
The way the system works, the initial predicted end-of-season results I will show here are based on predicted results for all games. At the end of each weekend, when all the previous week's actual game results are available, I change the predicted results for those games to the actual results and re-compute to get new end-of-season predictions. Thus at the beginning of the season, the predicted end-of-season results are likely to have the reliability indicated above. Each week, as I substitute actual results for predicted results, the predicted end-of-season results should be more reliable.
Further, after Week 5 of the season, I will stop using the pre-season predicted ratings as a basis for predicting future game results and instead will use teams' then current actual RPI ratings as the basis for predictions. This will coincide with the NCAA's release of its first official RPI ratings and ranks for the season.
Predicted Final Ratings and Ranks
Here are this year's predicted final ratings and ranks. They include the NCAA RPI, NCAA Non-Conference RPI, and Balanced RPI. For detailed explanation of those three sets of ratings, use these links:
In the table, the Adjusted RPI 2015 BPs is the NCAA RPI, the Adjusted NCRPI 2015 BPs is the NCAA Non-Conference RPI, and the URPI 50 50 SoS Iteration 15 is the Balanced RPI. The regions are based on the states where the teams are located and where the teams from their states play the majority or plurality of their games.
Predicted Final NCAA Tournament Automatic Qualifiers and At Large Selection Status
The table also shows which teams the program identifies as Automatic Qualifiers and which ones it identifies as having below 0.500 records and thus being disqualified from at large selection. The most likely at large selections are the Top 34 teams in the table that are not Automatic Qualifiers and not disqualified due to below 0.500 records.
Projected Final Conference Ratings and Ranks
No comments:
Post a Comment