This article is based on the actual results of games played through Week 7 (ending September 27) and simulated results of games not yet played, including conference tournament games. It projects where teams will end up at the end of the season, but before the NCAA Tournament.
Here is a link to an Excel workbook with the simulated full-season NCAA RPI and Balanced RPI ranks -- and other important information -- for teams, conferences, and regions.
2026 Article 8 has an explanation of the method I use to generate the simulated ranks and of the key information in the workbook. Article 8 also has instructions on the best way to download the workbook.
For purposes of evaluating the usefulness of the ratings the simulation currently uses to predict the results of games not yet played:
The table's reference to Team 1 is to the home team or, if a neutral site game, the team whose name comes first alphabetically.
The first data row in the table, 2026 Team 1 Actual Results, shows Team 1's actual results. The second data row, 2026 Team 1 Predicted Results, applies teams' simulated end-of-season RPI ratings to games actually played through September 27, to determine what Team 1's win, loss, and tie percentages should be if those ratings are correct. The third data row then shows the difference between those predicted percentages and the actual percentages. Thus, for example, the actual winning percentage for all Team 1s was 48.8%. According to teams' end-of-season simulated ratings, the winning percentage should have been 50.0%. The Actual Less Predicted Results row shows the difference between predicted and actual was -1.187%.
The last three data rows are similar, but instead of using teams' simulated end-of-season RPI ratings, it uses teams' actual current NCAA RPI ratings to determine what the percentages should be if those ratings are correct.
As you can see, the simulation's current ratings are a better match with actual results than are the actual current NCAA RPI ratings. Because of this, for now I will continue using the simulation's ratings to predict future game results rather than converting to using the NCAA RPI's current ratings. In the upcoming weeks, I will continue doing this until the NCAA RPI's current ratings perform better than the simulation's current ratings.
Starting this week, each week I'll highlight here some data from the workbook that may be helpful to those interested in how the NCAA RPI works. For this week, I'll cover some data related to regions.
I assign conferences to regions based on their scheduling patterns, i.e., Against which other conferences, over the years, have their teams tended to play their non-conference games? This has led me to organize the conferences into the following regions:
Autonomy (i.e., Power 4): ACC, Big Ten, Big Twelve, SEC
Middle: Horizon, MidAmerican, Missouri Valley, Ohio Valley, Summit
MidEast: Atlantic Ten, Big East, Colonial (now Coastal), Ivy
North: America East, Metro Atlantic, Northeast, Patriot
South: American, Southland, Southwestern, Sun Belt
Southeast: Atlantic Sun, Big South, Conference USA, Southern, United
West: Big Sky, Big West, Mountain West, Pac Twelve, West Coast
The following table is from the linked workbook's Regions page:
This table shows the proportion of games each region's teams play against opponents from their own region and also against teams from each other region. Among other things, this allows you to see how much "linkage" there is between regions. Looking at the West region, as an extreme example, its teams play 85.1% of their games "in region" and 7.7% of their games against Autonomy teams, for a total of 92.8% of their games within those 2 regions. That leaves 7.2% of their games spread among teams from the other regions. That means each of the West teams on average plays a little over 1 game per year against opponents from all the other non-Autonomy regions combined. In other words, the West region plays almost in isolation from the other regions except for the Autonomy region.
The next table shows the proportions of in-region ties:
Prior to 2024, the NCAA RPI formula for a team's Winning Percentage counted a tie as half a win. In 2024, the Women's Soccer Committee changed this to a third of a win, thus reducing the value of a tie. This table shows, for this year, who is likely to be hurt by and who benefits from this change. As you can see, the MidEast teams are the most likely to be hurt and the North teams are most likely to benefit.
The next table is about the NCAA RPI's basic defect, from a regional perspective:
As I have shown elsewhere, how the NCAA RPI evaluates a team overall is different than how it evaluates that team when it comes to that team's contribution to an opponent's strength of schedule. This leads to teams' NCAA RPI ranks being different than their NCAA RPI ranks as strength of schedule contributors. In the above table, the first highlighted column shows the region's teams' NCAA RPI ranks less their NCAA RPI ranks as strength of schedule contributors. A negative number means the NCAA RPI strength of schedule contributor rank is poorer than the overall NCAA RPI rank. A postive number means it is better.
Using the Autonomy region as an example, its teams' NCAA RPI strength of schedule contributor ranks on average are 29.4 positions poorer than their actual NCAA RPI ranks. In other words, if you play an Autonomy opponent, on average your own RPI is going to get credit for playing a signifcantly weaker opponent than the NCAA RPI itself says you should be getting. Since Autonomy teams mostly play other Autonomy teams, this means that the NCAA formula on average undervalues those teams' strengths of schedule.
The table thus shows which regions benefit from and which are hurt by this NCAA formula defect.
The purpose of the Balanced RPI is to fix this NCAA RPI formula defect. As the third column in the table shows, for all regions, there is almost no difference between how the Balanced RPI ranks regions' teams overall and how it ranks them as strength of schedule contributors. Thus the Balanced RPI data in the workbook's Regions page show what happens when you eliminate the NCAA RPI's defect. The following table shows what happens, from a regions perspective:
As you can see, Autonomy teams' average ranks suffer the most from the NCAA RPI defect, averaging 24 positions poorer than they should be as indicated by the Balanced RPI. Next come teams from the West, averaging 18 positions poorer; and then the MidEast averaging 3 positions poorer. All the other regions' teams, on average, are ranked better than they should be.
SIMULATED NCAA TOURNAMENT BRACKETS USING THE NCAA RPI AND USING THE BALANCED RPI
Here are currently simulated NCAA Tournament brackets, first based on the NCAA RPI and then based on the Balanced RPI. Following those is a table showing how the two compare. (NOTE: For conference tournaments, I have used simulated conference tournaments based on the NCAA RPI.) The information above, particularly the table just above, provides context for understanding what you will see in the comparison table.
Here is a key to the tables:
The first table shows the projected NCAA Tournament bracket based on the Committee's using the NCAA RPI:
The next table shows the projected bracket based on the Committee's using the Balanced RPI:
The final table compares the projected brackets. It shows the teams sorted first by region and within regions by conference so you can see how a change to the Balanced RPI would affect regions' and conferences' teams. Salmon highlighting means the team gets a poorer treatment by the rating system and green means the team gets better treatment. No highlighting means the treatment is the same.
NOTE: My weekly Comparison tables are preliminary and will evolve throughout the season, so it is not the details that are important but rather the overall impression in terms of the difference between using the NCAA RPI as compared to the Balanced RPI. In relation to the regions data in the first part of this report, the comparison shows the effects of the NCAA RPI's defect in relation to strength of schedule and what happens if you eliminate the defect.
No comments:
Post a Comment