RPI: Measuring the Correlation Between Teams' Ratings and Their Performance

Updated June 2025

How good are the NCAA RPI ratings at evaluating teams' performance?  How does home performance compare to away performance, in relation to teams' RPI ratings?  How well do conferences' teams  perform in relation to their conference's RPI strength?  How well do teams from different geographic regions perform in relation to their region's RPI strength?  How well do different variations of the RPI perform when compared to each other?  How well does the RPI perform as compared to other systems?

To answer questions like these, I developed a computer tool for testing how well groups of teams perform in relation to their ratings -- for example, home teams and away teams, teams grouped by conference, and teams grouped by geographic region.  I use the tool to evaluate not only the NCAA RPI, but also other rating systems.   I call the tool the "Correlator," since it determines the extent to which a group of teams' actual game results correlate with the teams' ratings.

This page is a technical explanation of how the Correlator works.  In the interest of transparency,  I give lots of detail, since I use the Correlator to show the nature and extent of the NCAA RPI's defects and I want those interested to be able to make their own judgments about whether the Correlator's analyses are correct.

For the explanation, I will use as an example how the 2024 NCAA RPI formula works as applied to the seasons from 2010 through 2024 with the No Overtime rule applied retroactively to 2010.  Whenever I refer to the NCAA RPI, that is what I am referring to.  The data base for that period has just under 44,000 games.

DETERMINING THE VALUE OF HOME FIELD ADVANTAGE

In order to know how teams perform in relation to their ratings, it first is necessary to determine the value of home field advantage.  A way to think about this is, suppose two teams will play each other.  Team A has a rating of 0.6000.  Team B has a rating of 0.6020.  If they play at a neutral site, Team B will be slightly favored.  If they play at Team A's field, however, then Team A might be favored depending on the value of home field advantage.  In other words, to determine the expected game result based on the opponents' relative ratings, one must adjust the teams' ratings to take the game location into consideration.

I use the Correlator to determine the average value of home field advantage.  To do this, I consider the most closely rated 3% of games -- a little over 1300 games.  These are games where home field advantage is most likely to affect game results and thus its value is the most measurable.  For less closely rated games, home field advantage still is there but is less likely to be needed to change game results and thus is less measurable.  Indeed when the rating difference is big enough, home field advantage is almost irrelevant to game results and thus is not measurable at all.  I use the most closely rated 3% of games rather than a lesser percentage simply to assure I am using a big enough data set.

As a starting point for understanding the Correlator, consider that there always are going to be games where the higher rated team ties or loses.  Let's assume that across all games the higher rated team wins 65.3%, ties 21.0%, and loses 13.7% of the time.  (These are the actual NCAA RPI numbers.)  Now, assume that Conference A's teams, in non-conference games where they are higher rated, win 65.3% of the games and, when they are lower rated, tie 21.0% and lose 13.7%.  Together these numbers add up to a "performance percentage" of 100%.  This means the conference's teams performed exactly as expected based on their ratings.  Now suppose they win 70.0% of the games where they are higher rated and win or tie 40% where they are lower rated.  These numbers add up to 110%, meaning they performed better than their ratings said they should.  In other words, they are underrated. 

Now, going back to measuring the value of home field advantage, the Correlator first calculates home teams' performance percentage when there has been no rating adjustment for home field advantage.  That calculation shows that in the most closely rated 3% of games, home teams have a performance percentage of 114.1%.  Conversely, away teams' performance percentage is 85.9%.  In other words, home field advantage is real and unless it is accounted for home teams will appear to be underrated and away teams overrated.

The next step is to determine the value of home field advantage.  To do this, I try out different rating adjustment values until I have homed in on an adjustment value at which home and away teams performance percentages are 100% -- exactly where they should be if their location adjusted ratings are just right.

In this table, the Home Addition Away Subtraction column shows possible game location adjustments to opponents' ratings to take home field advantage into consideration.  The way the Correlator does this is to add the amount to the home team's rating and subtract the same amount from the away team's rating, to reflect that one expects the home team will perform better than its rating when at home and the away team will perform more pooly when it is away.  (Alternatively, I could simply add twice the amount to the home team's rating, achieving the same effect.)  What the Correlator is looking for is the adjustment amount at which home and away performance is 100% -- in other words, teams are performing in accord with their location adjusted ratings.

To find the adjustment amount that produces a 100% performance percentage, the Correlator creates the following  table, which is an expansion of the above table:

Using this table as the data source, the following chart shows the relationship between home teams' performance percentages (vertical axis) and the amount of the game location adjustment (horizontal axis).  The blue markers are the data points for the performance percentages in the right hand column of the above table.

The dotted line on the chart is a computer generated straight line trend based on the data points.  The formula on the chart is for the trend line, and can be used to see the statistically expected performance percentage for any game location adjustment.  The R squared value is a measure of how well the trend line represents the data.  An R squared value of 1 would mean that the data points all are exactly on the trend line.  An R squared value of 0 would mean there is no relationship between the trend line and the data points, i.e., their distribution around the trend line is completely rancom.  The R squared value for the trend line in this chart, of 0.957, means that the trend line is highly reliable.

I use the trend line formula to produce another table, which is an expansion of the preceding table.  The Code column has a value to plug in as X in the formula to produce the trend line performance percentage (Result column) that goes with the adjustment amount in the Home Addition Away Subtraction column.  The chart thus shows in the right column the statistically expected performance percentage for each adjustment amount in the left hand column.  I then can scroll down the table until I find the 100% performance percentage and see what the corresponding adjustment amount is.

As you can see, for the 2024 NCAA RPI formula with no overtimes, the adjustment amount is ,0085.  Thus the home team gets an upward rating adjustment of 0.0085 and the away team a downward adjustment of 0.0085 to determine their appropriate ratings when calculating the rating difference between the two of them for purposes of determining their expected game result.

For the calculations underlying all of the following tables and charts, the rating differences between teams in games always have been adjusted based on game locations.

EVALUATING THE RATING SYSTEM

PRELIMINARY CALCULATIONS

The next step is a series of calculations related to how teams perform in relation to the location-adjusted rating differences in games.  The first part of this process produces the following table:

In this table, all games are broken down into groups of 500 based on how closely rated the opponents are.  At the top of the table are the 500 games with the greatest rating differences.  At the bottom are the games with the smallest rating differences.  The RPI Win Correct column shows the number of games in which the better rated team won.  The Tie and Loss columns show the number of games in which the better rated team tied or lost.  The very top row shows the number of games totals for the columns, which is overall performance by the better rated team without consideration of how closely rated opponents were.

The next table shows percentages of wins, ties, and losses, based on the numbers in the preceding table.  The very top row shows the percentages for all games.  And, the column on the right shows the median rating difference for each group of 500 games.

The next table is an expanded version of the above one.

This table provides the data for the following chart:

This chart shows the relationship between the better rated teams' win, tie, and loss likelihoods and the rating differences between the teams and their opponents.  Blue is for win likelihood, red is for ties, and grey is for losses.  The three formulas are for the trend lines, with the wins formula at the top, then ties, then losses.  The R squared values, as explained above, show how well the trend lines represent the data.  As you can see, the R squared values are high, which means the trend lines are a good basis for determining expected win, tie, and loss likelihoods for the different rating differences between opponents.

Starting with the preceding table, applying the trend line formulas results in the following table.  It shows, in the three Formula columns, statistically expected win, tie, and loss likelihoods in relation to the rating differences in the left-hand column

And, from this table, with some obviously needed manual adjustments at the very high and very low ends of the spectrum, comes the following Result Probability Table, which shows the win, tie, and loss probabilities for various rating differences between opponents:

Because the Correlator uses this table, it is important to know whether the table is reliable.  The following table compares overall  expected results (based on the Table) to actual results for a rating system.

In this table, the data rows show number of games and percentage of games the numbers represent.  The Actual columns, on the right, show the actual number and percentage of wins, ties, and losses for the higher rated teams.  The Likelihood columns, on the left, show the expected number and percentage of wins, ties, and losses determined by using the Result Probability Table and adding together all the win, tie, and loss likelihoods to reach the totals.

As you can see, the expected results using the Result Probability Table are almost identical to the actual results.  This verifies that using the Result Probability Table to see what teams' results should be, and comparing those results to what the results actually are, is a precise way to see if the rating system is properly rating teams.  Where this is useful is when comparing expected results to actual results for individual teams or for a group of teams, such as those in a conference or geographic region.

ALL TEAMS AS A GROUP

Drawing from the top of the second table in the preceding section, the following table shows the consistency of the NCAA RPI's ratings with actual game results when looking at all teams as a group:

The above table looks at how teams have performed in relation to their ratings.  The Correct column simply shows the percentage of time the better rated team won, without regard to the size of the rating difference with its opponent.  The Tie column shows the percentage of time the better rated team tied, and the Loss column is for losses.

In this table, as well as in the similar table below for the Top 60 teams, you can see that the column on the right the shows the percent of time the system's ratings match game results looking only at games that were not ties.  The reason for this column is that when using the Correlator to compare how different rating systems perform, for some other systems data are not available for all the years that are available for RPI ratings.  Since the percentage of ties varies from year to year, this means using percentages correct, incorrect tie, and incorrect loss does not quite give an "apples to apples" comparison.  Using the percent correct when disregarding ties allows a better comparison.  This especially is true when looking only at the Top 60 teams, since which teams are in the Top 60 varies from system to system, which ordinarily produces a different tie percentage.

TOP 60 TEAMS, PRELIMINARY CALCULATIONS

In the above material, the Correlator looked simply at how better rated teams perform in games, in relation to how closely rated the opponents are,  The Correlator also looks at how teams perform, in relation to where they stand in the rankings.  To do this, the Correlator looks at the ranking groups of 1 to 10, 11 to 20, and so on across the entire spectrum of rankings.  This produces the following table:

The next table expresses the numbers as percentages:

At the top of the table is a summary row showing the results for the Top 60 teams.  As you can see from the summary and from scanning the rows on the table, in general the better ranked the team, the more consistent its performance is with its rank (except at the very bottom of the rankings).  The reason for looking at the Top 60 teams' performance, at the top, is that historically all at large selections and seeds in the NCAA Tournament have come from the Top 57 teams in the NCAA RPI rankings.

TOP 60 TEAMS

Drawing from the top of the second table in the preceding section, the following table shows the consistency of the NCAA RPI's ratings with actual game results when looking at the top 60 teams:

INDIVIDUAL TEAMS, PRELIMINARY CALCULATIONS

On the RPI: Formula page, I showed that in the NCAA formula for calculating the RPI, a team's winning percentage has a 50% effective weight and its strength of schedule has a 50% effective weight.  Further, of the strength of schedule's 50% effective weight, 40% is accounted for by  the team's opponents' winning percentage and the other 10% by the team's opponents' opponents' winning percentage.  Thus when looking only at strength of schedule, 80% of its effective weight is opponents' winning percentage and 20% is opponents' opponents' winning percentage.  Suppose you calculate only teams'ratings as strength of schedule contributors and rank the teams accordingly.  Their SoS contribution ratings will be based 80% on their winning percentage  and 20% on their opponents' winning percentage.  Thus teams' NCAA RPI ratings and ranks are based on 50-40-10 effective weights of three elements but their SoS contribution ratings and ranks are based on 80-20 effective weights of two elements.  As the following discussion will show, this difference creates a problem.

The following table shows, for each team, data on its expected as compared to its actual win, tie, and loss numbers and percentages.  The teams are arranged by conference and, within each conference, by the difference between the expected and actual. results.   There is more explanation below the table.

In the above table, three important columns are the Actual Winning Percentage, Likelihood Winning Percentage, and Actual Less Likely Winning Percentage columns.  In the Actual Less Likely Winning Percentage column, a positive number means the team has performed better than its ratings say it should have performed -- in other words, has been underrated.  A negative number means the team has performed more poorly than its ratings say it should have performed -- in other words, has been overrated.

Take the time to scan down the table, looking at how each conference's teams as a group have performed in relation to their ratings.  Ideally, the Actual Less Likely Winning Percentage numbers for teams are as low as possible and are distributed randomly among teams.  So far as random distribution is concerned, however, the above table shows the distribution is not random.

For example, look at the ACC at the top of the table.  Every one of its teams has performed better than its ratings say it should have performed.  Compare this to the Colonial, for which 6 of its teams have performed better than their ratings say they should have performed and 7 have performed more poorly.  Go on to the Southland, where every one of its teams has poerformed more poorely than its ratings say it should have performed.  This is not a random distribution.

INDIVIDUAL TEAMS, ACTUAL LESS LIKELY WINNING PERCENTAGE DIFFERENCES

The following table draws from the one above:

This table shows the extent of the difference between (1)the team with the most positive difference between its actual performance and its likely performance based on its NCAA RPI rating, in other words the most underrated team and (2) the team with the most negative difference, in other words the most overrated team.  The Spread column shows the difference between the two performances.  The Over and Under Total column shows the amount by which all teams performances are either better or worse than their ratings indicate they should be.

INDIVIDUAL TEAMS,  MORE PRELIMINARY CALCULATIONS

Also in an ideal system, the rank the system gives a team is the same as how the system ranks the team as a strength of schedule contributor to its opponents.  Here is a partial table showing NCAA RPI ranks for teams as compared to their NCAA RPI ranks as strength of schedule contributors to their opponents.  This is only a partial table because the entire table is too large to reproduce here.

This table shows the teams whose names begin with "A," with the data from 2010.  (The entire table includes all teams, with data from 2010 to 2024).  The Rank for Rank to SoS Rank Comparison column shows the team's 2024 NCAA RPI formula rank.  The SoS Rank for Rank to SoS Rank Comparison row shows the team's 2024 NCAA RPI formula rank of the team as a strength of schedule contributor to its opponents.  The Rank Less SoS Rank column shows the difference between those two ranks.  A positive difference means the team's NCAA RPI rank is better than how the NCAA RPI ranks it as a strength of schedule contributor.  In other words, a positive difference means the RPI is giving the team's opponents less credit in their strengths of schedule than the team's NCAA RPI rank says they should be getting.  Conversely, a negative difference in the Rank Less SoS Rank column means the opponents are getting more credit in their strengths of schedule than the team's NCAA RPI rank says they should be getting.

INDIVIDUAL TEAMS, RATING SYSTEM RANK v SYSTEM STRENGTH OF SCHEDULE CONTRIBUTOR RANK DIFFERENCES

The following table draws from the above table (as expanded to include all teams for all years):

As this table shows, the average difference between a team's NCAA RPI rank and its rank as a strength of schedule contributor to its opponents is 31.3 positions.  The median is 24 positions, meaning that at least half of all teams have a difference of at least 24 positions.  The largest difference for a team is 177 positions.  The % rows show the percentage of teams that have NCAA RPI rank to SoS contributor rank differences of the number in that row or less.  Thus, for example, 16.6% of all teams have a rank difference of 5 positions or less.

This table is important in relation to whether it is possible to "trick" the NCAA RPI through smart scheduling.  If teams have big differences between their NCAA RPI ranks and their ranks as strength of schedule contributors, a coach scheduling a team's opponents would want to pick opponents not just based on what the coach thinks opponents' NCAA RPI ranks will be but also based on what their ranks as strength of schedule contributors will be.  Further, from an NCAA RPI perspective the coach will want to pick opponents whose ranks as strength of schedule contributors will be better than their actual NCAA RPI ranks.  The more teams that have significant differences between their NCAA RPI ranks and their ranks as strength of schedule contributors, the greater the opportunity to trick the RPI through smart scheduling.  Given that, it is worth noting from the above table that for the NCAA RPI, almost 2/3 of teams have rank differences of 15 positions or more.

CONFERENCES, PRELIMINARY CALCULATIONS

The discussion above has looked at individual teams' performance.  Now I shift to performance by conferences and regions, starting with conferences.

This table looks at conferences' performance in non-conference games.  The first three data columns show the actual numbers of games the conferences' teams have won, tied, and lost.  The next three columns show the conferences'  expected wins, ties, and losses based on applying the Result Probability Table to their games' rating differences.  The Conference NonConference Actual Winning Percentage column shows the conferences' actual winning percentages in non-conference games using the 2024 NCAA RPI's Element 1 formula for winning percentage.  The Conference NonConference Likelihood Winning Percentage column shows the conferences' expected winning percentages in non-conference games, based on applying the Result Probability Table to their games' rating differences, again using the 2024 NCAA RPI's Element 1 formula for winning percentage.  The Conference NonConference Actual Less Likely Winning Percentage column subtracts the likely winning percentage from the actual winning percentage.  In this column, if the number is positive it means the conference performs better in non-conference games than its ratings say it should, in other words is underrated in relation to other conferences.  If the number is negative it means the conference performs more poorly in non-conference games than its ratings say it should, in other words is overrated in relation to other conferences.

At the bottom right of the table are summary numbers for the Actual Less Likely Winning Percentage column.  The Over and Under Total is the amount by which all of the numbers in the column miss being 0% (with 0% meaning teams actual and expected performances are identical).  The Maximum number is the amount by which the most underrated conference is underrated; the Minimum number is the amount by which the most overrated conference is overrated; and the Spread is the distance between the Maximum and Minimum.  The Over and Under Total and Spread numbers are measures of how well the rating system performs as a rating system.  The lower these numbers, the better.

CONFERENCES, OVERALL PERFORMANCE FOR CONFERENCES

Drawing from the preceding table:

This table shows the overall extent to which the NCAA RPI discriminates among conferences based on the difference between a conference's actual winning percentage and its likely winning percentage.  It does not show the cause of the discrimination, only the amount.  The High column represents the difference for the most underrated conference.  The Low is for the most overrated conference.  The Spread is the difference.  The Over and Under column shows the total amount by which all conferences are either over- or under-rated.

CONFERENCES, MORE PRELIMINARY CALCULATIONS

As discussed above, the Strength of Schedule portion of the RPI formula values opponents' winning percentages (80% Strength of Schedule effective weight) far more than it values opponents' opponents' winning percentages (20% Strength of Schedule effective weight),  In other words, it values playing opponents who will have higher winning percentages against weaker opponents over playing opponents who will have lesser winning percentages but against stronger opponents.  So, which teams have higher winning percentages against weaker opponents?  They are teams in the upper rankings of conferences that are not top-tier conferences.  And which teams have lesser winning percentages but against stronger opponents?  They are teams not in the upper rankings of top-tier conferences.  Since all teams in the not-top-tier conferences play opponents from their own conference's upper ranks, one would expect teams from those conferences to receive strength of schedule benefits due to the Strength of Schedule high valuation of opponents' winning percentages.  And since all teams in the top-tier conferences play opponents from their own conference's not-upper ranks, one would expect teams from those conferences to receive strength of schedule demerits due to the Strength of Schedule low valuation of opponents' opponents' winning percentage.  This in turn should cause the underrating of teams from top tier conferences and the overrating of teams from weaker conferences.  The question is whether what in theory one would expect to be the case actually is the case.

The next table draws from the second table above and is a resource for answering this question:

This table has the average rating for each conference's teams matched up with the Conference NonConference Actual Less Likely Winning Percentage numbers from the preceding table.  Further, it has the conferences arranged in order from the one with the best average rating at the top to the poorest at the bottom.  Drawing on this table and expanding it produces the following table:

CONFERENCES, RELATIONSHIP BETWEEN CONFERENCE PERFORMANCE AND CONFERENCE RATING

From the preceding table comes the following chart:

This chart shows the relationship between conferences' average ratings and how their teams perform in relation to their ratings.  The conferences are in order of average NCAA RPI, with the best rated conferences on the left and the poorest rated on the right.  The vertical axis shows the extent to which conferences perform better than their ratings say the should (at the top) and more poorly (at the bottom).  The trend line shows the overall pattern of the data points, which is that stronger conferences are underrated and weaker conferences overrated.  The formula tells what the statistical performance is expected to be at any conference average rating level.  And as described above, the R squared value measures how well the trend line represents the data.  The R squared value here, of 0.684, is not as high as the values in other charts above, but still is high enough to indicate better rated conferences tend to perform better, in relation to their ratings, than the ratings say they should perform.  There may be other factors involved, but the NCAA RPI has a pattern of underrating stronger conferences and overrating weaker ones.  This is exactly what, in theory, one should expect.

This table is based on the trend line formula on the chart.  In the High column, the table shows the statistically expected performance of the best rated conference at the left of the chart.  Low shows the expected performance of the poorest rated conference at the right of the chart.  Spread shows the difference between these two numbers.  The Spread is a measure of the extent of the rating system's discrimination in relation to conference strength.

It is important to note that the above chart and table do not show why the rating system has a pattern of discrimination.  They only show that there is a pattern of discrimination.

CONFERENCES, MORE PRELIMINARY CALCULATIONS

Again, as discussed above, when the NCAA RPI formula evaluates a team as a strength of schedule contributor to its opponents' RPI ratings, the formula gives the team's winning percentage an effective weight of 80% and the team's opponents' winning percentage an effective weight of 20%.  Suppose there is a high level of parity within one conference and a low level of parity within another.  In that case, one would expect the high parity conference's teams to have winning percentages tending towards 0.500 and the low parity conference's teams to have winning percentages tending towards either well above or well below 0.500.  In that circumstance, given the NCAA RPI's  importance of a team's winning percentage to its strength of schedule contribution value, one should expect that the NCAA RPI formula will favor stronger teams from the low parity conference over stronger teams from the high parity conference simply because of the region parity differences, if teams schedule non-conference opponents smartly.

The following tables and chart look to see whether this is the happening.  As a measure of parity, they use the percentage of games within each conference that are ties.

This table has the conference in-conference percentage of games that are ties matched up with the Conference NonConference Actual Less Likely Winning Percentage numbers.  Further, it has the conferences arranged in order from the one with the highest percentage of in-conference ties at the top to the poorest at the bottom.  Drawing on this table and expanding it produces the following table:

CONFERENCES, RELATIONSHIP BETWEEN CONFERENCE PERFORMANCE AND CONFERENCE PARITY AS INDICATED BY PROPORTION OF IN-CONFERENCE TIES

From the preceding table comes the following chart:

This chart shows the relationship between conference actual performance in relation to ratings, when matched with the proportion of in-conference ties.  Conferences with high proportions of ties are on the left and with low proportions on the right.  It is a test to see if in-conference parity has an impact on whether conferences are underrated or overrated, using the proportion of in-conference ties as an indicator of in-conference parity.  The trend line suggests that high in-conference parity (on the left) causes the RPI to underrate a conference and low parity (on the right) causes overrating.  The R-squared value, however, is roughly 0.18, which is low.  A reasonable interpretation of this is that for the RPI, in-conference parity may cause some discrimination against conferences with high in-conference parity and in favor of conferences with low parity, but that other factors may have a more significant impact on conference actual performance in relation to ratings.

CONFERENCES, MORE PRELIMINARY CALCULATIONS

This table shows conferences' average NCAA RPI ratings and their average ratings under the NCAA RPI formula as Strength of Schedule Contributors.  In the right hand column, it shows the difference between the two numbers.  As you can see, the differences vary.

The next table matches up the differences with the actual less expected winning percentages of the conferences.

In this table, the conferences are arranged in order from the one with the greatest NCAA RPI rating less NCAA SoS Contribution rating at the top to the one with the least at the bottom.  By simply eyeballing this table, you can see that the conferences at the top perform better than the NCAA RPI ratings say they should and the conferences at the bottom perform more poorly.

The following table is an expanded version of the above table:

CONFERENCES, RELATIONSHIP BETWEEN CONFERENCE PERFORMANCE AND THE DIFFERENCE BETWEEN CONFERENCES' RATINGS AND THEIR RATINGS AS STRENGTH OF SCHEDULE CONTRIBUTORS

The following chart comes from the preceding table:

As the chart shows, there is a strong relationship between conferences' average NCAA RPI ratings less their average strength of schedule contribution ratings, on the one hand, and their actual performance based on their NCAA RPI ratings less their expected performance based on those ratings, on the other hand.  You can see this visually and the high R squared value of roughly 0.88 confirms it.  Remember, the earlier information on this page already shows that the NCAA RPI tends to discriminate against stronger conferences and in favor of weaker ones and against conferences with higher levels of in-conference parity and in favor of conferences with lower parity.  This chart shows why: It is due to how the NCAA computes teams' strength of schedule contributions within the RPI formula. There may be some other minor contributing factors, but the above chart seems clear that the main driver is the strength of schedule portion of the RPI formula.

In this table, the High column shows the trend-line-based actual less likely performance of the conference with the highest disparity between its NCAA RPI rating and its RPI strength of schedule contribution, at the left of the above chart.  The Low column shows the trend-line-based actual less expected performance of the conference with the lowest disparity at the right of the chart.  Spread shows the difference between these two numbers.  The Spread is a measure of the level of discrimination the disparity causes.

REGIONS, PRELIMINARY CALCULATIONS

Next, the Correlator goes through similar steps for teams grouped by geographic regions.  Each state's teams are assigned to one of four geographic regions.  The regions are based on where teams from the different states play the majority of their games.  You can find a map showing the regions at this website's RPI: Regional Issues page.  You also can find there a table showing the percentage of games each regions's teams play against opponents from their own and each of the other three regions.

The first region table is similar to the first table for conferences, above.

As the table shows,  when looking at games against teams from the other regions, there are differences in how the regions' teams perform in relation to their NCAA RPI ratings.  For example, the West region's teams perform better against teams from other regions than their ratings say they should.  In other words, teams from the West region, as a group, are underrated.  Teams from the Middle region, as a group, also are underrated, though to a lesser extent.  Teams from the North and South, as groups, are overrated.  This does not mean that all teams from the regions are overrated or underrated.  For example, if you look at the individual teams from the South and at conferences with all or most teams from the South, you will see that some of them are underrated notwithstanding that teams from the South as an entire group are overrated.

REGIONS, OVERALL PERFORMANCE FOR REGIONS

This table shows the overall extent to which the NCAA RPI discriminates among regions based on the difference between a region's actual winning percentage and its likely winning percentage.  It does not show the cause of the discrimination, only the amount.  The High column represents the difference for the most underrated region.  The Low is for the most overrated region.  The Spread is the difference.  The Over and Under column shows the total amount by which all regions are either over- or under-rated.

REGIONS, MORE PRELIMINARY CALCULATIONS

This table has the average rating for each region's teams matched up with the Region NonRegion Actual Less Likely Winning Percentage numbers from the preceding table.  Further, it has the regions arranged in order from the one with the best average rating at the top to the poorest at the bottom.  Drawing on this table and expanding it produces the following table:

REGIONS, RELATIONSHIP BETWEEN REGION PERFORMANCE AND REGION RATING

From the above table comes the following chart:

This chart shows the relationship between regions' average ratings and how their teams perform in relation to their ratings.  The regions are in order of average NCAA RPI, with the best rated regions on the left and the poorest rated on the right.  The vertical axis shows the extent to which regions perform better than their ratings say the should (at the top) and more poorly (at the bottom).  The trend line shows the overall pattern of stronger regions being underrated and weaker regions overrated.  The formula tells what the statistical performance is expected to be at any region average rating level.  And as described above, the R squared value measures how well the trend line represents the data.  The R squared value here, of roughly 0.43, is in the mid-range, but still suggests better rated regions tend to perform better, in relation to their ratings, than the ratings say they should perform.  There also may be other factors involved.

This table summarizes the trend information in the chart, showing the expected performances by the regions with the best and poorest average ratings and the Spread between those performances.   The Spread is a measure of the extent of the rating system's discrimination in relation to region strength.

REGIONS, MORE PRELIMINARY CALCULATIONS

As discussed above, when the NCAA RPI formula evaluates a team as a strength of schedule contributor to its opponents' RPI ratings, the formula gives the team's winning percentage an effective weight of 80% and the team's opponents' winning percentage an effective weight of 20%.  Suppose there is a high level of parity within one region and a low level of parity within another.  In that case, one would expect the high parity region's teams to have winning percentages tending towards 0.500 and the low parity region's teams to have winning percentages tending towards either well above or well below 0.500.  In that circumstance, given the NCAA RPI's  importance of a team's winning percentage to its strength of schedule contribution value, it is possible that the NCAA RPI formula will favor stronger teams from the low parity region  over stronger teams from the high parity region simply because of the region parity differences and how the teams chooses to schedule.

The following tables and chart look to see whether this is the happening here.  As a measure of parity, they use the percentage of games within each region that are ties.

This table has the average rating for each region's teams matched up with the Region NonRegionActual Less Likely Winning Percentage numbers.  Further, it has the regions arranged in order from the one with the best average rating at the top to the poorest at the bottom.  Drawing on this table and expanding it produces the following table:

REGIONS, RELATIONSHIP BETWEEN REGION PERFORMANCE AND REGION PARITY AS INDICATED BY PROPORTION OF IN-REGION TIES

The following chart is based on the preceding table.

This chart shows a positive correlation between the level of parity within a region and how the region's teams perform in non-region games in relation to their NCAA RPI ratings.  Looking at the R squared value of roughly 0.66, the correlation is reasonably strong but allows for other factors also affecting regions' performance in relation to their ratings.

The following table is based on the chart's trend line formula:

REGIONS, MORE PRELIMINARY CALCULATIONS

This leaves the question, as was the case with conferences, of what the underlying mechanism is that causes the uneven rating of regions.  The following tables and chart answer this question.

This table matches regions' actual less likely winning percentages in non-region games with the regions' average differencess between their NCAA RPI ratings and their NCAA RPI formula strength of schedule contribution values.  The next table simply is an expansion of this one.

REGIONS, RELATIONSHIP BETWEEN REGION PERFORMANCE AND THE DIFFERENCE BETWEEN REGIONS' RATINGS AND THEIR RATINGS AS STRENGTH OF SCHEDULE CONTRIBUTORS

From the preceding table comes the following chart.

Simply looking at the  chart, there is a high correlation between, on the one hand, the extent to which regions' teams, as groups, outperform or underperform in relation to their NCAA RPI ratings and, on the other hand, the extent to which teams' NCAA RPI ratings exceed their NCAA RPI formula strength of schedule contributions.  The high R squared value of roughly 0.93 appears to confirm this: The difference between how the NCAA RPI formula computes a team's rating and how it computes a team's strength of schedule contribution appears to be the main driver that causes the RPI to underrate some regions' teams, as a group, and to overrate other regions' teams, as a group.

The following table shows the Trend High, Low, and Spread numbers for the chart:

CONCLUSION

The information above shows how the Correlator works, using the 2024 NCAA RPI with no overtimes as an example.  It shows that the NCAA RPI has defects.  It does not show, however, whether it is possible to fix those defects while still using the NCAA RPI's basic structure or whether there are other rating systems that do not have those defects.  It is possible, however, to apply the Correlator to modified versions of the NCAA RPI and to other rating systems (if they provide enough publicly available and detailed data), to see how their performance compares to the 2024 NCAA RPI.  You can find comparisons of the 2024 NCAA RPI to a modified version of the NCAA RPI (the Balanced RPI) and to two other rating systems (Massey Ratings and KPI Ratings), on this website's RPI: Modified RPI? page.  You also can find a comparison of the 2024 NCAA RPI to the 2024 NCAA NonConference RPI at the RPI: Non-Conference RPI page.