Sunday, March 02, 2008
RPI: Not the Be All End All of Bracketology
R, P, and I. Probably the three letters that are thrown around the most during the first half of the month of March in the United States. While everyone is familiar with the term RPI, not everyone knows what it is and how much influence it actually has in the selection process for the NCAA men's basketball tournament. In this post, I'm going to show you how using the RPI alone would give the committee an inaccurate picture of the national landscape and would result in a radically different tournament field than a more balanced selection process would ensure. To do this, I'm going to compare the bracket I released Saturday morning with one based solely on the RPI after Thursday night's games, which was my cutoff point for the Friday night preparation of my bracket.
What is the RPI?
RPI stands for Ratings Percentage Index, a formula used by the NCAA to rate teams in various sports as a tool in the selection and seeding process for national championships. By nature, the RPI is not a basketball-specific formula, but since the actual formula used in each sport is secret, it can be assumed that each sport's committee makes tweaks to it to match their sport's particular characteristics. For example, since home court advantage is particularly important in basketball, it's pretty well known that there is a penalty built into the basketball RPI for losing at home.
The NCAA releases the RPI on Mondays during the season, but doesn't provide daily updates. For daily updates, there are a few sites that try to replicate the formula. Jerry Palm's College RPI, ESPN.com, and Kyle Whelliston's Basketball State require a subscription for at least some of their data. I use Ken Pomeroy's RPI listing and other statistics while putting my bracket together. He summarizes the basic RPI formula in this way:
"1/4*(Winning Percentage) + 1/2*(Opponents' Average Winning Percentage) + 1/4*(Opponents' Opponents' Winning Percentage)."
He offers a more detailed explanation here.
The RPI does a fair job of measuring a team's performance and even their strength of schedule. However, this doesn't mean it can't be exploited. When building a non-conference schedule a school can schedule games that they should easily win (boosting the 25% winning percentage portion of the formula) against teams who should play well in weaker conferences (boosting the 50% opponents' winning percentage portion while not causing too much harm to the 25% opponents' opponents' portion).
A high RPI rating can tell you if a team can win, it just can't quite tell you the types of teams that it's beating. The separate Strength of Schedule (SOS) rating does this, which is why, more often than not, you'll see a good sized disparity between a team's RPI and SOS.
The RPI formula also doesn't give easy access to two factors that are important in the selection process, recent performance (wins and losses in the last 12 games) and road and neutral performance. These are two factors that selection committee members must access and consider during the process.
The RPI Only Bracket
To create this bracket, I simply selected the 31 conference leaders based on record as my automatic qualifiers. Next, I took the top 34 teams in the RPI who weren't conference leaders to fill in the at-large spots. Since, 14 of the 31 conference champions were rated higher than #65 in the RPI, team #49 in the rankings, Florida, was the last team in.
In terms of seeding, I placed the teams in order from 1 to 65 in the S-curve. Tennessee (#1) was the top overall seed, and Morgan State (#133) and Alabama State (#235) were the two teams selected to play in the Opening Round game. I did have to make several adjustments to the actual bracket to meet bracketing principles (avoid conference matchups until at least the second round, avoiding regular season rematches, keeping teams within their natural regions when possible). For example, I swapped UCLA (#8) and Duke (#5) on the 2 line to keep them within their natural regions. I did the same with Kansas (#9) and Georgetown (#10) on the 3 line.
I did have to swap teams out of their natural seed lines in two instances to preserve bracket integrity. I swapped BYU, a true 6 seed, with Pittsburgh, a true 7, to move the Cougars to a Thursday-Saturday site for the first two weekends, as required. I also swapped VCU, a 12 seed, with Arkansas, an 11, to keep the Razorbacks out of Tennessee's region.
Bracket Differences
Here are some of the key differences between the two brackets, starting with who's in and who's out.
In the RPI Bracket, Out of the Full Projection
Dayton (10 seed)
Illinois State (10 seed)
Mississippi (12 seed)
Southern Illinois (10 seed)
Stephen F. Austin (12 seed)
In the Full Projection, Out of the RPI Bracket
UAB (13 seed)
Arizona State (12 seed)
Maryland (10 seed)
Syracuse (12 seed)
Virginia Tech (13 seed)
As you can see, only using the RPI to project the field brings in several teams who's recent performances have them only on the periphery of the at-large conversation (Dayton and Mississippi), don't have a great strength of schedule (Illinois State and Stephen F. Austin), and were on the verge of making it in heading into the weekend (Southern Illinois). In contrast, five teams, four of which are from major conferences, that have fairly good cases (well, for this year anyway) are left out.
These changes result in changes to the number of bids for each conference.
Conference Breakdown
Big East - 7 in RPI bracket (8 in full projection)
Big 12 - 6 (6 in full)
SEC - 6 (5 in full)
Big 10 - 5 (5 in full)
Pac 10 - 5 (6 in full)
ACC - 4 (6 in full)
Atlantic 10 - 3 (2 in full)
Missouri Valley - 3 (1 in full)
Mountain West - 2 (2 in full)
Southland - 2 (1 in full)
West Coast - 2 (2 in full)
Conference USA - 1 (2 in full)
Of course, this year it's hard to argue that the number of bids for these conferences will be far off from these totals (well, except for the Southland perhaps).
Last Four In
Florida
Ohio State
Mississippi
Stephen F. Austin
Last Four Out
Syracuse
Creighton
Western Kentucky
UAB
Next Four Out
St. Joseph's
Kentucky
New Mexico
Ohio
These categories are all based on the RPI. The Last Four In are teams rated 46-49; the Last Four Out 50, 52, 53, and 54 (51 rated Davidson was an auto bid); and the Next Four Out 55-58.
Most starkly, the RPI led to quite a bit of shuffling in the seeding. This is where teams' performance in their last 12 games and in road and neutral contests would've had the most impact.
Seeding Winners
Vanderbilt - 2 in RPI bracket/4 in full bracket
Michigan State - 4/6
Marquette - 4/5
Drake - 4/5
Butler - 5/6
Arizona - 6/9
Pittsburgh - 6 (true 7)/8
Clemson - 6/7
Kent State - 7/10
Miami - 7/8
UNLV - 8/9
South Alabama - 8/9
Oklahoma - 8/11
West Virginia - 9/10
Massachusetts - 10/11
VCU - 11 (true 12)/12
Belmont - 14/15
Austin Peay - 15/16
Portland State - 15/16
Seeding Losers
Kansas - 3 in RPI bracket/2 in full bracket
Stanford - 6/3
Indiana - 5/4
Notre Dame - 5/4
Washington State - 6/5
Purdue - 7/5
BYU - 7 (true 6)/6
St. Mary's - 8/7
Gonzaga - 9/6
Baylor - 9/8
Southern California - 9/7
Kansas State - 11/7
Mississippi State - 11/8
Texas A&M - 11/9
Arkansas - 12 (true 11)/11
Davidson -13/11
Florida - 13/10
Winthrop - 16/15
Robert Morris - 16/14
I'll be back with a full projection tomorrow evening. There will definitely be some changes after the weekend's action. Congratulations to Cornell, who's the first team in. A whole slew of teams sewed up their conference regular season titles yesterday, so they'll be included in the bracket until they either lose in their conference tourneys or claim the title.
E-mail me with comments or questions.
What is the RPI?
RPI stands for Ratings Percentage Index, a formula used by the NCAA to rate teams in various sports as a tool in the selection and seeding process for national championships. By nature, the RPI is not a basketball-specific formula, but since the actual formula used in each sport is secret, it can be assumed that each sport's committee makes tweaks to it to match their sport's particular characteristics. For example, since home court advantage is particularly important in basketball, it's pretty well known that there is a penalty built into the basketball RPI for losing at home.
The NCAA releases the RPI on Mondays during the season, but doesn't provide daily updates. For daily updates, there are a few sites that try to replicate the formula. Jerry Palm's College RPI, ESPN.com, and Kyle Whelliston's Basketball State require a subscription for at least some of their data. I use Ken Pomeroy's RPI listing and other statistics while putting my bracket together. He summarizes the basic RPI formula in this way:
"1/4*(Winning Percentage) + 1/2*(Opponents' Average Winning Percentage) + 1/4*(Opponents' Opponents' Winning Percentage)."
He offers a more detailed explanation here.
The RPI does a fair job of measuring a team's performance and even their strength of schedule. However, this doesn't mean it can't be exploited. When building a non-conference schedule a school can schedule games that they should easily win (boosting the 25% winning percentage portion of the formula) against teams who should play well in weaker conferences (boosting the 50% opponents' winning percentage portion while not causing too much harm to the 25% opponents' opponents' portion).
A high RPI rating can tell you if a team can win, it just can't quite tell you the types of teams that it's beating. The separate Strength of Schedule (SOS) rating does this, which is why, more often than not, you'll see a good sized disparity between a team's RPI and SOS.
The RPI formula also doesn't give easy access to two factors that are important in the selection process, recent performance (wins and losses in the last 12 games) and road and neutral performance. These are two factors that selection committee members must access and consider during the process.
The RPI Only Bracket
To create this bracket, I simply selected the 31 conference leaders based on record as my automatic qualifiers. Next, I took the top 34 teams in the RPI who weren't conference leaders to fill in the at-large spots. Since, 14 of the 31 conference champions were rated higher than #65 in the RPI, team #49 in the rankings, Florida, was the last team in.
In terms of seeding, I placed the teams in order from 1 to 65 in the S-curve. Tennessee (#1) was the top overall seed, and Morgan State (#133) and Alabama State (#235) were the two teams selected to play in the Opening Round game. I did have to make several adjustments to the actual bracket to meet bracketing principles (avoid conference matchups until at least the second round, avoiding regular season rematches, keeping teams within their natural regions when possible). For example, I swapped UCLA (#8) and Duke (#5) on the 2 line to keep them within their natural regions. I did the same with Kansas (#9) and Georgetown (#10) on the 3 line.
I did have to swap teams out of their natural seed lines in two instances to preserve bracket integrity. I swapped BYU, a true 6 seed, with Pittsburgh, a true 7, to move the Cougars to a Thursday-Saturday site for the first two weekends, as required. I also swapped VCU, a 12 seed, with Arkansas, an 11, to keep the Razorbacks out of Tennessee's region.
Bracket Differences
Here are some of the key differences between the two brackets, starting with who's in and who's out.
In the RPI Bracket, Out of the Full Projection
Dayton (10 seed)
Illinois State (10 seed)
Mississippi (12 seed)
Southern Illinois (10 seed)
Stephen F. Austin (12 seed)
In the Full Projection, Out of the RPI Bracket
UAB (13 seed)
Arizona State (12 seed)
Maryland (10 seed)
Syracuse (12 seed)
Virginia Tech (13 seed)
As you can see, only using the RPI to project the field brings in several teams who's recent performances have them only on the periphery of the at-large conversation (Dayton and Mississippi), don't have a great strength of schedule (Illinois State and Stephen F. Austin), and were on the verge of making it in heading into the weekend (Southern Illinois). In contrast, five teams, four of which are from major conferences, that have fairly good cases (well, for this year anyway) are left out.
These changes result in changes to the number of bids for each conference.
Conference Breakdown
Big East - 7 in RPI bracket (8 in full projection)
Big 12 - 6 (6 in full)
SEC - 6 (5 in full)
Big 10 - 5 (5 in full)
Pac 10 - 5 (6 in full)
ACC - 4 (6 in full)
Atlantic 10 - 3 (2 in full)
Missouri Valley - 3 (1 in full)
Mountain West - 2 (2 in full)
Southland - 2 (1 in full)
West Coast - 2 (2 in full)
Conference USA - 1 (2 in full)
Of course, this year it's hard to argue that the number of bids for these conferences will be far off from these totals (well, except for the Southland perhaps).
Last Four In
Florida
Ohio State
Mississippi
Stephen F. Austin
Last Four Out
Syracuse
Creighton
Western Kentucky
UAB
Next Four Out
St. Joseph's
Kentucky
New Mexico
Ohio
These categories are all based on the RPI. The Last Four In are teams rated 46-49; the Last Four Out 50, 52, 53, and 54 (51 rated Davidson was an auto bid); and the Next Four Out 55-58.
Most starkly, the RPI led to quite a bit of shuffling in the seeding. This is where teams' performance in their last 12 games and in road and neutral contests would've had the most impact.
Seeding Winners
Vanderbilt - 2 in RPI bracket/4 in full bracket
Michigan State - 4/6
Marquette - 4/5
Drake - 4/5
Butler - 5/6
Arizona - 6/9
Pittsburgh - 6 (true 7)/8
Clemson - 6/7
Kent State - 7/10
Miami - 7/8
UNLV - 8/9
South Alabama - 8/9
Oklahoma - 8/11
West Virginia - 9/10
Massachusetts - 10/11
VCU - 11 (true 12)/12
Belmont - 14/15
Austin Peay - 15/16
Portland State - 15/16
Seeding Losers
Kansas - 3 in RPI bracket/2 in full bracket
Stanford - 6/3
Indiana - 5/4
Notre Dame - 5/4
Washington State - 6/5
Purdue - 7/5
BYU - 7 (true 6)/6
St. Mary's - 8/7
Gonzaga - 9/6
Baylor - 9/8
Southern California - 9/7
Kansas State - 11/7
Mississippi State - 11/8
Texas A&M - 11/9
Arkansas - 12 (true 11)/11
Davidson -13/11
Florida - 13/10
Winthrop - 16/15
Robert Morris - 16/14
I'll be back with a full projection tomorrow evening. There will definitely be some changes after the weekend's action. Congratulations to Cornell, who's the first team in. A whole slew of teams sewed up their conference regular season titles yesterday, so they'll be included in the bracket until they either lose in their conference tourneys or claim the title.
E-mail me with comments or questions.
