Critiquing RPI and the Quadrant System

*A previous version of this article stated that game location does not affect RPI. This was originally the case, but was modified beginning in 2004.

As we approach Selection Sunday, everyone’s eyes become focused on the tournament bubble.  After every game with bubble implications, the question is once again phrased: who’s in, and who’s out?  Bracketologists can speculate as much as they want, but ultimately, the decision of who makes the 68-team field and as what seed comes down to the NCAA Tournament selection committee, a 10-member group of school and conference administrators.

The committee looks at a wide variety of factors while constructing the March Madness bracket, but one of the most valued and cited tools, which has been used since 1981, is the Rating Percentage Index, also known as RPI.  A new feature to the RPI this season is the quadrant system, which is intended to help categorize a team’s wins and losses.  Despite its heavy usage, though, the RPI and corresponding quadrant system have a number of flaws which severely diminish their value as metrics.

First things first: what exactly is the RPI formula?  It’s actually quite simple.  A team’s RPI is 25% winning percentage, 50% opponent winning percentage, and 25% opponents’ opponents winning percentage.  Since the latter two components make up a team’s Strength of Schedule (SOS), 75% of RPI is SOS. Additionally, a .6 modifier is applied to home victories, with a 1.4 modifier applied to road victories. The inverse is true for home and road losses.

If it wasn’t clear enough already, RPI only cares mainly about two things: wins and losses.  While that might seem reasonable at first, upon closer look, this creates several problems.  First off, RPI doesn’t factor in point differential.  The ability to win close games can be valued, but given identical records and strengths of schedule, I’ll take the team winning games by 30 over the one barely eking out victories any day of the week.

Another major factor is injuries and suspensions.  Texas Tech recently dropped four games in a row while Keenan Evans was battling a toe injury.  UCF and SMU come to mind as teams who have been decimated by injuries.  Notre Dame is an NIT-team at best without Bonzie Colson, but potentially a ranked one with him.  And of course, let’s not forget Minnesota.  Amir Coffey’s injuries and Reggie Lynch’s suspension-turned-expulsion made a game against the Gophers in conference play practically a gift, whereas they were ranked for the first five weeks of the season.  Yet, games against Minnesota then and now count as the same.

With RPI, who you play is often more important than if you win, as SOS makes up a larger portion of the formula than winning percentage.  This is how a South Carolina team that finished their regular season 15-15 can be ranked 74th, while a 21-13 Penn State team that just fell in the Big Ten semifinals and has been near the bubble over the past few weeks is 76th.  South Carolina’s SOS is 19, while Penn State’s is 73.

In the RPI, you can move up with losses, and go down with wins against poor opponents.  Of course, this should be possible.  A one-point loss like Louisville’s against #1 Virginia on Thursday (especially with its bizarre ending) shouldn’t harm the Cardinals, while a hypothetical two-point home victory against Chicago State, 1-28 against D-I opponents, should.  But again, the RPI doesn’t care about how a team plays in a win or loss; just the final outcome.

This is why more predictive metrics, such as the KenPom, Sagarin, and ESPN’s BPI have gotten more love in recent years, and will continue to as they continue to be refined.  The RPI is an extremely basic system and one which is outdated for our current analytical capabilities.   Some people hate the RPI and want it banished.  I’m not ready to take that stance quite yet – for its simplicity, it does a decent job.  Still, it shouldn’t pull as much weight as it currently does.

To the selection committee’s credit, they have moved to fix one of the RPI’s flaws (that it treats games the same regardless of location) with the new quadrant system, which grouping wins into categories based on a combination of opponent RPI and location as shown here:

If you follow college basketball regularly, this is why you’ve heard people talk about things like Q1 wins and Q3 losses.  For example, a game against the 40th ranked team in the RPI counts as a Q2 game if it happens at home, but a Q1 game if on the road or at a neutral site.

The quadrant system is primarily used to measure a team’s quality wins and bad losses.  It also rewards road/neutral wins and treats road/neutral losses more leniently.  I admit it’s a small improvement over judging teams by RPI Top 50 or 100 wins or RPI sub-150 losses, as has been common in previous years, but cutoffs are still as arbitrary and questionable as ever.  Personally, even though home advantage is a major advantage, having the same cutoff for Q2 home games as Q1 road games seems excessive.

To refer to my previous example, as of the time of writing, a home game against Penn State (RPI 76) qualifies as a Q3 game (and a potential “bad loss”, if you consider Q1/2 to be “good” and Q3/4 to be “bad”) while facing South Carolina (RPI 74) on the road would be Q1.  That’s insane.  Yet, the quadrant system is still just based off RPI, which, as I’ve explained, is heavily flawed.

It’s also highly volatile, and subject to change – even without Penn State or South Carolina playing a game.  Even a few matchups involving teams who played teams who played Penn State and South Carolina could easily alter their RPIs, changing what quadrant categories they fall into.

There’s no perfect way to measure how good a basketball team is, and there never will be, which means there will always be some subjectivity at play with the selection committee.  However, with the RPI’s flaws, it’s only reasonable to take some emphasis away from a metric prone to misjudgments.