Introducing the BPI

• Recommend0
• Tweet0
• Comments0
• Email
• Print

Rating college basketball teams is mainly about identifying NCAA tournament teams. The tournament itself decides who is the ultimate champion, but who deserves to be there?

There are arguments every year regarding fringe teams. ESPN Bracketologist Joe Lunardi will tell you who is likely to make it. Jay Bilas and other ESPN college basketball analysts want to be able to tell you who should make it. To that end, they asked the ESPN Analytics Department to help them. What we created is this, the College Basketball Power Index, or BPI for short.

Other power rankings already exist, the most prominent being the RPI, Sagarin Ratings, Massey Ratings and Ken Pomeroy's ("Kenpom") ratings. All of these methods are based upon the outcomes of games, their location -- home/road/neutral -- and the quality of opponents. Each one basically puts these together in slightly different ways and arrives at slightly different results.

[+] Enlarge

AP Photo/David SmithTray Woodall has lifted the Panthers since returning from injury.

RPI, due to its simplicity, tends to be the biggest decision aid for the NCAA tournament committee, even though it doesn't account for the actual scores of games. Kenpom and our BPI system both account for the varying pace that teams can play, which is an important analytical component of evaluating basketball teams. But we believe the BPI is a little more refined than any other existing power ranking.

There are a number of small details that we have in our methodology to make it reflective of a résumé for a tournament team -- these are pretty technical and many people won't be interested, so we won't go into detail, but we think they improve how the tool works.

On top of this, we decided to incorporate a little bit more information than the other power ranking systems use. In particular, we added a way of accounting for missing players.

If a team or its opponent is missing one of its most important players (determined by minutes per game) for a contest, that game is less important for ranking the teams compared to games in which both teams are at full strength.

Every season has teams for whom missing players can be important. In 2011-12, the Pittsburgh Panthers are a great example of how missing a player can make a difference in a rating.

The Panthers were on a roll with Tray Woodall at the point ... until he went down with injury after seven games. They won their first five without him, but then lost seven in a row with Woodall either on the sideline or playing limited minutes in a game he admitted he wasn't ready for.

After his return, Pitt showed much better play, not only winning, but beating good teams. By de-weighting those games in which Woodall didn't take the court, games that the Panthers generally were poor in, BPI sees a team that is stronger than its record reflects.

The games are not de-weighted so much that they don't matter -- because they are still reflective of other players on the team -- but they are weighted roughly 85 percent of a game at full strength. As a result, Pitt's ranking in BPI is more reflective of its full team's performance than in other methods. This affects only a few teams a season and the adjustment is still not as helpful to a team as simply winning, but it can be important in select examples.

Another way that BPI can rank teams differently than Sagarin or Kenpom is counting close games at home versus on the road. In BPI, a close win at home is better than a close loss on the road against the same opponent. This isn't necessarily true in other methods and, in methods that do that, they don't typically account for bigger wins. BPI gives marginally decreasing credit for bigger wins, with a 30-point win being only about 20 percent better than a 15-point win, not twice as good, which can happen in other methods.

By capturing blowouts, but not overweighting them, BPI credits the ability of good teams to easily beat poor teams without providing incentive to win by 30 when 20 is a safe margin. By capturing both blowouts and close games in this way, BPI summarizes a team's résumé for the NCAA tournament well.

By reflecting a résumé, BPI was not explicitly built to make predictions. But, in tests, its ability to predict appears to be as good or better than Sagarin or RPI at predicting results in the NCAA tournament and NIT.

Between the 2007 and 2011 NCAA tournaments, it picked 74.4 percent of the matchups correctly, whereas Sagarin picked 73.2 percent and RPI picked 71.9 percent. (Kenpom is more difficult to evaluate because its pre-tournament rankings are not available.) The average ranking of the NIT finalists was better in BPI than in Sagarin or RPI. Notice, of course, that many of these differences are small. The BPI is not a guaranteed way to pick a perfect bracket, but we do think it is the best power ranking available.

Ultimately, the College Basketball Power Index gives us a tool for rating teams that is more useful for ESPN than existing tools. It explains a team's wins and losses in a more detailed context and seemingly predicts future results as well as other tools, if not better. It also incorporates information about injuries and missed games that should be relevant for weighting the résumés of bubble teams on Selection Sunday. We feel confident that it will enhance coverage of college basketball for years to come.

Dean Oliver is ESPN's director of production analytics and previously helped construct Total QBR. He is the author of "Basketball on Paper" and has worked in the front offices of the Seattle SuperSonics and Denver Nuggets. His ESPN archives can be viewed here.

• Recommend0
• Tweet0
• Comments0
• Email
• Print