We’ve all seen the Ratings Percentage Index (RPI) quoted many times in basketball and baseball articles as the major metric the NCCA uses to determine at-large tournament bids and seeding. But how many of us really know what the RPI measures? How can a team win a game and have their RPI drop? To shed light on the RPI I’ll give a high-level explanation with particular reference to Maryland’s 2014 baseball season.
The RPI is a measure that the NCAA uses to adjust for the fact that not every team plays the same strength of schedule (SoS). It is intended to level the playing field, so to speak. The RPI is primarily used in basketball (Men’s and Women’s), baseball, and softball; all sports where there are dozens of games played. It is completely different than college football’s BCS rating. Which is not to say that the RPI is easily understood or without its critics as we shall see.
There are three major components making up the RPI: a team’s winning percentage, Opponents’ win %, and Opponents’ Opponents’ win %. That last component is not a typo. Quite simply it is the winning percentage of all of the teams that your opponents have played. Only games against Division I schools are counted. That’s not a trivial rule. In areas of the country (think Mountain West) where Division I schools are sparse, schedules typically include multiple games against non-Division I opponents. Since these games don’t count, Division I teams in remote areas tend to have win percentages biased towards 0.500.
Roughly, the NCAA calculates a team’s RPI as follows: (0.25*win percentage) + (0.50*Opponents’ win percentage) + (0.25*Opponents’ Opponents’ win percentage). The commonly cited SoS is based on the last two terms in the equation. The above formula results in a score for which teams are rank ordered. What’s initially bothersome is that the margin of victory isn’t considered. When Maryland swept Pitt last weekend it didn’t matter that they outscored the Panthers 32 – 6. The same benefit would have accrued if they won all three games by a 1 – 0 score. That’s why Boyd’s World and Warren Nolan have developed alternatives to the RPI that do take the above factors into account.
Maryland's 2014 season and the RPI
Let’s take a look at Maryland’s 2014 baseball season. The Terps have an overall record of 34 – 19 for a win percentage = 0.6415. Their opponent’s win percentage is 0.550. The opponents of Maryland’s opponents have a win percentage of 0.532. So theoretically the Terps’ RPI should be (0.160 + 0.275 + 0.133), which is 0.568. But home wins count 70% and road wins count 130%, so the Terrapins’ RPI is actually 0.563, currently 29th in the country.
Early on in the season when few games have been played a team’s RPI can vary widely from game to game. Conversely, towards the end of the season a team’s RPI shouldn’t change that much. This makes sense intuitively (Bayesian statisticians refer to this as getting more informative “priors”), but the first game of the season has as much influence on the RPI as the last game of the season.
For many people the most confusing aspect of the RPI is that a team can decrease their RPI even if they win, and increase their RPI when they lose a game. When Florida State beat Maryland in game one of their series 15 – 3, it was more beneficial to the Terps’ RPI than their win at Delaware. This is due to how heavily the SoS components are weighted in the RPI. So while scheduling teams such as U. Mass and St. Mary’s provides more games for the Terrapins, they’re no-win propositions. Our RPI will decrease if we beat them, but it will really take a hit if we lose to them, particularly if it’s at Shipley Field. Conversely, playing Florida helps Maryland enormously. The Terps’ RPI will rise even if they had lost all three games, but will jump up the more games that they win (Maryland took one of three games in Gainesville).
A good way to dissect Maryland’s RPI on a game-by-game basis can be found here at Boyd’s World. The numbers can be a bit daunting, but Boyd’s World color classification scheme simplifies things. Here are the effects of some games that Maryland played in 2014. The Terps faced JMU twice this year, winning on the road while losing at home. While the win didn’t change our RPI much, the loss lowered it. Maryland played U Mass in a three-game series and beat them decisively, yet the team’s RPI went down as a result. When the Terps defeated VCU twice it really helped Maryland’s RPI. However even a loss to VCU would have resulted in an increase in the RPI, just not nearly as much as winning did .
A better mousetrap
Confused? You’re not alone. I’ve asked an acknowledged RPI expert, Boyd Nation of Boyd’s World (www.boydsworld.com) three questions to dig a little deeper into the subject, which he was kind enough to answer. The parts bolded are my doing.
- How much, if any, does where the game is played (home/away/neutral) factor into the RPI? What about margin of victory?
In the winning percentage portion of the formula, which is 25%, home wins are counted as .7 of a win and road wins are counted as 1.3 wins, with corresponding amounts for losses. Margin of victory is not considered.
- How does Boyd's World proprietary Iterative Strength Rating (ISR) differ from the NCAA's RPI?
Rather than the simplistic two-step measures for Strength of Schedule that the RPI uses, the ISR implicitly (it's actually produced as a side effect of the algorithm) does a diminishing analysis several more steps out. The HFA (Home Field Advantage) adjustment is more correctly calibrated, and margin of victory is taken into account.
- Why does the NCAA value a team's SoS along with their RPI when the former already heavily influences the latter?
It's silly. I suspect that at least some committee members consider doing poorly against a strong schedule to be evidence of more moral fiber than doing equally well (with more wins) against a weak schedule.
Personally I would prefer a scoring system that includes some influence due to the margin of victory. Additionally, I think the present RPI inordinately rewards SoS. In non-revenue sports such as baseball, scheduling teams is based to a large extent on economic concerns, i.e. geographic proximity, particularly in midweek away games.