The 2019 NCAAs were held at PPG Paints Arena in Pittsburgh (Photo/Tony Rotundo, WrestlersAreWarriors.com)
This past season, the NCAA used seven key criteria for both selection and seeding of the Division I Wrestling Championships. Many of the criteria are familiar to wrestling fans, competitors and coaches such as winning percentage, quality wins and conference tournament placement. However, there is one seeding factor that might be confusing to the average onlooker.
RPI, which stands for rating percentage index, is a relatively simple calculation using the following formula.
As you can see the formula is based on three key factors: winning percentage, opponents' winning percentage and opponents' opponents' winning percentage. In theory, the formula provides an unbiased assessment of a wrestler since it measures one's success as well as their level of competition.
However, there is one glaring omission that stands out to anyone familiar with modern sports statistics. RPI does not take margin of victory into account. While lacking margin of victory is probably statistical mistake when it comes to most sports, it is actually exacerbated in the world of a wrestling. In the eyes of RPI, a one-point victory is worth just as much as a first-period fall.
Until recently RPI was also used for selection/seeding of the NCAA Division I basketball tournament. It was never really explicitly said, but the fact that it omits margin of victory was likely an appealing factor since it does not encourage teams to run up the score.
The Bowl Championship Series (BCS), which determined the title game for the NCAA Division I football championship from 1998-2013, used statistical rankings as a component of their selection formula. However, after fans and schools complained that the rankings encouraged running up the score, rankings that used margin of victory were banned in 2002. The move sparked backlash in the statistical community and many refused to continue their collaboration with the BCS.
Bill James, who is widely considered to be the godfather of modern sports statistics, responded in an article for Slate.
"This is very much like a situation in which a surgeon leaves a scalpel in a patient, and the hospital reacts by prohibiting surgeons from using scalpels. I understand that the point of the game is to win, not to score as many points as possible, and I certainly can understand football coaches saying, 'We want a system that emphasizes winning and diminishes the importance of the score.' That's reasonable.
But saying, 'We're not going to pay any attention to the score of the game, and, by the way, you can't pay any attention to whether it is a home game or a road game, either' -- that's just stupid."
Outside of ignoring margin of victory, there is another big issue with RPI. The measure rewards wrestlers more for competing against a tough schedule than actually winning. Only 25% of the calculation comes from a wrestler's results. The other 75% is based on their level of competition.
One of the reasons NCAA basketball moved away from RPI is the fact that the formula favored middling teams in large conferences over teams that dominated in mid majors. The college wrestling landscape is much smaller and more even. Even some of the smallest programs get the opportunity to regularly compete against the traditional powers. However, the issue is still prevalent in wrestling.
For example, let's take a look at Rutgers heavyweight Christian Colucci. He had a solid season for the Scarlet Knights after transferring from Lehigh. He went 15-13 on the season, but he went 10-13 against Division I competition and only 4-9 against Big Ten competition. While that is a respectable season, he was never ranked in the InterMat top 20.
Even though he did not have the most impressive record, Colucci did wrestle a very tough schedule. He faced off against David Jensen (Nebraska), Gable Steveson (Minnesota), Youssif Hemida (Maryland), Mason Parris (Michigan), Trent Hillger (Wisconsin), Sam Stoll (Iowa) and Derek White (Oklahoma State). He went 0-9 against these wrestlers, but RPI still rewarded him for facing them. In the final RPI rankings released by the NCAA, Colucci was ranked 26th. He was one of six wrestlers ranked in the RPI who were absent from the final coaches ranking at heavyweight.
The NCAA should not be shamed for using an impartial statistical measure as part of the selection/seeding criteria. However, there are a variety of alternatives, which could be used that provide a fuller and more accurate accounting of a wrestler's standing.
James came up with Pythagorean expectations as a way to differentiate between baseball teams that squeak past their opponents and teams that dominate their opposition. The formula listed below estimates the percentage of games a team "should" have won based on their run total.
If adapted for wrestling, Pythagorean expectations would certainly take margin of victory into account. However, it would be missing an element of strength of schedule. If the NCAA were to implement this metric for selection purposes, it would likely need to be paired with another measure.
When NCAA basketball moved away from RPI, they replaced the formula with NET. NET is custom metric composed of six factors.
What are the NET Rankings?�" NCAA March Madness (@marchmadness) November 26, 2018
Here's EVERYTHING you need to know. Be on the lookout for the first release 👀 pic.twitter.com/kdZwDEjFPS
Unlike RPI, it takes into account margin of victory and adds offensive and defensive efficiency. So far the reviews of NET have been positive. Adapting the metric for wrestling might prove difficult at the moment since an efficiency calculation would likely require more robust statistical tracking than is currently available in the wrestling space.
In 2017 researchers at the University of Iowa (of course) published a journal article about the ranking of collegiate wrestling. Kristina Bigsby and Jeffrey Ohlmann propose multiple alternatives or additions to the current selection/seeding process. One of the most interesting suggestions is PageRank.
PageRank is an algorithm devised by Google to rank web pages and named after co-founder Larry Page. It seeks to measure the importance of a website based on the quality and quantity of links to a certain page.
Bigsby and Ohlmann adapted the algorithm for wrestling via a formula that is a tad technical. They summarize the conversion by saying, "PageRank divides the importance or power in a network proportionally among the nodes according to number of links. In the context of wrestling, an individual accumulates power in the network by having few losses and many wins over wrestlers who also have few losses."
In addition to PageRank, Bigsby and Ohlmann also point out the potential value of using the Elo rating system. InterMat recently reached out to Bigsby and asked how a ranking like Elo would be an improvement over RPI.
"Most sports rankings focus on win/loss records and strength of schedule, but the magnitude (point differential) of a win is essential to determining victory type in wrestling, and should be a part of the ranking system," Bigsby said. "This is where Elo ratings are very promising. They consider both the quality of an opponent and the dominance of a win when ranking individuals. Elo is a probabilistic system that predicts what proportion of points will be won by each athlete. Something we did not explore in the paper was that this probability could be transformed into an expected point differential, and thus, expected victory type. The Elo system could be extended to predict which athlete will win a match, by how many points, and even how many team points he will earn. So Elo could be helpful not only in building a better individual ranking system, but in building a better team ranking system. I think that's a strong argument for Elo, or at least a strong argument for revising the current wrestling ranking system."
Arpad Elo created the rating system as a way to rank chess players. The rating has since been adapted for a variety of competitions including everything from international soccer to Magic the Gathering. If The Social Network movie is to be believed, Facebook founder Mark Zuckerberg even used an adaption of Elo to rate the attractiveness of classmates.
The calculation of Elo is a little bit more complicated than some of the other metrics listed here. If you would like more details on the behind the scenes math, this article provides a nice summary.
In its pure form, Elo does not take into account margin of victory. However, the formula is malleable and easily adapted. WrestleStat currently publishes rankings based on a modified version of Elo that uses the different types of wins to represent margin of victory, that is fall, technical fall, major decision and decision.
Greg Anderson, the founder of WrestleStat, stresses the importance of continually modifying and improving the algorithm, and he believes that a version of Elo could help the NCAA committee with selection/seeding.
"Elo is fairly useful since it was the baseline for the entire rankings process/algorithm. The algorithm has morphed quite a bit the last two-to-three years, dramatically once I started getting help from my 'algorithm guy.' Since he's taken over, there are a lot more math/equations/formulas involved to try and correlate it to wrestling better ... I think it would be beneficial to the NCAA that it is possible to get good rankings when you take out the human component. That's exactly what they're doing with RPI as well, and the more reliable services you use, the greater than changes you have of decreasing your variability."
Even though there are positives and negatives for any statistical ranking method, the NCAA should not shy away from their inclusion in the selection/seeding method for wrestling. However, it is time for wrestling to follow the lead of basketball and abandon RPI for a more robust system that includes margin of victory.