Scheduleball: Colorado State, Pitt exploit weaknesses of RPI
Teams should aim to assemble schedules that help secure NCAA tourney berths
Who teams play is just as important as which teams they choose not to play
Jamie Dixon has mastered turning decent efficient NCSOS into a top RPI NCSOS
In the science of non-conference scheduling, there exists what I call an "exploitable gap." Some teams are paying less for their high-RPI strength-of-schedule ratings than others.
Many factors affect non-conference schedules -- pressure to play local rivals, participation in multi-team events, travel and timing restrictions -- but the over-arching goal for a good team should be to assemble a schedule that helps it earn an NCAA tournament bid and boosts its seeding in the bracket.
The NCAA tournament selection committee uses the RPI formula to assess teams' non-conference strength of schedule (NCSOS). Two-thirds of RPI's NCSOS is based on the raw winning percentages of a team's opponents, and the other third is based on the raw winning percentages of opponents' opponents. A team vying for an at-large NCAA tournament bid is best off having a respectable NCSOS rank and a number of wins over RPI top-25, top-50 or top-100 teams. While the selection committee has stated that RPI is just one of many tools it uses, the fact remains that schedule strength is viewed predominantly through the RPI's lens.
The problem is that it's a warped lens. Seventy-five percent of the RPI formula is about strength of schedule (SOS), and because the RPI uses the flawed metric of raw winning percentage to assess SOS, it fails to provide a true measure of the quality of opponents. The truest measure available is kenpom.com's NCSOS ranking, which creates a pythagorean winning percentage based on opponents' adjusted efficiency, and even adjusts for home/neutral/road situations, which the SOS portion of RPI does not.
As one might expect, significant gaps sometimes exist between teams' NCSOS rankings according to kenpom.com and the RPI. And therein lies the exploitable opportunity: It's possible to obtain a high NCSOS according to RPI -- which is all that really matters -- without playing a non-conference schedule that's highly difficult.
RPI-savvy coaches have been manipulating schedules for years, but mostly with the goal of obtaining the highest RPI possible. Exploiting the efficiency-vs.-RPI gap is playing next-level Scheduleball. I've yet to hear a coach discuss doing this intentionally, but by creating and analyzing a database of the past five seasons of NCSOS data, I was able to identify two prime cases of Scheduleball success.
In bubble discussions this past March, there was no debate about Colorado State. Of the 115 bracketologists tracked by The Bracket Matrix, all 115 had the Rams in the field as an at-large, even though they were a 20-11, fourth-place team out of the Mountain West -- a profile that isn't typically a lock for inclusion.
But the RPI loved Tim Miles' team. There, it ranked 29th in part due to its No. 26-ranked NCSOS, which was by far the best in the Mountain West. (The next-highest NCSOS was UNLV's, at 131.) The Rams' resume included three wins over top-100 RPI teams outside their league (Montana, Colorado, Denver) and three wins over top-50 RPI teams in-conference (San Diego State, New Mexico and UNLV).
From that perspective, Colorado State looked like a lock. Not only did it receive a No. 11 seed in the NCAA tournament, it was also spared from having to play in the "First Four." On Selection Sunday, there was no griping from the punditry about the Rams' seeding ... but there was no congratulating them about their feat of NCSOS gapping, either. Let me explain:
While Colorado State's NCSOS was No. 26 according to the RPI, it was No. 130 according to adjusted efficiency -- an incredible 104-spot gap. From 2008-2012, there were only five bubble teams that pulled off 100-plus-spot gaps, and only the Rams had done it to move to the edge of the RPI's top 25. They had managed to play a non-conference schedule that, in actuality, was middle-of-the-road, but it was so well-tuned to exploit the RPI's flaws that, in the eyes of the selection committee, it looked elite.
|Bubble Teams With 100+ Gaps from Efficiency NCSOS to RPI NCSOS, 2008-2012|
How did Colorado State pull this off? It helped that Miles, who used the NCAA bid as a launchpad to the Nebraska job, likes to study RPI numbers, and that CSU's then-athletic director, Paul Kowalczyk, was experienced in RPI-boosting from his previous gig at Southern Illinois. They combined smart forecasting with a bit of luck:
To start, half of the Rams' 14 non-conference games were against top-100 RPI teams. "Tim liked to try to predict in advance," Kowalczyk said, "how many top-50 and top-100 wins he might get from inside the conference, and then from there, build a non-conference schedule with what you might need to qualify as an at-large team."
Three of those top-100s -- Montana, Southern Miss and Northern Iowa -- were teams that would go on to have strong records in lesser conferences, and thus be valued higher by the RPI which uses raw win%, than they were by kenpom.com's efficiency ratings. (Montana, for example, wasn't a top-100 team in efficiency, but it ranked No. 74 in RPI.) Scheduling these types of teams is essential to any gapping strategy.
Miles also scheduled a game at Duke, which was no-risk, all-reward. The Blue Devils were ranked No. 20 in efficiency, but they were No. 5 in RPI and had the No. 2 overall RPI SOS. "We always try to play marquee teams that won't hurt our RPI," Miles said. "There's far more [RPI] damage in losing to a poor team than losing to a good team."
As I noted earlier, one-third of RPI NCSOS is based on the winning percentages of your opponents' opponents -- so it helps to play teams that have difficult schedules. Five of Colorado State's non-conference foes ranked in the top 30 of RPI NCSOS: Duke (2), Northern Colorado (8), Texas Southern (10), Northern Iowa (13) and Southern Miss (29).
Whom the Rams didn't play was just as important: They only had one 250-plus RPI opponent, Northern Colorado (262), and only three others in the 200-250 range. It's best to avoid the 250-plus RPI teams altogether, and, as Kowalczyk said, "You're almost better off scheduling a Division II team at home, which doesn't count against your RPI, than a really bad D-I team." This is most definitely true, and the Rams put it into practice, playing Nebraska-Omaha in Fort Collins rather than bringing in a D-I team that would've been a drain on RPI.
Miles' team had to go out and win games, too. But by creating seven opportunities for top-100 victories outside the Mountain West, and assembling the No. 26 RPI NCSOS without actually playing a high-efficiency slate of opponents, they put themselves in excellent position to earn an at-large bid. Theirs was the ultimate gap-exploitation schedule. If their efficiency NCSOS (130) and RPI NCSOS (26) rankings had been reversed, the Rams would've actually had to sweat on Selection Sunday.
As another phase of this study, I separated out the coaches who were with the same team from 2007-08 to 2011-12, and looked at their non-conference schedules over that entire span. The five with the most consistent ability to exploit the efficiency/RPI NCSOS gap are below, in order of average gap per year:
|Coaches with top EFF/RPI NCSOS Gaps, 2008-2012|
Pitt's Jamie Dixon is rarely lauded for his scheduling -- if anything, the knock on him has been that he doesn't schedule hard enough when he has an elite team -- but here, he looks shrewd. He's the best coach in the country at consistently turning a mildly efficient NCSOS into a respectable NCSOS in the eyes of the RPI, and thus, the selection committee. As he says, "I look at kenpom statistics -- I love all the stats he does -- but in scheduling, the NCAA is gonna look at RPI. So I care about RPI."
Dixon's year-to-year gaps break down like this:
Sometimes a gap won't matter -- like last season, when the Panthers nose-dived and went 5-13 in the Big East. They had no shot at the NCAAs. But in 2010, when Pitt earned a No. 3 seed despite having zero marquee wins outside the Big East, it was in part due to Dixon's manipulation of the 158th-best efficiency NCSOS into the 49th-best RPI NCSOS.
What Dixon likes to do for his home guarantee games, he says, "is play the teams that we think are the best picks to win the non-BCS conferences." These are the best "gap" teams, because they're beatable despite having high RPI returns. In 2010, Dixon beat five of them in Wofford (69 RPI), Wichita State (43), Kent State (47), Ohio (95) and Robert Morris (129). He only had one 250-plus RPI opponent (Youngstown State, at 271), either, and so it didn't matter that he played just one marquee game (against Texas) and lost it; the Panthers were in good standing due to their choices of non-BCS opponents. Despite their efficiency profile suggesting they were the quality of a 7-8 seed, they were a No. 3 on the strength of their RPI.
It's almost always better to play a mid-major that'll go on to have 20-plus wins than it is a cupcake mid- or low-major, or a basement-dweller from a major conference. Dixon still adheres to this philosophy; last season he hosted Long Beach (34 RPI), Robert Morris (99) and Wagner (92), and this season he has Detroit as well as either Robert Morris or Lehigh, depending on how the NIT Season Tip-Off Bracket plays out.
And he says there's an added benefit to playing guarantee games in which you might have a chance of losing: "Those are the teams that other [BCS-conference] schools don't want to play, so not only do you get a higher RPI by scheduling them, you can also possibly schedule them for less money. It comes down to supply and demand, and there's just less demand for those teams."
Not only is he gaming the RPI, he's saving his school money on its guarantee-game budget. High-value teams at low prices: That is next-next-level Scheduleball.
Penguins squeak past Ducks in shootout
Johan Franzen scores two as Red Wings pound Devils in Detroit