Oklahoma Football: New S&P+ numbers kinder to Big 12

College football statistics maven Bill Connelly tweaked his S&P+ efficiency ratings during the 2018 season to enhance the predictive prowess of the model. With the offseason in full swing, he has applied the changes to his numbers for past seasons.

For kicks, I looked back at the previous iteration of Bill’s statistics for the Big 12 from the 2015, 2016 and 2017 seasons to compare against his updated numbers. (Yes, this is what I do for fun.)

In the process, I teased out a few observations that are worth considering in the broader picture of the Oklahoma Sooners and the Big 12 as a whole.

In the past, S&P+ has underrated the Big 12 as a whole

To be fair, you could probably say the same about each of the Power Five conferences. Turning the dials sent Group of Five teams tumbling down the overall rankings, replacing them with representatives of leagues such as the SEC and Big 12.

Although the adjustments didn’t produce major changes in ‘15 and ‘16, Big 12 teams moved up roughly 11 spots on average in the national rankings in ‘17 (seen above). Seven of the league’s 10 teams saw their overall ratings improve at least marginally by updating the numbers, while one held steady.

Defenses in the Big 12 have been better than they have been given credit for

Are the conference’s defenses uniformly great? Not at all. Precisely one team, the Texas Longhorns in 2017, fielded a unit that ranked in the top 15 in Defensive S&P+ in the three years.

On the other hand, Big 12 teams saw their Defensive S&P+ rankings climb an average of nearly 14 spots in both ‘16 and ‘17. This contributed heavily to the upgrades in overall S&P+ in ‘17.

I asked Bill for insight on what happened.

“The defensive ratings were pretty skewed in the Big 12, and it was causing S&P+ to underestimate the conference by a decent amount on average,” he said. “So when I installed that extra layer of conference ratings adjustments, it was pretty [Big 12]-friendly.”

Continuing a theme, Mike Stoops was (slightly) better than many thought

Rob Ferguson-USA TODAY Sports

Okay, okay, take a deep breath. Mike Stoops did not do a great job in his second run as OU’s defensive coordinator. Far from it, actually.

However, the new formula indicates OU’s defense played better than initially thought in two of the three seasons. Of note, OU’s derided defense moved from a humiliating 101st in the nation in Defensive S&P+ in ‘17 to 43rd following the adjustments.

Of the seven Big 12 teams that saw their defensive rankings improve in ‘17, OU’s 58-spot leap proved to be the biggest of all. As a result, OU’s defense came in fourth in the conference that season, not eighth.

The previous season witnessed a similar move, albeit on a smaller scale. OU went from 55th overall to 31st, which ranked second in the conference.

Unfortunately, we’re really talking about window dressing for a program that aspires to win national championships. Rankings in the 30s and 40s on either side of the ball aren’t going to cut it. Not to mention, OU fell to 84th overall in Defensive S&P+ in 2018.

Lincoln Riley made a necessary move when he gave Stoops the ax in the middle of the 2018 season. He would have been justified doing it even before then. I’m simply saying the updated S&P+ stats reflect a little better on Stoops’ tenure now.

The miraculous defensive turnaround of the Iowa State Cyclones in ‘17 looks less so

After the latest round of adjustments, the Clones ranked 88th nationally in Defensive S&P+ for ‘17, down 57 spots from their previous ranking of 31st. No team in the Big 12 came close to matching ISU’s downgrade in the update.

If Bill’s new analytics are to be believed, the Cyclones actually had the second-worst defense in the Big 12 in 2017, not one that was almost on par with the Texas Longhorns and TCU Horned Frogs as the league’s best.

This puts some people (including me) in the pontificating class in an intellectual pickle. The ‘Clones improved from 3-9 in 2016 to 8-5 the following year. After the ‘17 season, ISU’s D received plenty of credit for the turnaround. The Cyclones’ shiny Defensive S&P+ metrics were cited in some circles as a sign that head coach Matt Campbell and defensive coordinator Jon Heacock had devised a formula for slowing the prolific spread offenses of the Big 12.

Instead, the revamped S&P+ numbers suggest the Cyclones actually got worse on defense in 2017 from the prior year. Meanwhile, ISU’s Offensive S&P+ ranking climbed from 71st to 13th in ‘17.

It is worth noting that ISU checked in at No. 28 in Defensive S&P+ in 2018, so it’s not off-base to say the Clones were on the right track in ‘17. The problem is that the adjusted metrics from ‘17 don’t support that claim. You could just as easily argue improvement at quarterback with Kyle Kempt or losing one fumble the entire season were the keys to ISU’s surprising ‘17 campaign.

Analytics and college football are still feeling things out

Out of some sense of machismo or proud ignorance, football culture has largely balked at the analytical revolutions sweeping through sports such as baseball and basketball. The resistance to such nerdery is now fading as the old guard gradually cedes control to a new generation familiar with big data and moneyball.

In the process, though, we’re also learning more about the massive challenges posed by trying to create effective statistical models for college football. The action in football doesn’t fit traditional modeling techniques as readily as other sports. The lack of solid connectivity between teams’ data sets complicate matters even further. Taking that into account, it shouldn’t come as a surprise when refinements to what are still nascent models result in fairly significant swings in results.

S&P+ offers a useful tool for analyzing college football, but the shifts resulting from the latest round of adjustments should serve as a reminder to take care in drawing hard-and-fast conclusions from advanced stats without understanding the context and caveats around the numbers. We’re still so early in the game with college football analytics that it’s easy to get overzealous in how they’re applied.