In taking a closer look at the work that ESPN does with their FPI and Bill Connelly does with his S&P+ I found a golden nugget that I hadn’t really studied closely before. The nugget was in Connelly’s work. The vast majority of these two analytical approaches to “forecasting” is based on returning production. There are some elements included which try to account for recent history. Connelly doesn’t use the most recent year; he uses year 2 through year 5 in history.
For recent history, I’ve found that getting a little weird predicts pretty well. This number isn’t a strict five-year average — last year’s ratings already carry heavy weight from the returning production piece. Instead, what you see below is a projection based solely off of seasons two to five years ago. Recent history doesn’t carry much weight in the projections, but it serves as a reflection of overall program health. We overreact to one year’s performance sometimes.
So our program health is being evaluated by Connelly based on the performances we had in 2014, 2015, 2016, & 2017. So while this is the second year with Pruitt at the helm, Connelly’s approach sees it still as Butch Jones’ program. ESPN FPI, on the on the other hand, gives points for a returning head coach. We didn’t have a returning head coach last year but we do this year.
The remaining component of these analysis approaches is recruiting impact. In Connelly’s work he’s found that weighting the historical recruiting classes are more predictive than just using a simple average. This is the golden nugget. The 2016 recruiting class in his analysis will only contribute 3% to our recruiting rating this year. The 2017 class, the five-star hearts class, will contribute 15% to our score for recruiting impact. The 2018 class, Pruitt’s year zero class, will also contribute 15% to our recruiting score. A whopping 67% or 2/3rds of the recruiting impact on our team this year will come from the 2019 class. In total 82% of the recruiting impact in Connelly’s analysis for predicting our season will come from recruits that Pruitt has brought on board. Whether that’s reality or not from a standpoint of who we put on the field, that’s the sort of math that these prediction models are using.
The reason I was looking at this more closely is Devo I think has raised some legitimate concerns regarding our defense for the coming season. ESPN thinks we’ll have a top 20 defense, projecting us at #19 but Connelly sees it different, positioning us on defense all the way down to #49. Connelly likes our offense at #13 and ESPN FPI has it at #18. So why the big difference in views regarding our defense?
Connelly states that he is using historical performance not including the most recent year to assess program health. From a program health standpoint he has us ranked as having a program state that ranks as the #33 best in the country based on our performances in the 2014, 2015, 2016, & 2017 seasons. Any credit we get from jumping from #82 to #49 in total defense from 2017 to 2018 is in his approach already accounted for in the returning production numbers. Our 2016 defense was ranked #95 for total defense in the country so I think those two years are likely skewing the analysis by not accounting for the change in coaching staffs and the improved performance on that side of the ball in 2018, something that ESPN’s FPI apparently rewards us for.
One last point. Devo, along with many others, have raised specific concerns about our defensive line this year. Others continue to harbor ongoing reservations about our offensive line heading into this season. Connelly suggests based on his analysis of the vast amount of data in his database there is little correlation between either line and the performance of the unit on that side of the ball. This certainly could be a flawed analysis and he admits that it is contrary to what one would expect. As a standalone data point I can easily see how positive and negative correlations would tend to cancel each other out if you were trying to isolate performance just based on the line of scrimmage.
Each unit on a football team is composed of 11 men. Skill position players have measurably the most impact on these mathematical models but I suspect that a team’s running game and passing game with good skill position players and a good offensive line is going to more often vastly outperform a team with good skill position players and a mediocre offensive line. The same applies to the defensive side. I think it was Muschamp when he was at Florida who said that the best pass defense was a good 4-man pass rush. I’ve heard over the years that a defense is really only going to be as good as its secondary and Connelly agrees that defensive backs have more impact on the performance of a defense than any other position group on that side of the ball. Putting this altogether I find myself in agreement with Devo in this respect, that while I think we have the potential for a really good secondary this year, it will be all that much better if we can field a decent defensive line. jmo.
Edit to add that I think most people would tend to agree that in our 3-4 scheme that having a good defensive line will also likely directly contribute to the improve performance of our primary skill players (playmakers) against our opponent's running game, our linebackers.