George Carlin noted in the first monologue in Saturday Night Live history that football, when stripped to its essence, is a game of ground acquisition. That analogy came to mind when I was trying to come up with a statistical power rating of college football teams.
Those of you that have read or followed my work with adjusted stats will recall that adjusted yards gained and allowed per play are the single best stats in terms of correlation to points scored and allowed. But as I worked with those figures it quickly became obvious that any regression model would not be accurate enough to properly approximate team strength if it didn't include turnovers as well. I had been focusing on per play stats but the nagging thought in my mind that I had never extended per possession analysis kept bugging me. Putting those two issues together with the knowledge that turnovers per possession had always seemed more logical as a relevant stat than turnovers per play prompted me to run two different regressions, one based on per play stats and one based on per possession stats.
In the end the statistical significance of the two different models were nearly identical with the possession basis having an extremely slight edge. The edge was small enough to be insignificant, but I decided to proceed in that direction because in the end a team's offensive and defensive success comes down to what they do with each possession. Going 80 yards for a score in 10 plays or 6 plays results in six points. But perhaps more importantly, that analogy and that turnovers make more sense from a possession point of view won out in the end.
What it shows
YMPPs - Yardage Margin per Possession - Adjusted yards gained per possession minus adjusted yards allowed per possession
NPM - Net Punting Margin - Adjusted net punting average minus adjusted net punting average allowed
TYMPPs - Total Yardage Margin per Possession - Sum of the above two figures
TMPPs - Turnover Margin per Possession - Adjusted turnover margin per possession (previously available on AdjustedStats.com)
FPP - Field Position Power rating - Formula-based rating derived from each season's statistics-to-power linear regression
Essentially what this shows is the best teams at winning the field position battle. The tables are sortable by each statistic if you want to view, for example, just the yardage margins or any of the individual statistics. It could even be visualized as a ranking of each team if football was played without end zones on a field of infinite length and football truly was a battle of only territorial acquisition. Line the ball up at a point and play a specified number of possessions. If the ball is on their side of that starting point after the number of possessions are complete then you win. If not, you lose. Don't ask me about breakaways.
If you review the numbers, you should quickly notice that outstanding special teams play on a team's punt units can make a big difference; take a look at Cincinnati's line for an example. And that aspect of the game is something that is usually not given enough attention, so I don't necessarily view that as a bad thing.
What it doesn't
The final number ignores kickoffs and kickoff returns as those are the result of scores and not field position exchanges. It also obviously ignores red zone percentage and actual scores. It is intended only as a different look at team performance and I'll be following it to see if any undervalued or overvalued teams can be identified. Yes, I know West Virginia lost to Syracuse tonight, but there's no model in existence that saw that score coming. And I realize Maryland is a huge WTF at #9, but that's the way it goes. If you think that's crazy, check out 2007 Kansas State. In 2007 the Wildcats went 4-7 against FBS teams despite averaging 22.53 yards per punt return and scoring 5 touchdowns on punt defense.
Some other weaknesses off the top of my head are the effect of the end zone on net punting margin as a team will gain a benefit from their opponent punting into the end zone when their defense gave up a few more yards. Also, luck in turnovers isn't quantified (e.g., Kansas State has recovered all 8 of their offensive fumbles this year), but that could be added in the future by adjusting each team's turnover numbers to equal interceptions plus half of fumbles/fumbles forced instead of all fumbles lost. At this point I don't plan on doing that but it may be added in the future.