# A Word On Advanced Metrics, College Football, and Texas A&M

## Thoughts On Texas A&M Football and Advanced Metrics

Earlier yesterday evening, I was minding my own business and browsing Twitter when I saw SEC Mike putting out some stats for the conference, ranking all the teams (including Texas A&M football, of course) according to “offensive scoring efficiency” and “defensive scoring efficiency.” All two members of my consistent readership here will know that I often reference statistics that are not exactly top line box score numbers when I’m talking about matchups and analyzing previous games, so I spend a certain amount of time thinking about these stats. That is not said to show off some kind of bonafides—as a matter of plain fact, “spends a certain amount of time thinking about stats” is not even anything approaching a credential—but rather my point is to say that I see someone putting out a statistical ranking without saying what numbers they’re using to measure it, I wonder what the calculation is. It especially makes me wonder when the lists themselves don’t comport with my priors (which is not to say that my priors are always correct; rather more generally that people are more likely to question purported facts that seem out of step with their perception), which this one definitely didn’t. So I replied to Mr. Michael and asked!

His reply was to say that the scoring efficiency number he used was total plays divided by total points. This is not all that dissimilar to points per play, which I have used before—in fact, it is literally an inversion of that statistic! However, I have a qualm with the presentation here; first of all, if my calculations are correct, SEC Mike did not filter out non-offensive touchdowns in these numbers. I’m not saying non-offensive touchdowns gained or given up have no significance when discussing the quality of a team, but rather that an offense that gives up a touchdown to the opposing defense, such as that of Texas A&M football vs. Arkansas or Auburn, should not cause their defense to look worse as a result. The Aggie defense gave up only 3 points to Auburn and only 16 points to Arkansas. That is what the *defense* allowed to those teams.

This brings me to a related point. Let me say first that this is intentionally ridiculous in order to illustrate my point; I realize the scenario I am putting together is not realistic. Imagine the greatest defense ever: every play is a TFL, turnover, whatever. At a maximum, they only face three plays per drive whenever an opposing offense trots out to face them. Now imagine that this perfect defense is paired with the worst offense ever. This offense is bad enough that no matter what they do, they keep giving up scoop-and-scores and pick-sixes. If you don’t filter out non-offensive touchdowns, this steel-wall defense will be ranked, by this metric, as a pretty poor one; they haven’t faced many plays compared to the points they give up.

Now let’s consider another scenario; this one much more realistic. Defense A faces five different drives; four of them are three-and-outs, and the last one is a long touchdown on a busted coverage. Five drives, thirteen plays, seven points allowed. Defense B faces five drives from the same offense; they are less of a lockdown unit and allow several conversions, with the offense moving more reliably into scoring position. However, a missed field goal by the offense and a couple of negative plays to knock them out of range for a kick in some other cases where they drew near make it so that they also only score on one of these five drives, a long one that ends in a two-yard touchdown. Let’s say Defense B faced a total of 35 plays on these five drives. So five drives, 35 plays, seven points.

Analyzing these two scenarios with the plays/points metric can be deceiving. Purportedly, that metric shows defensive quality by showing how many plays on average it takes for opposing offenses to equal one point (which by the way is a much poorer way to show that than the inverse—football is not scored one point at a time, but it *is* played one play at a time); so a higher number is better here. So in this case, Defense A has a defensive scoring efficiency of 13/5, or 2.6, whereas Defense B has a defensive scoring efficiency of 35/5, or 7. Even though Defense A was smothering the opposing offense for most of the day, this metric has it as the inferior defense by a factor of a little over two and half.

Here’s the point: this metric does not account for explosive plays. That one play that Defense A gave up that was broken for a touchdown was consequential enough to rate Defense B’s performance as over two and a half times better, even though both teams gave up the same amount of points. Due to the prevalence of explosive plays in any given game, this causes way too much noise for this to be instructive as a standalone measure of scoring efficiency.

Let’s contrast this to points per drive, a similarly simple way to calculate scoring efficiency. Since we’re talking about defense, a *lower* points per drive number is preferable. Using the earlier scenario, since both teams scored seven points on five drives, both defense A and defense B would have a rating of 5/7, or 0.714. Now, since both teams gave up the same amount of points in the same sample size here, and this measures scoring efficiency, that makes more sense. Of course, it does not account for how well Defense A was playing on a down-in and down-out basis, but it doesn’t necessarily need to since it’s a measure of *scoring* efficiency. The point with the previous example is that Defense A was *actively penalized* for basically just being a better defense—obviously you would rather give up 3-and-outs than long sustained drives.

Ultimately, to measure the quality of a team (or even a single unit!) in a sport like football, you need a multiplicity of statistics given the matchup-oriented nature of the game; not only that, but you need a multiplicity of opponent-adjusted statistics. College football especially is a sport with a very small sample size and very limited cross-pollination between teams that are often being compared—many times, Ohio State and Texas A&M football (for example) don’t even have one single common opponent in a given year. That’s why some of these statistical profiles get so big and complex: to help account for these constraints. Even then, no one system is perfect. Individual statistics are best used to evaluate specific matchups, and even then the methodology behind these statistics needs to be examined. What are the strengths and weaknesses of a given stat? What could it fail to account for?

There’s also a difference between predictive and descriptive statistics. You can get some descriptive value out of pretty much every statistic—it’s just that with some of them, the thing you’re describing is almost useless for any practical purpose—but not every statistic has predictive value. Here’s what I mean: taking pure scoring (points per game) can tell you what a team *has done*, but it is not necessarily the best way to project forward what they *will do*. There are unrepeatable, random aspects of football that can affect a univariate analysis such as scoring in a big way. Instead, it is better to use numbers that best correlate to reliably creating points over a large sample size. This is the use of some of these more esoteric-sounding statistics such as success rate, EPA, etc.

At the end of the day, it’s unhelpful to say you’re ranking teams by something as broad as “scoring efficiency” and use a single number as a catch-all for that concept. There are some “all-in-one” numbers out there, like opponent-adjusted EPA/play margin or an SP+ rating, but those are really complex, as I just said, and therefore don’t quite resonate as much with people. The better way, in my estimation, is to preview one team versus another and give statistics about their specific strengths and weaknesses.

## Evaluating the 2023 Texas A&M football team

So what about Texas A&M football? Which statistics tell us the most about the Aggies? Well, as I wrote earlier, there are a couple of all-in-one statistics that attempt to paint a picture of who A&M is—some of which I have written about as they come across my timeline.

To paint a fuller picture, however, here are a few numbers that might be helpful in understanding just who this Texas A&M football team is so far this season. By my points per play calculations, which filters out defensive and special teams scores, the Aggies are the fifth-best offense in the conference (0.499 points per play) and the fourth-best defense (0.290 points per play allowed). By a points per drive metric, the Aggies rank the same: they again are the fifth-best offense (2.753 points per drive) and the fourth-best defense (1.394 points per drive allowed). Applying some slight opponent adjustments to these numbers (I use a simple averaging method) is illuminating, however: The Aggies then are the third-most efficient offense and the third-most efficient defense by both points per play and points per drive. UGA is the only team in the conference that is ahead of the Aggies in all of these categories; LSU is better on offense but horrible on defense, whereas Alabama is better on defense but deficient on offense (the USF game probably still affects them there).

Those are some simple efficiency numbers. To close out, I’ll give some quick-hitters that I think best capture where Texas A&M football is strong and where they are weak.

## Strengths for Texas A&M Football

- 52% average YPC allowed: the Aggies have held opponents to only 52% of their rushing averages on a yards per carry basis. Essentially, if a team came in averaging 5 yards per carry, this number suggests the Aggies would hold them to 2.6 yards per carry. That’s the best number in the conference by a wide margin.
- 3.89 sack-adjusted YPC allowed: this number controls for the fact that college football statistics count sacks as rush attempts and sack yards against rushing totals. Filtering those “rush attempts” out gives us the sack-adjusted YPC. This number for Texas A&M football is once again the best in the conference, speaking to how much the rush defense has improved.
- -.095 PPA/Rush allowed: PPA has the same theory underlying it as EPA; basically what this number means is that, on average, each rushing play that an Aggie opponent attempts actually makes it
*more likely that the Aggies will score next*, rather than the team that actually has the ball. That’s how good the rush defense has been. - 25% opponent third downs converted: hopefully this doesn’t take too much explanation! Opponents only convert 25% of third downs on average against the Aggies.
- 16.72% TFL rate: The defense for Texas A&M football creates a tackle for loss on 16.72% of the snaps they play. Tops in the conference and in the nation!
- 47% third down conversion rate: Texas A&M football has converted 47% of the third downs they’ve faced. That ranks third in the SEC.
- 47% opponent-adjusted success rate: also third in the conference!

## Weaknesses for Texas A&M Football

- 118% average passing explosiveness allowed: I use the explosiveness metric from collegefootballdata.com which is based off of PPA; basically, the theory behind this number is to measure explosiveness as an answer the question “when you have positive plays, how big are they on average?” This number I’m referencing for Texas A&M football means that teams are 18% more explosive than their averages in the passing game when they play the Aggies, so it’s what we’ve all seen: the Ags allow big passing plays too often. This is the worst mark in the conference by this metric.
- 0.145 PPA/Rush: each A&M rush on average doesn’t create much scoring value. I’d chalk this up to the lack of explosiveness on A&M’s rushes so far this year, as their success rate (percentage of rushes with a positive PPA) is actually one of the higher marks in the conference.
- 4.393 Points per Red Zone Trip: We all know that another struggle for the Aggies this year has been converting in the red zone. That showed up massively against the Crimson Tide and was very consequential for the game. This number puts them at 10th-best in the conference, only above Miss. State, Kentucky, Auburn, and Vandy.