This offseason, I’m devoting some time to highlighting some “mistakes” in conventional football wisdom and explaining how we can improve on them. This week, I’m here to discuss “game control”. Specifically, I’m going to address time of possession.

This particular stat came to light multiple times in TCU’s 2019 season, as the Horned Frogs lacked explosion and tried to rely on a ground-and-pound game, “controlling the ball” and shortening the game. Coach Patterson is not unique in his affinity for “slowing down” and establishing the run; many a college football coach (I see you, Pat Fitzgerald) believes that holding on to the ball for longer periods of time, in and of itself, will increase the probability that a team wins the game.

At first blush, it makes sense, right? If I hold onto the ball more, I milk the clock, and my team will “be in control,” providing fewer opportunities for the other team to score, especially if I have a lead.

Controversial Air-Raid God/New Mississippi State Coach Mike Leach begs to differ:

People that value time of possession, they just sit behind the center and pretend they are gonna tickle his ass, and maybe they are and maybe they’re not and just wait for the clock to tick by. You’re just pretending you’re a smart football coach if you’re doing that. If what you do in practice has any value whatsoever, you’d probably be focused on executing plays.

The problem is, of course, with that word “opportunities.” If a team holds the ball for 8 minutes and never cross the fifty-yard line, they didn’t really create a scoring opportunity. Opportunities in college football aren’t based so much on time as they are drives.

Time-of-possession is a misleading stat because it insinuates that all drives are productive drives. What is the first rule of football analytics? Not all stats are created equal - in this case, not all minutes of possession are created equal.

Let’s go to a corner solution: Imagine my team, Team A, holds the ball for 59 minutes of a game, because Team B fumbled on the first play of every drive. Did I win the game because I held on to the ball more? Well, no, I won the game because Team B had no opportunities to score - not because they didn’t have time to score, but because they didn’t move the ball into a scoring opportunity. Sure, time becomes a limiting factor in college football games, mostly at the end of the halves, but the real measure of game control is not what you do with your time, but what you do with your drives.

Bill Connelly tracks scoring opportunities (defined as having first down inside the opponent 40 yard line) in his Box Scores, and often the sheer number of scoring opportunities, along with points per scoring opportunities, provides a pretty good measure for who won the game. I think we can go a little further into understanding season-long team performance by borrowing from hockey analytics.

Corsi measures the ratio of productive shots (at even strength, meaning no power plays/advantageous situations) between two teams in a hockey game. It was invented by analyst Tim Barnes and named after some guy with an awesome mustache. Corsi is a measure of productive puck possession - how much of the game’s productive puck possession did your team account for? Corsi has been found to be a robust predictor of future performance in NHL games, and the concept holds up in college football: instead of focusing on how long you held the ball, why don’t we focus on how productive you were when you had the ball?

Corsi is calculated simply as Shots + Blocks + Misses/(Shots + Blocks + Misses + ShotsAgainst + BlocksAgainst + MissesAgainst). In football, that translates nicely to Scoring Opportunities/(Scoring Opportunities + Scoring Opportunities Allowed). It’s a measure of game control in that it tells us in a single statistic what percentage of the productive drives you accounted for in a game.

Before I dive into the data, a couple disclaimers. Corsi on it’s own will be a little tricky to sort out, because the ratio will be close to 1 if you win 49-0 with 7 scoring opportunities to 0 for your over your opponent, or it will be close to 1 if you win 3-0 with 1 scoring opportunity to 0 for your opponent. It’s a measure of game control and relative, not absolute performance. A low Corsi could mean your defense got beat up or that your offense couldn’t get anything started. As with most single-number stats, it adds value in context.

Corsi vs. Time of Possession

In college football analytics, we mostly care about stats that are predictive, that is, stats that tell us something about what we can reasonably expect from a team in the future. No stat will perfectly inform us as to a team’s quality, because each game is full of randomness. A good stat, though, should give us a reasonable idea of how a team might perform in the future. To that end, let’s compare time of possession and Corsi in predicting win percentage for FBS teams.

This plot shows the relationship between time of possession and win percentage. There’s a correlation here! It’s slightly positive, indicating, as we know, that teams who hold the ball more win games more. That says nothing about the fact that, of course, the better team in a game will end up holding the ball more. Time of possession doesn’t cause winning, it’s a result of winning! But, still, there is a lot of variation, and the line is only slightly positive. In fact, the R-Squared is .2371, indicating that only 23.71% of the variation in win percentage is explained by time of possession.

On the other hand, if we plot Win Percentage against Corsi Index, we get a much steeper positive relationship, and an R-Squared of .5586. Corsi Index explains more than twice as much the variation in win percentage that Time of Possession does. So, not only is Corsi a more sensible stat, it performs better that TOP!

If we filter the Corsi Index for garbage time, then we have some slight movement, and an R-Squared of .5926.

Having established that the time-adjusted Corsi Index is more connected to meaningful on-field performance and is a better predictor of future wins, let’s document teams by Corsi.

Grading Teams by Corsi Index

The top ten college football teams according to time-adjusted Corsi:

1. Clemson 77.9%
2. Ohio State 73.9%
3. Georgia 65.7%
4. LSU 64.3%
5. Utah 64.1%
6. Boise State 63.4%
7. Michigan 62.4%
8. Florida 60.6%
9. San Diego State 60.4%
10. Notre Dame 60.4%

Well, all ten of those teams are pretty good football teams, with 9/10 in the top 30 of SP+ rankings this season. The Corsi Index, not adjusted for opponent, paints a pretty solid picture of the most dominant teams in the country. Think about what this tells us - Clemson, in the regular season, accounted for almost 78% of the scoring opportunities in their games! This stat says nothing about finishing these drives (an extension for another day), but instead tells us who even had a shot. Teams facing Clemson, on average, had 1 scoring opportunity, 1 productive drive, to Clemson’s 3! Now, it wouldn’t have mattered if an opponent held the ball for 8 minutes each drive, because Clemson would’ve still had more productive drives.

Let’s look at the Big 12:

1. Oklahoma 59.0%, 15th
2. TCU 57.8%, 17th
3. Iowa State 56.6%, 29th
4. Baylor 54.8%, 34th
5. Kansas State 53.4%, 43rd
6. Texas Tech 52.1%, 57th
7. Oklahoma State 48.8%, 86th
8. West Virginia, 47.5% 92nd
9. Texas 47.5%, 93rd
10. Kansas 40.1%, 119th

Not surprising - most of the “good” teams are at the top. This ranking tells us a few things off the bat: TCU - very good defense! The Horned Frogs were 27th in scoring opportunities against. Shockingly, the Horned Frogs were 23rd in scoring opportunities for - TCU was set up for success with a solid defensive foundation and an offense who could move the ball between the twenties. Obviously, the disparity between their actual performance and their Corsi Index highlights an excruciating inability to finish drives.

Oklahoma State, at 7th, and Texas at 9th in the Big 12 both seem a little low, and that highlights a limitation and a possible extension: Texas’s offense was 10th overall in SP+ this year - they moved the ball well and were explosive through the air, but their defense was 68th. Perhaps that number would balance out if we added in big play scores to the scoring opportunities in the Corsi Index. Oklahoma State suffered from a similar problem.

Conclusion

What have we learned? Time of possession is a bad stat because it cannot distinguish between “productive” and “unproductive” possession - a three and out that takes 35 seconds and a 75 yard touchdown bomb both factor in the same. By borrowing the Corsi Index from Hockey, we can measure game control not by “who had the most possession time” but instead by “who had the most productive possessions.”

The Corsi Index should be used in context with other advanced metrics to better understand what happened in a game. By diagnosing game control properly, we can see if an offense has a disconnect between finishing drives, field position, or offense/defense. The Corsi Index provides a nice contextual measure of how well an offense and a defense performed together to “control” a game.

There are some limitations, most notably the “big play” problem, but the raw Corsi Index provides a nice foundation for understanding what really happened in a game, predicting future performance, and a solid step forward from using time-of-possession metrics.