I’m always trying to find ways to take advantage of being last.
Football analytics have unquestionably lagged behind that of other sports, which stinks from a “building a depth chart of nerds” perspective but certainly helps from the perspective of generating ideas. The first iteration of my S&P ratings was basically a stab at an OPS-style number for football. The Five Factors are a blatant, shameless spin-off of basketball’s Four Factors. Et cetera.
I wanted to quickly look into a soccer concept: game state. It’s an enormous thing for soccer.
As we all know, the score does change a team’s tactics. When a team is losing they must push forward and try to get the ball and keep possession. When a team is winning they are prone to sit deeper as defense become a priority.
Because there are so few scores in a given soccer match, your tactics will tend to shift dramatically if you’re up a goal, down a goal, up two, tied, etc. Obviously it’s a little bit different in college football — the average team scores around 27-28 points in a given game, so being up or down a single score probably won’t change much about your tactics, especially early on.
But those tactics do change at some point.
Out of pure curiosity, I pulled the average college football run rate by scoring margin and quarter from 2006 to present. At what point does play-calling begin to shift?
If you squint, you can see a small change immediately. A team up by a small margin in the first quarter runs 53.1 percent of the time, while a team down by a small margin runs 51.8 percent of the time. That 1.3 percent difference increases to 2.4 percent in the second quarter, 2.8 percent in the third, and a whopping 29.5 percent in the fourth.
(Note, I don’t have full clock data — as in, the official time left in each quarter — for every play, otherwise I’d break each quarter into halves, and we’d see an even greater set of splits.)
One thing I enjoy here: you can see when the white flag gets waved. One teams are down 35-plus, they want to get the game over with as much as the leading team does.
In soccer, game state data can get a little bit funky thanks to the simple fact that, while your tactics can indeed change when up or down a goal, you’re also more likely to be leading when you’re the better team. It makes things a little bit messy. So game state data can be very interesting in terms of when you’re tied, up one, or down one, but if you’re, say, down three goals, it’s probably because you’re simply outmanned.
With that thought in mind, here are teams’ success rates by game state from 2006-present.
Even though teams that are ahead run more, they’re also pretty successful at it ... likely because they’re simply better than their opponents. But the interesting part comes with the fourth-quarter data.
If you’re up by one possession in the fourth quarter, you’re likely a) running the ball a ton and b) doing so unsuccessfully. If you’re up two possessions, you’re running the ball even more and you’re slightly more successful at it (possibly because you’re the better team). But that’s the one area where success rate and leads don’t match up nicely. It’s the one time where clock management becomes as or more important than making good plays with the football.
The bolded areas of those two charts, by the way, signify some changes I’m thinking about making with my garbage time definitions. For a long time, I’ve defined garbage time as anything that takes place when a team is up by more than 28 in the first quarter, up by more than 24 in the second, up by more than 21 in the third, or up by more than 16 in the fourth. That’s the definition that long ago seemed to best coincide with good predictiveness. But that has long felt too narrow. When I get a moment, I’m going to tinker with this definition:
- All first quarter plays count, no matter the score.
- Garbage time doesn’t kick in until a team is up 36 in the second quarter.
- Garbage time doesn’t kick in until a team is up 26 in the third quarter.
- Garbage time doesn’t kick in until a team is up 20 in the fourth quarter.
Teams themselves seem to treat games as winnable (or losable) as long as the game is within those bounds, so I should too. As long as it maintains predictiveness for my measures, anyway.
One other use for this data by the way: As I’ve mentioned before, my goal is to slowly shift more toward expected success instead of straight success rates.
The range of expected efficiency and expected explosiveness based on down and distance are different animals.
(By the way, consider that wording a harbinger. The next step in adjusting my S&P+ ratings, probably after this coming season, will be taking expected success into account instead of simple success rate. But that’s a different post.)
While down, distance, and field position are major drivers of expected success — as in, certain downs-and-distance combos (second-and-1, for instance) are likely to produce high success rates, others (say, third-and-12) are not — game state clearly is, as well.