I just split this article into two. I need to do a lot more research for the second portion (I'll likely publish that one Tuesday evening), so I'm gonna take this one long hot-take and turn it into two long articles. You'll still need to read this one to fully understand what I'm saying in that one, but I don't think I want to put it all into one article. Before we talk about the rest of it, we need to have a long discussion about turnover margin. This is the turnover margin article.
Any discussion of turnover margin needs to start by talking about turnover luck. Not actual turnovers (we'll get to those), but turnover luck. I've probably written about this subject more than any person on earth who doesn't have a college football analytics site but... one more article won't hurt. People often recoil at the word "luck", so let's start there.
Forced fumbles aren't luck. The Peanut Punch isn't luck. Your tailback carrying the ball with one hand and getting it punched out 13 times isn't "unlucky." But recovering fumbles above a certain percentage is luck. Yes, you can teach your guys how to fall on fumbles instead of a failed scoop that just gives the ball back to the fumbling team. You can move the needle a little bit by training your team how to spot and recover fumbles. But overall, the percentage of fumble recoveries every season will level out right around 50/50 - 50% of the time it's recovered by the offense, 50% of the time it's recovered by the defense.
Now, that's not to say it's the same on offense and defense. I've written at least three articles about that very thing. When a quarterback fumbles a snap in the offensive backfield, according to this study (which I've linked below), the offense recovers that fumble 75.6% of the time. When a receiver fumbles a ball beyond the line of scrimmage, the defense recovers that fumble 60% of the time (and 17.3% of those fumbles are only retained by the offense because the ball rolled out of bounds). But when you combine every fumble scenario (muffed punts and fumbles that bounce out of bounds and and everything else), the expectation is right around 50/50.
Here's a chart from that article. Of all fumbles, 6.7% go out of bounds, 19.1% are recovered by the fumbler, 24.1% are recovered by someone else on the offense, and 50.1% are recovered by someone on the defense. 49.9% of the time it stays with the offense, 50.1% of the time it goes to the defense.
Again, these are NFL numbers in this study, but I'd expect college and high school to be very similar. Defenders fall on more fumbles, but offenses have the luxury of A) the ball rolling out of bounds and B) the ball falling in front of the quarterback where he can grab it immediately with no defenders in the area, so overall, the numbers come out 50/50. Fumbling more is obviously a problem -- fumble eight times in a season and you can expect to lose four; fumble 30 times in a season and you can expect to lose 15 -- but the second number "should" be half the first number.
And so that's how most NERDstats record fumble luck. If there have been 18 fumbles in your games and you've recovered 9 and your opponents have recovered 9, it's just... a typical season. If -- and I've used this example so much you should know it by heart -- you're 2008 Oklahoma and you've recovered 16 of the 18 fumbles in all of your games, you're not "just coached so well that you can spot the fumbles the moment they pop out." You're extremely fortunate that an oblong ball bounced your direction 16 out of 18 times.
Last year, while leading the country in interceptions (which put us third nationally in turnover margin), we were actually a bit unfortunate with fumbles. There were 34 fumbles in our 13 games last year (both teams combined). We recovered 14 of them our opponents recovered 20. Those numbers should have been 17/17, so yes, we had a little bit of bad fumble luck (and a fair bit of good interception luck).
So far in 2023, there have been nine fumbles in our four games. Our opponents have recovered eight. We've only recovered one (the fumble Dylan Rosiek forced against Kansas, recovered by Clayton Bush). On offense, we've fumbled only three times this season (Altmyer getting the ball swatted out as he was attempting to throw + Hank Beatty's catch and fumble in the FAU game as well as the Josh McCray fumble in the Penn State game), and all three were recovered by our opponents. Our opponents have fumbled twice as many times, but they've fallen on their own fumble five of those six times. Well, four of the six with the Toledo fumble frustratingly rolling out of bounds.
I know that you might want to scream "that's because our coaches don't teach our players to fall on fumbles!", but statistics have shown that teams that are "good at it" never sustain being good at it. Some seasons you recover more, some seasons you recover less, but it always moves back towards the middle. It's simply an oblong ball hitting a flat surface.
Against Wisconsin in 2019, a Jake Hansen forced fumble chose to hit someone in the foot and deflect directly to Isaiah Gay, ending a Wisconsin game-clinching drive and allowing us to start our comeback, eventually winning on a field goal. At Kansas a few weeks ago, on their opening drive, their tailback fumbles and it bounces directly to a teammate. Oblong ball, flat surface, you win some, you lose some.
And we've been losing a lot of them. I think we'll start winning some. It doesn't mean we will - some team is going to finish this season recovering only 19% of all fumbles and it might be us - but over time, the magnet of 50/50 should do some pulling back towards the center. In our last 16 games, we've recovered 15 of 43 fumbles in our games. I'd love to see us balance that out by recovering 28 of the next 43 fumbles.
(Yes, coin flip people. I know that after 10 "tails", the next coin flip is still 50/50 heads/tails. I'm not expecting 28 out of the next 43. I am expecting 22 or 23 out of the next 43, though. Wildly unlikely that 15 of 43 continues.)
The other half of "turnover luck" isn't as clear as 50/50 fumble recoveries. I wrote about this last week. Generally, you intercept around 22% of all "passes defended." That means tipped passes, batted-away passes, and, yes, passes grabbed with two hands by the defensive back and returned the other way.
The general concept: if you're getting hands on the ball consistently, be it something simple like Zachary Tobe knocking the ball out of the receivers hands in the endzone or Johnny Newton getting his fingertips on a pass going over his head - the statistics say you'll intercept 22% of those. A team with 1 PBU the entire game (and no other passes defended) is likely getting surgically picked apart. A team that gets their hands on 14 passes is probably very close to picking some off (or, simply grabbing a floater after a tipped pass). 22% is the standard.
Our defensive numbers so far:
Toledo - 4 passes defended, 1 interception
Kansas - 2 passes defended, 1 interception
Penn St. - 6 passes defended, 0 interceptions
FAU - 12 passes defended, 1 interception
The total: 24 passes defended, 3 interceptions. The expectation at this point using 22%: 5.28 interceptions.
Again, this one isn't as crystal clear as the fumbles. It's more... guidance. The "but Altmyer threw the pass right to the guy" argument is valid. Tracking how close the defenders are to the ball is a good way to find an average, but there are interception-prone quarterbacks who blow right through that stat with wildly errant passes and throws into triple-coverage where it's bound to be at least deflected.
Still, it's a stat that can guide us. And in games so far this year:
Illinois - 24 passes defended, 3 interceptions
Opponents - 22 passes defended, 7 interceptions
We should have 5.28 interceptions. Our opponents should have 4.84. Should be +0.44. Instead, -4.
Let's pause there for a bit and I'll come back to it. I want to switch over to Post-Game Win Expectancy for a second. Specifically, expected scoring margin.
I don't think I need to explain this one. I don't think there are many people reading this who haven't read my last three start-off-the-week NERDstat articles. PGWE simply says "over the years, a team with these stats beats a team with those stats XX% of the time." It looks at explosiveness (explosive plays win games), it looks at efficiency, and yes, it looks at expected turnovers instead of actual turnovers. It's looking for 50% fumble recovery rate and 22% PD-to-INT ratios.
Here's our Post-Game Win Expectancies so far this season:
Toledo - 83.8% (play the game 100 times, Illinois wins 84 times, Toledo wins 16 times)
Kansas - 19%
Penn St. - 21.3%
FAU - 96.9%
"Robert," you say. "You yourself said that if Hank Beatty doesn't make that tackle after the fumble, we probably lose to FAU. How can you then agree with a stat that says Illinois wins the FAU game 97 out of every 100 times?"
Because these are predictive stats. And they're saying that if Beatty hadn't made that tackle, it would have been one of the three times that FAU could have beaten us out of 100 tries with those exact statistics. We had 510 yards of offense while holding them to 341. We were better on third down, better in the redzone, and won time of possession. Statistically, that's a game we should win going away.
Why didn't we? Why was it a 6-point win? Mostly these two stats:
- Three fumbles in the game and FAU recovered all three (two when they were on defense, one when they were on offense).
- We got our hands on 12 passes but only had one interception (in the endzone, on 4th down, when a bat-down would have been more-or-less the same as an interception).
Again, PGWE is looking at what should have happened with turnovers in the game. Fumbling twice is bad, but when you fumble twice, on average, it's only going to hurt you once. And we shouldn't only have one interception if we get at least one hand on 12 passes. On average, teams that get their hands on 12 of the opponent's passes come away with 2.64 interceptions. The formula (again, I'm paraphrasing) looks at those statistics and says "in the history of college football, the team with these stats beats the team with those stats by 17 points." We won by 6.
And that's why I wanted to talk about Turnover Margin today. That's why I'm going to suggest tomorrow that we might be able to win the Big Ten West. Using PGWE, here's the expected margin for our four games so far:
Toledo - win by 16.1 (won by 2)
Kansas - lose by 2.5 (lost by 11)
Penn St. - lose by 4.7 (lost by 17)
FAU - win by 17 (won by 6)
Have we been great? Of course not. Have we killed drive after drive with penalties? 100%. Do we have the same lockdown defense as last year? Absolutely not. Should we be 2-2 at this point. Yes, it says so right there. Using PGWE, we should be 2-2. We should have won the two G5 games by 16 and 17 and only lost to Kansas and Penn State by a field goal, but still, that's 2-2. We're 2-2.
I'm just saying that statistically, this is not a "oh man, we're pretty close to 0-4" season. In fact, it's much closer to "oh man, we're not that far off from 4-0". With even average turnover luck -- not "average turnovers", I'm saying that if we recovered 50% of fumbles and intercepted 22% of passes defensed while also seeing the same averages on the other side of the ball -- then we would have won the Toledo and FAU games handily, lost to Kansas by 3, and lost to Penn State by 5. And you'd be much more encouraged.
Why do we feel the way we feel about this season? Mostly because we're 128th nationally in Turnover Margin. We were 3rd nationally last year and everything felt amazing. We're 128th this year and nothing feels amazing. But how much of that is "deserved"? Let's look at another P5 team right next to us in turnover margin (Virginia) and see how they compare in these stats.
Using interceptions, Virginia is about the same as us. Both teams should be even or +1 in interceptions but Illinois is -4 and Virginia is -5. But fumbles is where we we really see how the whole "deserving" thing shakes out.
- Fumbled 3 times in the four games so far, opponents have recovered all 3. Expectation: -1.5. Actual: -3.0.
- Forced 6 fumbles, only recovered 1. Expectation: +3. Actual: +1.
- With fumbles so far, we should be adding 1.5 to our turnover margin number and instead we're -2.
- Fumbled 6 times in their four games so far, opponents have only recovered 1. Expectation: -3. Actual: -1.
- Forced 2 fumbles, recovered 1. Expectation: +1. Actual: +1.
- With fumbles so far, Virginia should be subtracting 2 from their turnover margin number and instead they're even.
So yeah, Virginia probably deserves their ranking on the turnover margin list. Without some fumble recovery luck, they'd likely be dead last in turnover margin. Illinois? Let's turn it into a number.
Fumble margin if we were hitting the averages on fumble recoveries (as described immediately above): +1.5
Interception expectation from our passes defended vs. opponent's passes defended: +0.44
Expected turnover margin: +1.94
Actual turnover margin: -6
I should pause to note that this is my own number. I don't know the specifics that go into Bill Connelly's formula. I'm just using his principles to find the variable between actual and expected using 50% fumble recoveries and 22% passes defended. He might have additional layers he builds in, I don't know.
But if you're wondering why even Iowa analytics guys are tweeting nice things about Illinois (shoutout to Timmy D. for linking this on Slack)...
Illinois was 0/4 in fumble recoveries and dominated per play stats. People just checking scores are going to undervalue them.— Hawkeye Analytics (@iowaanalytics) September 24, 2023
Maryland was probably a little better than MSU, but +4 in TO (2/2 fumble rec) widened the score. Iowa should beat MSU, but I'm not sure they're horrible. https://t.co/kZq76E6HCI
...this is why. People just checking the scores are going to undervalue Illinois.
Some teams constantly put the ball on the turf. They also see defenders get a hand on pass after pass meaning that interceptions are imminent. And on defense, they can't force a fumble and they can't even get close enough to the ball to get a hand on a pass. Those teams deserve their negative turnover margin.
We don't. Statistically, we're -6 when we should be +2. And that's just to get back to "even." With a little luck (just a tiny bit of turnover luck) we'd be +3, +4, or higher. And these games would have played out a lot differently.
So does that mean we have a real chance in the Big Ten West? I think so. How?
Check back tomorrow evening or Wednesday morning for Part II.