Some TOs are more equal than other TOs

In our second installment of "Inside the Numbers," we use data from every college football game of the last three seasons to examine the implicit assumption that a turnover is a turnover -- that fumbles and interceptions hurt your team equally. We know what the talking heads think, but what does the data say, and what does it all mean for your Cardinal?

In trying to build a computer ranking for college football teams, I've gathered the stats from every regular season college football game played over the last three seasons. Over 2,000 games is a sample size plenty large enough to draw conclusions from, and some of the lessons learnt are rather counterintuitive.

Some of you stat nerds might have heard hints of this, but next time you see that turnover margin graphic on television, think again. The graphic implies all turnovers are equal, for your team coughing up two fumbles while making one interception is a -1 on the day, same as throwing two picks and forcing one fumble. However, the data show throwing interceptions hurt a team far more than coughing up fumbles, as we're about to show – and discuss what it might mean for Stanford this season.

First, let's look at the data. In all games over the last three seasons (about 2,100)…

Teams with zero fumbles: 1077-795 (.575, reference)
Teams with one fumble: 790-771 (.506, 12 percent worse)
Teams with two fumbles: 305-350 (.465, 19 percent worse)
Teams with three fumbles: 83-115 (.419, 27 percent worse)
Teams with four fumbles: 11-31 (.275, 52 percent worse)

(Note: Though only 2,100ish games were played the last three seasons, each game has two teams, so the wins and losses above will sum to 4,200, which was counterintuitive to me at first. Well, actually, not quite 4,200 because of teams with five or more fumbles.)

Anyways, we see that your win odds with four fumbles are pretty stark, but that is a small enough sample that maybe that the .275 mark will bounce back over the next few years. Otherwise, however, even if you fumble away the ball three times, you're only 27 percent less likely to win than you were at kickoff. For Stanford, I think this is great news, because the Card are a run-dominated team, and would therefore be expected to fumble more. So go crazy Toby and fumble if you must – just not in the red zone please! And defense, not that you're ever supposed to try for the strip instead of just wrapping up, but here's that much more reason to just worry about bringing the guy down.

However, interceptions are a different story:
Teams with zero interceptions: 1152-516 (.691, reference)
Teams with one interception: 799-694 (.535, 23 percent worse)
Teams with two interceptions: 246-525 (.319, 54 percent worse)
Teams with three interceptions: 56-228 (.197, 71 percent worse)
Teams with four interceptions: 13-80 (.140, 80 percent worse)

If you want to know who won the game without actually checking the score, just looking at how many interceptions your team threw (and forced, for if your opponent threw three picks, it's winning less than 20 percent of the time on average). The dropoff is steep, as with each additional interception, your winning percentage just dropped about 200 basis points (.200, if I'm watching my CNBC correctly).

For Stanford, the implications for next year aren't so clear-cut. I think this table suggests what an impact poor quarterbacking did have on last year's squad, and with a presumed upgrade at the position, maybe there's reason to think Stanford can pick up a few more wins. This year's receivers also figure to be better at achieving separation. Then again, the presumed upgrade at quarterback is a freshman, and the biggest improvement quarterbacks tend to make after their freshmen year is, you guessed it, cutting down on their interceptions. So the interception total could be higher or lower next year, but it looks like whichever way it moves will be a major key to Stanford's season. And hey, at least we look golden come Andrew Luck's sophomore season.

Similarly, despite returning nearly all its key guys, last year's defense forced far fewer picks than the Scott Shafer unit of two years ago, which appeared to go 60-yard bomb/interception/60-yard bomb/interception on repeat for sixty minutes. So maybe last year was just an aberration and this year's squad will snag more picks, and Stanford's probability of winning will skyrocket accordingly. Then again, maybe scheme is the culprit, which seems likely given that the players from two seasons ago and last year were also all the same. So if the Card do stay in a more reactive, less attacking scheme, maybe they'll continue to struggle to force interceptions. While the correspondingly fewer 60-yard bombs may make such a strategy worthwhile, these numbers do suggest that forcing at least a few picks will be a major key on the path to bowldom.

Now the Stanford graduate in you (hey, any Bootie's a Stanford grad to me) is screaming out, "But wait, Daniel!" (Well, maybe not quite like that, as it sounds like we're in a Hugh Grant romantic comedy all over again.) But you could be saying, quite accurately, that teams that run more will fumble more, teams that pass more will throw more picks and teams that run more tend to win more of their games to begin with, all of which are true. Thus, maybe the different winning rates are just a function of worse teams passing more, and thus being more likely to throw interceptions than an equally bad run-dominated team. (In statistical terminology, we're wondering if team quality is a confounding variable.) Luckily, I correlated whether a team won with its fumbles per play and its interceptions per play, the "per play" part controlling for teams choosing to run or pass more. And, fortunately for our theory, interceptions per play explained nine times the variance in winning as did fumbles per play, suggesting that interceptions are more important after all.

Making the trend stronger yet is that interceptions are actually more frequent than fumbles in college football, with the average team in our dataset throwing one interception and losing 0.85 fumbles per game. The intuition here is this: say we covered the ball with baby oil so that teams regularly fumbled all the time, say twenty times per game. With forty fumbles per game, we wouldn't expect any one lousy fumble to matter all that much. So that teams actually force about 15 percent more interceptions than fumbles, yet interceptions have two to three times the effect of fumbles on winning probability only reinforces our argument, which we can express in Tarzan terms, the better for our Thunderdomers to remember it after five halftime Tonga Juices:

Fumble? Meh.
Interception? ARRRRGGHH.

As to why this is, your guess is as good as mine, and I'd love to hear some explanations. I imagine interceptions are easier to return for touchdowns, but that happens pretty infrequently, so I don't think it would explain the magnitude of difference. Teams are more likely to run early in the game and try to wear out the other team, but come crunch time, strategy shifts to okay, we need to march down this field in three minutes and score or it's over, screw making the other guys tired. In the clutch, teams open up their playbooks and pass. So maybe interceptions come at more important times in games than fumbles do, but again, I don't think that's the whole story.

Best I can come up with is this: fumbles happen pretty randomly, it's not like running an iso versus off-tackle versus a draw is going to dramatically change the probability of coughing it up. Picks, on the other hand, are more dependent on play-calling, and if you're losing by two touchdowns, you have to start taking more risks than you would otherwise, and so you're going to throw more picks when you're already likely to lose.

Whatever the case, we'll keep crunching the numbers at Boot HQ in search of a better understanding of this game. Meanwhile, a good mental formula to keep in mind when you're watching a game would be to count every interception twice, so a team with two fumbles lost and one interception forced is actually even on the day, not minus one. Most importantly, if Stanford football wants to be somewhere sunny come Christmas break, both sides of the ball need to forget about the fumbles – and start focusing on avoiding (or grabbing) as many picks as possible.

Are you fully subscribed to The Bootleg? If not, then you are missing out on all the top Cardinal coverage we provide daily on our award-winning website. Sign up today for the biggest and best in Stanford sports coverage with (sign-up)!

The Bootleg Top Stories