By Kurt Bullard
Not all turnovers are created equal.
Just ask Utah wideout Kaelin Clay, who last year had just torched the Oregon defense for what appeared to be a 78-yard score before he Desean Jackson’d the ball at the one-yard line, letting Joe Walker of the Ducks pick up the ball in the end zone and return it 99 yards for a touchdown, effectively costing the team 14 points in one play.
In addition, as previous HSAC research has shown that interceptions have different costs based on field position and down. For example, throwing a pick-six from a first-and-goal at the one essentially is a loss of 14 expected points, as was demonstrated above with the Clay fumble or the Kurt Warner pick-six at the end of the first half of Super Bowl XLIII. On the other hand, an interception could theoretically increase the expected point value for a team if a quarterback on a 4th-and-20 from his own 20 threw an interception that ended up at the opposing team’s 5, as he would have effectively just completed a really good punt.
With this in mind, I wanted to investigate if there was cluster luck in interceptions in the NFL, akin to how there is cluster luck in offense in baseball, an idea partially inspired by Sam Bradford’s two red zone picks this past week against the Saints. I’m operating under the assumption that turnovers randomly happen across the field and that quarterbacks can’t control when and where they throw INTs, nor the length of the interceptor’s return. Under this assumption, you’d expect the average “cost” of an INT to come from the same distribution of EPA and that, over multiple simulations of seasons, each QB would be equally affected. But due to randomness, the cost of an interception won’t end up being the same for each team in a 16-game sample.
Using NFL Savant data from 2014, I looked at all interceptions that occurred over that season. I calculated the expected points of the team before the interception and then found the expected points after the play for the team that had thrown the interception, from which I could find the change in EP due to the interception. I then aggregated the “costs” for each team over the entirety of the 16-game campaign.
A quick side note: To calculate EP, I used the expected points framework from Austin Tymins’s article evaluating the expected point value of sacks. The framework comes from a smoothed variation of the Advanced Football Analytics expected points model which comes from historical data. The smoothing was done using a simple moving average, which in most cases is very close to a linear trend based on field position, down, and yards remaining for a 1st down.
After looking at all of the interceptions from last year, it turned out that the average “cost” of an interception in terms of expected points was just under 4 points, registering at 3.98. Knowing the value of an average INT, I could then see how unlucky teams were over the course of the season by subtracting the expected points a team actually lost due to INTs from how many points a team “should” have lost to picks.
It turns out that the impact of this randomness isn’t too large between teams. The most a team was helped by favorable INTs was the Lions, who only notched a total of 12 ”luck” points over the season. The Falcons were hurt the most, almost losing 20 points to bad luck on INTs. But on the whole, very few teams were severely impacted by the randomness. As I posited in the beginning, QBs probably don’t have control over where on the field they throw INTs, so it makes sense that a lot of the team’s “luck” points are fairly close to zero.
But on the whole, interceptions do have a huge impact; the difference between the expected points that the Falcons lost due to picks compared to the Chiefs was just under 63.
The worst INT of last season was thrown by next week’s likely starter for the Bills, EJ Manuel, who was picked off on 3rd-and-3 at the Texans’ 12 by none other than J.J. Watt, who snatched the ball of the air in the flat and returned it to the house.
Two INTs actually had “positive” impacts for a team last season. The so-called “best” interception was thrown by Josh McCown on a 4th-and-17 from his own 47 on a deep ball that was picked off at the ATL 9 by Kemal Ishmael, effectively registering as a great punt that added .57 expected points.
So over the course of the season, each team should expect to concede a few bad runbacks on interceptions. The best way to limit the impact of interceptions, then, is not to throw them.
Data used for this post can be found in our GitHub here.