≡ Menu

Throwing deep against Charles Tillman can be hazardous to your passer rating

Throwing deep against Charles Tillman can be hazardous to your passer rating.

Last off-season, I produced an exhaustive analysis of fumble recovery rates. With 13 years of play-by-play data at my disposal, I thought it would be worthwhile to take a closer look at interceptions. You can skip to the results section if you like, but let me start with a few disclaimers.

Interceptions are tricky to analyze. Interception rates are very inconsistent from year to year, so much so that completion percentage alone may be a better predictor of future interception rate than actual interception rate. But putting aside randomness, there are two other big factors that determine interception rates: the score in the game and the length of the throw.

Just about every quarterback will throw more interceptions when his team is trailing in the fourth quarter. Situation plays a huge role in football, and that’s true when it comes to interception rates, too. Similarly, quarterbacks are much more likely to be intercepted on deep passes than short ones. One thing I wanted to look at was how much league-wide interception rates varied over a wide range of circumstances.

Unfortunately, there still is a bit of bias in the data. The best quarterbacks are most likely to be winning and the worst quarterbacks are most likely to be losing. That means to the extent that trailing teams throw more interceptions than leading teams, the results are probably slightly overstated. Still, I think getting a sense of the league baseline over hundreds of thousands of throw — even if not evenly distributed — can be a useful exercise.

The Results

From 2000 to 2012, there were 6,689 interceptions thrown. Here’s the breakdown with respect to the points differential (i.e., points scored minus points allowed) for the offensive team immediately before the interception. For example, teams trailing by more than four touchdowns have thrown 110 interceptions over the last 13 regular seasons. That accounts for 1.6% of all interceptions thrown over that time period:

Category
INT
Perc (%)
Trail > 28 points1101.6
Trail 22-28 pts2573.8
Trail 15-21 pts6199.3
Trail 8-14 pts116117.4
Trail 1-7 pts187428.0
Tie Game107916.1
Lead 1-7 pts97814.6
Lead 8-14 pts4006.0
Lead 15-21 pts1532.3
Lead 22+ points580.9
Total6689100.0

But looking at interceptions alone isn’t enough: interception rate is the more important stat. This next table shows the average interception rate in each of those categories.1 Teams trailing by 22 to 28 points have accounted for 2.9% of all passes attempts in the NFL since 2002; but as we saw from the table above, 3.8% of all interceptions occurred in this situation. That’s because the interception rate is pretty high when teams trailed by 22 to 28 points: passers trailing in that range threw one interception every 25 passes:

Category
Pass
Perc
INT
INT Rate
Trail > 28 points28571.3%1103.9%
Trail 22-28 pts64462.9%2574%
Trail 15-21 pts162747.4%6193.8%
Trail 8-14 pts3217714.7%11613.6%
Trail 1-7 pts5602125.6%18743.3%
Tie Game4131718.9%10792.6%
Lead 1-7 pts3823417.5%9782.6%
Lead 8-14 pts163247.5%4002.5%
Lead 15-21 pts62382.9%1532.5%
Lead 22+ points28621.3%582%
Total218750100%66893.1%

We can also break interceptions down by quarter. Here’s how to read the next table: Teams have thrown 32.8% of all interceptions in the 4th quarter despite “only” 27.6% of all passes coming in that quarter. That’s because passers have a 3.6% interception rate in the fourth quarter.

Quarter
INT
Perc
Pass
Perc
Rate
1119917.9%4636321.2%2.6%
2181427.1%6212128.4%2.9%
3145621.8%4884622.3%3%
4219332.8%6040927.6%3.6%
5270.4%10110.5%2.7%
6689100%218750100%3.1%

I also have data on pass direction — short left, short middle, short right, deep left, deep middle, and deep right — but unfortunately I only have that for half the data set. Still, that gives us over 100,000 passes to analyze. Here, “deep” means anything over 15 yards from the line of scrimmage. And as you can see, “deep middle” is the most dangerous pass a quarterback can attempt. 500 interceptions were thrown with that designation, which accounts for 14.9% of all interceptions that have directions attached to them in my database. That might not sound like a high percentage, but only 5,949 passes — or 5.1% of all passes — were thrown in the deep middle portion of the field. The interception rate on passes 15+ yards down the middle of the field was 8.4%.

Direction
INT
Perc
Pass
Perc
RATE
Short Left57917.3%3311128.5%1.7%
Short Middle57217.1%2230919.2%2.6%
Short Right66219.8%3869533.3%1.7%
Deep Left47914.3%79516.8%6%
Deep Middle50014.9%59495.1%8.4%
Deep Right55316.5%82837.1%6.7%

Now that we have a sense of the base rates, let’s start combining the data. This next table shows the number of pass attempts (in the left columns) and the interception rate (in the right columns) in each quarter broken down by score. There’s a big difference in fourth quarter interception rates between teams trailing by a touchdown or less and teams leading by a touchdown or less. Teams trailing by 1-7 points threw 16,111 passes in the 4th quarter and had an interception rate of 4.18%. Meanwhile, teams leading by 1-7 points in the 4th quarter threw 6,751 passes and had an interception rate of just 2.53%.

Score
1(Att)
2(Att)
3(Att)
4(Att)
1(INT%)
2(INT%)
3(INT%)
4(INT%)
Trail > 28 points01377032017--5.113.134.02
Trail 22-28 pts95961936390505.73.933.76
Trail 15-21 pts1583292465181731.273.863.164.2
Trail 8-14 pts172590677909134742.963.253.593.94
Trail 1-7 pts115431766410660161112.862.983.224.18
Tie Game252498389339133392.542.832.422.82
Lead 1-7 pts685714966964567512.382.622.62.53
Lead 8-14 pts7766015584036911.292.392.792.25
Lead 15-21 pts411650275317944.882.732.362.29
Lead 22+ points53451358115402.031.692.43

Now, what if we combine the score and quarter data with the information we already have on the distance of the throw? That brings us to the most complicated — but also the coolest — table. This groups all throws by either short or long, but otherwise presents the same data (although as noted earlier, the sample size is half as big as in the other set).

The table shows the interception rate for every quarter/score/distance combination. Each cell shows the interception rate and the number of pass attempts in that situation in parentheses. For example, take a look at the intersection between “Trail by 8-14 points” and “4Q/Deep.” In that situation, there have been 1,398 passes thrown and the interception rate was 9.5%. Meanwhile, look at the “Lead 8-14 points” and “4Q/Short” cell. There were 1,506 short throws in the fourth quarter when a team had a 8-14 point lead, and the interception rate was just 1.3%. Obviously, situation matters significantly, and not all fourth quarter throws are created equally.

Score
1Q/Short
1Q/Deep
2Q/Short
2/Deep
3Q/Short
3Q/Deep
4Q/Short
4Q/Deep
Trail > 28 points--% (0)--% (0)1.2% (82)13.3% (30)2.8% (357)2.9% (68)2.6% (936)9% (223)
Trail 22-28 pts0% (6)0% (3)3.4% (261)10.7% (56)2.5% (954)10.8% (203)2.7% (1831)9.6% (374)
Trail 15-21 pts0% (65)8.3% (12)2.3% (1401)9.4% (340)2.7% (1943)5.4% (479)2.5% (3512)10.5% (783)
Trail 8-14 pts1.7% (775)8.6% (187)1.9% (3900)6.6% (899)2.8% (3381)5.9% (798)2.1% (5581)9.5% (1398)
Trail 1-7 pts1.9% (5187)6.5% (1164)1.8% (7442)7.3% (1816)2% (4452)7.2% (1143)2.4% (6404)9% (1974)
Tie Game1.8% (11042)6.6% (2238)1.6% (3591)6.6% (833)1.7% (1390)5.3% (320)1.8% (1353)6.4% (346)
Lead 1-7 pts1.8% (3032)5.8% (692)1.6% (6324)4.5% (1567)1.6% (4261)6.8% (939)1.5% (3039)5% (621)
Lead 8-14 pts1.4% (345)3.5% (85)1.4% (2586)4.8% (650)1.5% (2474)5.9% (562)1.3% (1506)5.3% (321)
Lead 15-21 pts0% (19)25% (4)1.2% (684)5.9% (169)1.6% (1267)4.5% (290)1.9% (782)5% (159)
Lead 22+ points0% (4)0% (1)1% (197)4% (50)1.1% (638)3.6% (137)2.2% (593)5% (101)
Total1.8% (20475)6.5% (4386)1.7% (26468)6.3% (6410)2% (21117)6.3% (4939)2.2% (25537)8.4% (6300)

This is one of the reasons I find the concept of Game Scripts really important to understand. And it’s another reason not to get too obsessed with last year’s interception numbers. The Chiefs, Cardinals, and Jets all finished in the bottom three in both interceptions and interceptions rate. All three teams acquired new quarterbacks in the offseason, so they won’t represent good case studies. But Kansas City, Arizona, and the Jets were also just bad teams last year, and bad teams tend to throw more interceptions because they’re trailing.

As a group, they threw 60 interceptions on 1,576 passes in 2012, giving them a 3.8% interception rate. But one quarter of those interceptions came in the second half when trailing by between 8 and 21 points. Those three teams threw just 350 passes in those situations, giving them a 4.3% interception rate when passing in the second half and down by 8-21 points. To the extent that parity exists in the NFL, the snowball effect when it comes to interceptions won’t persist from year to year. That means if those teams improve, they’ll be less likely to be in high-risk situations, giving us just another reason why interception rates fluctuate wildly every season.

  1. No play-by-play database is 100% perfect, including mine: they do no perfectly match up with my yearly numbers. However, I have no reason to think that there’s a bias in any direction with regards to any errors in the database, and it is well over 99% accurate. []
{ 17 comments }
  • Bob June 3, 2013, 2:39 am Reply
    • Chase Stuart June 3, 2013, 9:24 am

      Thanks Bob!

      Reply
  • Danny Tuccitto June 3, 2013, 3:36 am

    OK, smart guy, here’s the plan:

    1. Figure out the “stickiness” of air yards.
    2. Figure out the “stickiness” of passing identity (or game scripts).
    3. Predict next season’s INT rates (incorporating any concomitant regression to the mean you find, of course).

    Reply
    • Chase Stuart June 3, 2013, 9:26 am

      I think figure out the stickiness of completion percentage, and then we’re good.

      I still can’t figure out Joe Flacco’s low interception rate, though.

      Reply
      • Richie June 3, 2013, 12:35 pm

        Flacco is easy. You know he is going to throw either 10 or 12 interceptions every year.

        Reply
        • Richie June 3, 2013, 12:36 pm

          Flacco’s interception totals are almost as fun as Adam Dunn’s HR totals.

          Reply
        • James January 8, 2014, 7:09 pm

          Whoops!

          Reply
      • Andrew Carroll June 14, 2013, 9:25 am

        I’ve been wanting to do something like this for a while! Great work on this post.

        The Joe Flacco stuff is confusing. His high aDOT, combined with his low completion percentage, should have him throwing way more picks.

        It’s at this point that film study might be necessary. I know FO has Flacco notched for more than a fair share of “dropped interceptions.” We cannot predict defensive incompetence, but it is one way in which Flacco has lucked his way into a low INT rate.

        Also, having big receivers who go up and fight for balls can help. As a 49er fan, I can speak for a guy like Randy Moss who, last year, prevented an interception or two just by purposefully causing a foul or otherwise fighting for a ball. Not all quarterbacks have a guy like that.

        If aDOT is well-correlated year-to-year, and if we can reasonably predict a quarterback’s comp%, then we should be able to arrive at an “expected” INT rate for every quarterback in the league.

        It’s not perfect, but, for example, Alex Smith will have an INT rate below the league average this year. Based on his aDOT and his high comp% we have to feel comfortable saying that. Plus, he purposefully takes sacks instead of making risky throws.

        INT rates are random, but there has got to be a way to break them down in every way possible to find trends and figure out something predictable in them. What you’ve done here is a big step. This info is awesome. Thanks.

        Reply
  • Kibbles June 3, 2013, 10:35 am

    Not that hard to figure out, Chase- random is random.

    If you wanted a better read on how situation affects int%, you could always compare QBs. “Player A throws B int% when leading and C int% when trailing, so his rate goes up D% where D = C-B. if you average all values D for every QB in the sample, you get E%, which is the average impact of situation on int%. Knowing E and knowing which situations player F faced, I calculate that he threw G more/fewer interceptions than would have been anticipated from a league-average QB.”

    I wonder if G wouldn’t have a much higher correlation from season to season than plain old int%.

    Reply
    • Chase Stuart June 3, 2013, 10:43 am

      Shockingly, I followed this. I suspect there’s nothing you can do to get a high correlation percentage with future INT rate, because of how random interceptions are. But I like and agree with your method.

      Reply
  • Mike Wolfe June 3, 2013, 2:12 pm

    Did you exclude Hail Mary’s at the end of the first half? I would have expected to see an uptick in the 2Q Deep column across all point differentials as a result of that phenomenon. Nice work, by the way.

    Reply
    • Chase Stuart June 3, 2013, 6:14 pm

      I did not, Mike, but that’s probably a good idea.

      Reply
  • ron Taylor June 13, 2013, 2:30 pm

    Chase– Where did you get the 13 years of play-by-play data??
    Ron

    Reply

Leave a Comment