clock menu more-arrow no yes

Filed under:

3rd Downs: Strategy and Leverage

New, 6 comments

This week, Parker looks in depth at TCU’s third down attack and attempts a measure of how important a given play is.

TCU v Iowa State Photo by David Purdy/Getty Images

Third downs can serve as a proxy measure for a team’s overall efficiency. TCU’s 2017 offense, ranked 39th in overall S&P+ last year, but ranked 23rd in overall efficiency, and an impressive 18th on 3rd down S&P+. The TCU offense converted 91 of 199 third down attempts last year (45.7%), good enough for tenth in the nation.

This week, I’m asking and answering two questions:

1. How did TCU attack third downs?

2. Are all third downs created equal?

TCU Third Down Strategy by the Numbers:

Total Opportunities: 198

(This number involves filtering out all penalties or timeouts, and so we lose one attempt where the opposing team committed a penalty, resulting in 198 instead of 199).


84 attempts (42.4% of plays) for a total of 345 yards (4.1 average), with a maximum value of 42, and a minimum of -6.

Success Rate: 45 conversion/84 attempts = 53.5%


114 attempts (57.5% of plays) for a total of 1022 yards (8.96 average), with a maximum value of 70, and a minimum of -12.

Success Rate: 50 conversions/114 attempts = 43.9%


3rd and 6+:
Total Success rate: 38.9%
20 rush (21% of plays, 74 yards, 3.7 average, 6c/20a = 30% success rate)
75 pass (78.9% of plays, 833 yards, 11.1 average, 31 c/75a = 41.3%)

3rd and 5 or less:
Total Success Rate: 56.3%
64 rush (62.1% of plays, 271 yards, 4.2 average, 39c/64a = 60.9%)
39 pass (37.8% of plays, 189 yards, 4.8 average, 19c/39a = 48.7%)

1st Half:
Total Success Rate: 51.5%
36 rush(34.9% of plays, 172 yards, 4.7 average, 22c/36a = 61.1%)
67 pass (65.1% of plays, 567 yards, 8.5 average, 31c/76a = 46.3%)

2nd Half:
Total Success Rate: 44.2%
48 rush (50.5% of plays, 173 yards, 3.6 average, 23c/48a = 47.9%)
47 pass (49.5% of plays, 455 yards, 9.68 average, 19c/47a = 40.4%)

Score Differential:

Ahead Big: +14 or more
Total Success Rate: 50%
24 rush(54.5% of plays, 65 yards, 2.7 average, 12c/24r = 50%)
20 pass (45.5% of plays, 260 yards, 13.55 average, 10c/20r = 50%)

Ahead Small: +13 to +4
Total Success Rate: 43.3%
29 rush (43.3% of plays, 159 yards, 5.5 average, 18c/29 = 62.1%)
38 pass (56.7% of plays, 198 yards, 5.2 average, 11c/38a = 28.9%)

Even: +3 to -3
Total Success Rate: 52.8%
13 rush (36% of plays, 31 yards, 2.4 average, 7c/13a = 53.8%)
23 pass (64% of plays, 206 yards, 8.9 average, 12c/23a = 52.2%)

Behind Small: -4 to -13
Total Success Rate: 52.9%
11 rush (32.4% of plays, 26 yards, 2.4 average, 4c/11a = 36.3%)
23 pass (67.6% of plays, 283 yards, 12.3 average, 14c/23a = 60.8%)

Behind Big: -14 or more
Total Success Rate: 41.2%
7 rush (41% of plays, 64 yards, 9.1 average, 4c/7a = 57.14%)
10 pass (59% of plays, 65 yards, 6.5 average, 3c/10a = 30%)


TCU’s offense was particularly potent in terms of third down success this year. That was driven by a strong rushing attack to control a big lead, and a fierce passing attack to chip away at a small deficit. Additionally, TCU tended toward rushing success in the first half on third downs, indicating either a conservatism early on, while outcomes are still broadly determined, or perhaps a favoritism of the pass when the game is on the line later on. TCU passed better on third and short, which falls in line nicely with last week’s post about symmetry and speed. After some mechanical babbling below, I’ll revisit TCU’s 3rd down tendencies in light of high-stakes.


Tom Tango is credited with creating the concept of a leverage index in baseball. The concept is simple yet elegant. From Fangraphs’ invaluable sabermetrics library:

Leverage Index is essentially a measure of how critical a particular situation is. To calculate it, you are measuring the swing of the possible change in win expectancy.
You take the current base-out state, inning, and score and you find the possible changes in Win Expectancy that could occur during this particular plate appearance. Then you multiple those potential changes by the odds of that potential change occurring, add them up, and divide by the average potential swing in WE to get the Leverage Index.

Baseball is lightyears ahead of football analytics in these derived valuations - the best I could find about some kind of win probability was a Massey Chart, and while many people create expected points added metrics, many of them are shaky and there is not the consensus as in baseball. (Neither is there the transparency - the college football world is extremely closed in terms of data and metrics; individuals leading the field prefer to keep their data and methodology proprietary... as opposed to... letting people use it? I don’t know, it seems to work well in baseball. Tangent, whatever.)

So, stumbling down that path, I am here working through a rough concept on “Leverage” in college football, here starting small with just the sample of third downs from TCU’s 2017 season. My simple model awards weight to the following four categories: time left in game, scoring margin, distance needed, and field position.

I can justify those four relatively easily; For time and scoring margin, towards the end of close games, plays have more capacity to swing the win expectancy as opposed to earlier plays, plain and simple. Field position is weighted by a central measure; the biggest swing in expected points generally comes from moving into or out of a scoring opportunity or a “pinned back situation”. Thus, plays occurring in “no-man’s land” (own 35 to opponent 41), are weighted more heavily, as they contribute heavily towards increasing expected points. Finally, and weighted the least, is distance needed - converting third and thirteen is going to be much more valuable than converting third and one, given the relative expectation of success in each situation, all else equal.

With that out the way, I’ll conclude with a brief rundown of the top five highest leverage third downs of the season. Note that the leverage index measure is not necessarily cardinal - i.e., the import of degree of separation is subject to independent valuation.

Top Five Highest Leverage Third Downs of the Season, Analytically:

  1. WVU (32.4 Leverage Index): 8:50 left in the 4th quarter, TCU and WVU are tied. TCU has the ball, facing third and 2 from their own 43 yard line. TCU passes, successful.
  2. WVU (32.2 Leverage Index): 7:05 left in the 4th quarter, TCU and WVU are tied. Same drive as above. TCU faces third and 1 from the WVU 41. TCU runs. Unsuccessful.
  3. Iowa State (23.9 Leverage Index): 1:25 in the second quarter, Iowa State leads 14-0. TCU faces third and 9 from their own 35. TCU passes to the RB. Unsuccessful.
  4. SMU (21.6 Leverage Index): 14:38 in the fourth quarter, TCU leads SMU by six. TCU faces third and 8 from the SMU 47. TCU passes to the RB. Unsuccessful.
  5. Arkansas (20.6 Leverage Index): 11:22 in the fourth quarter, TCU faces third and thirteen from their own 41, leading Ark by 7. TCU passes to the RB. Unsuccessful.

Rounding Out the Top Ten:

  • Iowa State 4th quarter, 3rd and 12 from the TCU 46. [Run, unsuccessful]
  • West Virginia 4th quarter, 3rd and 7 from the WVU 31. [Pass, successful]
  • Arkansas, 2nd quarter, 3rd and 1 from the TCU 42. [QB Run, successful]
  • Stanford 4th quarter, 3rd and 5 from the STAN 15 (15.5 Leverage index) [Pass to RB, unsuccessful]
  • Stanford 4th quarter, 3rd and 14 from the STAN 37. (15.4 Leverage Index) [Run, Unsuccessful]

Some brief observations

TCU really liked the RB pass on a meaningful third and short, and it was unequivocally unsuccessful. This could mean a couple of things that the raw PBP data won’t capture: could be great coverage on a telegraphed play, could be a check down forced by pressure, could be a broken play. It could also just mean that TCU’s offense trusts its backs. Just casually, this kind of affirms that TCU does better passing in obvious run situations.

So, I’ll end on a note of further inquiry: Is this a decent list? How did the leverage index line up with the reality of the important third down moments of the season? Are we missing any key plays? What factors might have contributed to leverage that we have omitted? As always, your comment feedback is welcome.