Running In Style: Vikings Game-Tying Drive

In a passing league, consistently running the ball is a mistake. Consistently running the ball well is both difficult and rare.  Down 21-14 in the 3rd quarter, the Vikings used a tandem backfield as a crutch, supporting and propelling them over Carolina.  The Vikings, who lead the league in Run EPA, relied heavily on both Toby Gerhart and Adrian Peterson.  The two RBs were featured on 9 of the 13 plays, 6 of which were successful (2 registered a neutral EPA of -0.06).  Here is a look at the evolution of the game-tying drive - which increased the Vikings' chance of winning by 27% - using our Markov model.

Minnesota ran the ball on each of the first 5 plays, slowly decreasing their chance of a punt from 54.4% to 16.7% and increasing both TD and FG odds.  They then turned to Christian Ponder on back-to-back unsuccessful pass plays, rocketing the punt probability back up to 45.6%.  On the biggest play of the drive, Ponder converted on a 3rd-and-10 completion to Visanthe Shiancoe.  Realizing their previous mistake, the Vikings went back to the run and ultimately scored on a 9-yard give to Adrian Peterson.  

The drive expected points tell the same story.  The Vikings methodically increased their chances of scoring until the switch to the passing game on play 6.  Even though the 3rd-down passing conversion was the most significant play on the drive, you have to wonder whether they would have been backed into such a hole if they didn't go away from the run.

While there is certainly an element of game theory when it comes to play-calling, there is something to be said for consistent efficient play.  If it ain't broke, don't fix it.

Keith Goldner is the creator of Drive-By Football, and Chief Analyst at - The leading fantasy sports analytics platform.  Follow him on twitter @drivebyfootball or check out numberFire on Facebook

  • Spread The Love
  • Digg This Post
  • Tweet This Post
  • Stumble This Post
  • Submit This Post To Delicious
  • Submit This Post To Reddit
  • Submit This Post To Mixx

3 Responses to “Running In Style: Vikings Game-Tying Drive”

  1. Will says:

    Keith, why do you use a smoothed curve to plot these results? There is no meaning to play 3.7 or 4.2, and there should be no expectation of continuity between plays. Bars would probably be the best format, but if you insist on lines, at least let them indicate the discrete nature of the underlying process.

    And a nit..."absorption probability?" These are not generic state transitions, they are "drive results" or "drive outcomes" or "likelihood drive ends in..." or something that indicates that these numbers have a meaning.

  2. Keith Goldner says:

    Interesting comment about using a more discrete plot, I'll definitely look at it. I like the smoothed curve because I think it does a better job highlighting the evolution and patterns of the drive.

    As for absorption probabilities, it is based on an absorbing Markov chain, and that is technically where the calculation comes from. It is the probability that the chain (or drive) is absorbed (ends in) that specific drive-ending state.

  3. Tarr says:

    I agree that smoothing the curve isn't meaningful or useful. I disagree that bars would be best, though - I think an unsmoothed line plot would be by far the easiest way to view these results.

    While I get the origin of the "absorbtion" term, it seems like unnecessary technical jargon that surely confuses some of your audience. "Drive outcome probability" is both a precise and accurate term in this case, and it is much more clear to the semi-layman, I think.

    It would be interesting to see some further research on your argument at the end of this piece, that they should have stuck with the run the entire drive. Of course, looking at the results of one specific drive isn't very meaningful, due to variance (and the fact that passing is inherently higher variance than running). But with enough data, the results could be interesting.

    Brian has done some research on optimal run/pass balance in the past. I'd love to see the Markov model applied to this. When play-calling is pushed more heavily towards the more efficient choice, to what extent does this hurt the efficiency of that choice?

Leave a Reply