Filed under:

Statistical In-Ferentz, Week 12: In Search of Lost Time

Considering that it was Iowa's first road win in a year, that Marcus Coker went for 139 yards, Marvin McNutt went for 151 and two touchdowns, and that James Vandenberg looked as accurate and efficient as he has all year, would it seem ungrateful to say that the Hawks could have won this game by 30 points? Or impolite to acknowledge that they also could have lost the game by 10 without some very timely plays? It was a very strange game, but it's remarkable looking back how much just a few plays swung the course of events. Here's a list of the most decisive plays in the game by their impact on win probability* (click to enlarge):

* The win probability model is the same one used in previous columns and the figures used are derived from the Advanced NFL Stats calculator. The usual caveats apply about the generic NFL WP model not totally applying to the college game.

The most decisive plays that can possible usually involve turnovers near the end zone, because that's where expected points can be taken off the board and unexpected ones put on. Iowa, unfortunately, had two such red zone turnovers against Purdue, and they really could have cost them the game if subsequent events had gone a little differently. Mika'il McCall's fumble at the Purdue 6 yard-line probably cost Iowa touchdown, a dominating 14-point lead, and .14 WP, while James Vandenberg's fumble in the end zone gave Purdue a touchdown, turned a 7-point lead into a tie, and took .19 WP off the board. If it had been a more normal game and Purdue didn't seem equally committed to giving the game away, we would be doing some serious ruing today. Deep, deep rue, I tell you. Thankfully, the Keystone Kops philosophy was on display by both teams, and Purdue gave back almost as much in WP as Iowa did with their own inexplicable, unforced mistakes: a fumbled kickoff return that cost the Boilermakers .12 WP and a bizarre semi-fake punt that was blocked and cost them .07 WP. That still puts Iowa up (or down) in the inexplicable mistake category by .14 WP, suggesting that Iowa's mistakes were a little more costly than Purdue's.*

* And that's without including Marvin McNutt's dropped touchdown pass, which was a mistake, but not quite as big as you might think. If he had scored at that point, Iowa would have been up 14 and had a WP of .90, but after the subsequent missed field goal, their WP dropped to .72. Since the missed kick itself was worth -.08, and the failed conversion on 3rd and 2 that led to the field goal attempt was worth -.06, the drop only cost Iowa -.04 WP.

The real hero of the game for Iowa, at least on defense, was Tanner Miller. Both of Miller's interceptions stopped Purdue drives that could have cut the Iowa lead to single digits. Those two interceptions were by themselves worth .19 WP for Iowa, which must be up there with any other player on the team for this game (I didn't calculate the WP totals for every player, but Marvin McNutt must have accumulated at least +.30 in WP; it's another question whether it makes sense to attribute credit to one player, whether a defensive back or a wide receiver or a quarterback in such a collaborative sport as football).

The other defensive hero of the game, at least in terms of recording a crucial play, was Shaun Prater, for his sack of Robert Marve in the third quarter. Overall, Iowa was mediocre at getting Purdue off the field on third down, allowing them to convert a 3rd and 10, a 3rd and 11 and even a 4th and 17, but Norm Parker chose an excellent spot to bring some outside pressure. Looking at how the play set up, it is actually quite reminiscent of the corner blitz that Minnesota used to force a fumble against Iowa earlier this year, down to the fact that the motion man happened to cross the formation to the side opposite the blitz and the play-call forced the quarterback's eyes away from the blitzer:

The man in motion actually winds up being open in the middle of the field, but Marve doesn't have time to see him, because Prater is on him in a flash. The result is a sack and a fumble, recovered by Purdue at their 35. The play was a combination of a big missed block by the back for Purdue, a nice confluence of offensive and defensive play calls (from Iowa's point of view, at least), and good timing by Prater not to tip his hand too early. This blitz not only ended Purdue's drive, it forced them well outside of field goal range (they were against the wind in this quarter, so it's doubtful they would have kicked it from even the 30).

Here's a graph of WP story for the game. You can see that Iowa was the favorite throughout, but that there were points in the second quarter where the Hawks could have put their foot down and become prohibitive favorites with a good play here or there: Michael Meyer's missed field goal and the fact that Iowa got zero points out of the muffed kickoff return is one such point; McCall's fumble near the goal-line is the other.

Eventually Iowa found its way back to a dominant position in the game, and it started with a crucial 4th down conversion. Iowa had a 4th and 1 from the Purdue 36 with about three minutes left in the first half. A conversion would be worth .11 WP, while a failure would cost only .04 WP. The alternative, a punt to, let's assume, the Purdue 10, would actually be worth zero WP. The break-even probability to go for it in that situation works out to be just 27%, which seems like a very reasonable bet for just one yard. Iowa got the conversion, eventually scored a touchdown, and just as importantly, gave Purdue the ball back with too little time to start a drive at the end of the second quarter. You could argue that Kirk Ferentz missed a couple of other equally good chances to go for it on fourth down: first, on the field goal attempt that Meyer missed, and second on a 4th and 1 from the Purdue 40 later in the second quarter. I'll spare you the gory details, but the break-even percentage for the first works out to be 50%, and for the second, 25%. The bar for the conversion from the 15 is a little high, I suppose, but the other one seemed just as good as the conversion the Hawks eventually tried.

Formations

I've gotten away from charting Iowa's offensive formations recently, but I decided to look at it again this game. Nothing too shocking, but interesting nonetheless:

There are a couple formations where Iowa seems to tip their hand by formation, but it's not all that surprising: the two tight end, two running back set is, of course, a heavy running formation, and the three-wide shotgun set is used for passing. Otherwise the coaches kept things pretty balanced all in all, and even maintained good balance in terms of running and passing on early downs:

• First downs: 15 pass plays/17 run plays
• Second downs: 12 pass plays, 13 run plays
• Third downs: 11 pass plays, 3 run plays
And it's not too shocking that Iowa passed so much on third down: they had 6.5 yards to go on average in those situations.

Bonus strategic questions

Here are a few other general strategic questions that I thought might be worth looking at.

#1) What in Buddha's name was Ken O'Keefe thinking with that play-action pass out of his own end zone?

I actually appreciate the daring nature of the play-call, but there were three problems with a play-action pass in that situation: 1) it took way, way too long to develop, 2) it forced Vandenberg to turn his body away from the pass rush, leaving him blind and vulnerable, and 3) given the situation in the game and the strength of the two teams, a low-risk strategy would have been better. Here's an example of a passing play from the end zone done right:

Notice the use of the shotgun, which buys Brady a few precious microseconds of time, and the option for a very quick read (or throw-away). Iowa's play seemed to take ten times as long as this play. And in terms of the game situation, Iowa was up seven and certainly seemed to be the stronger team. There's a general rule for underdogs that they need to do whatever they can to upset the normal course of events to win the game, i.e., they need to introduce high variance (big risk/big reward) plays whenever they can. This implies a converse rule for favorites, too: they need to do whatever they can to reduce the high variance plays to a minimum. This is all 20/20 hindsight -- if the play had worked, we'd all be calling Ken O'Keefe a genius now -- but in the context of a game where Iowa seemed to have Purdue's offense under wraps and Iowa's own offense was working well, doing the boring thing in this situation might have kept things slightly more in the low-variance, Iowa wins universe than passing.

#2) Why the boring Coker-Coker-Coker-punt offense in the 4th quarter?

The other interesting strategic question, and one that seems very relevant to Iowa's success in close games, concerns their 4th quarter offense. The Iowa coaches actually did a pretty good job on their first drive in the fourth quarter of mixing in run and pass, and managed to take a few minutes off the clock and pin Purdue at their own 9 yard-line. The second drive, however, it was pretty clear that Iowa just wanted to run the ball and burn clock. They ran three straight plays to Coker out of their power formation (1 WR, 2RB, 2TE) and punted, having taken two and a half minutes off the clock. This left Purdue with 4:28 left, the ball at their 49 (thanks to a predictably poor punt into the wind), and 10 points to make up. They got uncomfortably close to succeeding. But who knows? Maybe taking those two and a half minutes off the clock were crucial? Is there a way to think about this question more concretely? If we make a few simplifications, I think there is.

Simplification #1: let's assume that the "run Coker" option is bound to produce exactly what it did here: three straight unproductive run plays and a punt to, say, the Purdue 35 (it was easy to know the punt would be bad, but there was no way to know it would be that bad). This option would leave Purdue with an 8% chance of winning the game in the generic model.

Simplification #2: let's assume the other option is to pass the ball three straight times, and that there are two immediate results of this strategy: 50% of the time the team is forced to punt with only 15 seconds having run off the clock, and 50% of the time, the team gets a first down at the 50 yard-line. As it turns out, punting after only 15 seconds would improve Purdue's chances to 10%, while getting that first down at the 50 would improve Iowa's chances to 97% (or decrease Purdue's to 3%). It's hard to say if this is how a pass-only strategy would break down (I actually think assuming it would produce a first down only half of the time is slightly pessimistic), but it gives you a general sense of the payoffs and costs of "going for the first down" versus "burning the clock." Even a series that burned the absolute minimum amount of time on the clock winds up not improving Purdue's chances all that much in this scenario (2%), while getting the first down worsens their chances by a fair bit more (5%). If you take the expected value of the "all passing" strategy (.5*.10+.5*.03) you can see that it makes Purdue's chances of winning slightly worse (6.5%).

There are some big simplifications here (I'm ignoring the possibility of an interception for one), and maybe the percentages wouldn't hold up in reality, but the basic lesson I take from this is that taking the time off the clock is a strategy for very late in the game, not midway through the fourth quarter. The benefit of actually moving the ball and keeping possession is generally higher than whatever slight gain one accrues by running purposely unproductive plays and taking an extra minute off the clock. And running predictable run plays can actually prove to be counterproductive in terms of burning time, because it guarantees that the offense's drive will last only three plays rather than five or six or more. And while passing with the lead may seem like the kind of high-variance strategy I just recommended that favorites avoid, there's high-variance and then there's high-variance. Simply running one's normal offense in the middle of the field is just not what I would call "dangerous." Sure, there is a chance of interception any time you pass, but it's pretty small, and besides, there's a chance of fumble any time you run. The entire debate is actually reminiscent of the "should you try to win the game before overtime" question that we all argued about after the Iowa State game. On the one hand is a "risky" strategy -- passing the ball down the field in the hopes of scoring/extending the drive -- and on the other a "safe" strategy -- running the ball to get to overtime/burn clock -- with surprisingly low pay-offs. It's understandable that coaches fear the "risky" strategy, because it is the kind of thing that columnists love to criticize after the fact (ask Mike Smith), but in this specific case, where Iowa stood the chance of ending the game with a long drive and instead decided to burn off an extra minute of clock and give the ball back to Purdue, you can't say the "safe" strategy was actually the best strategy. Purdue's chances of winning had they not fumbled the ball out of the end zone at the end of the game were not great -- recovering an expected onside kick happens only about 20% of the time in the pros -- but if they had recovered, time would not have been a concern: they would have had 1:38 left and only 30 yards or so to get into field goal range to tie the game. To tie a game where Iowa was, at one point, a 97% favorite!

#3) Who benefits from receiving the kick first?

One last topic that has come up in comments a few times: does it make sense for Iowa to choose to receive the opening kick so often?  In the generic model, the team that receives the opening kick starts the game with a WP of .52, suggesting that there's a slight edge to receiving to start the game. And, indeed, historically, 52% of teams that receive the opening kick in the NFL go on to win the game. I couldn't find any similar figures about college football, though, so it's hard to say if that result carries over. It's kind of strange that there is any advantage at all -- the benefit of receiving in the first half seems to be balanced pretty symmetrically by kicking in the second half -- and I can only speculate about why one exists. The best guess I can make right now has to do with the fact that there's a surprisingly large bonus in the generic model to scoring first: a team that goes up by 7 in the first quarter jumps up to roughly a 70% favorite. So maybe, just possibly, the fact of taking the lead first has some effect on the play of the other team? Maybe some teams start to press, either early or later (if the lead is maintained), and do atypical things to "get out from under" the lead?

Or maybe it has to do with the number of drives each team winds up getting in a game: if a team receives the opening kick, it is probably slightly more likely to have one extra possession in the first half and one fewer possession in the second half. If the team that receives to start the game uses its first, say, 9 possessions well, the game may already be out of reach by the 10th possession, meaning that teams that kick to start the game may be slightly more likely to have completely meaningless, non-game-deciding last minute possessions. We're only talking about 2% here, so it doesn't have to be a huge effect. Anyone else out there have other theories?