Kendall Argues

After Baylor went for it on 4th and 4 from its own 36 yard line with 8:06 left and down 17-14, a segment of the fanbase has been apoplectic. After Baylor failed to convert, the opposition went wild. “You punt and trust you defense.” “You can’t give Oklahoma State the ball there.” “I don’t care what the analytics say, it’s the wrong call.”

While we can do a long math explanation that breaks down expected point values, the reality is that the situation and context are more important. And those factors show Aranda made the right move going for it.

Fourth down decisions and most of life benefit from Bayesian reasoning. The idea is that you take an original belief and then adjust the belief with new information.

I’m going to use the 2016 election results as an example because we’re all familiar with it. I’m not doing this to argue that you should have voted for Hillary Clinton, Donald Trump or Gary Johnson. I’m just showing how Bayesian reasoning works with a subject we can all understand, before showing how it supports Aranda’s decision. So please don’t tell me that Trump or Clinton are evil in this article. We have plenty of other avenues to argue one or both of them are terrible or the greatest thing ever.

Before the 2016 election, the best expert gave Clinton a 70% chance to beat Trump. The argument for Clinton winning: Barack Obama was fairly popular, which gave a Democrat a good chance to win another term for the party. The economy was doing well. Clinton led in the polls. And Trump had high unfavorable ratings.

Trump still had a 30% chance though. Parties suffer a penalty trying to hold the White House for three terms. Clinton was also very unpopular. The Democrats were bleeding support among white non-college voters too. If the Democrats struggled with those voters in one state, it would be especially tough for a winning Democratic coalition because their victorious coalitions from 1992-2012 depending on winning decent percentages of white non-college voters in Pennsylvania, Ohio, Michigan and Wisconsin. Lose much support with that demographic, and suddenly one small error would lead to a Trump presidency.

So on election night 2016, we started with a prior or belief that Clinton had a 70% chance to win. But once we got results from a few counties in Kentucky, Indiana and Florida, we saw that Clinton was doing way worse with non-college white voters than Obama. So we had to adjust our prior thought for a 70% Clinton win to account for that decline. As we got more states—eventually counties in Michigan, Pennsylvania and elsewhere—we knew that Trump was going to win. His better two-party performance with white non-college voters was enough to propel him to the presidency.

That’s a long journey that hopefully won’t lead to us debating the relative merits of any of these candidates, but it shows how Bayesian reasoning works. We start with a premise informed by a bunch of data and theories, then we adjust it to reality. Although we had a fairly decent sample of prior American presidential elections, we eventually adjusted those theories to actual votes and could update our theory. Before too long those that relied upon, “Clinton won’t decline that dramatically from Obama with non-college whites” had to adjust their prior and see what Clinton’s chances were with that drop. It turns out it was enough for Trump to win.

Similarly, football coaches can do the same with 4th down decisions. Aranda enters a game with 4th down probabilities and when the Bears want to go for it. But far from the idea, “Aranda was a robot for the chart” he considers the time, situation, offensive effectiveness and defensive effectiveness. Those factors weighed in favor of going for it.

Going into the game, Aranda knew Baylor was 8-of-9 on 4th downs. He knew how likely he felt his team could gain four yards on a play. And he had a feel for when his team wanted to go for it.

With 8:06 left, Aranda could start using Bayesian reasoning. Rather than rely upon just his thoughts entering the game, he now knew that Baylor wasn’t moving the ball effectively most of the night. Rather than argue that makes field positioning more important, possessions are much more important with 8:06 left. After failing to convert, Baylor had just two possessions the rest of the way. If Aranda punts, Baylor has to outscore Oklahoma State by three points over just two remaining possessions. And if Oklahoma State takes the ball 40 yards back, the Cowboys have more field to drive longer, which eats up more clock, and might mean Baylor really only has one possession left.

The “punt in that spot” folks would argue that Aranda should have adjusted his prior to notice this was now a field position game. Baylor could punt and trust its defense to give Baylor good field position. But this hides the issue, “What value does good field position have for Baylor?” The reality of going for it on 4th and 4 is simple: eventually Baylor had to move the ball four or five yards to beat Oklahoma State. If the Bears couldn’t do that, they were going to lose. Baylor couldn’t and lost. After taking over from its own six, Baylor didn’t gain a single first down. That’s partially because Baylor lost yards on a 3rd and 3 run. That wasn’t because Baylor was pinned deep. It’s because Baylor couldn’t move the ball. After the game, Aranda said, “It’s one of those things to where we want to be aggressive and be that team that’s hard to stop, it starts with us as coaches.” The Bears were aggressive; it just didn’t lead to enough first downs.

The “trust your defense folks” also ignore that going for it is trusting your defense. Aranda trusted that if the offense failed, the defense wouldn’t surrender anything. And sure enough, Baylor forced a three and out and got the ball back. If your defense can’t stop a team over 35 yards, it’s not a sure bet they can stop anyone over 70 yards. The Bears knew the defense had decent odds to get a stop there—again updating the math on when it’s appropriate to go for it—so going for it wasn’t a guarantee Oklahoma State would score.

The final possessions show what the real issue is: Baylor’s offense was abysmal. The Bears scored on a 55 yard touchdown run from Smith, and then had one long drive that ended in seven points. On the other 11 possessions, Baylor gained 133 yards—barely more than a first down a possession. That’s just not good enough to beat anyone. But it highlights that Baylor needed as many possessions as possible to hope to break something. Not going for it would have given away another opportunity to score.

We could look at a long fourth down chart or expected points added, and going into the game, most of those shows Baylor should have gone for it. But actually looking at the situation, using Bayesian reasoning, shows going for it was even more imperative in this game. I asked Aranda about going for it more on fourth down after the Kansas game, and he mentioned, “When (as a defensive coordinator) the offense stayed out it’d be, ‘okay here we go, these guys.’ It’s cool to be on the other side of that.”

Baylor has some real issues to address right now. The secondary isn’t holding up in some passing situations and isn’t locating the ball well. The offensive line couldn’t protect Gerry Bohanon long enough to take deep shots effectively. The Bears have had zero rushing attack the last six quarters. And the receivers haven’t done the best job fighting for certain balls. These aren’t fatal issues over a full seasons, but they were on Saturday.

The Bears don’t have an issue going for it. The decision actually worked, and the situation called for going for it. Aranda knows they have plenty work on. But he doesn’t need to work on being more conservative on fourth downs. If you prefer conservatism that doesn’t operate anymore, The Weekly Standard archives are still up (I did like The Weekly Standard back when it existed!). Just because Aranda went against convention doesn’t mean he’s wrong. It means conventional wisdom is wrong, and a better analysis—Bayesian reasoning—shows he was right.

Travis Argues

All any of us can do at any point in time is, given the information at hand and the goal in mind, make the best decision possible. There is no benefit of hindsight in probabilistic decision making—it’s either smart to buy the lottery ticket or it isn’t, regardless of whether you win or not.

With 8:40 left in the game and down 3 points, Baylor went for it on 4th and 4 from their own 36 yard line. It didn’t work. But it was the right decision. Here’s why.

The job of the head coach with every single decision he makes is to give his team a better chance to win the football game. When the play-clock is running low and you’re not gonna get the ball snapped in time, why do you call a timeout? Is it just because that’s what you do? No. It’s because you’ve determined that spending one of your 3 timeouts makes it more likely that you’re gonna win the game as opposed to setting your offense back 5 yards. You can apply this reasoning to every single decision a coach makes: Every play an offensive or defensive coordinator makes should make it more likely to win the game; Kicking it deep on a kickoff instead of onside kickoff should make it more likely for you to win. Etc. Etc. It’s turtles (make it more likely that you win) all the way down.

Now, to Aranda’s choice. You can’t analyze literally every factor that goes into the decision, but you can analyze the major ones. Let’s do that one by one, and afterward determine the best choice.

Decision 1: Punt the Ball

Best case scenario: Don’t take that too literally—the best case would be a punt that the OSU player picks up and fumbles in the endzone that Baylor recovers for a TD. I’m talking best case for within 99% of what’ll happen. So, best case, Baylor gets a booming punt and is able to pin OSU pretty far back in their territory. The ball was on Baylor’s 36, Isaac Power is averaging 40 yards per punt, let’s say he rips off a great 50 yard punt or so and OSU starts with the ball at their own 15 yard line. Then Baylor gets a 3 and out, with a couple of runs and a pass and OSU has taken around 90 seconds off the clock. OSU punts back, maybe it’s not a great one, 35 yards or so, and Baylor starts with 1st and 10 from their own 45 yard line with 7:00 left on the clock.

So, given this best case, Baylor is starting with the ball in the essentially the exact same position they would be if they converted on the 4th down prior to punting.

Worst case scenario: Again, don’t take too literally—worst case Baylor fumbles the snap and OSU picks it up and runs for a TD. But a more likely version is something like this: Baylor punts, OSU methodically marches down the field taking a lot of the clock (something they had already down 3 or 4 times this game), and scores a TD. Now you’re looking at Baylor getting the ball back with ~3 minutes left down two scores. The game is more or less over.

Of course, there are a lot of scenarios captured in-between those two. Something like, “Baylor gets off a good punt, OSU gets a few first downs and punts, Baylor gets the ball back at their own 20 with 5 minutes left” might be the most neutral, even probable outcome of a punt. So how desirable is that? All of this stuff—and I mean all of it—has individual probabilities attached, which is why these binary “yes/no” decisions are so difficult. It’s the job of the coach to roughly approximate these probabilities in his head and determine the best choice.

Decision 2: Go For It

Of course, this is what Baylor opted to do. Let’s look through the possible outcomes here.

Worst Case Scenario: You don’t get it (again, caveats aside for rare possibilities of catastrophe). With 8 minutes left, OSU starts with the ball at your 36 yard line up 3 points. At that point they either: A) go 3 and out and are forced to punt, B) get a first down or two and get a FG, or C) score a TD which can’t take up much time because of the short field. So, worst case you’re looking at Baylor getting the ball back down 10 points with 6 or so minutes left. But, more likely given how Baylor’s defense has played, you’re probably getting a 3 & out or giving up a FG.

Best Case Scenario: You convert. To convert, Baylor had to reach their own 40 yard line. The expected points when you have the ball on your own 40 are decently high. Furthermore, Baylor had just scored TDs on its previous two possessions. They had some offensive momentum. Best case, you get that first down and are able to march down the field for the go ahead TD, or even get a few more first downs to get a FG.

The Total Inputs

There is a reason people have attempted to generate models for “predicted points” in football It is very difficult to sum up all of these inputs to determine the correct choice. There are so many variables! I won’t get too deep into those stats here to avoid people’s eyes glazing over, but I hope it is at least intuitive that you can assign “predicted points” given where you have the ball—i.e., you can predict to score more points (over time, multiple possessions) starting with the ball on your own 40 vs your own 10.

Just to throw out a few numbers, when you average out all the results of all CFB teams, teams who have the ball 1st and 10 on their own 40 on average score about 2.5 points. To compare, your own 20 you’re looking at about 1.8 average points. Start on the opposing 20? 4.5 points.

But again, I’m not gonna deep-dive the predicted points stuff here. I think they’re helpful, but for many people it just makes their eyes glaze over and it isn’t persuasive. Instead, think of it like the major inputs I went through above.

Remember, if you punt, the best case scenario is that you get the ball back 1st and 10 right around where you went for it on 4th down.

We have to think about how many possessions are left in the game, too. Prior to the failed 4th down conversion, Oklahoma State had 10 possessions and had gone 3+ minutes on 4 of them. As said in the worst scenario above, if they had a long drive which ended in TD it effectively ends the game because Baylor simply won’t get 2 more possessions without an onside kick. However, the worst case scenario for the failed fourth down conversion is OSU score a TD but it takes far less time because the field is so much shorter. Now being down two scores isn’t nearly as bad because you can naturally get 2 more possessions.

I’ve had a couple different people say to me that it wasn’t smart to go for it because the offense wasn’t doing well up to that point. I think this belies the point. Your offense’s ability should affect your decision insofar as it affects how likely it is that you convert. But this late in the game, where there are only 1 or 2 possessions left, it isn’t nearly as relevant. Baylor was down. To put it quite simply—they couldn’t win the game if they didn’t score. To punt the ball is simply to kick the can down the road, hoping you can hold them and then score on the next possession. If you can’t convert on a crucial 4th and 4, you probably don’t deserve to win the game. In this game, Baylor couldn’t and they didn’t. Hoping for a defensive score or a freak special teams play is just defeatist. You can say that you think Baylor scoring off a turnover or crazy special teams play was more likely than Baylor converting on 4th and 4, I’d just strongly disagree. Baylor’s offense was bad against OSU, but it wasn’t so bad that converting there was highly unlikely.

Let’s not forget in all this that Baylor had just scored on 4th and 2 from it’s own 45 yard line just a few possessions previously. If Baylor had failed to convert there, a real possibility given how Baylor had been playing previously and how OSU stacked the box, fans would be even more upset. But again, to get mad at a decision you have to get mad at the inputs at the time, no benefits of hindsight. I haven’t seen any fans complaining about this decision, despite the fact that Baylor had way more time left in the game (4:30 left in the 3rd Q), was “only” down 2 scores, and 4th and 2 from the 45 really isn’t that different than 4th and 4 from your own 36.

To wrap this up, you can hyper-analyze every single possibility and allocate a % probability for each of them. That’s what all of us do every time we make a decision, even if we aren’t thinking about it that way. Of course, a coach can’t do that in the heat of the moment either. Instead, he has to have a general sense and feel for the game to roughly approximate the probabilities. On that 4th down, Aranda determined that going for it gave Baylor a better chance to win the game. He was right.

In football you don’t play to keep the game close, you play to win. Against Alabama this past weekend, Ole Miss went for it on several 4th downs because they knew they had to score to win. When they failed and gave Bama the ball with better field possession it made the score uglier but that didn’t matter—there are no “close losses” in the conference standings. If Ole Miss had punted on all those 4th downs, they probably don’t lose by as much but they eliminated any possibility of them winning. And that’s all that matters—Wins and Losses. Aranda made the right choice — the choice which gave Baylor the best chance to win.