The Probabilities in Harbaugh’s Questionable in-game Decision Making

The super aggressive, analytics forward nature of John Harbaugh and the Ravens 4th down decision making, may have cost them a game, once again. I say may have because with four minutes and change left to play against the Bills, any number of things could have played out in that time frame.

Let’s jump right in. The following win probabilities come from Pro-Football Reference’s Win Probability Calculator, or from Number Fire’s live, real time win probability. So for the “what if” scenarios, only the PFR calculation will be used, which takes into account the point spread of the game, probably as a way to measure who the stronger team may be.

Discuss your thoughts on this topic on our message board.

It’s 3rd and Goal. The Ravens, three-point dogs in this game, have the ball on the Buffalo 4-yard line. It’s the 4th quarter. 4:59 to play. Game tied at 20.

The Ravens win probability by PFR is 70.2%. Number fire, 76.1%. Lamar Jackson scrambles up the middle for about a yard and a half. You can’t put half yards in the calculator, so we’ll say he got one yard.

It now becomes 4th and goal, with three yards needed to cross the goal line. Decision time. Their new win probability is PFR, 65%. Number Fire, 75.6%. Number fire didn’t penalize them hardly at all. But Pro Football Reference took over five percent away for not scoring on 3^rd down. Interesting.

Here is where John Harbaugh is faced with the decision. We know the options and we’ll get to those probabilities shortly. What played out was going for the touchdown, not getting it, and salt in the wound, an interception which brought the ball out for the Bills to start on their own 20-yard line instead of pinned back on their own 3.

It’s 1st and 10, Buffalo on their own 20, 4th quarter, 4:09 to play, tie game. Ravens WP = 46.2% Number fire 47.3%. The Ravens would not be the favorites to win the game anymore, nor did they even get the ball back in this one. Their win probability fell by 18.8%, or 28.3% depending on which publication you go with.

What else could have happened?

Ravens kick the field goal and take a three-point lead, the Ravens WP = 62.4%. Let’s go back. Before the snap on 4th down it was a WP of 65%. Ravens WP would have gone down 2.6% if they make the field goal. Further, PFR considers the point spread. If you flip the script and make the Ravens a three-point favorite, like they are this Sunday night against the Bengals (for some reason) their WP goes up from kicking the FG to 69.8%. I could buy this being a factor when you’re looking a college football where some games feature a very wide gap in talent, talking 30 and 40+ point spreads. But not at the NFL level. If you set the calculator to a pick em’ Vegas line, the Ravens still increase their WP ever so slightly with a FG, up to 66.2%. That’s probably what you should be using in a game like this. Jets versus Chiefs, okay, maybe you take some of the likely 17.5 point spread into account.

Is there a flaw in this system? In what world is taking the lead, any lead, with four minutes to play a bad thing? Is simply giving the ball to the other team with anything less than a touchdown a losing proposition?

I’m going to keep the -3 Vegas line in the Bills favor in the calculations moving forward, even though I think it makes for wild swings among pretty much equal teams.

If the Ravens don’t convert but don’t give Buffalo, the touchback with the interception. Buffalo starts on their own 3-yard line. Ravens WP = 52.3%. In this case the WP dropped 12.7%, but the Ravens still have the higher probability of winning.

The difference in taking the easy points with the Tucker chip shot, and not converting without turnover is 10.1%. But the probability of winning is still in your favor by 2.3%. Is that all John Harbaugh cares about? 2.3%? As in, if you do this 100 times (which you won’t, would still be a small sample size. you basically get to do this exact thing once) you would be still expected to win 2.3 more times even if you screw it up every single time?

If the Ravens score a touchdown, assuming they kick the PAT (with this group of decision makers, who the hell knows), Ravens WP is 83%. It would have gone up 18%.

While the Ravens may use a different win probability calculator than PFR (I should hope they are with Ivy League computer programmers on staff) John Harbaugh basically wanted his win probability to change in his favor by 18%, rather than decline by 2.6% with the field goal, even if it meant the risk was a drop in win probability by 12.7% by not converting at all. But he forgot, or was not informed in his headset, the variable of the interception which made it a decline in win probability of 18.8%.

The risk did not outweigh the reward. I don’t fault Lamar Jackson for throwing the interception. He shouldn’t have been out there in the first place. It should have been Justin Tucker and the field goal unit.

Win probability calculators shouldn’t carry so much weight. It takes nothing else into account except how many teams in that situation are expected to win their game. A computer has no feel for the game. No sense of how the teams perform, other than the Vegas line. WP goes down slightly with the field goal because you are then giving the other team the ball. But wouldn’t you like to give the other team the ball with a lead rather than not? Let the opposition be the ones pressured into playing mistake free football to earn the chance to retake the lead.

Harbaugh, constantly finds himself in these head scratching moments of trying to win a game in one spot when a game isn’t even over. That’s what chaps my hind about him. I really think he tends to believe that those acts of aggression win you the game, in that instant. Or I feel like he is so blinded by the reward that he refuses to see the risk. Like the two point try in the Packers game last year. Ravens could have taken a one-point lead. But Aaron Rodgers had more than enough time to move the Packers into field goal range for a game winning field goal anyway. Ravens failed the attempt, and the Packers kneel out a one-point victory, rather than kicking the extra point, maybe entertaining the possibility of overtime, more football, where you can still win.

What if the interception here is a pick 6 and Buffalo takes a 7-point lead? Ravens WP then, 11.6%. A 53.4% decline from the start of the play. When they talk about game changing moments, that certainly could have been one. These things should be accounted for if you are going to be a slave to the numbers. 

The Bills had time to operate. A lot of it. If the Ravens did get the seven points, when the Bills march all the way down the field to the Ravens 11-yard line with a 1st and 10 coming and 1:50 left, you’re win probability in that spot is only 67.9%. Only 2.9% better, than where you started, way back when you had to make the 4th down decision in the first place! When Devin Singletary takes it down to the 3-yard line on that play. Ravens WP dropped to 59.1%. Worse than where you started.

What played out.

Ravens 65% WP before the snap on the 4th and goal.

Jackson INT – WP = 46.2%

Bills drive the field, Oweh tackles Singletary at the Ravens 3 – WP = 8.3% (If he lets him score, WP = 8.5%. Hardly a difference).

Bass FG – WP = 0.0%

And here was Harbaugh’s response post-game.

“”Right. Well, I felt like it gave us the best chance to win the game because seven [points], the worst that happens is if they go down the field and score – and I think we’ll get them stopped – but if they go down the field and score a touchdown, the worst thing that can happen is you’re in overtime. But you kick a field goal there, now it’s not a three-down game anymore, it’s a four-down game. You’re putting them out there, you’re putting your defense at a disadvantage because they’ve got four downs to convert all the way down the field and a chance to again score seven, and then you lose the game on a touchdown. So, then the worst thing … The other thing you think you’re going to get the ball at the two-yard line, so I’m very confident in the defense’s ability to stop them down there with the ball on the two-yard line, so we have them backed up if we didn’t get it. It didn’t turn out that way, unfortunately, and we lost the game. So, hindsight, you could take the points, but if you look at it analytically, understand why we did it.””

It sounds like someone who was thrown a curveball question (which this wasn’t) and the response is spew out enough words that hopefully the right answer comes out somewhere, and it doesn’t. “…look at it analytically, understand why we did it.” Sounds like, “we’re smarter than you, you wouldn’t understand.” Yeah. I understand that I rather have a lead, any lead, than not, when the other team has the ball toward the end of the game.

In my opinion, win probability machines are fun little tools to play with, but not to be used for in game decision making. It’s fun to look back on in hindsight. To look and say a team that is a 3.5-point favorite that just took a 21-point lead with 26 seconds left in the third quarter has a win probability of 99.9%. This was the Ravens in Week 2 against the Dolphins by the way. Right before everything that could go wrong did and the Dolphins, by the calculator, did something a team would be able to do once in 1,000 tries.

I don’t think it is a tool to be used to drive in game decision making. Harbaugh didn’t want his win probability to drop 2.6% by kicking the field goal. Even though your win probability is going to drop when the other team gains a few yards on their ensuing drive anyway.

I’m done with Harbaugh’s manner of managing a game. It doesn’t work. Especially when he spends half the game looking like a casual observer rather than a coach. Like letting the play clock run out on a Tucker field goal try from 46 yards, that makes it a 51 yard try, in the slop. Tucker, of course nails it anyway cause that’s what he does. But sorry. You can’t be casual observer, rely on your coordinators type of head coach, but also play the I’m smarter than you card when someone drops probabilities in your earpiece that you blindly follow.

Play the scoreboard. Not the calculator. Football is a game of small sample sizes. Unlike the analytics used in baseball with 162 games a year, hundreds of plate appearances per player, thousands of pitches thrown by pitchers where you can draw some conclusions and have the long-term schedule to see things play out. You don’t have that in football.  You have that moment, right there, and that’s it.

A friend texted me after the game and said, “This team might be good if they weren’t so smart.” Truer words have never been spoken.