Sports

What the “Analytics” Debate Around the NFL Is Missing

Math worship is not to blame for the Ravens’ run of gut-wrenching losses.

Harbaugh speaking with a hooded jacket and Ravens hat on, on the sideline
Head coach John Harbaugh of the Baltimore Ravens prior to the game against the Pittsburgh Steelers at Heinz Field on Dec. 5, in Pittsburgh. Joe Sargent/Getty Images

Three times in three weeks, the Baltimore Ravens have scored fourth-quarter touchdowns, opted for two-point conversion tries instead of extra-point kicks, and missed. They lost each week by a combined four points, and two of the losses came down directly to missed attempts that likely would’ve won the games in regulation. This being the NFL fishbowl, Baltimore coach John Harbaugh has become the latest flashpoint in a holy war between those who believe in the purported “analytics” of going for two and the traditionalists who prefer nearly automatic kicks. It has not been a good national conversation, as a mind-numbing CBS studio segment last week exemplified. That was just a spearhead for similar exchanges that have unfolded for years on social media, in living rooms, and at sports bars. One the one side, there are spreadsheeting dorks who don’t understand real football; on the other, cave people     who can’t grasp smart strategy.

Advertisement
Advertisement
Advertisement
Advertisement

There’s a better way to talk about football, but I’m afraid it’s impossible as long as “analytics”—a poisonous, polarizing word that makes everyone retreat to one side or the other—are the lens through which we view coaching decisions. It’s a useless term. Analytics are everything and nothing. They are as simple as teams counting how often a team runs left out of a certain formation and as jargony as any number of formulaic acronyms most people have never heard of. Somewhere along the line, every two-point or fourth-down decision became a matter of “analytics”—even though very few of them seem to be the result of coaches crunching win probability models in real time and making calls based on a model’s percentages.

Advertisement
Advertisement

There are a couple of situations where math should make a coach’s decision obvious, but the Ravens’ decisions the past few weeks haven’t really been about “analytics” at all. They’ve been about Harbaugh evaluating his team in ways that sound more old-school than new, arriving at football decisions rather than mathematical ones, and then watching them not work out. The outcomes have been bad for the Ravens, but they haven’t come down to math nearly as much as they’ve come down to football rationale. If we accept that aggressive coaches are just coaching football rather than attempting a full-on Excel takeover of the sport, we’ll understand them better and unlock a more productive way to evaluate their decisions in critical moments.

Advertisement
Advertisement

On Sunday, the Ravens trailed the NFC-leading Green Bay Packers by 14 points in the fourth quarter. Baltimore scored two touchdowns to nearly pull even, the latter with 42 seconds left. It was an improbable comeback; backup quarterback Tyler Huntley was having the game of his career with Lamar Jackson injured, and the Ravens had well over a dozen absences as a result of other injuries and COVID-19 protocols. They were playing against Aaron Rodgers, and their depleted secondary had struggled to stop him and his receivers for most of the game before some critical late stops. The Ravens were missing five key defensive backs at kickoff and lost another during the game. Harbaugh did not like his team’s chances in overtime against a much healthier opponent, so he tried to win the game on a two-point conversion. It didn’t work.

Advertisement
Advertisement
Advertisement
Advertisement
Advertisement
Advertisement

“I was just trying to get the win right there,” Harbaugh said. “I think our chances of winning right there were better than they were in overtime, maybe, if you calculate it out.” That sounds a little mathy, to be sure, but the critical thing is that Harbaugh was trying to weigh various factors confronting the Ravens in real time. He was not just looking at a model that told him to go for it and following its orders blindly. He realized he was facing Rodgers with a thin roster and a QB of his own who had little experience, and he thought going ahead on one play (and then stopping Rodgers for 42 seconds rather than an entire overtime) was his best bet. “Give Aaron Rodgers as limited a window as possible to beat my dilapidated secondary and backup QB” is much more of a gut-level call than anything some kind of Two-Point Conversion Supercomputer would’ve advised him to pursue.

Advertisement

The week prior, the Ravens scored a touchdown to get within nine points of the Cleveland Browns, with nine minutes left in the fourth quarter. They failed on a two-point attempt here, too, prompting the CBS segment that decried “analytics” in general for taking too much influence over Harbaugh’s decision-making. But if that decision was about analytics, it’s thanks to the broadest possible definition of the word. When a team trails by 15 points in the fourth quarter, as the Ravens did before scoring that touchdown, it’s a virtual guarantee that they’ll need two touchdowns to tie the game and that one of them will require a two-point conversion. There’s no inherent reason that a team is likelier to convert after the second touchdown than the first, so trying the two-pointer after the first score, as Baltimore did in Cleveland, is the obvious move. This isn’t about advanced math—15 points are 15 points—but about a basic concept of time. If you miss early, you can adjust your strategy and try to extend the game to get another possession. If you miss late, you’re screwed. It’s not analytical at all.

Advertisement
Advertisement
Advertisement

“It’s pretty much standard, really a nondecision,” Harbaugh called it. “You do it at that point in time because you’re going to have to win a two-point conversion. So you understand if you get it or don’t get it early where you’re at, going from there, how many possessions you’re going to need, and what you’re going to have to do. If you wait until the last two-point conversion and you don’t get it, the game is over, you’ve lost.”

Again, the play didn’t work. The coach and players will get criticism for that in professional sports. It just doesn’t have a lot to do with analytics. And, perversely, the strategy worked exactly as it was supposed to, even in a 24–22 loss. Having missed that early two-pointer, the Ravens knew later in the quarter that they needed to hurry up, score, and recover an onside kick to have a chance. They did score and did recover the onside kick, and only a late stop by the Cleveland defense prevented a near-miraculous Baltimore win. (Update, Dec. 20, 2021, 2:30 p.m.: In a weird shift, Harbaugh didn’t follow this strategy against the Packers after trailing by 14 rather than 15, but he’s gotten less blowback for that than going for it in general.)

Advertisement
Advertisement
Advertisement

A week before that, the Ravens were in Pittsburgh to play the Steelers. Jackson led a last-minute touchdown drive to pull his team within a point, and Harbaugh called a two-point try for the win. That one made sense, too, for reasons that do not require a mathematics degree to comprehend. The Ravens’ badly injured secondary couldn’t stop Ben Roethlisberger in the second half, after the team’s top cornerback, Marlon Humphrey, joined several of his teammates on the injury report. “We were pretty much out of corners at that point in time,” Harbaugh told the press, and he saw a chance to circumvent a shortage of bodies at a critical position. “It was an opportunity to try to win the game right there.” Again, it didn’t work.

Advertisement
Advertisement
Advertisement

The play call was ideal. Mark Andrews, one of the best tight ends in the NFL, was wide open. Jackson threw him a decent pass, but Steelers edge defender T.J. Watt, one of the most dominant players in football, disrupted him just enough to send the pass slightly wide. Andrews couldn’t corral it. It sucked for the Ravens, and Harbaugh wasn’t beyond criticism. But the idea was sound, and the play was right there. Sometimes, in the NFL, shit simply happens, and it may well have happened to the Ravens in overtime, too.

Advertisement

These losses might say bad things about the Ravens’ leadership. The Steelers are the dictionary definition of mediocrity, and perhaps Baltimore should’ve played well enough to not need a dramatic finish at all. The Browns aren’t special, either. Against the Packers, on the failed two-point conversion, maybe Huntley should’ve been coached up well enough to see an open receiver in the back of the end zone instead of trying to squeeze a missile into the hands of a tightly covered Andrews. Harbaugh makes more than enough money to get flack for it whenever the Ravens lose, especially when his decisions don’t pan out. But to frame those decisions as analytics-driven risks is wrong on two fronts. For one, they were as much about the football context as anything a calculator would tell Harbaugh. For another, playing overtime is its own huge risk. It just unfolds over a longer stretch of plays than a neatly confined two-point conversion attempt.

Advertisement
Advertisement
Advertisement
Advertisement

None of this is a defense of Harbaugh as a savant who sees the whole chessboard while his luckier peers get by on old conventions. Along the same line, Los Angeles Chargers coach Brandon Staley is not a tortured genius because his barrage of fourth-down attempts against the Kansas City Chiefs did not work out for him. But it’s worth remembering that aggressive coaching strategies cut both ways and are not always based on 400-level math. Earlier this season, Harbaugh made a game-altering (and game-winning) fourth-down decision against the Chiefs by simply asking the mega-talented Jackson, “Hey, Lamar! You wanna go for this?” That call was rooted in a desire to leave the game in Jackson’s hands rather than those of the opposing QB, Patrick Mahomes, and that one worked. That Harbaugh’s recent aggression hasn’t worked isn’t a reflection on data-based decision-making as much as on a different truth, which soldiers on all sides of the analytics war can agree on: Football is really hard.

Advertisement