Which teams got it right on fourth-down and 2-point conversion calls in Week 13 of the 2021 NFL season -- and which teams got it wrong? The Next Gen Stats analytics team uses the Next Gen Stats Decision Guide powered by AWS to break down the numbers behind the decisions that shaped the game.
John Harbaugh's aggressive 2-point try falls flat
FOURTH QUARTER: With 12 seconds remaining, the Ravens (trailing 20-19) elect to go for 2 and the win instead of opting for an extra point and a tie. Lamar Jackson's pass to Mark Andrews falls incomplete.
The NGS Decision Guide recommended the Ravens kick the extra point and play for overtime (with a difference of 7.6 percentage points in win probability value between the two decisions). Harbaugh and his team of analytics experts apparently had other ideas.
Here's how our numbers break down:
When accounting for the strength of both teams in 2-point conversion tries and similar situations, the Ravens had a 48 percent chance of converting if they went for 2, while Justin Tucker had a 95 percent chance of making an extra point in Heinz Field conditions. (Note: the NGS field-goal probability model does not take into account kicker strength, so the true extra-point estimate would presumably have been even higher than 95 percent for the immortal Tucker.) If the game had gone into overtime, the Ravens would have had a 55 percent chance to win, according to the NGS win-probability model.
Analytically speaking, in order for Harbaugh to justify his decision to go for 2, it would have been driven by a priori knowledge of the probability of success of their 2-point play design. The Ravens certainly found a way to get their best pass-catcher open on the play -- Andrews was open by 4.4 yards when the pass originally hit his hands. If not for T.J. Watt's disruptive pressure, Jackson and Andrews would have connected on a high-probability play-fake in a "do-or-die" scenario.
In Harbaugh's postgame press conference, the Ravens head coach noted his decision to go for 2 was influenced by their lack of healthy cornerbacks (Marlon Humphrey left the game with what is likely a season-ending shoulder injury). Given that information, it is not unreasonable to think the Ravens had less than a 50 percent true chance of winning in overtime. Paired with a high-probability play design (Andrews was open), playing for the win on the road in Pittsburgh was not as unreasonable as our pre-play call numbers suggest.
In comparison to what Steelers coach Mike Tomlin termed the Ravens' aggressive strategy, Pittsburgh was more conservative with fourth-down decisions. Tomlin faced four situations in which he had 4 yards or fewer to go (1 yard, 2 yards, 4 yards, 4 yards) and the numbers narrowly recommended going for it (by between 1 and 2 percentage points in win probability). The Steelers punted all four times.
Nick Sirianni matches numbers, keeps offense on the field
The Eagles and the analytics were in sync when it came to optimal fourth-down decision-making. Out of five situations where the NGS Decision Guide made a recommendation by 1 percentage point or more in win-probability value, Sirianni and the Eagles got all five decisions right, including on two go for it scenarios during the same late third-quarter drive.
THIRD QUARTER: With 2:52 left and the Eagles (leading 27-18) facing a fourth-and-4 on the Jets' 48-yard line, Philadelphia draws Jets linebacker C.J. Mosley offsides, converting for a first down as a result of the penalty.
This was only the third time this season that an offense successfully drew a defense offside to earn a new set of downs. Conversely, there have been 23 delay-of-game penalties called when the offense was still on the field on fourth down. That equates to roughly a 1:10 ratio of success for offenses in this scenario through the first 13 weeks of the season.
We had this play as a go for it by 1.8 percentage points in win-probability value, just under our "strong" go for it threshold. It did appear the Eagles were going to go for it here, whether or not they could get a defensive player to jump.
THIRD QUARTER: With 1:24 left and the Eagles (leading 27-18) facing a fourth-and-1 on the Jets' 34-yard line, Gardner Minshew converts with a 2-yard quarterback sneak.
Less than 1:30 in game-clock time after their first go-for-it decision, the Eagles correctly keep the offense on the field again, this time with fewer yards to go, and with the team being in what is considered field-goal range. The NGS Decision Guide had this as a strong go for it by 2.8 percentage points. Sirianni was wise to keep the offense on the field in a high-probability situation (70 percent conversion rate), taking into account the score differential (9 points) and the impact of a field goal (increasing the lead to 12 points) versus preserving a chance to score a touchdown (increasing it to 15 or 16 points).
Robert Saleh correctly keeps the offense on the field at the goal line
When you consider the score differential (trailing by 2 points), it could appear on the surface that kicking a near-automatic field to take the lead would have been the wise choice. However, at this point in the game, early in the second quarter, this is not true. With so many possessions left in the game, the first-year Jets head coach made the optimal call, as indicated by the NGS Decision Guide, which pegged this as a go for it situation by 3 percentage points.
Kliff Kingsbury continues to be rewarded when going for it
The Cardinals not only own the best record in the NFL this season, but they also have the third-highest fourth-down conversion rate of any offense league-wide (67%). Their fourth-down success is even more notable when starting quarterback Kyler Murray has been healthy; Arizona has converted on 8 of 9 fourth-down tries (89%) with the star quarterback at the helm. Murray and his favorite target -- DeAndre Hopkins -- missed the Cardinals' previous three games, and on the duo's return Sunday, Kingsbury made sure to hand Murray the keys on Arizona's first drive.
The NGS Decision Guide found this fourth-and-2 situation to be a toss-up, favoring going for it by 0.4 percentage points. Kingsbury stuck with his fresh QB and dialed up a mesh concept over the middle of the field that successfully had James Conner open at the sticks on a crossing route. However, Murray had other plans, releasing the ball as soon as he hit the back of his drop and connecting with Hopkins on a 20-yard fade route. The pass had a lower chance of being completed (40%) than our estimated pre-snap conversion probability (51%), which does highlight the riskiness of Murray's decision, relative to the situation. But Murray saw a matchup he liked, and his connection with Hopkins can make even 50-50 balls a plus proposition -- it certainly does not tend to favor the defense.
Dan Campbell's aggressive call goes awry (but doesn't ruin Lions win)
Campbell once again displayed an aggressive mindset in short-yardage situations on the fourth-down decision described below, which was one of those right process, (SEVERELY) wrong outcome plays. Our numbers suggest Campbell was right to keep his offense on the field here, though the NGS Decision Guide would have recommended a different play call.
FOURTH QUARTER: With 4:08 remaining and the Lions (leading 23-21) facing a fourth-and-1 from their own 28-yard line, Jared Goff is sacked and fumbles, with the Vikings recovering the ball.
With the winless Lions nursing a tenuous late fourth-quarter lead (NGS estimated they had a 58 percent chance of securing their first victory of the season at this point), Campbell faced a difficult decision. Conventional wisdom might have suggested the Lions should have punted the ball to protect their advantage. However, our numbers would disagree, recommending a strong go for it by 5.7 percentage points of win-probability value. While a fourth-down failure would substantially hurt the Lions' chance to win (carrying a 34 percent win probability), a successful conversion would boost the team's chances of winning to 68 percent. Key to that win-probability boost is the high probability of the short-yardage conversion.
The Lions' 68 percent chance of converting on fourth-and-1, agnostic of play call (run or pass), increases to 72 percent if they run the ball and falls to 59 percent if they pass. To say the result here -- Goff dropped back to pass and faced pressure immediately, resulting in a strip-sack -- was suboptimal would be an understatement. Minnesota scored just six plays later. Still, the Lions mounted a last-second comeback to finally break into the W column.
Vic Fangio goes for it three times on a single drive
If we had named a most improved decision maker in our midseason superlatives article, it would have been Fangio, whose Broncos have an optimal call rate of 86 percent on fourth downs when the NGS Decision Guide finds an advantage of at least 1 percentage point in win-probability value this season, good for second in the NFL. In 2020, Denver ranked dead last in these same situations (55 percent optimal call rate).
The Broncos faced the Chiefs as heavy underdogs on Sunday Night Football, and Fangio kept his offense on the field on three consecutive fourth-down situations on a single drive in the second quarter. This was just the second time a coach has gone for it three times on the same drive in the first half of a game since 2016.
SECOND QUARTER: With 1:12 remaining and the Broncos (trailing 10-3) facing a fourth-and-2 from the Chiefs' 8-yard line, Javonte Williams gets stuffed for a 1-yard loss.
After extending the possession twice with previous successful fourth-down decisions, the Broncos had eaten up over 11 minutes of game clock across 19 plays on a drive that started at their own 3-yard line. Fangio and Co. followed the underdog playbook to a T -- reducing the amount of possessions in the game and being (judiciously) aggressive on fourth down to maximize variance of outcomes. Going for it on fourth-and-2 was predicted to be a likely conversion (57%). However, the Chiefs called a well-timed run blitz, and Williams had just a 17 percent chance of picking up the first down when he received the handoff, according to our expected rushing yards model, which is significantly less than our pre-play model input.