Some of you might like the footy. Others of you might really hate the footy because America (but please spend some time watching our golden boy, Christian Pulisic, tear up the English Premier League). Regardless of your opinions on soccer, there is a concept from it that I think can be useful in describing the 2019 Green Bay Packers and, in turn, helps us forecast a bit for the 2020 Packers.
Before I get into the American football, I have to explain some stats and concepts from soccer. First, we’re going to start with expected-goals, or xG. xG is a pretty simple statistic: it measures the likelihood that any given shot turns into a goal. The closer and more centered you are on the goal, generally the higher the xG is. A tap-in from five yards has a better probability than a shot from 25 yards. If a shot has a 20 percent chance of being a goal the xG for that shot is 0.2. If it has a 5 percent chance, the xG for that shot is 0.05.
In each game, you can add up the total xG for each team to see which team “created the better shots/took the better shots.” Soccer is a low-scoring-event game. Absolute barnburners like the Liverpool/Arsenal Carabao Cup game this past fall still only had 10 goals, which, compared to something like basketball, is hilariously low. However, American Football is also a relatively low-scoring-event game. With this, they both have a pretty high degree of variance within individual games.
The big-picture view on expectation
This leads me to the bigger picture on xG. Using xG for an individual game is only mildly useful. There’s still too much variance. But if we zoom out over the course of a half-season or full-season, xG and its sister stat expected-goals allowed (xGA) can tell us more about a team than their actual goals scored/goals allowed.
Generally, the better teams are going to have the largest expected-goals differential or xGD. This works most of the time. For example, in the 2017-18 English Premier League season, Manchester City ran up an astounding 100 points, racking up 32 wins to just two losses and four draws. They also racked up a hilarious xGD of +58.3. There are times when it doesn’t work as well, such as the 19-20 Premier League season. Liverpool have already clinched the title with 93 points, 21 ahead of Manchester City. City didn’t get worse though — they have a +53.7 xGD, and Liverpool’s is only +28.5. What is going on?
This isn’t meant to be a deep dive into soccer stats, but a concept from soccer may be applicable to football here. Liverpool have only faced eight deficits all year long. They spend the vast majority of their time either tied or leading the game. The next closest is Manchester City at thirteen. What teams do when they’re up matters, too. Liverpool does a phenomenal job of controlling games. When they’re up, they don’t drop points. Part of the reason their xGD might not be as outstanding as Manchester City’s has some to do with the fact that City’s team is a steamroller of attacking beauty, but also a lot to do with the fact that Liverpool don’t spend as much time needing to put the pedal to the floor.
Liverpool isn’t even the prime example of this. Jose Mourinho’s Chelsea teams would often “park the bus” after getting a lead, and spend the remaining time defending deep in their own half, essentially spending entire halves playing “prevent defense.”
But what about the pigskin?
That brings me finally back to American football. And I want to talk about the Green Bay Packers and their own “xG.” It’s no secret that the Packers’ point differential in 2019 was pretty mediocre for a team with a 13-3 record. I’m not here to argue that this was a 13-3 true talent team. It wasn’t. They had a pretty easy schedule of quarterbacks that helped quite a bit and they were good in one-score games, both things that are prime for regression in 2020. Their point differential ranked 5th in the NFC last year at +63. The gap between Green Bay and the #4 team, Minnesota, was 41 points. The Packers’ Pythagorean record, which works off of point differential was 9.7-6.3. The Packers were a hilarious 9-1 in one-score games.
To make a long story short: that is A LOT of data showing Green Bay’s record last year was a fluke. And that doesn’t appear to even be wrong, but HOW right it is is actually pretty important.
The big counter to all of this data is a look at NFL games by win probability, rather than points, from FiveThirtyEight. By using win probability instead of point differential, they projected Green Bay for a 12-4 record. That’s still worse than their actual record, but not three-plus wins worse.
Controlling the game
I hypothesize that this difference between the two has a lot to do with game script. The Packers often did a great job at getting out to leads early, ranking fifth in first-quarter points per game and sixth in opponents’ first-quarter points allowed per game. From there it was often a “hold on for dear life” situation. To illustrate this, here are ESPN’s win probability charts from each of GB’s one-score regular season games in 2019. (Note: one score is defined as eight points or less):
While GB had an insane number of one-score games, according to win probability the games weren’t quite as close as the scorelines show. Green Bay was fortunate to pull both Lions games out of a hat, but there are also games against Carolina and Washington where the final score feels closer than the games ever were by win probability.
So why did I spend several paragraphs talking about soccer and providing you a ton of pictures to look at? To talk about game state. When a team is winning, they are more likely to play conservatively than when they are losing. Passing rates decrease with a lead. Passing is more efficient than running. If you’re doing the efficient thing less, the game likely gets closer. As you can see below, Green Bay was not afraid to throw the ball in situations where the game was still in reach.
When GB did get a lead, though, they ran the ball quite a bit more. Green Bay’s pass rate in two score games (9+ points) on 1st and 2nd down fell to 14th. What Green Bay did wasn’t the football equivalent of “Parking the Bus,” but they spent a good amount of time in those scenarios. Green Bay spent the 6th-most first and second down offensive plays in the NFL with a lead. Green Bay wasn’t all that good at playing offense with a lead though, particularly in the passing game. Green Bay’s 47% success rate when passing with a two-score lead ranked well below the 52% league average. Their yards-per-attempt number also looked poor at 6.6 compared to the league average of 7.6 in those situations. Rodgers averaged -.04 EPA per play in those situations as well. The Packers offense sucked once it got ahead. As a comparison, Rodgers’ EPA per play on pass attempts in one-score situations was +.11, and when behind by two-plus scores was +.12.
I don’t know if this is Rodgers intentionally being over-conservative with a lead, Matt LaFleur’s playcalling getting too conservative with a lead, a reflection of not being that good, a myriad of other possibilities, or a combination of all of that, but the Packers need to be able to step on the opponent’s throat when they have them down. Parking the bus may have worked in 2019, but it’s asking a lot out of the Football Gods year-over-year.
The Packers certainly weren’t a true-talent 13 win team last year, but they may have been a bit better than their Pythagorean record shows. The predictiveness of using win-probability based “close games” is yet to be seen, but it may indicate while some regression is inevitable in 2020, that the bottom won’t fall out on this team.