Yesterday, a friend sent me these quotes from Philadelphia Union coach Jim Curtin, pulled from an interview he gave to the Inquirer after Philly’s 2-0 win over NYCFC on Saturday:
“Here in Philadelphia, we don’t care about possession. Possession without purpose is meaningless, and we’re OK with teams — I think in probably every game we’ve played so far, they’ve led in the possession numbers, but possession doesn’t win games.”
The article by Union beat writer Jonathan Tannenwald sets Curtin’s Union up as a counterpressing side in the Klopp / Rangnick model: a high press, quick attacks, fairly direct passing. It’s probably a little early, after four matches played, to really test whether that description fits this season’s Union, but hey, might as well try. Here are some takeaways from that match and from the Union’s season so far.
It’s a dynamite start to 2022
Jim Curtin has been around longer than fbref have been publishing MLS data, so I went as far back as I could. Here’s a rolling average of the Philadelphia Union’s expected goals for and against, going back to the 2018 season:
I use a ten-game rolling average to account for the kinds of radical swings that can happen game to game; if you have, let’s say, a hapless D.C. United side at home one week and MLS Cup-winning Columbus Crew away the next, the lines will jump all over the place. Here, you can see that the Union played their best football under Curtin in early 2019 (when Brenden Aaronson came into the team) and then in a 2020-2021 stretch that ends in July (Jamiro Monteiro’s last appearance before his summer leave of absence was 08 July vs. NY Red Bulls, and, neatly, everything trends down after that point).
As you can see, the Union ended 2021 hot and started 2022 on a tear. The attack is clicking as well as it has since we got good in 2019, and the defense is allowing less than one expected goal a game. We’ll see how sustainable it is, but right now, the Union look great.
Possession numbers explain nothing here
I wanted to test Curtin’s specific claim — that possession numbers mean nothing to us. Curtin is going against accepted wisdom here; a 2021 study in Portuguese journal Motricidade found that, in the UEFA Champions League at least, teams with more ball possession won 49.2% of their matches, drew 23% of the time, and lost 26.5% of the time.
However, in this limited case at least, the numbers do seem to bear out what Curtin’s saying. Here, I’ve charted a ten-game rolling average of the Union’s possession percentage against their expected goals difference over the same period:
Overall, these variables have next to no correlation (r = .0675), meaning that the Union’s possession numbers don’t explain their performance over time.*
Funnily enough, the one game this year when Philly “won” the possession battle was the 1-1 draw against Minnesota in week one (53% possession); in the following three games, all wins, Philly had 36% at Montreal, 31% against San Jose, and a surprisingly low 28% at NYCFC. Each of those wins was, as Tannenwald notes, “deserved” according to StatsBomb’s xG models (via fbref) — the Union actually created the least threatening chances in the Minnesota match (1.0 xG from 19 shots) when they had the majority of the ball, and they created much better chances (2.2 xG from 13 shots) in the NYCFC match.
Game state matters
When I say “game state”, I mean the scoreline at a particular moment in the match. Game-state effects are obvious when you’re watching a match live (“they grabbed the lead early and parked the bus to protect it!”) but can kind of disappear when you’re scrolling through match reports (“how the hell did they score if they never had the ball?”). Studies have shown, for example, that Leicester City’s 2015/16 title can be at least partially explained by game state; Leicester got lucky in that they often scored first, forcing teams to open up against them in search of an equalizer, which was really dangerous to do against Jamie Vardy and Riyad Mahrez.
mlssoccer.com has (I don’t know how I never saw this before) possession stats for five-minute intervals as part of their match reports. Here’s the chart for NYCFC vs. Philly:
Having watched the match live on TV (admittedly on my second screen, I had Middlesbrough v. Chelsea on the first), it didn’t seem to me like we had 27 or 28% of the ball, and this match timeline explains why: in the buildup to the first goal (scored in the twelfth minute) and the second goal (scored in the 33rd), Philly genuinely controlled the match. After the second goal, they had almost none of the ball, and they created almost nothing, because, well, why bother? You’re up 2-0 away, none of the shots New York are taking are particularly threatening, you’ve got an established back four and the best goalkeeper in the league. There are, of course, plenty of teams who’d handle the situation differently (Guardiola teams, yeah, but Swansea famously played for “defensive possession” during their last stay in the Premier League — they sort of just kept the ball in their own half all the time so that nothing happened), but, to turn Curtin’s phrase on its head: if you don’t need a goal, then you don’t need the ball.
That’s not to say the Union sat back and did nothing for the whole second half. One of the best measures of defensive engagement we have is passes per defensive action, or the number of passes attempted by the attacking team divided by the number of tackles attempted, interceptions made, & fouls given away (in short, the number of attempts, regardless of their success, to win the ball back) — a low PPDA is active defending, a high PPDA is passive defending. Philly’s PPDA for Saturday’s match was 12.37, which is pretty low — the league average in 2021 was 10.21 by my calculations. The Union’s rolling average has hit twelve plenty of times over the past few years, though it’s generally been trending downward (read: we’re pressing more intensely) since the start of 2019:
Game state and defensive positioning
There really aren’t a ton of second-half highlights in the video posted by the MLS YouTube channel after the match. It’s a 7:51 video, and the second half starts at 5:34. This feels about right; I don’t remember much happening in the second half, and the xG chart shows a few low-value NYCFC chances and nothing else. The fact that the Union had a comfortable lead mostly explains why. Not rocket science, but here’s an example: the ball gets turned over in midfield, and, because he always does, Union midfielder Jose Martinez moves up to put pressure on the ballcarrier:
The awkward Yankee Stadium camera angles briefly make this look like a dangerous situation. The Union’s diamond midfield is spread way out, with Alejandro Bedoya out on the right wing and Leon Flach caught upfield. If fullbacks Kai Wagner and Nathan Harriel were in their usual advanced positions, this’d be trouble. A quick pan of the camera reveals a very compact back four with everything in front of them:
Here’s a shot from the Minnesota game where the Union, chasing a go-ahead goal, basically have seven players up in the attack; that’s Wagner at the top left calling for the ball as he makes a late run into the penalty area.
Wagner is a very useful attacking fullback, hitting 3.71 crosses per 90 in 2021. He’s been in the upper echelon of attacking fullbacks in MLS for a couple years now & was recognized as such with an MLS All-Star game spot in 2021. The Union are at their best when he’s up providing width on the left flank, drawing the opposition fullback out so the left striker, normally Sergio Santos, can run into the channel. If we’d needed a goal against New York City, Kai probably would’ve been up here:
but we didn’t need a goal, so he wasn’t up there, and the move ended with Taty Castellanos trying to rip the ball out of Andre Blake’s hands for some reason. And that’s how you win a game with 27% possession: get a few goals early and then hang out and enjoy the spring day.
* if you’d prefer, the Union’s game-by-game expected goals difference has an r = .2642 correlation with their game-by-game possession stats — this is a very very weak positive correlation but worth mentioning because it’s a different number from the rolling averages one