Austin FC 2023: System of a Down (Year)

Choose your preferred pop culture lede:

1. I’m just sitting in my seat and waiting for some goals! (to change these games)

2. Josh Wolff definitely has a method, but is it a system or anything resembling a process?

During the 2023 season, the Austin FC online realm was full of discussion about Josh Wolff and his “system”, “style” and any other term that essentially stands in for “how this team is working together to try to score goals.” To make sure we are all starting from the same point, I will give my basic understanding of the Wolff/Austin FC system: The goal is to disorganized the opponent with the ball, while maintain high levels of possession and using that possession to create high-value chances at shots on goal. Austin has also attempted to maximize set pieces, frequently utilizing short corners and long throws, but that’s not what we’re discussing right now.

The above-mentioned discourse included, and still includes, discussions of whether the style of play and focus on possession are even possible in a league like MLS, with its roster rules and limits and the other vagaries associated with the top of the American soccer “pyramid.” A related discussion asks questions like “does the Wolff system not work or are the players not good enough?” But that’s for a different time as well. The question at hand: 

Does style of play or "System" matter in MLS, when it comes to chance creation and chance suppression?

As I have said, and will continue to say, soccer is a fluid and complex game with little structure commanding what teams do within the basic bounds of not using your hand/arm (shirt sleeve?) and keeping the ball within the giant rectangle while working with your teammates to put the ball in the goal. That makes it hard for my practical brain that enjoys precise definitions to analyze without a little help from concrete data. And we have access to some data!

The two measures I’ll use to approximate play style are Passes per Sequence and Direct Speed, as provided by Opta/Stats Perform on their The Analyst MLS 2023 Stats webpage. 

Passes per Sequence (PPS) is exactly what it says on the box: the average number of passes in a sequence. As far as I understand, sequence is a stand-in for possession, as in, this is the average number of passes that occurs when a team has possession until it loses possession, including scoring a goal. Direct Speed (DS) is equally, uh, direct: a measure of how quickly a team progresses the ball upfield in meters/per second. These seem to be decent proxies for playing style, which I will confirm with myself by telling you that STL CITY and NY Red Bulls were 1st and 4th in direct speed and 28th and 29th in passes per sequence. Austin was 27th in direct speed (of course) and 11th in passes per sequence. (The slowest team by this measure of Inter Miami, which added the most famous slow walking footballer of all time in the midseason.)

So I took the PPS and DS for every team in MLS and charted them against the non-penalty expected goals (npxG) AND non-penalty expected goals allowed (npxGA) to see what impact the speed. I used the non-penalty numbers because I’m assuming penalties have little to do with play style, but I supposed that’s something else that could be investigated. Then, because once I start I just keep asking questions, I charted the PPS and DS for the expected goals earned and allowed PER SHOT to hopefully determine if a playing style can help create (or allow) BETTER shots instead of just MORE shots.

This idea and methodology was somewhat inspired by the work of Los Nerdes Verdes. Check out their great work at Capital City Soccer.

Before I really get into the charts, here is the link to all of the work, including bigger versions of each chart and the source data: Direct Speed and Passes Per Sequence Impact on npxG and npxG/shot

(PPS and DS via The Analyst and all other numbers via FBRef) 

Editor’s Note: I am my own editor, so this is a blanket apology for anything an editor would fix. 

Writer’s Warning and Plea: What’s coming up is somewhat tedious to parse but I found it pretty interesting.  Also, I am mostly self-taught on stats, so please point out any errors in thinking that might be learning opportunities.  Someone who knows better please help with Correlation vs. Coefficient of determination (R-squared) with regard to these numbers. I think correlation is better to use for analyzing what I’m investigating here.

Direct Speed

DIRECT SPEED vs. npxG - Correlation = 0.3 / R-squared = 0.094

The first chart has the highest correlation; in general, teams that move faster are bit better than teams that move slower. However, the best attacking teams fall a little closer to the median in direct speed. Austin is (very) slow in direct speed and middle of the pack in npxG. It feels right that Inter Miami is last since adding the world’s most famous sports walker, but I assume they’ll move a little faster overall next year.

DIRECT SPEED vs. npxGA - Correlation = -0.15 / R-squared = 0.026

There’s a slight negative correlation with speed of play and non penalty expected goals allowed, which means that teams that play forward faster in attack tend to give up less total xG. I’m sure the effect is significant, except to maybe say that teams that defend and counter quickly (like Minnesota and Portland) will fit nicely in a metric comparison like this. Austin was just out there moving slow and giving up chances (and the most shots in the league).

Aside: The Red Bulls season is a pretty weird one. They were obviously very good at limiting chances (1st in npxgGA) and goals (tied for 5th in GA) but underperformed overall xG by 15 goals while receiving a bang-on neutral season from their goal-keeping. If they had been a little less unlucky (not even actually lucky!), the story of MLS 2023 might have been ENERGY DRINK SOCCER but instead it was just LOOK HOW GOOD STL CITY IS. For the record, Red Bulls were just a bit more pressy at the top of the Passes Per Defensive Action table, with a 9.4 to STL’s 10.2 PPDA.

DIRECT SPEED vs. npxG/shot - Correlation = -0.08 / R-squared = 0.008

This is the smallest correlation in this entire set. The trend line is basically flat. How fast teams attack has basically nothing to do with the quality of their average shot. I’m not even sure any useful info can be pulled from this illustration. There doesn’t seem to be any correlation between overall npxG and npxG per shot, which suggests that good teams are good at getting lots of chances, not just getting good chances. Austin (0.103 npxG/shot) was above average (0.098).

DIRECT SPEED vs. npxGA/shot - Correlation = -.1159 / R-squared = 0.013

There’s a bit more correlation here, with teams that attack faster *maybe* giving up slightly worse shots, but it’s a similar story to previous sets.

Passes Per Sequence 

Passes Per Sequence vs. npxG - Correlation = 0.095 / R-squared = 0.009

Passes Per Sequence vs. npxGA - Correlation = 0.096 / R-squared = 0.009

These metrics came out very similar, so much so that I checked everything a fee the to make sure everything was correct. There seems to be some unusual yin/yang with a few teams where the distance from the trend line is similar but flops all the way to the other side in the opposite metric. Which is to say, good teams are good and bad teams are bad and sometimes it all averages out. Austin was closer to the middle of the pack than I expected, but I guess that’s what happens when you play a high possession system that doesn’t have a lot of great high possession players.

Passes Per Sequence vs. npxG/shot - Correlation = 0.15 / R-squared = 0.024

Another small correlation (relatively large for this sample), but it seems that teams that pass more might be a little more likely to have higher xG, which, again makes sense. All things being equal, you have to better at soccer to pass more than if you pass less. A fun one here: NYCFC had the fewest PPS and LA Galaxy had the most (Puig, I guess) while having identical npxG/shot. In a related story, I might become a soccer nihilist.

Passes Per Sequence vs. npxGA/shot - Correlation = 0.17 / R-squared = 0.029

This is the second highest correlation in the set and in maybe the thing I’m most interested in finding, aside from my broad conclusion (coming soon). There’s a slight correlation between how many passes a team has in its attack sequences to the quality of the shots it gives up. 

I have two thoughts from this info. First, defensive possession is definitely a thing (I’ve watched a lot of Brendan Rogers in my life) but I would’ve assumed that we would see more of the effects from that in the overall npxGA, not the per shot numbers, which makes me a bit skeptical of the usefulness of this revelation. Second, its possible (likely?) that teams who possess the ball well and/or for a decent length of time are better able to keep a shape in and for defensive transition, due to rest defense positioning.

Passes Per Sequence vs. DIRECT SPEED

Included for reference. It also appears on The Analyst’s MLS page but this one is in the same format as the others. A fairly obvious correlation here: teams that play more passes move slower.

Conclusion

Now we’re back to the initial question: Does style of play matter in MLS, when it comes to chance creation and chance suppression?

The answer seems to be a resounding "Probably Not."

(My only caveat is that the Energy Drink Soccer teams were all at least league-average in most of the metrics here, which supports my theory that a rigidly outlined system is an advantage in MLS. WIth the state of the roster rules, there a huge difference in talent across a single roster and a rigid system helps mitigate the gaps in talent from player to player whether within the first XI or on the bench.)

Playing fast vs playing slow or playing a lot of passes vs playing few passes did not, in 2023 at least, have a significant impact on a team’s ability to generate chances overall or specifically good chances, nor did those factors have a significant impact on. Team’s ability to suppress chances overall or specially good chances against.

This conclusion is not revolutionary, but it’s great when the data can back up something that feels intuitive already.

What all this mean for Austin? I think it means that Wolffball can work. The overall talent level needs to be upgraded, as well as the team speed, but the “system” itself is certainly viable in the league, and might be asset on its own IF the right players are involved. I’m certain that if the results don’t improve, Josh Wolff’s system might only be able to dream in total darkness.