What can we say about Austin FC’s defensive performance in 2021? As for me, I’d probably start with “dumpster fire” and then progress from there. Only two teams allowed more goals in MLS, and only three teams had more expected goals allowed. Were it not for Brad Stuver playing like the second coming of Lev Yashin for half a season, things would have looked even worse. But apart from Stuver’s heroics and Jhohan Romana’s development into a halfway decent player, bright spots along the defensive line were few and far between. But (cue “30 for 30” voiceover) what if I told you that, by one metric that has been gaining quite a lot of traction in soccer analytics circles in recent years, one of Austin’s defenders was one of the five best center backs in MLS in 2021? I know that I kind of gave away the answer with the name of this post, but it’s true: from a certain data-driven perspective, Julio Cascante was Austin FC’s defensive MVP for the entire season. Before you ready the torches and pitchforks, allow me to introduce you to a concept called “goals added,” and how this bizarre reality came to be.

**What Is Goals Added?**

Simply put, goals added (or g+) is a metric for evaluating soccer performance in a more holistic way than simple expected goals. Developed by a group of statisticians at American Soccer Analytics (ASA), g+ rates every action taken on a soccer pitch in terms of the percent change of a goal being scored or conceded. The benefits of g+ have been espoused by just about every soccer analytics outlet with which I am familiar (Messrs. Goodman and Caley of *The Double Pivot*, Ryan O’Hanlon of ESPN and *No Grass in the Clouds*, and Ted Knutson of StatsBomb to name just a few), and I haven’t been able to find any credible detractors to debunk its science. You can find ASA’s description of their methodology here, and an interview that O’Hanlon conducted with one of its creators here. But perhaps it will be easier for me to describe g+ by way of an example and, because I want an excuse to watch this video another few dozen times, let’s take a look at the first Austin FC goal scored at Q2 Stadium.

The clip starts with Hector Jimenez carrying the ball through the middle third near the right touch line, and then attempting a line-breaking pass to Cecilio Dominguez in a more central position about 25 yards from goal. In the entire history of soccer, this exact pass combination has been attempted thousands of times by players in these exact positions and, thanks to the millions of hours of video evidence and petabytes of positional data collected by companies like Opta in the past few years, we know exactly how many times this precise pass combination leads to a goal scored or conceded. The exact numbers are proprietary, but let’s say for the sake of argument that, when Jimenez started his dribble, Austin FC had a 3% chance of scoring and a 1% chance of conceding in the resulting run of play. When Jimenez found Cecilio with a completed pass in a more dangerous position, Austin’s chance of scoring increased to 5% while their chances of conceding decreased to 0.5%. As a result, the pass is worth the change in the likelihood of a goal scored: (0.05 – 0.005) – (0.03 – 0.01) = 0.025 goals added. These goals added are then divvied up between passer and receiver based on the relative difficulty of completing the pass (where trickier passes reward the passer more than the receiver). We could then perform the exact same calculations for every other pass in the sequence (Cecilio to Gallagher, Gallagher to Pochettino, Pochettino back to Gallagher) and derive goals added values for all of them, rewarding each player for each positive action that led to Gallagher’s goal.

The obvious benefit of a measurement like goals added is that it credits build-up play in ways that expected goals or expected assists do not. Jimenez wasn’t credited for an assist on this pass, nor was he credited with an expected assist as his pass didn’t immediately result in a shot. But thanks to the mountains of historical data that statisticians now possess, we can quantify exactly how much this one pass affected the scoring probabilities of both teams, and can reward it appropriately.

But there’s a secondary benefit, which separates g+ from other stats like xG: by measuring the relative change in a goal being scored, it can appropriately reward the defense for decreasing the likelihood of being scored against. Just as a completed pass in the final third may increase the attacking team’s chances of scoring, a successful tackle or interception by a defensive player may decrease the chance of a goal being conceded. Goals added considers six distinct defensive actions (tackles, interceptions, blocks, clearances, recoveries, and contested headers) and bundles them all into what ASA labels “interruptions.” Any interruption which decreases an opponent’s chance of scoring is then credited to the defensive player as a positive goals-added contribution. For instance, if Nick Lima runs down a breakaway attacker and dispossesses him of the ball with a sliding tackle, and this tackle decreases the opponent’s chance of scoring from 20% to 2% while increasing Austin’s chances of scoring from 1% to 3%, then Lima would be credited with (0.20 – 0.01) – (0.01 – 0.03) = 0.21 goals added. In this way, it’s easy to think of g+ as a stat like Wins Above Replacement in baseball, which considers a player’s contributions in both offense and defense to determine his relative value. So now that we have a clearer understanding of what g+ measures, let’s figure out exactly why this model seems to love Julio Cascante so much.

Here is the sortable table of goals added in MLS in 2021. Since five out of the six actions used to calculate g+ are offensive actions, the metric can tilt largely towards offensive players. To account for this, the g+ chart defaults to “goals added above average,” which compares players in similar positional groups to their peers and then normalizes the results. Among the leaders are lots of names you would expect, like league MVP Carles Gil, Sounders offensive juggernauts Joao Paulo and Cristian Roldan, and best XI CB/destroyer of worlds Walker Zimmerman. However, you don’t have to scroll much further down to see Cascante’s name, comfortably ahead of such titans as Ezequiel Barco, Sebastian Blanco, and Tajon Buchanan. This naturally leads to the question: what the hell? Was Julio Cascante honestly better than Hany Mukhtar and his 16 goals and 10 assists in 2021? If I’m being honest, probably not. As much as I’d like to believe that Nashville would jump at the offer of a Mukhtar-for-Cascante trade straight up, I doubt they’d even pick up the phone as soon as they saw Claudio Reyna’s name pop up on the caller ID. But still, Cascante’s numbers tell a far more interesting story than what most Austin FC fans might have thought at first glance.

**Julio Cascante, Very Weird Player**

*Chart courtesy of American Soccer Analysis*

A quick perusal of Cascante’s goals added above average chart confirms a lot of what we’d expect from the eye test: a player who is mediocre as both a passer and a receiver, who adds minimal offensive value on setpieces, who doesn’t dribble and isn’t very adept at tactical fouls. But then you sort the table by “Interrupting” and suddenly you realize that he’s a top 5 CB when it comes to decreasing the likelihood of a goal being scored. By ASA’s calculations, Cascante’s net goals added through interruptions is 1.68 above average, meaning his defensive actions prevented 1.68 more goals than they caused compared to the average CB. While these numbers absolutely do not seem to pass the eye test, it’s at least worth further investigation. Maybe I just blocked from my memory all of those awesome goal-saving tackles and clever interceptions? Let’s see what Cascante’s per 90 stats have to say about his defensive performance.

*Chart courtesy of Football Reference*

Oh. Oh dear. He’s well below average in just about every defensive category per 90 minutes when compared to other center backs in MLS. Football Reference doesn’t provide data on recoveries per 90, but every other stat that goes into calculating interruptions (tackles, interceptions, blocks, clearances, and contested headers) paints the picture of a player who doesn’t actually interrupt the game much at all. So again the question must be asked: what the hell?

**Finding Meaning in Small Data Sets: The Analyst’s Dilemma**

I contend that there are two possible explanations of why Julio Cascante was the g+ darling of Austin FC in 2021: either he made very few mistakes defensively over the course of the season (and thus very few chances to push his g+ into the negative), or that his handful of positive contributions had an outsized effect on decreasing the other team’s chances of scoring. If the former is true, and he really did minimize mistakes over the course of the season, then one would reasonably expect that to be reflected in his underlying numbers. However, I don’t really see this being true with the data available. Of the top 10 CBs rated by interruptions above average, Cascante has the worst percentage of tackles won, the worst percentage of successful pressures, and the most errors that directly led to an opponent’s shot (all numbers per Football Reference). In short, he was not particularly immune to making errors in the defensive third.

So if he wasn’t great at limiting his own mistakes, then his positive contributions must have come at particularly opportune times. Without watching all of his 2,400 minutes played while making note of match state and other factors, it’s difficult to measure his relative value in high-leverage goal-scoring situations. Still, there’s one stat that could unlock the mystery of why the goals added model is enamored with Julio Cascante: he blocked an incredible number of shots.

If you take a gander at Cascante’s per 90 stats posted above, you’ll see that he’s in the 55th percentile among center backs at blocks, averaging 1.71 per 90. Not terrible, but certainly not exemplary. But if you look at the number of Cascante’s blocks that were blocked *shots*, it paints a far rosier picture. Over the course of the season, he blocked 35 shots (1.31 shots per 90). No other center back among the top 10 in interruptions even comes close to that.

In fact, Cascante was third in *all of MLS *in blocked shots, and second in blocked shots per 90 (for players with at least 2,000 minutes played).

Some of this is purely situational: the more shots you concede, the more opportunities you have as a defender to block the shot. Austin allowed 514 shots in 2021, second-most in MLS behind lowly Cincinnati FC. In fact, Cascante’s teammates Jhohan Romana and Matt Besler were also in the top 25 in blocked shots, with 25 and 22 respectively. It cannot be said strongly enough that Austin FC was a putrid defensive team. But without all of these blocked shots, it may have been even worse.

Intuitively, a blocked shot is an incredibly valuable defensive action. A blocked shot usually occurs when the offense’s chance of scoring is at its peak, and plummets the odds of a goal to effectively zero. Block enough shots, and the goals added model will reward you handsomely.

So now we have one data point which suggests that Julio Cascante was a valuable player in 2021. This then raises two questions: is it meaningful, and is it repeatable? To answer the former, I’ll once again defer to the sharp analytical minds of the folks at American Soccer Analysis. Their g+ model was very impressed by Cascante’s performance, enough to rank him 24th in all of MLS in goals added above average. The question then becomes whether his success in 2021 is predictive, and that question is quite a bit harder to answer. Cascante did one thing really, really well, and that one thing was defensively very valuable, but we’re still talking about a mere 35 discrete defensive actions over the course of an entire season. Can we reasonably predict him to have the same defensive impact in 2022 and beyond?

Football Reference doesn’t maintain a blocked shot leaderboard season over season, and doesn’t have any shot-blocking data for MLS prior to 2018, so we’ll have to start with a cohort of strong shot-blockers in 2021 and then look back through their player profiles to determine if shot-blocking is a repeatable skill or mostly the result of statistical variance. There were 34 players with at least 20 blocked shots in 2021. Five of these players were new to MLS this season (or, in the case of Geoff Cameron, returning to MLS after a long absence), so they will be excluded from further analysis. The remaining 29 players have played a combined 90 seasons in MLS since 2018. Below is a chart of their blocked shots per 90 minutes in 2021 compared to their overall blocked shots per 90 since 2018.

Player | 2021 | Career |
---|---|---|

Jonathan Menseh | 1.35 | 1.03 |

Julio Cascante | 1.31 | 0.96 |

Justen Glad | 1.16 | 0.87 |

Chris Mavinga | 1.13 | 0.66 |

Matt Besler | 1.13 | 0.79 |

Rudy Camacho | 1.11 | 0.82 |

Bill Tuiloma | 1.09 | 0.9 |

Andy Rose | 1.09 | 0.99 |

Antonio Carlos | 1.01 | 0.84 |

Dario Zuparic | 1.00 | 0.88 |

Henry Kessler | 1.00 | 0.71 |

Xavier Arreaga | 1.00 | 0.67 |

Oswaldo Alanis | 0.97 | 0.84 |

Miles Robinson | 0.90 | 0.92 |

Ranko Veselinovic | 0.90 | 1.12 |

Robin Jansson | 0.89 | 0.82 |

Bakaye Dibassy | 0.89 | 0.83 |

Andrew Farrell | 0.88 | 0.63 |

Jonathan Bornstein | 0.85 | 0.57 |

Jakob Glesnes | 0.82 | 0.69 |

Danny Wilson | 0.81 | 0.95 |

Mauricio Pineda | 0.79 | 0.69 |

Maxime Chanot | 0.76 | 0.81 |

Ilie Sanchez | 0.72 | 0.45 |

Auston Trusty | 0.70 | 0.75 |

Anton Walkes | 0.68 | 0.72 |

Yeimar Gomez Andrade | 0.68 | 0.61 |

Julian Araujo | 0.65 | 0.49 |

David Romney | 0.64 | 0.69 |

A cursory glance at these numbers suggests that career blocked shots per 90 was a relatively decent predictor of blocked shots in 2021. The correlation coefficient across the data set is r = 0.61, which suggests a positive correlation between career performance and performance in 2021. There’s even evidence to support that shot-blocking is a skill that can be developed over time, as 11 of these 29 players have improved their shot-blocking numbers every season since 2018. But what’s perhaps most encouraging for Austin FC fans is that there’s no evidence that a shot-blocking season like Cascante had would be a statistical fluke. Of the 29 players within the sample, 16 of them have had at least one season with at least one blocked shot per 90 minutes. The lowest subsequent blocked shots per 90 was posted by Danny Wilson, who had 1.02 blocks per 90 in 2018 and a still-quite-impressive 0.74 blocks per 90 in 2019.

So what does this all mean, 2,700 words and a half dozen charts later? It means that blocking shots is good, and players who block a lot of shots have a positive impact on their team’s chances of winning. It means that shot-blocking is an ability that may be developed over time into a repeatable skill, and that one’s ability to block shots doesn’t seem to vanish overnight. And perhaps most encouragingly for Austin FC, it means that one of their most derided players may have had a greater impact on the team’s meager success in 2021 than what one may have ever guessed. So here’s to Julio Cascante, shot-blocking king of McKalla. Long may he reign.

Great Observation and In Depth Analysis; Pennypacker!!