Is Danilo finished?

Today, I'm diving into a bit of personal research inspired by recent conversations on X (formerly Twitter) regarding Juventus defender Danilo. Being a Juventus fan, I've noticed a lot of debate over his recent performances, particularly criticism suggesting he's been a liability in defense.

Observers pointed out that Danilo has struggled to keep pace with opponents, contributing to defensive lapses, including conceding two penalties and receiving a red card in just four games. Naturally, this got me wondering if there's a way to quantify whether his presence on the field is actually undermining Juventus's defensive stability.

To investigate, I designed a method to measure the impact of opponents' actions specifically within Danilo’s defensive area. Here’s my approach - with Danilo as an example -:

• Convex Hulls: For each game Danilo played in Serie A, I created convex hulls outlining his area of influence on the field. This is defined as the zone where most of his activity takes place during his minutes on the pitch, and consequently his area of responsibility.

• Opponent Actions: I focused only on opponent actions occurring within Danilo’s convex hull, and only in the minutes he was actually on the pitch.

• Action Value Analysis: By aggregating and normalizing the value of these opponent actions per 1,000 events, we can gauge how much scoring threat or quality has been generated in Danilo's zone. To refine this, I only included actions in open play, excluding set pieces where individual defensive responsibility is less clear.

Before analyzing my results, I wanted to confirm that this approach has genuine relevance to football analytics.

To validate this metric, I analyzed data from non-goalkeeper players in the Premier League across the 2022/23 and 2023/24 seasons. I included only the games in which the players started the match. This means 15183 different entries corresponding to one single player-match-convex hull-opponent’s value, which required 6 hours of modeling - just in case you wondered why I’ve being so strict with number of season and players.

We then aggregate all the games data to find the total atomic VAEP produced across the season in their responsibility area, area of play, convex hull or however I’ll call it going forward. To level things out I decided to normalize the A-VAEP sum for 1000 events from the opponents, but I’ve then proceeded to exclude all the players didn’t face at least 1000 events, as their numbers could be impacted from limited minutes, actions from the opponents.

So we call this final metric VAEP_padj.

Finally, to assess the predictiveness of the VAEP_padj metric, we retained only players who met the 1,000-event threshold in both seasons, so that we can calculate correlation between the values of 216 players across 2 seasons:

Calculating the correlation between the two seasons’ scores yielded a correlation coefficient of 0.62, indicating a substantial level of predictiveness. A correlation of this strength suggests that VAEP_padj has a meaningful ability to identify players who consistently either mitigate or allow high-quality opponent actions within their defensive zones.

After reviewing the top-performing players from each season according to VAEP_padj, the metric’s rankings aligned well with players we’d qualify as strong defensive performers, further validating its relevance and accuracy.

2022/23 season’s results:

And 2023/24’s:

With this process, VAEP_padj has proven effective at identifying defenders who limit opponents' high-value actions season after season in the Premier League, indicating its potential applicability in broader football analysis contexts.

To address the initial question — whether Juventus defenders, particularly Danilo, undermine the defensive phase — we applied the VAEP_padj model to Serie A's 2024/25 season data. By focusing on center-backs with at least 500 events in their defensive areas (given that we're only about a quarter of the way through the season), the findings highlighted some intriguing trends.

Defenders like Acerbi and Hien stood out as two of the best center-backs - which seem to be recognized by observers too so far -, and Lucumí's presence as an underrated performer - considered how is data and his media attention align - also reinforce my thoughts on the model's ability to identify defensive quality. So, the metric seems solid and effective for Serie A.

However, Juventus center-backs — Kalulu, Gatti, and Bremer — ranked in the lower half of the list. More strikingly, Kalulu placed among the bottom 10 players, while Danilo appeared in the top 10 If we don’t consider the event number threshold:

Despite Juventus boasting the best defense statistically in Serie A (first in non-penalty expected goals [npxG] allowed, with just 6.2 npxG against), their defenders’ areas allow a surprising number of high-value opponent actions.

This discrepancy suggests that Juventus's defensive quality might stem more from a tactical approach that limits scoring success than from individual defensive prowess.

Possible Explanations:

• Atomic VAEP’s Limitations: Since the Atomic VAEP model doesn’t account for the success or failure of each action, it means that while Juventus defenders may allow high-quality chances in their zones, opponents often fail to capitalize, likely due to Juventus’s defensive structure and individual defensive ability.

• High-Risk, High-Reward Opponent Strategy: Juventus’s playing style could be encouraging opponents to attempt higher-risk, higher-reward actions that have a lower probability of success. By forcing opponents to rely on such actions and using their money to build the squad with strong individuals, Juventus could be choking opponents’ offenses. This may also explain why, against richer and well built teams (like Inter or Stuttgart), Juventus faced more significant threats, as these teams are better equipped to execute low expected completion actions but with high rewards waiting.

Tracking this metric throughout the season will be critical to confirm these findings as Juventus monopolize the ball in their games, which clearly impacts and impacted opponents actions so far.

I’d expect to see in the future if Juventus’ defender VAEP numbers will change dramatically or if I’ll find a better explanation. Until Milan’s, Manchester City’s and Aston Villa’s games I wait.