Did Celtic perform better against Ferencvaros this year? Why the numbers don't tell the whole story

This question may seem silly at first, but the purpose of this exercise is to highlight some of the limitations of data collection, the importance of small margins within single games and how game states can be important in performance analysis.

All data and related metrics are sourced from Wyscout. This introduces the first concept: data collection. Each data vendor collects using people to review game videos and log every single event. This introduces issues such as human error and dispersion in interpretation. The vendors deploy various peer review oversight, but my experience has been that data quality varies.

In addition to data collection, the vendors also deploy different terminology, characterize events somewhat differently, and have their own models for metrics like expected goals (xG). The result is that the data and related metrics vary for each game across vendors. With the two games being compared in this analysis, this is an important concept.

Some of the more common metrics suggest that Celtic had a greater degree of control of the game last year.

Possession was higher by four per cent and just one corner and two crosses were conceded, which compare to seven and 17 this year.

Celtic dominated last year in deep completions, which are non-cross passes that target within 20 metres of the opposition goal. However, successful entries into Celtic’s penalty area were comparable, and just one more shot was conceded this year.

In attack, Celtic were more active last year. They took more shots, were accurate with more crosses and had eight more deep completions. Passes per defensive action (PPDA) is a proxy for pressing intensity, with last year’s game showing lower for both teams, suggesting higher pressing intensity versus this year’s.

From an xG perspective, this is where data collection and related issues enter the picture.

This shot map shows the first attempt from Ryan Mmaee as being off centre to the left of the goal when, in reality, it was at the front post. This misrecorded location will likely have a significant impact on the assigned xG value, which was 0.75.

As this screenshot shows, not only was Mmaee at the front post, but Joe Hart was well positioned and the crossed ball was well off the ground. As far as can be discerned, Wyscout’s xG model does not account for the height of crossed balls at the point of contact for a shot or the location of defenders. In contrast, StatsBomb’s xG model had the same chance at 0.29, which to me seems far more realistic.

READ MORE: Celtic 'do their thing' to exorcise Ferencvaros demons... but victory was more than a revenge mission - Sean Martin

Following that notion, it is notable that StatsBomb had the xG at 3.54 v 0.67 in favour of Celtic, with npxG 2.78 v 0.67. That npxG differential of 2.11 is significantly larger than the 1.39 for last year’s game in Wyscout’s model although that is not an apples-to-apples comparison.

So Celtic’s performance on Tuesday was better, right? Not so fast – what about game-state?

In last year’s match, Ferencvaros scored a very low probability chance via an angled 25-yard screamer into the far side netting in the seventh minute. Through the first 56 minutes of that game, they had just 0.05 xG on three shots, though one resulted in the huge goal. Celtic had 1.23 in xG on 17 shots, with seven higher than the one Ferencvaros scored.

Conversely, in this year’s game, Ferencvaros had that 'big'Mmaee chance in the third minute, which was recorded at about 10-times higher quality. In addition, Cameron Carter-Vickers initiated some contact - it was marginal but could have worked out differently had he been just a little more forceful.

Through the first 56 minutes of Tuesday’s game, the two teams had comparable xG up to that point, on eight shots for Celtic versus five for Ferencvaros, with Mmaee’s chance by far the highest quality for either team. What would have happened had that shot been scored?

As this screenshot shows, with a tie score they were caught with five players in a high press in the sequence of play which resulted in Jota’s incredible pass and Kyogo Furuhashi’s brilliant run, touch and finish. For a side which had been very successful in frustrating Celtic up until this point by not pressing, instead soaking up pressure with periodic counter-attacks, what would the game-state have been had Mmaee scored?

Almost immediately after Kyogo’s goal, this happened:

Anyone who has not watched the replay of this should focus upon Anthony Ralston’s heroic effort to close down Mmaee over the last 10 yards – just tremendous. And perhaps as importantly, it prevented a shift in game-state. Would Ferencvaros have been as open in the final half-hour had they pulled the score back to 1-1?

Declaring a definitive conclusion on which game involved a “better” performance would not be objective. This analysis suggests that the games were actually fairly similar in terms of how each team performed, with critical differences in finishing quality and game-state.

Rather than make a declaration, the main point of this exercise was to try and offer how “data” is distinct from analysis. Fine margins.