Fiorentina have conceded 20 goals this season. 9 of them have been in the final 15 minutes. That’s 45% of the goals they’ve allowed coming in the final 16.7% of matches. That’s almost 3 times as many as you’d expect from a random distribution. Because that seems statistically unlikely, I wanted to look a lot more closely at the team’s inability to keep the ball out of their own net late in the game. Moreover, because it seems like the players regularly run out of gas in the closing stages, I wanted to see if there was any correlation between running and late concessions.

The first thing I did was make some tables, because that’s fun. I decided to match the order to the standings because I think that helps display any clear relationships between numbers and success. And, as always, these numbers mean nothing without context. I’m interested in seeing how they back up what we’ve seen from Fiorentina this year and trying to figure out what that says. It’s another way of understanding the game, not a way to replace actually watching.

The data

I looked at 4 statistics for every team in Serie A here: total goals conceded, goals conceded in the final 15 minutes, average distance run per game, and passes per defensive action (PPDA). PPDA counts how many passes a team’s opponent completes in their own half for every defensive action—challenge, pressure, foul—the team performs in that half; a lower number means that a team is pressing hard and high, which generally requires more running, although the total running numbers also rely on numerous other factors.

A quick note about ordering the data: I decided to display the teams in order of the standings because I thought it was most important to see if any of these numbers, by themselves, correlate with a position in the table (spoiler alert: not especially). There’s also no way to create a dynamic table in the publishing software here, so if you’re interested in getting this data, you’ll have to recreate it yourself, although that shouldn’t take very long.

Goals conceded numbers from SoccerStats. PPDA numbers from Understat. Distance run from Serie A.

Observations

The average Serie A team concedes 25.4% of its goals in the final 15 minutes. That’s higher than a random distribution, and it also makes sense: defenses are more tired and more prone to mistakes later in games, especially since defenders are subbed off far less frequently than attackers. Exhaustion makes a big difference to a team’s ability to defend, as it requires focus and awareness of both teammates’ and opponents’ positioning. It’s really hard.

Also, the final 15 minutes are when teams tend to start going for broke to get a result. The conventional wisdom is that if you need a goal, you try to keep it tight and then throw numbers forward towards the end of the game. That means that some games open up in their final stages, which means that goals are more likely to occur.

But! It’s not as simple as that. After all, this increased rate of goalscoring depends on game state. If there difference between the scores is ≤1, one or both teams might try to score. If one team is already losing by 3 goals, though, neither the winner or the loser is really going to go all out to score. So this is very much a situational thing.

I didn’t really see a lot of correlation between distance covered, PPDA, and late goals conceded, which really surprised me; I thought that those would be linked very closely, because (and this is partly based on my own, deeply amateur experience) it’s harder to defend well when you’re physically and mentally exhausted after running very hard for 88 minutes, but these data, at least, don’t lead me to that conclusion. If nothing else, it’s a nice reminder that these players possess absolutely superhuman fitness and that we take it for granted.

In terms of Fiorentina stuff, I was astonished to see that the Viola aren’t among the league leaders in distance covered and PPDA. My explanation there is partly down to discipline: having spent more time playing with 10 men than any other team in the league, it makes sense that those games would mess with the club’s running totals. After all, when you’re a player light, you want to sit deep and keep your shape out of possession, which usually means less running and a lower PPDA.

I also wonder if games like the one against Sampdoria, which was basically sewn up by halftime, lead to some weird variance here; maybe Fiorentina ran and pressed really intensely in the first half and then relaxed a bit after the break. I feel like this is a matter of sample size and random variance based on when teams score more than anything else.

Conclusions

So these numbers look pretty random, right? Napoli and AS Roma are 2 of the 3 best teams at preventing goals scored in the final quarter hour and they’re both in the top 5. But the other team that’s really good at preventing late goals is Udinese, who are near the bottom. There are only 3 teams that concede 40% or more of their goals late on and they’re pretty evenly distributed through the middle of the table.

Distance covered is maybe even more random. First place Napoli run less than anyone, while mid-table Lazio are track stars. The teams with a higher PPDA don’t even run all that much more. I went back after the fact and looked at how possession stats related to distance covered and found it to be somewhat unrelated as well; Napoli don’t run at all and see more of the ball than anyone, while Sassuolo, who rank second in that category, are 16th.

There’s a slight correlation of PPDA with position in the table. The teams with the lowest stats in that department are mostly clustered towards the bottom of the standings, although 13th-place Torino throw a wrench in that, seeing as how they work way harder without the ball than anyone else. Inter Milan and AS Roma don’t work especially hard either and are doing just fine.

Conclusions

There’s a lot more work that someone a lot smarter than I am could (and probably is) doing on this kind of stuff. I think the primary thing that’s missing from this analysis is game state: figuring out how much teams run in neutral or 1-goal-deficit situations and how that impacts when they concede goals. Unfortunately, I don’t think I’ve got access to that data, so here’s hoping that someone with an Opta or StatsBomb subscription is wondering the same thing.

For Fiorentina, though, I can safely say that I didn’t figure out a goddamn thing here. I learned that a lot of things that I thought were strongly correlated—distance covered, PPDA, possession, and conceding late goals—are anything but. That makes me think that Fiorentina’s late concessions run deeper; after all, this was a problem with the team last year as well, to the point where I did a lot of work like this for an article I ultimately scrapped.

Either way, it’s pretty clear that this is the team’s biggest weakness. Leaking goals at the ends of games is a massive problem and Vincenzo Italiano is definitely going to work on it. While my dip into the numbers here hasn’t turned up anything remotely useful, I do think that a big part of the problem is a failure to adjust to game state more than a problem with the system or the players. However, measuring game state is something that’s really hard to do if you’re a layperson, so let’s hope that someone at Viola HQ is looking into it.