2024-09-09, 11:54 AM - Word count:
What a fun season! My first being totally plugged in. I wanted to take a look at some of the season end numbers and see what we could learn about how teams performed. Warning: I’m fairly new to football manager and soccer metrics in general so there is likely an issue or two in the data handling, but hopefully we can learn something or at least take a fun walk through the season just completed.
A quick note about the data since this an area I might have made a mistake. I integrated league table data from the portal with some views from the season end file of stats and player ratings. I used the league observer short list so I think it has total visibility. Stats are across all competitions as far as I can tell, and seem to line up based on some spot checks – I did build back up from a player level aggregation to the team level and then indexed to a per 90 basis which causes some slight distortions to other sources at a team level aggregation, but I think the results are still useful in a directional way at least. If you do notice any numbers that seem unusual let me know.
And now the high level results. Reykjavik U. and Schwarzwälder FV again separated themselves from the rest of the pack, though the final league finish was a reverse of the prior year with Reykjavik on top, though they again suffered a brutal defeat in the cup this time to Buenos Aires. After the top 2 teams there is a cluster of 4 closely matched clubs. And then the final two teams a fair bit back, but both building toward the future, with São Paulo even notching a win this go around!
![[Image: image.png]](https://i.ibb.co/Lds3fxB/image.png)
In addition to the league points and cup finish I added a handful of offensive stats across all competitions on a per 90 basis, heat mapping shows across the selected stats the cream tends to rise to the top almost universally, notching more goals, assists, shots, clear chances created (CCC), crosses, and dribbles. I was surprised by the degree of the separation. Shots in particular I thought would be closer, though that maybe is a sign of my lack of general soccer knowledge. I could buy that better teams tend to get better quality shots, but was surprised to see top teams shooting 2 or 3 times as often as bottom teams. I guess having more of the ball and having more skilled players on said ball allows you to generate more offensive actions across the board. Crosses and dribbles stand out a little bit, especially among the middle tier, and may be an area for future investigation to see how differences in offensive philosophy (being more direct, or working more centrally) might affect some of the outcome metrics.
You might have noticed the top two teams tied in xG averaging 3.2 per 90 across their competitions, but experienced a significantly different conversion rate of xG into Goals(0.88 goals per xG for Schwarzwälder and 1.03 goals per xG for Reykjavik) which drove a 0.5 goals scored delta between the clubs that had to contribute to who finished on top. Given that may have been a driver in how this season turned out lets investigate a bit more.
I’m still building intuition around the expectation modeling that drives xG, particularly in football manager. I understand the delta between xG and actual goals scored might be attributed to a sustainable skill of the player, but there also could be some systematic over or undervaluing of xG coming from certain game contexts (for example maybe header xG is undervalued so Reykjavik’s higher cross per 90 count gave them more opportunities for more easily converted xG), or it could simply be variance. In the context of this league I think player skill contribution is worth exploring since there is an attribute under our control, finishing, that in theory should increase the conversion rate of xG to goals.
Below I show the formation from the end of season file for each team and also some player level stats sorted by the number of shots taken from most to least. In yellow I highlighted the players in the two forward most strata, presumably the most skilled finishers. In green I highlighted midfielders or wide backs, and finally in red I highlighted center backs.
Here’s the data for Reykjavik:
![[Image: image.png]](https://i.ibb.co/RpW6qYK/image.png)
And here it is for Schwarzwälder:
![[Image: image.png]](https://i.ibb.co/8z2WQT5/image.png)
I bolded the xG/shot for what I classified as central attackers. These central attackers tend to have the highest xG/shot. People that tend to hang out toward the edge of the box and take long shots end up with lower xG/shot. I would like to be able to separate the shot type but will need more experience and possibly other data sources to really separate out the different sorts of shots, for now the metrics show the blend which might give some misleading results, but at a high level I think we can say that from an xG perspective the most high quality shots come from central attackers, and for the most part these players invest in higher finishing skill.
Now to discuss some differences. First of all, one of the more surprising results was the relatively higher number of shot attempts from center backs for Reykjavik. My guess is that they are utilized as targets of corners, though I can’t confirm. At the end of the day it doesn’t matter so much whether a center back or another position provides these shot attempts so long as your team strategy accommodates this, but I felt it was worth mentioning. The central midfielders then see a similar but opposite contrast being used more frequently by Schwarzwälder whose midfielders are positioned in the CM strata of this attacking mentality tactic. Reykjavik’s central midfielders are positioned in the DM strata of a balanced mentality tactic and are among the least utilized shooters on the team.
The other primary difference to discuss is the success Reykjavik had in funneling their shots toward the players most heavily invested in finishing. The players with 18 and 20 finishing have clear cut advantages in the number of shots taken. I think that is generally a good metric to see with TPE deployed toward game actions your player participated in most commonly. And we see some indication that it improved results achieving some of the better results with regard to goals per xG (G/xG). Schwarzwälder on the other hand had their leading shot taker with 5 finishing. It’s worth noting that between Hakkinen and Bull the extra investment in finishing does not appear to have had a positive impact on G/xG. But I think the over-arching point of trying to understand the sorts of shots their CWB is taking that might better be distributed to a player with a higher investment in offensive finishing.
I should emphasize than these sorts of metrics won’t let us say what the optimal strategy is. And the “blame” shouldn’t be assigned to any single player, there are a number of factors at the player and team level that might lead to a sub-optimal shot distribution. At the end of the day certain positions will tend to take more of those low probability shots when the offensive possession struggles to break down the defense (it’s worth noting again how few shots the lowest performing clubs put up). But I think noting that this season the team that finished on top got better outcomes on the xG they generated. And we’ve seen some possible reasons why: funneling shots toward highest finishers, and possibly creating more shots through crosses or header attempts in the box. Players and teams should give some consideration to how they can avoid long shots with low xG from players with poor finishing.
Let me know what you think.
A quick note about the data since this an area I might have made a mistake. I integrated league table data from the portal with some views from the season end file of stats and player ratings. I used the league observer short list so I think it has total visibility. Stats are across all competitions as far as I can tell, and seem to line up based on some spot checks – I did build back up from a player level aggregation to the team level and then indexed to a per 90 basis which causes some slight distortions to other sources at a team level aggregation, but I think the results are still useful in a directional way at least. If you do notice any numbers that seem unusual let me know.
And now the high level results. Reykjavik U. and Schwarzwälder FV again separated themselves from the rest of the pack, though the final league finish was a reverse of the prior year with Reykjavik on top, though they again suffered a brutal defeat in the cup this time to Buenos Aires. After the top 2 teams there is a cluster of 4 closely matched clubs. And then the final two teams a fair bit back, but both building toward the future, with São Paulo even notching a win this go around!
In addition to the league points and cup finish I added a handful of offensive stats across all competitions on a per 90 basis, heat mapping shows across the selected stats the cream tends to rise to the top almost universally, notching more goals, assists, shots, clear chances created (CCC), crosses, and dribbles. I was surprised by the degree of the separation. Shots in particular I thought would be closer, though that maybe is a sign of my lack of general soccer knowledge. I could buy that better teams tend to get better quality shots, but was surprised to see top teams shooting 2 or 3 times as often as bottom teams. I guess having more of the ball and having more skilled players on said ball allows you to generate more offensive actions across the board. Crosses and dribbles stand out a little bit, especially among the middle tier, and may be an area for future investigation to see how differences in offensive philosophy (being more direct, or working more centrally) might affect some of the outcome metrics.
You might have noticed the top two teams tied in xG averaging 3.2 per 90 across their competitions, but experienced a significantly different conversion rate of xG into Goals(0.88 goals per xG for Schwarzwälder and 1.03 goals per xG for Reykjavik) which drove a 0.5 goals scored delta between the clubs that had to contribute to who finished on top. Given that may have been a driver in how this season turned out lets investigate a bit more.
I’m still building intuition around the expectation modeling that drives xG, particularly in football manager. I understand the delta between xG and actual goals scored might be attributed to a sustainable skill of the player, but there also could be some systematic over or undervaluing of xG coming from certain game contexts (for example maybe header xG is undervalued so Reykjavik’s higher cross per 90 count gave them more opportunities for more easily converted xG), or it could simply be variance. In the context of this league I think player skill contribution is worth exploring since there is an attribute under our control, finishing, that in theory should increase the conversion rate of xG to goals.
Below I show the formation from the end of season file for each team and also some player level stats sorted by the number of shots taken from most to least. In yellow I highlighted the players in the two forward most strata, presumably the most skilled finishers. In green I highlighted midfielders or wide backs, and finally in red I highlighted center backs.
Here’s the data for Reykjavik:
And here it is for Schwarzwälder:
I bolded the xG/shot for what I classified as central attackers. These central attackers tend to have the highest xG/shot. People that tend to hang out toward the edge of the box and take long shots end up with lower xG/shot. I would like to be able to separate the shot type but will need more experience and possibly other data sources to really separate out the different sorts of shots, for now the metrics show the blend which might give some misleading results, but at a high level I think we can say that from an xG perspective the most high quality shots come from central attackers, and for the most part these players invest in higher finishing skill.
Now to discuss some differences. First of all, one of the more surprising results was the relatively higher number of shot attempts from center backs for Reykjavik. My guess is that they are utilized as targets of corners, though I can’t confirm. At the end of the day it doesn’t matter so much whether a center back or another position provides these shot attempts so long as your team strategy accommodates this, but I felt it was worth mentioning. The central midfielders then see a similar but opposite contrast being used more frequently by Schwarzwälder whose midfielders are positioned in the CM strata of this attacking mentality tactic. Reykjavik’s central midfielders are positioned in the DM strata of a balanced mentality tactic and are among the least utilized shooters on the team.
The other primary difference to discuss is the success Reykjavik had in funneling their shots toward the players most heavily invested in finishing. The players with 18 and 20 finishing have clear cut advantages in the number of shots taken. I think that is generally a good metric to see with TPE deployed toward game actions your player participated in most commonly. And we see some indication that it improved results achieving some of the better results with regard to goals per xG (G/xG). Schwarzwälder on the other hand had their leading shot taker with 5 finishing. It’s worth noting that between Hakkinen and Bull the extra investment in finishing does not appear to have had a positive impact on G/xG. But I think the over-arching point of trying to understand the sorts of shots their CWB is taking that might better be distributed to a player with a higher investment in offensive finishing.
I should emphasize than these sorts of metrics won’t let us say what the optimal strategy is. And the “blame” shouldn’t be assigned to any single player, there are a number of factors at the player and team level that might lead to a sub-optimal shot distribution. At the end of the day certain positions will tend to take more of those low probability shots when the offensive possession struggles to break down the defense (it’s worth noting again how few shots the lowest performing clubs put up). But I think noting that this season the team that finished on top got better outcomes on the xG they generated. And we’ve seen some possible reasons why: funneling shots toward highest finishers, and possibly creating more shots through crosses or header attempts in the box. Players and teams should give some consideration to how they can avoid long shots with low xG from players with poor finishing.
Let me know what you think.
![[Image: 0HMDG8L.png]](https://i.imgur.com/0HMDG8L.png)
