Calculating a Goal Difference per 90 Metric for the S28 Draft Class

[iltornante]

I have been looking at the stats for the most recent Academy season, and trying to come up with a universal metric that can be applied to any player to approximate their impact on team goal difference per ninety minutes they played. Sort of like a more advanced plus-minus statistic, which takes things other than actual goals into consideration.

The idea is fundamentally flawed for various reasons -- which I will not get into here -- but I gave it a try anyway, because I thought it would be interesting to look at. 

In the end, my weightings were as follows:

• Non Penalty xG per 90 is weighted at 0.7
  
• Actual Non-Penalty Goals per 90 is weighted at 0.3
  This combination rewards clinical finishing by ensuring that a player scoring in line with xG gets credited +1 per 90 for every goal scored
  
• Assists per 90 is weighted at 0.4
  This ensures the majority of the benefit goes to the player actually scoring the goals, but assists are still valuable 
  
• Open Play Key Passes per 90 is weighted at 0.08
  An average open play shot has an xG of around 0.1; so we reward the playmaker regardless of their teammate's finishing, to make sure good creators are not penalised for having wasteful teammates 
  
• Dead Ball Key Passes per 90 is weighted at 0.03
  Set piece deliveries have a much lower conversion rate on average (somewhere around 3-4%) so this number is lowered to prevent artificially inflating the value of players taking set pieces
  
• Progressive Passes per 90 is weighted at 0.01
  Moving the ball closer to goal increases the chances of a goal being scored. In modern spatial models, an average completed pass into the middle third adds around 1-2% chance of scoring. This model does not have pass position breakdowns, so it assumes all progressive passes are equal, and assigns a conservative baseline.
  
• Dribbles per 90 is weighted at 0.015
  Beating a defender carries more inherent threat than a pass forward, as it typically destabilises the defensive structure by pulling a second defender out of position or creating a quantitative overload (e.g. 2v1, 3v2, etc). In most modern models, a successful dribble typically increases scoring probability by around 2%. Again, this model does not have the required granularity to correctly award this, so a conservative baseline is used instead.
  
• Turnovers Won per 90 is weighted at 0.008
  This is a reverse-engineered metric from Attempted Presses and Press %, which serves two purposes. Firstly, it wipes out the opposition's scoring opportunity (valued at around 0.01-0.03 xG against on average). Secondly, it increases the possibility of a counterattack via a pass or shot. Once more, the position in which the ball is won is not known, but this weighting means a ball-winning player who averages 4 successful presses per 90 adds around a third of a goal over the course of an academy season. This allows a central midfielder or fullback's defensive contribution to be quantified without artificially inflating their score.
  
• Mistakes Leading to a Goal are weighted at -1
  Essentially, every mistake leading to a goal counts exactly as is. If you contribute +2 goals per 90 but make a blunder leading to a goal every 90 minutes, your overall contribution should be +1 GD per 90.
  

With the explanations out of the way, I have calculated this metric for every academy player with at least 630 minutes played this season, to attempt to eliminate small sample-size outliers. As you might expect, five of the top ten are PS Jakarta players, given their general domination of the Academy league. Three of the other five are from Stockholm IK, and the remaining two are both from Victoria Falls.

Unsurprisingly, Franco Maestro features near the top of the list (in third) with an approximate +0.94 GD per 90, thanks to his clinical finishing -- 1.28 NPG + Assists per 90 (2nd in the sample). Notably, of the top five, Maestro is the only player with more than 400 TPE. Fellow Stockholm forward Michele Honcho (+0.62 GD per 90) is ninth, with his contribution of 0.86 NPG + Assists per 90 offset by a respectable 2.32 turnovers won per 90 -- higher than you'd expect for a striker. The third Stockholm name in the list is Emmanuel Bocandé (+0.67), largely coming from his frequency of key passes (4.75 per 90) which was higher than any other player in the sample.

Jakarta midfielders Micool CXXXII (+0.96), Carter Black (+0.87) and James Hogan (+0.82) occupy second, fourth and fifth place, respectively. Micool's score reflects elite goalscoring as well as creation -- NPG + Assists per 90 sum to 1 -- while also producing a large number of key passes (3.91 per 90; third highest in the sample) as well as a solid defensive contribution (6.03 turnovers won per 90). Black and Hogan also have similarly high numbers for NPG + Assists (1.28 and 1.00 respectively) but have noticeably lower defensive contributions than Micool. The remaining two Jakarta players in the top ten are Galadriel-Garfunkel Electabuzz-Zooboomafoo (+0.81) and Alyssa Barnes (+0.75), sixth and seventh, respectively. Barnes' 0.71 assists per 90 is second highest in the sample, reflecting a strong creative influence on her team, while GGEZ has one of the most balanced overall sets of metrics, with NPG + Assists per 90 at 0.93, while also displaying strong progressive ball movement -- 2.71 OKPK per 90 (6th) and 2.63 dribbles per 90 (=2nd). 

If you've been paying attention, you should now know the rankings for our two Victoria Falls players. In tenth, Char Zard (+0.6) with a combined 0.8 NPG + Assists per 90 headlining his stats for the season. But perhaps the big surprise is the player at number one. Victoria Falls fullback Eghosa Nwachukwu (+1.11 GD per 90) riding high a full 0.15 goals per 90 ahead of Micool in second place. The rightback contributed 1.29 NPG + Assists per 90 (1st) as well as 4.43 OPKP per 90 (2nd), 3.21 progressive passes per 90 (3rd), and 6.11 turnovers won per 90. All this despite having the second lowest TPE among the top ten.