Opponent ability adjusted stats
We've all played kickabout games where it is horrendously unbalanced. One team is 17-0 up in 5 minutes. Someone on the winning team will then magnanimously suggest swapping one of their best players for one of the worst (usually me) on the losing team. It normally doesn't take much of a swap for the game to become a lot more competitive.
And a funny thing happens. The players on the poor team suddenly look a lot better. They have more time on the ball, they have a player (usually me) they can easily bypass, they have a player they can put pressure on and dispossess (usually me) and they have a good player on their team who can hold onto the ball and play off.
If you were measuring stats for the games you would see a huge discrepancy from pre and post the team rebalancing.
Obviously, at the professional level, you are not going to have unfit, middle-aged men, dry-heaving after 10 minutes, as opponents. Unless you are playing West Ham.
Then think of the stats padding advantage that good team players get.
Think of two identically skilled players both in a league with 18 teams. 6 are good, 6 average and 6 awful.
The player on the good team will face 5 good, 6 average and 6 awful opponents.
The player on the awful team will face 6 good, 6 average and 5 awful opponents.
Almost 6% of the games will take place against either a team you should thrash or a team you will be thrashed by.
For a defender, this could mean two 5-0 wins effortlessly batting aside your opponent or two 5-0 defeats run ragged by Messi.
For a striker, this could present the opportunity to rack up the goals against relatively poor performers.
If all this sounds a bit familiar it is because I wrote a Statsbomb article showing that playing on a dominant team does tend to boost your overall stats.
What I didn't do was offer any solution.
I think there are various things you could do:
Apply a "dominance adjustment factor" to stats to scale up or down how we'd expect those outputs to look on various teams. What this factor looked like could be done by looking at the output differences from players who had moved between teams.
This could be quite interesting from a recruitment point of view. If we scouted a 20-year-old on a poor team and could "age-adjust" and "relative dominance" adjust their stats we might find that we would expect them to perform significantly better on a more dominant team as a 21-year-old based on how every other player in our data-set has performed under similar circumstances.
Adjust the sample so that you create "mini-leagues" of good/average/awful and see how their performances varied.
Perhaps even just double count the performance against the closest match within the league so that the awful teams get 1 extra game against awful teams and the good teams have an extra game against good opponents put into their stats.
I don't know if it will make a huge difference but the perennial problem of stats based scouting is finding good players on relatively poor teams. This could help.
And a funny thing happens. The players on the poor team suddenly look a lot better. They have more time on the ball, they have a player (usually me) they can easily bypass, they have a player they can put pressure on and dispossess (usually me) and they have a good player on their team who can hold onto the ball and play off.
If you were measuring stats for the games you would see a huge discrepancy from pre and post the team rebalancing.
Obviously, at the professional level, you are not going to have unfit, middle-aged men, dry-heaving after 10 minutes, as opponents. Unless you are playing West Ham.
Then think of the stats padding advantage that good team players get.
Think of two identically skilled players both in a league with 18 teams. 6 are good, 6 average and 6 awful.
The player on the good team will face 5 good, 6 average and 6 awful opponents.
The player on the awful team will face 6 good, 6 average and 5 awful opponents.
Almost 6% of the games will take place against either a team you should thrash or a team you will be thrashed by.
For a defender, this could mean two 5-0 wins effortlessly batting aside your opponent or two 5-0 defeats run ragged by Messi.
For a striker, this could present the opportunity to rack up the goals against relatively poor performers.
If all this sounds a bit familiar it is because I wrote a Statsbomb article showing that playing on a dominant team does tend to boost your overall stats.
What I didn't do was offer any solution.
I think there are various things you could do:
Apply a "dominance adjustment factor" to stats to scale up or down how we'd expect those outputs to look on various teams. What this factor looked like could be done by looking at the output differences from players who had moved between teams.
This could be quite interesting from a recruitment point of view. If we scouted a 20-year-old on a poor team and could "age-adjust" and "relative dominance" adjust their stats we might find that we would expect them to perform significantly better on a more dominant team as a 21-year-old based on how every other player in our data-set has performed under similar circumstances.
Adjust the sample so that you create "mini-leagues" of good/average/awful and see how their performances varied.
Perhaps even just double count the performance against the closest match within the league so that the awful teams get 1 extra game against awful teams and the good teams have an extra game against good opponents put into their stats.
I don't know if it will make a huge difference but the perennial problem of stats based scouting is finding good players on relatively poor teams. This could help.
Comments
Post a Comment