Read on Twitter
THREAD: What is Impact Score (IS) and how does it work?
It's been great to have gained a large number of followers in the last few days, and with that has come plenty of questions about our main metric, Impact Score. This thread will be a detailed dive into what it is (1)
It's been great to have gained a large number of followers in the last few days, and with that has come plenty of questions about our main metric, Impact Score. This thread will be a detailed dive into what it is (1)
The concept of IS is to give a singular top down perspective on a player that gives some idea of how good a level they play at. It was created agnostically so that it can be used to compare players past and present, in any league globally (2)
It's obviously not a perfect measure, and can't account for everything. BUT, I do feel it's a good initial starting point when trying to decide if a player is good. It can't tell you why a certain score is high/low, that's where further scouting is needed (3)
So, the two main factors that go into IS are a player's minutes played and at what level. Minutes count not just in their quantity, but also in % of available minutes a player has been picked for (so discounting injury) (4)
The level of minutes takes into account various factors. These include the league/competition played in, their teammates, the opposition, the age group and, over the long term, at what points per game a team are picking up with said player in the lineup (5)
As an example, let's look at Lionel Messi. He has 58144 minutes in my database. When looking at those minutes and in what context they were played, our algorithms deduce he plays approximately at a 2.479 PPG standard. Very good, as we'd expect!!! (6)
This is then where we calculate his IS, by comparing this number to the average player in the database. This is scaled so that 100 is average. 120 is therefore 20% above average, 80 is 20% below, similar to OPS+ in baseball (7)
When we do this for Messi, his IS comes out at 155.236, meaning he's over 55% better than an 'average' player! Very impressive. Note, this means an IS can change with time as new players are added, but not massively due to a large player sample. (8)
We can then make further adjustments to this IS to account for other factors. For example, in our database, Messi is an attacker, so we can just compare him to attackers. When we do this, his IS becomes 160.287. This shows the average attacker is worse than... (9)
... the average player, as the IS has gone up. We can also adjust so we just look at Messi's club mates at Barca. This gives an IS of 129.817. This highlights how much above average Barca's squad is, but that Messi is still a standout at +30% (10)
This feature of IS allows us to spot good players on poorer teams, and even passengers on good ones. It can also be used to give a yardstick of how successful a potential transfer might be. (11)
The IS of older players does on average tend to be lower, due to a combination of less minutes resulting in more variance, and a lower overall standard of competition devaluing some of those minutes. Our reliable cutoff for IS is 5000 mins. (12)
We believe anyone with a sample of more than 5000 mins has reached roughly their 'true' IS and it is unlikely to fluctuate massively in future. Under 5000, it may change more, but we're still confident in the predictive power of IS. (13)
Currently, IS is calculated for players retired and current, in leagues the world over, including youth tournaments, non league and even grassroots. I believe this kind of top down approach is a great leveller for initial comparisons for players (14)
IS then becomes one of 3 factors in our player comparison tool which I use to do the historical teams threads. The other two are g90/cs90, and team defensive quality (as we try to isolate this to the individuals). We then find the lowest difference... (15)
... between the target player and all the other players. The IS of players are also used in our simulator. How this works is we simulate 5000 games between the two lineups. For each game, we simulate each minute for 90 minutes. (16)
In each minute, we compare defence against midfield to see whether a chance has been created. If so, we then compare attack against goalkeeper to decide whether that chance gets scored. This is repeated for the remaining 89 minutes (17)
After the 5000 games, we can then find probabilities for home/draw/away. Other than for keepers, the def/mid/att scores are a combination of quality and quantity. For example, the midfield score is the sum of the IS of all the midfielders... (18)
...multiplied by the average IS for these midfielders. This means fewer but better midfielders can beat more but weaker defenders etc. The simulation is constantly backtested against old lineups and tweaked in order to improve accuracy (19)
I hope this thread has given you a better understanding of how we can use IS. If you do have further questions, please get in touch and I'll be more than happy to help you out. Hope this was useful @GoalAnalysis @FootyScribblers @FootyThreads_
