Midfield Rating System
5 min read

Midfield Rating System

Midfield Rating System

This is something that I have been kicking around for a while starting with a pretty simple rating system that used the 13 stats from my radars, the z-scores for each to create an overall score.

This system produces solid results at getting players that fill out radars but I think it misses the ability to try to find players that fit a certain profile and role. I have been trying to think through how to better present data like this and came up with breaking things up by general skill/action that a player does.

The results of that are what I call my "stats dashboard" graphic. This shows the familiar polar type charts (aka Crab Cakes) but for stats that are grouped together.

We are going to use Granit Xhaka as an example throughout this article.

I think that this type of graphic helps show at a glance what a player is good at. I still like the other types of graphics but this gives more stats and more context in comparison to these:

So the next logical step is to take this concept and apply it to ratings. I got a lot of inspiration from this from Liam Henshaw and what he did with his rating system.

Henshaw Analysis player ratings — methodology, discussion & examples
Recruitment in football is difficult. There are hundreds of matches every week. How do scouts know what games to go to? Well, that all…

Stats Used

For this midfield rating system, I have broken up the midfield skills into six skills/attribute types. They are passing, carrying, ball-winning, aerial, creating, and shooting.

Each stat used is given a z-score to illustrate how far a player is from the mean by standard deviations. This is helpful because it will give players that are very good or very bad more and less credit than simply doing percentiles.

The stats in the passing category are progressive passing, pass efficiency (actual passing rate compared to expected passing rate), final third entries, forward pass %, total progressive passing distance, and Goal Probability added passing.

The stats in the carrying category are progressive carries, turnover efficiency (how often a player loses the ball compared to expected based on where they touch the ball and how many dribbles they attempt), total progressive distance carried, dribbles completed, dribble success rate, and progressive passes received.

The stats in the ball-winning (I am using this term over defense here because I think that is a more accurate description of what we are measuring) category are pressures, pressure regains, successful tackles, interceptions, blocked passes, ball recoveries, and fouls committed (where things are reversed so fewer fouls is better).

The stats in the aerial category are aerials won and aerial duel win rate.

The stats in the creating category are assists, shot-creating actions, open play key passes, open play expected assists, and deep completions (not from a cross).

The stats in the shooting category are open play shots, non-penalty expected goals, non-penalty goals, xG per shot, and on-target rate.

You may have different thoughts on the stats that go into this and I would love to hear your thoughts (oh.that.crab at gmail dot com to reach me) but I think that these stats paint a good picture and with weighting, these can be adjusted up and down based on the role that we want to try to measure.

Weighting

This section was the one that I struggled with the most. I think it is pretty obvious that certain actions are more important/more valuable than others and should count for more. The question is how much more?

To go through this I have decided that the best way to do this is with a flexible weighting system that I can adjust based on the role template. For this proof of concept, I have started with what I am calling the "Attacking 8" role, with what Kevin De Bruyne does for Manchester City as the ideal (no surprise he tops the list in the rankings for this).

I will add additional roles as time allows and update the weighting used here as those come available.

Attacking 8 Weights

Passing:  progressive passing (25%), pass efficiency (25%), final third entries (15%), forward pass % (10%), total progressive passing distance (15%), and Goal Probability added passing (10%). Overall weight 30%.

Creating: assists (7%), shot-creating actions (30%), open play key passes (13%), open play expected assists (30%), and deep completions (20%). Overall weight 25%.

Ball Winning:  pressures (10%), pressure regains (20%), successful tackles (20%), interceptions (15%), blocked passes (15%), ball recoveries (10%), and fouls committed (10%). Overall weight 19%.

Carrying:  progressive carries (20%), turnover efficiency (21%), total progressive distance carried  (20%), dribbles completed (15%), dribble success rate (19%), and progressive passes received (5%). Overall weight 13%.

Shooting: open play shots (20%), non-penalty expected goals (25%), non-penalty goals (20%), xG per shot (20%), and on-target rate (15%). Overall weight 10%.

Aerial: aerials won (55%) and aerial duel win rate (45%). Overall weight 3%.

These stats then feed into an overall rating that is also calculated using z-scores but with an overall rating that is represented as a percentile so that things look pretty on the 0-100 scale.

In the visualization used to represent this information, things are shown as both the percentile rank (left on the crab cake) and as to where they fall on the z-score distribution for the stat (right).

This helps show some interesting things, for example, Granit Xhaka here comes out as a very good passer landing in the 92nd percentile. Looking instead at where he lands in the distribution he is well to the right of the average midfielder but he is still closer to the average midfielder than he is to the leader (Cesc Fàbregas from 17/18).

Overall this matches with my expectations having watched Xhaka and I think the combination of both the percentile ranks and distributions gives important context for the overall rankings.

Limitations

No single number score is going to be perfect and I am fully aware that these are not that. I would be careful in using this, with the thought that this helps to create shortlists for further analysis and comparisons.

The position filters are not always perfect for these so some players do not fit well into these. Players are asked to do different things and that will obviously change their statistical output. The quality of teams and the league that players play in will change things (on that front this only uses players from the "top 5 leagues") as well as the tactics that the teams use.

Another thing that I want to add (and I am working towards that) is a weight towards playing time. Right now the cut-off is 900 minutes, with 60% or more coming in midfield, I would like to add in the percentage of minutes available played as actually getting on the field is quite valuable but that is not fully captured here.

There is also the actual weighting used here and how that can change things. People will certainly have different priorities and opinions on things. I have done my best here to show what I have done but other choices are certainly valid.