Today I have a challenge go through. Done by FC Rstats however the submission was a few weeks ago and I did it in a rush and I don’t think it was my best work. Therefore this is a re hash of my submission so could end up with different results.
The challenge is simple suggest the top 3 players to replace Nikita Parris. Currently Parris plays football for Manchester city and she is leaving. I have to find the 3 best players to replace her. Lets have a look at Nikita’s headline stats from this season taken straight from google:
We are looking for someone who can replace 19 goals and 7 assists. This puts her second in the top scorers list and 4 in the top number assist list. This is going to be a challenge finding players of a similar standard. First things first I need to create a profile of Nikita in order to compare it to other players to find the players most similar. I have the Statsbomb event data to do that. Let’s first look at the goal scoring
I thought the first thing to look at after the goals was the total expected goals and we can see Nikita ranks pretty high, however this doesn’t really tell you anything. A much better metric is the expected goals adjusted for the opposition faced. If you generate a lot of expected goals against weak opposition that’s worth less than against good teams.
Adjusting the Expected goals for the quality of the opposition the shot was against shows that, although not top, Parris has had a large amount of shots at a high expected goal rate. What’s also interesting is Miedema who was top of the total xG chart is now near the bottom of the average per shot showing she generally takes a lot of shots worth lower expected goal values. Maybe that suggests she wont be the best person to replace Nikita as last season wont be sustainable.
This xG value can be further adjusted by the number of miutes the player plays. How much xG should were expect from a player per 90 minutes they play. I think this is the best most representative value of expected goals and its what will be used to select our possible replacements.
I also need to find a player who is similar with where they receive the ball and pass from on the pitch. Below you can see all the position on the pitch which Nikita has received the ball.
Clearly they are mostly in the oppositions half. In order to have this as a metric i’m going take the average point
On average over the season Parris has mostly touched the ball high up the pitch in central areas so we are looking for a player that touches the ball in similar areas. I wont use position for the clustering however I will use it to compare our final 3 players.
Moving onto the length of passes which Nikita players and you can see from the histogram she mostly does short backwards passes. Possibly because she is generally higher up the pitch and therefore more likely to find a pass backwards.
On the list of expected assist per 90 minutes Nikita doesn’t perform as highly as she does for other metrics. This is definitely a metric that I will use to cluster my players.
Above you can see a summary of Parris’s final 3rd passes and you can see she is incredibly active. There are a lot of short passes on the left side as well as a lot of passes going into the box. Overall she has a pass success rate of 66.1%. which for the amount passes seems to be a good rate. Final 3rd pass completion rate will be used in the clustering.
The bee-swarm plot above shows each players final 3rd pass completion with the players at the extremes completing very few passes.
Finally looking at dribbling, another area attacking players generate offence. Below you can see a histogram of all players dribble completed percentages.
With a relatively small amount of data it i a roughly normal distribution. As there is vast differences in the amount of dribbles players have attempted i’m going to build an empirical bayes estimation model in order to normalise the data so all players can be compared. With the normalised dribble
Above there is a summary of all the players estimated dribble percentage. On the whole it looks like it isnt one of Nikitas strong skills. Could this be an area we could find a player who is generally better at dribbling.
Thats it for the this first part in finding a replacement for Nikita Parris. We have seen how she compares to other players for a variety metrics and came up with metrics I can use to carry out K-means clustering to find players similar and therefore the best ones to replace her. Look out for the next blog which I go through the code for the clustering and identify the 3 possible replacements. Thanks to Statsbomb and FC Rstats twitter account.
[…] https://theparttimeanalyst.com/2019/05/30/replacing-nikita-1/ […]