Thread by @Jhdharrison1

THREAD: It’s the #FIFAWWC final today! So it is 1 year since I started a project analysing the goalkeeping at the #FIFAWorldCup2018. So about time I tweet the findings! It is on distribution & how we can go beyond pass completion! #footballanalytics #Lloris #Pickford #Alisson

Firstly, pass completion is the only distributive metric that regularly gets any airtime in the mainstream media.
Websites like @WhoScored & @Squawka regularly compare goalkeepers using pass completion.
#FIFAWorldCup2018 #Goalkeeping #Distribution

Given the amount of data websites like @WhoScored & @Squawka have you would imagine it is a pretty robust metric?

However, if you think about it it feels like there should be many factors which effect pass competition which have nothing to do with a GKs distributive ability.

In order to investigate this, & possible improvements to the analysis of goalkeepers & their distributive ability I watched every single minute of the #FIFAWorldCup2018 & logged every single action by a goalkeeper, including many statistics which are not publicly available.

The #FIFAWorldCup2018 showed that pass completion has basically nothing to do with a GKs distributive ability. The plot of pass completion vs fraction of passes long is very strongly correlated with an R^2 value of 0.8381!

R^2=0.8381 means 84% of the variance in pass completion by GKs at the #FIFAWorldCup2018 can be attributed to the amount of long balls they play. Thus pass completion is useless when it comes to finding which GKs are good with their feet. It just picks out what team they play for!

To remove this bias the obvious thing to do is to bin by distance however the distances to chose are not immediately obvious. @OptaJoe chose two bins under 35yrd & over 35yrd. This creates some issues as not all passes over 35yrd are equal, & neither are all passes under 35yrd.

This is were being a GK helps!

I know there are 3 main passing techniques:

Sidefoot for under 20 yards

Clip for 20-50 yards

Strike/Drive for over 50 yards

Therefore I hypothesised that these bins would be more informative than the @OptaJoe’s bins.

At the #FIFAWorldCup2018 I found that passes over 50 yards are harder than passes between 50-35 yards to a statistical significance of 10 sigma! (Which basically means they are very different to a strong statistical significance & therefore should not be binned together)

I also found that passes under 20 yards are much easier than passes between 20- 35 yards to a statistical significance of 8 sigma! Again highlighting that they are very different types of passes and should not be binned together.

This reinforced my hypothesis that using 3 bins for distance (short, midrange, & long) is better than simply short & long because it takes into account the different techniques used by goalkeepers & thus can probe strengths & weakness when using the different techniques.

It also stops the bias I have discussed in previous tweets from cropping up. For example #Alisson & #Kepa look incredible at long balls simply because they play a lot of passes over 35 but under 50 yards compared with other goalkeepers. (They are also very good at those passes!)

The plots of long pass completion, midrange pass completion, & short pass completion vs fraction of passes long shows that the correlation is indeed removed by the binning. YAY!

So if we simply look at pass completion in these bins does that remove the majority of the bias? Do we now pick out the best goalkeepers with their feet? Sadly not. There is another major issue with pass completion. Simply getting a touch to a pass is enough to complete the pass.

So I decided to create a metric called pass success. As I wanted to measure how useful a pass was to the team as it is obvious that not all completed GK passes are “good” passes. A GKs job it to maintain possession as high up the pitch as possible & the metric should reflect this

I decided that a successful pass is one in which:

• The pass is completed

• At some point after completion the GKs team has possession under no direct pressure

This rules out passes which are fired into targetmen who aimless flick on the ball as positive passes

It also penalises bad decision making as players who regularly play into pressure poorly will not make as many successful passes

Watch a game & apply the rule & see it yourself

When success is compared to pass completion we immediately see that completion does not equal success! This effect is stronger for longer passes than shorter passes as one would expect.

Therefore I believe to attempt to accurately pick out which GKs are best with their feet we must use success rather than completion as it levels the playing field for GKs who don’t have a huge target man or super mobile forwards who always get a first touch regardless of quality.

Success & Distance are not the only important factors. Pressure on the ball is also important. When a GK is under pressure as they receive the ball their passes tend to be more wayward.

Using the #FIFAWorldCup2018 data I find that being under pressure causes passes to be successful less often for long passes, midrange passes, and short passes. The statistical significance of the results are 2, 4, and 9 sigma respectively.

This means GKs who are forced to play under pressure more often are unfairly represented when looking at success percentage. So we must also bin by whether the distribution events were under pressure.

So we have found distance binning is important, pressure binning is important & completion doesn’t always mean the pass was useful for the team. So when looking at distribution with this in mind who comes out on top?

Well the kings of long passing at the #FIFAWorldCup2018 were:

1. @kschmeichel1 26/90 successful
2. #Akinfeev 23/94 successful
3. @JPickford1 15/66 successful

👑👑👑 👏🏻👏🏻👏🏻 they consistently made useful passes over 50+ yards more than anyone else!

The kings of midrange passing at the #FIFAWorldCup2018 were:

1. @rpatricio1_ 14/17 successful
2. @MatyRyan 15/21 successful
3. @NavasKeylor 9/12 successful

👑👑👑👏🏻👏🏻👏🏻 Patricio, Ryan, & Navas showed incredible consistency with their midrange clips top GKing!

The kings of the short passing under pressure at the #FIFAWorldCup2018 were:

1. @JPickford1 9/9 successful
2. @Alissonbecker 7/7 successful
3. @thibautcourtois 6/6 successful

👑👑👑👏🏻👏🏻👏🏻 #Liverpool is clearly a breeding ground for top footwork under pressure!

These results matches up very well with the eye test. And also highlight that if we were to simply look at pass completion we would find it to be horrendously misleading. Let’s compare distribution king @JPickford1 who has 75% pass completion to @YannSommer1 who has 76%.

The radars are split into distance & pressure situations & we use success rather than completion. The axes are percentile ranks based on the means & standard deviations of the #FIFAWorldCup2018 data. The inner black ring is average & above/below it is better/worse than average.

We can see @JPickford1 is superior in all aspects of distribution other than midrange under pressure. However the number of passes of each type both players play mean that their pass completions look similar! (Sommer often plays short passes & is generally under less pressure)

CONCLUSION:

• Pass completion is an awful metric for GKs & only tells you the teams style of play.

• Passes should be binned into 3 distance bins + pressure bins to remove some of the bias.

• Using pass success rather than completion filters out unhelpful passes

FUTURE WORK:

I feel there is one more key variable which defines how useful/difficult a pass is, and that is the number of players played through. After watching countless games I believe it is possible to make a robust GK distribution model.

This model would split passes by:

• If under pressure
• Length of pass
• Number of players played through
• To what degree successful

(I hope to use this in future work if I get the time!)

If you managed to make it to the end of this 32 tweet monster thread, thank you for taking the time to read it.

If you have any questions or are interested in a particular goalkeeper’s performance at the #FIFAWorldCup2018 just let me know!