The right footed left forward

This is a brief study I started working on this summer but never had time to get ready enough to publish, until now. So, here it is!

I had an idea to study if there is any significant difference in goal conversion for right footed players in the left part of the pitch in relation to left footed players shooting from the right part of the pitch. The idea would not only be that a right footed would player get more opportunities to score from the left, supported by the argument that players better can cut inwards when playing with the odd foot, but also that the angle to the goal would be somewhat better thus increasing the goal conversion.

Said and done, with my dataset from Superettan and Allsvenskan dating a few years back to now I looked at all shots made from players where I know if they were left or right footed. I divided the offensive pitch into 4 parts from right to left and then looked at the shot conversion in the parts for right, and left-footed players individually.

This is what happened:
The total number of shots I looked at was: 10935

On the small right hand side:

Small right hand side

A left footed player scores 9,3% of the time
A right footed player scores 8,1%

On the large right hand side:

Large right hand side

A left footed player scores 14,7% of the time
A right footed player scores 13,7%

On the small left hand side:

A left footed player scores 8,6% of the time
A right footed player scores 9,8%

On the large left hand side:

A left footed player scores 13.0% of the time
A right footed player scores 14.2%

So I’d say that the results are quite conclusive, and not at all surprising.

Then I started thinking on how I could use the results. I always have thought of using some player attributes in my xG model. So now when I had gathered some data I did just that, added the foot attribute to my xG model. One way of looking at how adding an attribute to an xG model affects the model is by doing a regression on the comparison of xG in relation to actual goals scored. And doing that had this result for a subset of teams XG-G included in my dataset: (Trace 0 without foot attribute, Trace 1 with)

Trace 0, y, Trace 1, y, Trace 1, y - fit, Trace 0, y - fit

The difference is not big, saying that the foot attribute in general is worth much less to the possibility to score than for instance the distance to the goal. The funniest thing was that the above plot took 30 hours to compile for my MacBook! And then really saying much…

Another way of evaluating the impact of a newly added feature is to look at the feature importances. I’ve done that in the past when for instance adding game state to my xG model. And personally I like that approach much better. Looking purely at R² from a regression when the impacts are so small really doesn’t say anything.

So when I looked at the feature importances with the foot attribute (last in the list) added the result was:

[ 0.34383995 0.30124342 0.02760246 0.00277241 0.04683145 0.00419405
 0.12862736 0.07352959 0.06102812 0.0103312 ]

That tells us that which foot a player uses for a certain shot has a 1% impact on the chances of it becoming a goal, in general. It may differ a bit depending on location but around 1%. That makes sense when looking at the results I saw looking at scoring conversion.

The next thing I now had to take into consideration was whether to include this parameter to my xG model. And this is something I’ve been thinking about the last months. At times I have thought to exclude it since it is harder to maintain and update in relation to the impact it actually has. On the other hand it really HAS an impact.

I guess the answer on wether to include a parameter such as foot would be to ask yourself what you want to have your xG model for. If you want to use it for a descriptive purpose, adding a foot property would add some value for sure, We know we had a left footed forward playing to the right, so we’re expecting him/her to have a higher conversion rate.

But if the model is to be used for predictive causes, where we don’t know or can influence the preferred foot of the players playing, then maybe such a parameter should be ruled out.

And following this reasoning I decided to include which foot a player prefers to my xG model.

Published by

Ola Lidmark Eriksson

Football analyst/programmer https://blog.stryktipsetisistastund.se/

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s