Is that the Thierry Henry finishing?

Some players have a trademark goal. For Totti, it was the chip, for Juninho (Pernambucano) it was the vicious and deceptive free kick. But one that’s always stuck in my mind is Thierry Henry’s.

You know the one.

HenryLeeds.gif

 

Having seen (and spent far too much of my childhood years trying to emulate) this, I have often wondered whether shots struck with the inverse foot (that is the right foot when on the left of the pitch, and vice versa) are more likely to be scored than others.

Using StrataData, we can investigate this empirically for shots classified as a ‘good’ or ‘great’ chance in the wide areas of the pitch (L*/R*).

stratazones

When we look at the scoring in these areas, an interesting pattern appears:

p1

We can clearly see that over this period of time, shots taken with the left foot from the right hand side of the pitch were scored at almost twice the rate of those from similar positions elsewhere on the pitch.

Moreover, we can drill further down into the data to a finer level of detail. What we find is that this pattern persists in shots classified by Strata Analysts as ‘Great Chances’ (13 scored from 16!) as well as ‘Good Chances’:

Rplot05

Why on earth is this happening?

We can get a clue when we look at the volume of shots in each of these buckets. It is immediately clear that there are far fewer shots being attempted in these situations on the right than on the left:

p2

 

Given the small number of shots being taken, it may be tempting to blame it on a small subset of elite players (Robben?) skewing conversion with their finishing skill. However, this goes against what we know about finishing skill. What’s more, only five players took more than a single shot with their left from the right and did not contribute enough to skew the sample (Robben, of course, took the most, scoring three from eight shots).

An alternative explanation for this disparity is the relative rarity of left footed players in the population. Because there are more right footed players (it is estimated that between 87 and 92% of the world are right-handed although this effect is obviously diminished at the highest level of football), it may be more likely that players will cut in from the left onto a favoured right foot to shoot, whereas those on the right are more likely to cross the ball or take it on their right.

Screen Shot 2016-04-20 at 05.09.25

Could it be that right footed players cutting in on the left are more shot happy and therefore take more shots from poor locations (c.f. Coutinho, P)? While we have hopefully accounted for some degree of chance quality in the StrataData definitions (‘good’ and ‘great’ chances), this could be having an effect.

Interestingly, if this is the case, it does not clearly manifest itself in the shots’ location. Almost all the wide shots are taken from the L2 and R2 buckets (see above) and if we isolate shots from these equivalent zones, the conversion does not change by more than 1% in any of the four groups identified above (L/R foot; L/R side of the pitch). It is possible that the shot locations could be skewed within these zones; however, it seems unlikely that they could be driving such a large effect.

In a similar way, the effect we are seeing may well be a result of selection bias. On the right of the pitch, players may tend to use their (often weaker) left foot only when presented with a very good opportunity to do so. This seems like the most satisfying explanation so far; however the effect remains very large and I’m open to other ideas.

A final possible explanation that ought to be considered is that of the definitions. Perhaps chances from the right with the left foot are disproportionately considered poor chances (‘Attempts’), leaving fewer unconverted shots in the ‘Good’ and ‘Great’ chance categories. Again, it seems unlikely that such a large effect should come from a bias like this (it should also be noted that each match of StrataData is verified by two additional analysts).

(02/05/16): Recently Paul Riley investigated an alternative explanation investigating the different patterns of goalkeeper positioning between right and left handed ‘keepers. This is a really nice piece and nicely shows how public work can lead to collaboration and development ideas that otherwise would not be possible: https://differentgame.wordpress.com/2016/04/30/left-right-left-right/.

Expected goals and collapsing the wave

Given the way it is generally calculated (at least publicly), it easy easy to think of expected goals (xG) as an attribute that applies just to shots. When we see an xG map or an interactive, it’s easy to fall into this way of thinking. In reality, it makes more sense to think of every action on the pitch having an expected goal value. In fact, to would probably be more accurate to think about the fact that at any point at which the ball is in play, there is a given probability that a goal is scored within the next n minutes of play.

This is in line with recent work done by Thom Lawrence at Deep xG investigating the average time taken for a shot to be produced at different points in a game.

In this way, when we take an xG value at the time of a shot, we’re really just looking at one dimension of a much larger picture (data show is entirely fictional and used for illustrative purposes):

fake-xg.png

This has repercussions for how we evaluate chances in a broader sense. For instance, when multiple shots happen in a passage of play, how do we evaluate the chance? Perhaps the simplest answer is to sum the xG values. However, if you do this, you can end up with an expected goals total for a single passage of play that is more than one. Clearly, you cannot score more than one goal in one passage of play.

A frequently proposed alternative, then is to take the total probability of a goal being scored from each of the successive shots. In other words, the probability of the first shot being a goal plus the probability of the second shot being a goal multiplied by the probability of the first shot being missed:

xg-tree.png

However, when we view expected goals as an attribute of shots, as opposed to a continuous variable that’s constantly fluctuating, I can’t help but feel that this seems an improper way of evaluating a series of connected shots or chances, that is ultimately an imperfect way of thinking about goals and probability in football.

 

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