## Shot conversion in the Argentine Primera División in the last 3 years.

In another blog I am keeping track of my work on a expected goals model. All variables and assumptions are largely based on other people’s models, but here I’ll discuss something I have not yet seen (or missed..): the underlying shot numbers. While working on the analysis for the bigger model I found a lot of interesting statistics on the shot conversion rate (goals/shots) per, for example, game state, number of shot in the game, time in game, and more. Where for the model I will dive deeper into the numbers and use statistical testing these are largely just descriptive tables. But no tests, no significance levels. And yes, each of the paragraphs would need a blog of its own..!

Note: All penalty shots and own goals have been deleted from those numbers.

###### Basic numbers

In the following table you see the general numbers of all shot data.

* see here what the angle to goal is: the ϴ-angle from the image.

Here’s the distribution:

A total of 326 players took 25 shots or more over the last 3 years. 11 of them did not score at all, Dinamo Kiev loanee and 2015 top scorer Marco Ruben has the highest conversion rate with 31,5% (17 no-penalty goals out of 54 shots) . Here is the distribution of all players. Half of the them converted between 6,3% and 13,7% of their shots, the distribution is quite concentrated and has a few outliers on the right (high) side:

###### Conversion rate per shot angle

The most important metric is the angle from the shot location to goal (that ϴ-angle as described above). Here’s a bar chart with the average conversion rate per degree on a 5-degree rolling average: the conversion for angle aº is taken for  a – 2,5 < a < a + 2,5.

The chart looks quite nice and I am not to worried about some of the outliers between 167º and 115º as they are taken over very small samples (e.g. the 33% conversion you see at 122º is the result of 3 goals from 9 shots).

###### Type of assist and shot conversion

As type of passes (eg. through ball, cross, etc.) are not given by the data provider so I had to calculate each of them based on the coordinates of the pass: the origin and destination. All matches from which I have data are entirely uploaded to YouTube and were thus available for me to check. I distinguished 8 types of passes: a “forward pass” would be in a lot of cases comparabla to a through pass as it is a vertical pass forward. However from the data I cannot see the position of the defenders and thus call it a real through pass. The same applies to “forward pass sidewards”, but then diagonal. A “corner” is clear, a “cross” as well, but an important note is that I only took into account passes from 15 meters or closer to the goal line. “Change of sides” is a long ball to a player close to the sideline. “Pass” is a rest category of passes I was not able to assign to any other category.

A table with all types of assists with the average shot distance, shot angle and conversion rates. Important to note is that the distance and the angle (which is dependent on the distance) of the shot differ per each type of assist. For this reason conversion percentages are not directly comparable. This is why I added the average conversion% for the average angle of the angle for each type of assist:

From this table you can see that the “forward pass” – as mentioned, comparable to the through pass – has the most positive impact on the shot. Normally a shot with an angle of 18,86 would be scored in 9,6% of the cases. After a “forward pass” that rises to 13,7%. The corner, with generally a lot of defenders around, has a negative effect on goal conversion.

Game state and shot conversion

The game state is defined as the goal difference of the game of the team of the shooter. If Carlos Tévez shoots at the moment that Boca Juniors is winning 2-1 the game state for the shot is +1. The table:

An interesting table: the conversion rate increases quite significantly when the team of the shooter is leading the game. And in this case that is not subject to a better shooting position as they are all pretty much the same for each game state. A very logical explanation is that the team that is winning the game is better than the other team (probably). However, as if often argued the shot conversion rate when standardized for shot angle and shot distance does not differentiate ‘good’ and ‘bad’ players or teams on the long term.

###### Time in game and conversion

From this table it can be seen that the conversion rate slightly increases over the course of the game, with unsurprisingly a very small peak towards the end of the game when there are probably some tired defenders and maybe some fit substitutes trying to get that winner. On the other hand the highest goal conversion is found between minute 5 and 10. A pretty weird peak, so maybe not so manymeaningful results here..

###### Number of shot in game

You have those players (Depay and Cristiano Ronaldo come to my mind) that keep shooting and shooting. Maybe not so stupid and annoying as it seems: based on the following table the chance of scoring increases per shot. There does not seem to be an interacting effect of the shot angle, as that variable remains quite constant per shot.

Something I did notice is that only the better players make it to 5, 6 and 7 shots in a game. For the ones that see Argentine football: e.g. Tévez, Mora, Caraglio, Romero, Benítez, Sand, Rinaldi, Simeone.

###### Time since LAST POssession opponent and CONVERSION

The time since the opponent last had possession  (in seconds) since the shot. This can be an important measure as if the team that takes the shot has recently recovered the ball this indicates their opponent has not yet had the time to regroup and get in position to defend. Something that points in that direction can be seen here:

###### Historical conversion to current conversion

A lot has been said about the repeatability of shot conversion rates. The general tendency is to say that a high conversion rate (when standardized for shot location) is not a skill and it will regress to a general mean. However the following table suggest otherwise:

The conversion percentage increases rapidly based on the increase in past conversion rate. This is not only due to the improving shot location as can be seen from the added two columns on the right.

Another notable statistic is that players with a higher past conversion rate find better shot locations (closer to goal / better shot angle). And contrary to what I said in the paragraph above this image: players with a better past conversion rate (probably due to better shot locations) seem to be better players: they find better shot locations.

###### CONVERSION FROM SHORT DISTANCES

I set up the following table in order to check whether my assumption hold true that shot angle did not matter was much when the shot location is very close to the goal. Following this data it seems to be true: the conversion for a given angle is much higher for very close shots.

###### Conversion from year to year

When plotting the conversion rate of players in year t and year t-1 it looks like there it is an almost unrepeatable statistic (for 2013 to 2014 the R2 = 0,03 and f0r 2014 to 2015 R2 = 0,07):

Here’s the Tableau worksheet for the graph above.

Note: all data comes from DataFactory.