# Assists must be replaced by expected assists

### Assists must be replaced by expected assists

An assist in football is generally attributed to the player making the final pass to a member of his team following which the team member receiving the pass scores a goal.

**Which player contributes the most to the team’s attacking performance?**

The **Stat Squabbler** says…

… the table suggests that Player A makes the greatest contribution to his team’s attacking performance by setting up the most goals. (Let’s assume they have all scored the same amount of goals so that the question is answered in terms of the *assists* metric.)

Example 1

Player gets an assist because their teammate dribbles past several players and scores.

Example 2

Player gets an assist because their teammate dribbles past several players and scores.

In the two examples above, we can clearly see that not only is the *assists* metric hugely influenced by the player’s teammate, but is, in fact, **wholly dependent** on whether their teammate scores or not. No matter how big (deserving) or small (undeserving) the individual’s contribution is, the metric is ultimately measuring the goalscorer’s ability to find the back of the net. Using the two examples above, the table below defines two metrics to identify whether or not an individual’s contribution is recognised/measured.

The intention of the *assists* metric is to measure an individual’s contribution for their team’s goal. In theory, if someone scores, they will have likely been helped by a teammate, who, in turn, gets credited with an assist – a reward for their contribution. There are inherent flaws with this metric. First, *how* much an individual helps varies vastly (see below). Second, it is wholly dependent on their teammate’s contribution too, i.e., their teammate has to score in order for the assist to arise.

A more accurate measure of an individual’s contribution to their team’s goals is undoubtedly that of *chances created*. The table at the beginning now has a row added showing the *chances created* by each player.

Another way to highlight the role of the teammate in the *assists* metric is by looking at the *(all) chances created* metric. Despite Player B creating 56.7% (34/60) more chances than Player A, Player B has been awarded fewer than half (7) of the assists than Player A. Looking at their teammates conversion rates, again, highlights the distortion teammates have on the *assists* metric. Player A’s teammates convert 28.3% (17/60) of their chances provided by Player A. Player B’s teammates convert 7.4% (7/94) of their chances provided by Player B. Player C’s teammates convert 9.4% (3/32) of their chances provided by Player C.

However, the *(all) chances created* metric can be further broken down to *BIG chances created* and *chances created*. A BIG chance is defined on the PremierLeague website as *‘providing an opportunity where the receiving player would reasonably be expected to score and manages to get a shot away*.’ It must be pointed out, according to this definition, that a player may not be rewarded/credited when in fact they should. For example, a player may create a ‘big’ chance but their teammate decides to take extra touches, losing the ‘big’ chance, and consequently failing to register a shot – so, despite their teammate losing it, it was still created! Notwithstanding this flawed definition/reasoning, chances should be further categorised into *big chances created* and *chances created* (i.e., not a big chance).

The table below shows each player’s *(all)* *chances created* metric broken into* chances created* and *BIG chances created*. For example, Player A’s total of 60 is broken into *chances created* (45) and *BIG chances created* (15).

You may have assumed that Player A created more BIG chances as he created less chances altogether than Player B but still has many more assists. Whilst it may be the case sometimes, it does not apply in this scenario. Though, Player A does have a higher proportion of *BIG chances created,* 25%, (15/60) than Player B, 16%, (15/94), this is evidence of metric distortion, i.e., giving underserving credit to some players and not the deserving credit of others.

One of the principle drivers of a sporting metric is to provide a single statistical measure to convey performance level. However, the *(all) chances created per 90* metric does not factor in the ratio of *chances created* and *BIG chances created*, i.e., a player is not rewarded for creating more BIG chances. Also, *the BIG chances per 90* metric does not factor other chances created. Therefore, replacing the *assists* metric with two metrics is thus not optimal, nor is using just one of these metrics as important data(performance) is omitted. Though it does paint a much-improved picture on the player’s attacking contribution. Now, can we combine the *chances created* and the *BIG chances created* metrics somehow, and account for our teammates’ contributions? The answer is yes.

There are two ways to do this. One way is to combine the metrics together after ‘weighting’ the *BIG chances created* metric in order to reflect its greater contribution to a goal being scored relative to the *chances created *metric and then divide by the appropriate ratio. To calculate the weighting, add up all the *chances created* in the league (3192) for the previous season and the amount of goals (380) that were scored from these chances. Meaning it takes, on average, 8.4 (3192/380) chances to be created for a goal to be scored. Do the same for *BIG chances created* (784) and goals (280), meaning it takes, on average, 2.8 (784/280) BIG chances to be created for a goal to be scored.

Now, looking at the ratio of *BIG chances created* to *chances created* (8.4/2.8), we can see that one BIG chance is ‘worth’ the same as three chances created (conveniently a whole number – it is fictitious data). Therefore, the weighting for a big chance will be 3.

Second, we sum together the *‘weighted’* *chances created *and the *‘weighted’* *BIG chances created *to find a single total for *‘weighted’ (all) chances created*, which factors in the types of chances the players created.

Now using the average number of chances to score (from the previous season) we can calculate the amount of assists a player can **expect** to achieve from the chances they created for their teammates. To calculate this, we do the *‘weighted’* *(all) chances created *divided by the average number of chances needed to score. So, Player A can expect 10.71 (90/8.4) assists for the season. To note, we could have reached the same result (30/2.8) if we ‘weighted’ differently, i.e., *chances created* x 1/3 and *BIG chances created* x 1. The relationship between the two quantities is the same, in other words proportional.

The difference between the actual *assists* metric and the *expected assists* metric highlights the huge distortion that is measured in the *assists* metric. Player A has been credited with 37% more than his performance deserves. Whereas Player B and Player C have received less credit than their performances deserve, by 110.9% and 50.7% respectively.

The other way to calculate the expected assists is to divide a player’s chances created by the correct ratios and then to sum the answers. To create the ratios for the league based on the previous year (or two..), it is the same as earlier.

Now, divide the player’s chances created by the correct ratios. In the example below, Player B created 79 chances and created 15 BIG chances. As illustrated, Player B can expect their teammates to score 9.40 of the 79 chances they created (79/8.4) and to score 5.36 of the BIG chances they created (15/2.8). Then, just sum the two totals to find the number of assists you would expect the player to receive when removing the variability/contribution of their teammates.

Returning to the original question: **Which player contributes the most to the team’s attacking performance? **

Football is a team sport and virtually all individual metrics are influenced by their teammates’ performances. Also, there is no doubt that there is a positive correlation between *chances created* and *assists*, i.e., in general, the more chances created by a player the more assists they will have (see Case Study below). However, a teammate’s performance must be removed as much as possible from an individual metric. Therefore, the *assists* metric must be replaced by *expected assists *metric. An assist is wholly dependent on a player’s teammate, regardless of the player’s contribution beforehand. By using *expected assists*, the metric is more accurately measuring an individual’s contribution to goals by:

**Removing**a**teammate’s contribution**[metric distortion].**Removing**a**fluke**or**a non-skilful action**by the player being rewarded [metric distortion].**Making sure a player is attributed when influencing the game**despite what their teammate does [better measurement].**Using a bigger sample of data**to measure performance/influence, i.e., more chances are created than assists [better measurement].**Weighting chances proportionally**to value the ‘worth’ of BIG chances [better measurement].**Using the league’s average of chances needed per goal**[better measurement].

**Even better if** a player’s position is considered, on the basis that a player is more likely to create chances further up the field, i.e., closer to the opposition’s goal. So, once a player’s position is considered, they should then be ranked according to the number of *expected assists per 90 *metric from that position. This will enable more informed judgements to be made when comparing players against other players.

For example, Player C creates less chances than Player B. However, Player C is a defender (D) whereas Player B is a midfielder (M). Player C has created the most chances as a defender and Player B has created the 3^{rd} most chances of a midfielder.

**Quick Case Study: Premier League’s Top Assists in 2018/19**

The Premier League website only provides data on *BIG chances created* so admittedly there is data missing and does not allow me to fully calculate expected assists using the method explained in this article. However, what the first two columns demonstrates in the table below is that whilst there is a correlation between the number of *BIG chances created* and the number of *assists*, it evidently highlights that this is not a perfect correlation and that the *assists* metric has a lot of distortion (behind it). For example, both Salah and Silva created more than twice as many BIG chances than Pogba, yet still have fewer assists.

The final two columns are taken from *The Expected Goals Philosophy* by James Tippett who has calculated the *expected assists* (xA) and the *expected assists per 90* (xA90) metrics for the top 10 players in the Premier League last season. Whilst the method used is different to the one (briefly) explained in this article, it nonetheless, uses the fundamental concept that the outcome of the pass is standardised with the average player. So, whilst Eden Hazard got most of the plaudits as the best playmaker, Ryan Fraser was actually the best playmaker last season. Looking further using the xA90 metric, Hazard only ties 4^{th}, with Ryan Fraser, Leroy Sane, and David Silva. In other words, given that they play the same amount of time, Fraser, Sane and Silva all create more than Hazard.

** **The** Stat Squabbler **concludes:

- That the
*assists*metric**must**be replaced by the*expected assists*metric because the*assists*metric is wholly dependent on a teammate’s performance and is hugely distorted. - Should chances be weighted further, i.e., in a game, or team, that is not producing many chances, should a chance created be worth more than a chance created in a game, or team, that creates them aplenty?

Do you agree with the Stat Squabbler? *Assists* must be replaced by *expected assists.* *Comment below.*