Beyond Goals and Assists: Quantifying Player Impact with VAEP

Introduction

When looking to quantify a player’s performance, traditional statistics often fall short. Statistics such as goals and assists rarely paint a full picture, as they do not provide any information regarding the relative contributions of either player. Compare, for example, these famous assists by Busquets and De Bruyne:

Both players would be awarded 1 assist for their contributions.

Additionally, more complex metrics such as xG and xGOT are flawed. Whilst they can help to identify the quality of chances created and shots taken, they are largely redundant when considering the performance of non-attacking players.

Even newer metrics designed to address these flaws are not perfect; xGChain (developed by Hudl), for example, aims to quantify the impact of players involved earlier in the possession chain. However, all passes are valued equally in this framework, meaning again information regarding the relative contributions of each player is lost.

VAEP is a metric which addresses this.

What is VAEP?

VAEP — or Valuing Actions by Estimating Probabilities — was developed by SciSports and researchers from KU Leuven in 2019.

The core idea of VAEP is that every action a player takes will either increase or decrease the probabilities of (a) scoring or (b) conceding in the immediate future. To estimate this impact, VAEP evaluates several contextual factors, including the action’s location, body part used, direction, and the sequence of preceding actions.

In its simplest terms, VAEP is calculated as:

(Change in probability of scoring within the next 10 actions)
— (Change in probability of conceding within the next 10 actions)
following a given action.

For example, a forward pass which increases the probability that a goal is scored by +0.4 and reduces the chance the team concedes by 0.2, will be assigned a VAEP value of 0.6. Conversely, losing the ball in a dangerous area may change the team’s probability of scoring by -0.3, and increase the probability of the team conceding by 0.5. This would produce a VAEP value for this action of -0.8. Therefore, at face value, a negative VAEP value indicates a reduced chance of scoring / increased chance of conceding, and a positive VAEP value indicates the opposite.

By assigning a value to every action in a possession sequence, VAEP captures contributions that traditional metrics often miss. This makes it particularly useful for identifying players whose impact is subtle but consistently positive.

Comparing players by their average VAEP per action provides a way to quantify the best players in each position, offering a more comprehensive measure of on-ball contribution beyond output.

Methodology

Here, I looked to use VAEP to identify the most valuable players in each position in the EFL Championship. Using the soccerdata and socceraction packages in Python, event-level data for the EFL Championship was retrieved from WhoScored. Playing times and positions were extracted from FotMob.

A VAEP model was trained on a dataset of over 1.7 million actions from the 2022/23 and 2023/24 seasons. This was then validated on a dataset of over 800k actions from the 2024/25 season. Finally, the trained and validated model was tested on a dataset of over 500k actions from the 2025/26 season so far.

For each analysis, I compared the average (mean) VAEP / action and the actions per 90 performed by each player. Using these two statistics, we can assess both the value and the volume of actions performed by any given player. In all cases, only players who have played a minimum of 450 minutes this season were considered.

The Results

In all cases, the VAEP / action (y-axis) is measured against actions / 90 (x-axis). The graphs are split into 4 quadrants, with the vertical division representing the median actions/90 among the cohort. The horizontal division sits at 0.0, the threshold at which a player’s actions can be considered entirely neutral.

Furthermore, to assess overall performance, VAEP/90 was calculated. This combines both measures to determine which players add most value overall. For each position group, notable performers in either metric will be highlighted, alongside the best overall performers in each position.

Forwards

Although Zan Vipotnik ranked 1st in VAEP/action with a value of 0.0134, his low number of actions/90 (26.8; 6th lowest) limits his overall impact. This performance suggests Vipotnik may be viewed as a conventional poacher, adding significant attacking value despite lesser involvement on the ball. Conversely, Nathan Broadhead and Nestory Irankunda impress with the highest actions / 90 values among all forwards in the league (60.3 and 51.7, respectively), yet their overall impact is limited by their modest (yet positive) VAEP/action valuations.

Two of the division’s most elite performers — Mathias Kvistgaarden and Brandon Thomas-Asante — combine high volume and high impact, sitting comfortably in the top-right quadrant. These two players can be recognised as complete forwards.

On the other hand, Kyogo Furuhashi stands out for all the wrong reasons, registering the lowest mean VAEP/action (-0.0117) while also recording the 3rd lowest actions/90 (24).