FIFA World Cup 2026: Analysis of Expected Goal Contribution (xGC) Model

The Athletic launched its World Cup tracker this week, which we highly recommend checking out. Here, you can find predictions of each team's chances of advancing to each stage, as well as the most likely Round of 32 matchups, the current points table, and many other fascinating predictions and scenarios. The purpose of this article is to explain how we generate the team strength that powers this simulation, which we call Expected Goal Contribution or xGC. We call it a "system" rather than a "model", because it is a set of multiple models that together measure the game, starting from an individual pass or dribble to assessments of players and teams. This model is for anyone who wants to know what is likely to happen in an upcoming football match, as well as why. Which USMNT players should I pay attention to? Who is expected to provide the most value to their team, and in a given game, who is performing below or above expectations? The goal was to create a system that explains why teams are good, not just which ones are good. Expected Goal Contribution depends only on two main principles of the game of football: 1. Teams are made up of players. They are, of course, made up of other things too (coaches, strategies, systems, collaboration, cultures), but the players themselves are the key. 2. Better players help increase their team's chances of scoring (or prevent the opposing team from doing so) badly. If you can get behind these principles, our system is actually quite simple.

Here are: 1. We estimate the probability of each team scoring a goal at every point in the match. 2. We give “credit” to players who engage in actions that increase their team's chance of scoring a goal. 3. We "penalize" players who are usually in close proximity to actions that increase their opponents' chances of scoring. 4. We use an Elo-style rating system to aggregate and adjust these goal contributions, giving us offensive and defensive player ratings. 5. We create a team's offensive and defensive ratings by taking a minute-weighted average. 6. We take advantage of the difference between a team's offensive rating and their opponent's defensive rating (and vice versa) to determine how many goals each team can expect to score in the upcoming match. Of course, there are a lot of nuances to each step, which I'll explain in more detail below. However, for those who are satisfied with this high-level explanation, I want to move on to the good stuff – how does this system evaluate each team's roster at the start of the World Cup? We have Spain and France as the top teams of this tournament, which coincides with the valuations of the markets which consider these two sides as clear favourites. Below you can see how our ratings compare with FIFA's latest rankings. Here you can see that while we are generally in sync with the FIFA world rankings, there are some significant disagreements. Argentina won the last World Cup in 2022, and

Currently on a winning streak of 18 matches. However, xGC estimates that even with Lionel Messi, their roster does not hold the same threat as teams like France and Spain. Norway – by contrast – is participating in its first World Cup since 1998, and is therefore in the back half of the 48 qualified teams in the FIFA rankings. But xGC rates Erling Haaland as one of the top forwards at this tournament, while Alexander Sorloth, Martin Odegaard and Antonio Nussa all also rank very favourably. Below is an overview of the top five forwards, midfielders and defenders on the World Cup roster according to xGC. I use these position groupings quite extensively: Forwards consist of central forwards, wingers and attacking midfielders; Midfielders include central, left, and right midfielders; And the defenders include wing-backs, full-backs and centre-backs. Spain's Lamine Yamal is highly rated by football observers and our models. Okay, we've avoided specifics long enough. Below I will go through each step of this process in more detail. To estimate goal probability, we need to assess how much a player increases his team's chances of scoring a goal. This is not a new idea, first written about by Sarah Rudd in 2011, and has since been known as Expected Threat, VAEP, g+, and OBV, among other names. These all effectively do the same thing, and our implementation is nothing new. We simply built a linear model to estimate

We can calculate how likely a team is to score a goal, with the information available to us (e.g. where the ball is on the pitch, how long the team has had possession with the ball, what the pace of the game has been in the last few seconds, etc.). If we divide the pitch into zones, and take the average of the goal probability estimates from all of them, this is how likely a team is to score a goal from each zone. Once we have goal probability, calculating goal contribution is straightforward. We simply find the difference in goal probability at the beginning of that event (like a pass or carry) and the end of the event, and this becomes the goal contribution added by the event. For offensive players, it is very simple to assign this goal contribution to the players involved in the incident. Carries are assigned solely to the player who advanced the ball, passes are shared between the passer and receiver, and shots get credit for the xG of the shot they took, which determines what their goal probability was before the shot was taken for their team. Defensive attribution is more complicated, because for a given on-ball action, we don't actually have data on who marked the players involved in the play. We have an idea of ​​where each defender typically defends on the pitch (e.g. tackle, interception, challenge, etc.), and we use that spatial information to "penalize" defenders according to their relative "responsibility" for where the offensive action occurred. This means that an attack on the left wing is more punishing for a right-back than for a central defender. "

The "expected" part is the goal contribution portion that is created using an Elo-style updating system. We start each player with a league average rating, we generate an expected goal contribution based on their rating for their upcoming match and the rating of the opposing team, and then we adjust the player's rating based on what was expected compared to the observed goal contribution. The more a player performs above or below what was expected, the more we update their rating and thus their The final piece of the puzzle here is to go from a player-level xGC to a minute-weighted average of who is going to start for each team. But, how do you know it's the right one to use? For three reasons: In the game-analytics world, we are measuring something within the player, not just their rating. If we start it cold and generate it from only one half of the season, it has a much stronger correlation with their rating in the other half: players with higher xGC are usually paid more. are transfermarkt prices, which means

That we are measuring something that the wider football community considers valuable. Our xGC values ​​correlate with Transfermarkt values ​​with a Pearson's coefficient of 0.65. Perhaps a more intuitive statistic, if we randomly select two players from our sample, the player with the higher net xGC has a higher Transfermarkt value 80 percent of the time. Roster xGC predicts future match scorelines: Teams with higher roster xGC (the collection of player xGC) generally score more goals and win more than teams with fewer goals, and when we use roster xGC to predict how many goals a side will score, we find that our predictions are very well calibrated – meaning, if we predicted zero goals for 100 matches with a 30 percent chance. So, in 30 of those matches, there were zero goals. So, our player ratings are stable, they correlate with crowd intelligence, and they are useful in predicting match outcomes. And this is it. This is xGC, the system driving our World Cup tracker predictions and forecasts. Please let us know in the comments what you think about this system, and don't forget to follow our predictions as the tournament progresses.