Earlier this month, FloSports was excited to announce a historic, multi-year broadcast partnership deal with D.C. United, marking the first agreement FloSports has had with an American “Big Five” professional sports league. Alongside live streaming of games, with both English and Spanish broadcasts, we are committed to providing high-quality, data-driven MLS content.

While soccer analytics is still in its infancy, especially in comparison to higher-volume, discrete-action sports like baseball or basketball, major strides have been made in the collection and analysis of data in the beautiful game. The history of a sport like soccer has left some traditionalists reluctant to change, but those who have been able to embrace the data revolution have been able to find a major competitive advantage. 

In a game as fluid as soccer, with so many variables interacting at once, it is no simple task to parse through all the noise. While teams have made use of soccer analytics to help with game strategy, injury prevention and roster building, much of the added value in all of these fields comes from improved player evaluation and the metrics that define them. In this article, I am going to give a quick overview of some of the most important metrics of soccer player evaluation, how they are calculated and how they are utilized. 

Expected Goals (xG)

Expected goals is the most popular and most cited advanced metric in soccer analytics. While goals scored has long been a way to evaluate a player’s effectiveness, the relative infrequency of the event (in 2018, there were 3.22 goals scored per game in the MLS) leads to a high variance from player to player, leading goals scored to be a very unreliable indicator of a player’s actual impact. 

Expected goals is a count of how many goals a player should have been expected to score based on the quality of their chances. While there are many different models, most are based on the distance of the shot and how the shot was taken. The further away a shot is taken, the less likely it is to go in. Additional variables for distance can include the shot’s angle, height and distance from the center of the goal mouth. How the shot was taken is a bit more complicated and can include variables like foot versus head, dominant versus non-dominant foot, as well as type of the play that the shot takes place on. For example, a shot taken off a corner kick decreased the expected goal probability by .622 in comparison to a shot taken from the exact same place if it was not off a corner kick, according to the American Soccer Analysis Model. This can largely be attributed to the fact that a player will have much more space against a transitioning defense when the ball is in free play.

Expected goals don’t tell us everything, however. The metric’s primary value is in determining how well a player sets himself or herself up to score. A player who has a large discrepancy between their goals scored and expected goals scored could be suffering from a case of particularly bad luck, or may be very good at getting into position to score, but poor at finishing. This metric can also be helpful in determining the relative strength of a goalie, and how frequently he or she can stop high-percentage shots. 

Expected Assists (xA)

It takes two to tango, and on the flip side of the expected goals metric is the expected assists metric. In this case, it is designed to determine how dangerous an attacker is, and how well they can set their teammates up to be in position to score. The expected assists statistic uses the same metrics as expected goals, but assigns them to a passer. Therefore, if a through ball leads to a chance with a .62 xG, the one taking the shot is awarded .62 xG, and the passer is awarded .62 xA. 

Score Effects

Not all shots are created equal, and oftentimes in-game situations change players’ risk calculus in terms of shot selection. A team losing by one or two goals is more likely to take a higher volume of low percentage shots in order to try to even the score. A leading team is more likely to form a “defensive shell,” and hold possession, as scoring another goal is less valuable to them than protecting another goal from being scored. This is problematic when we are comparing teams’ shots on targets, saves and expected goal numbers at the end of the game. We often find that teams trailing by a goal tend to outshoot their opponents, leading to lower conversion and efficiency percentages. Leading teams score 53% of goals and concede 47% of goals, despite taking fewer shots. If you aggregate enough statlines, it would appear that teams who take fewer, more efficient shots are more likely to win. But, in reality, this only appears to be the case because decision-making changes with score. Score effects allow us to take time and place into account in order to determine a team’s “true” scoring efficiency rate. 

More Advanced Metrics

With a baseline of expected scores that takes into account time and score, you can build a majority of most advanced player evaluation models. For example, a player’s scoring threat would be his or her expected goals per 90 minutes, while the player’s finishing ability would be actual goals scored minus expected goals. Neither is necessarily a better indicator of a player’s overall value. When and how to use players who are constantly a threat to score versus those who are particularly good at finishing in low-threat situations is a tactical decision that is up to the discretion of the manager.

Key passes and big chances, metrics collected by Opta, are related to the expected value of a pass/shot. A key pass is defined as “the final pass or pass-cum-shot leading to the recipient of the ball having an attempt at goal without scoring.” A big chance is “a situation where a player should reasonably be expected to score, usually in a one-on-one scenario or from very close range.” The cut-off for what an “attempt at scoring” or “should reasonably be expected to score” is defined by the expected goal probability of that shot. 

More advanced player-evaluation metrics that attempt to distill a player’s impact down into one number are usually measured in xG equivalents. How do completed passes, assists, turnovers and shots relate to the overall expected goal of a possession? This can be aggregated across games or looked at on a per minute/per possession basis to understand the effectiveness of certain players.

Soccer analytics may be a relatively new development, but they are certainly here to stay. While there is no replacement for watching the games live, with these metrics as the baseline, we hope to be able to bring new insights to our readers about players, strategies and trends in soccer, this season, and beyond.

Fri Lavey is a writer and consultant who studied economics and statistics at Harvard University. He has previously worked as a data analyst for the NBA, the MLB, and the Philadelphia 76ers.