As most of you are aware, we at SportsKPI, believe in analyzing teams and players above the aforementioned variance. We aim to evaluate players based on their overall performances over their end product. This motto of ours leads us to constantly look for innovative metrics which may encapsulate and portray player performances better. One such metric we are working on is Expected Goals, (xG) in short.  Please read through our previous article to know the basic context of xG.

Expected Goals (xG)

Football is a low-scoring game. More often than not, games are decided by a margin of 1 or 2 goals. All it takes is one shot, one fumble, one error in defensive organization and we end up with a goal. This low-scoring nature of the beautiful game brings in a possibility of variance wherein a striker may have been lucky to score a goal or unlucky not to. This variance may make us believe that a striker is good because of the number of goals he/she has scored or we may assume that a striker is bad because of the lack of goals he/she has scored. While such evaluations made by us may be true, it just might be the variance at play.

How many good chances did a team create? How many half-chances? Just how “good” were they? How many good chances did they concede, and so on? These are intuitive football questions. When you’re following a match, you’re watching the creation of chances, getting excited when it appears a scoring chance might be conjured for your team, or getting worried when the other team is building one. We all watch for something like “expected goals.” Managers and players create tactics aimed at creating good chances and preventing them from their opponents.

Expected goals is a metric for estimating the quality of chances that a football team creates or concedes in a match. Expected Goals basically give us the probability that any shot hit by an attacking player will lead to a goal. It is calculated by taking many factors such as distance from the goal, shooting angle, Type of Assist, etc., into account. A shot that has an xG of 0.3 means that the shot would probably end up as a goal 30 times out of 100 (in other words – 30% of the time).

Why it matters?

1. It is the single best predictor of future team performance than actual results, goals scored and conceded, or even shots scored and conceded.
2. Gives a more objective evaluation of a team’s defensive and attacking performance.
3. Helpful in identifying players for possible ”cheap” recruitment who have been underperforming in terms of goals scored but have good xG numbers.
4. Identifying possible strong areas of attack for the opposition and neutralizing them.
5. Enhancing attacking performance by exploiting areas of attack that might be of high xG value.

Case Study – Manuel Lanzarote:

Manuel Lanzarote’s performance in ISL 2017-18 was a classic example of a big overperformance. He scored 12 goals in 17 matches for FC Goa. On the surface, this number looks extremely good but if someone had looked at his shot map and xG values for the season, the story is a bit different. He had his xG value for the season at around 6 and his shot-map is displayed below.

ATK signed him in hopes that the numbers would stay the same but given this xG value and shot-map, it was likely that his numbers would fall the following season. That’s exactly what happened. Lanzarote “regressed to the mean” scoring only 5 goals for ATK in the ISL 2018-19 season.

There are quite a few tales of under-performance and over-performance in relation to xG every season. Armed with the tools of statistical and analytical modeling, better decisions can be made by coaches and recruiters alike to ensure that team performance is enhanced and improved and the aforementioned “variance” is gotten rid of as much as possible.

Here are a few xG values for top-scoring players in ISL 2017-18:

State of Development

We believe that performance evaluation metrics such as xG are the future of football and would help teams a lot in how they perform on the pitch. We are currently developing our xG model which is showing promising results and are looking to make the model better in the coming weeks. As of now, the accuracy is at 89.3% and the error is at around .24. With time, the numbers and predictions will improve.