THIS PAGE IS UNDER CONSTRUCTION.
Shot-based stats and metrics
Total Shots Ratio (TSR)
It expresses the dominance of a team on the basis of shots taken. TSR is useful, as it has a high correlation with goal difference and points. Moreover, it is repeatable across seasons and can be used to predict team performance.
However, TSR is biased. It only counts shot quantity, and not shot quality. For example, a team that took 5 shots in the 6-yard-box will be rated the same as a team that took 5 shots from the halfway line. Surely the first team’s better going forward than the second?
Expected goals solves this problem.
It doesn’t stand for anything. It’s the online handle of Brian King, the maker of the metric, who used it in hockey. PDO measures luck. The peculiarity of PDO is that PDO numbers regress to the mean. A high score means your luck will soon run out and a low score means you’ll put up better results soon. The way to measure PDO is 1000(Shot on target conversion % + Save %). The average PDO is 1000, and all teams’ PDO regress to that number.
However, an issue with PDO is that it only takes shot quantity into account and not shot quality. To read a critique of PDO, take a look at Ted Knutson’s excellent article at Statsbomb.
Expected Goals (xG, ExpG or xGoals)
Football’s favourite stat. Usually, it is shortened to ExpG or xG. It is a model that assigns a goal probability to a shot. For example, a shot from the six-yard box will have a higher chance of going in compared to a shot from 30 yards out. So, the xG for a shot from the six-yard box will be higher than the one taken from 30 yards out.
Many people have their own xG model. Although shot location is the most important factor, different people use different criteria, such as whether the shot was a header, whether it came from a cross, game state, whether it came from a counter-attack and so on. The most popular is Michael Caley’s, which uses many, many variables. Paul Riley, Colin Trainor and Constantinos Chappas, and Sander Ijtsma have good xG models of their own as well.
The main reason for which the stat is frequently used is because it takes into account not only shot quantity, but quality as well. xG is also versatile, as it is used for prediction, evaluating forwards, and evaluating goalkeepers. Here are a few metrics that are derived from xG:
Expected assists (xA)
The amount of assists a player should have played. A few new ways of calculating this have emerged.
Individual player stats
Per 90 minutes (p90)
Used instead of per game or totals in stats. Most young players get less game time compared to players in their later twenties. So, dividing player statistics by games played will favour the older players. It’s better to divide the stats by the number of minutes played and then multiplying that by 90. This gives a more accurate measure of skill.
Shot assists (or key passes) + goal assists.
A non-penalty goal is a goal which wasn’t from a penalty. Why should we look at NPG90? A penalty shot has an 80% chance of becoming a goal. Though some players are ‘penalty specialists’, like Franck Ribery, Mario Balotelli, and Cristiano Ronaldo, many players can score from the penalty spot. If we use goals to evaluate players, we’re also using the easy penalty goals that anyone can score. Hence, it’s rational to exclude penalties from goal numbers.
Possession-Adjusted Defensive Statistics (Padj)
The problem with defensive stats is – okay there are a lot of problems with individual defensive stats. But one of them is the fact that teams that keep the ball better will score less on on-the-ball defensive actions like interceptions and tackles. So, first a base possession rate of 50% possession should be divided by the amount of possession the team in which the defender we’re looking at conceded. Then, that number should be multiplied by the number of tackles or interceptions he made. So if a team controlled 60% of the ball, and one of their center-backs completes 10 tackles a game, his possession-adjusted tackles number is 12.5.
(This page may be updated once in a while).