First, it’s important to lay down a framework for defining and understanding the numbers that are involved in sports analytics. Traditionally, we refer to the numbers recorded in sports as “stats.” The problem with using this term in the context of sports analytics is that “stats” is already shorthand for the field of statistics, which is one of the many disciplines that sports analytics draws many of its practices from. So, we’ll refer to a single type of data as a metric and a single value of a metric as a data point. These are not terms unique to sports analytics by any means, but it will be helpful to use these designations as the topics get more complex.
While metrics will be presented in an assortment of formats and can be the result of complicated calculations, there are three fundamental types of metrics to be familiar with: counts, measurements, and points.
Counts
A count is the most basic kind of metric. It is, as it sounds, a count of the number of times something happens within the rules of its respective sport. Examples include 3-point attempts in basketball, passing attempts in American football, singles in baseball, shots in hockey, and so on. It and points metrics are the only metrics that are recorded as a direct result of the rules of the game that’s being played.
While there may be more nuanced sports conversations one could have about the total number of times somebody did something (i.e. a basketball player taking too many or too few shots), the data points themselves are particularly pure and void of complication.
This is because the only possible bias in count metrics is whether the actual recording of the event was done in error. This can be more or less of a concern depending on the subjectivity of the event. For example, in basketball, if the player takes a shot from behind the three-point line, it is recorded as a three-point attempt. The only possible bias in that data point is if the player actually had his foot on the line but nobody noticed. A more subjective example would be a turnover in hockey, because possession of the puck is not always obvious. If a hockey player is being pressured by an opposing player and lets the puck drift away from them, but one of their teammates ends up being there to grab it, one scorer may count that as a pass and another may count that as a turnover and then a recovery. Both a pass and a turnover are events allowed by and defined in the rules of the game.
Note that metrics that count the number of times an event occurs that is NOT outlined in the rules of the game, and is therefore some other subjective distinction made by an individual, should not be treated as a count metric. An extreme example of this would be to watch a baseball game and count the number of times the third baseman makes a good or a bad throw to first base. This is something that is obviously subjective and doesn’t fit the framework of a count metric that we laid out. Categorizing throws from the third baseman as good or bad based on some information is something else entirely, but doing it merely on an individual’s intuition would place it in one of many categories that will be discussed in later sections.
Measurements
A measurement metric is a quantitative measure of something that occurs during a game. Examples include shot distance in basketball or hockey, average receiver speed in American football, and spin rate of the ball in baseball. The main distinguishing feature of measurement metrics is that they are direct measurements of tangible actions, not occurrences of players satisfying a condition outlined by the rules.
Perhaps the most important quality of measurement metrics is that they are perfectly and completely objective but also the most void of inherent value. They are objective because they are a pure measurement of a physical action. It is not reliant upon the rules or any human recording something, these measurements are done with computers that are programmed to record measurements. That being said, because they are perfectly objective and not concerned with the rules of the game being played, they are also void of any inherent value. This often makes it more difficult to work with and use measurement metrics, because an individual has to analyze massive amounts of measurement metric data, along with their own intuition, in order to contextualize the individual values or averages drawn.
To use NBA basketball as an example, the NBA arenas track the GPS location of every player on the floor. From that raw tracking data, you could compute the average distance between the player and the hoop when they are on offense. And say Player A has an average distance from the hoop of 10 ft. That data point has no inherent value. It requires basketball context in order for it to become meaningful. If we were to say instead that Player A is a point guard and is, on average, 2 ft. closer to the hoop than the league average for all other point guards, now we can use it to start to draw conclusions about Player A. Maybe that suggests that they attack the basket more than the average point guard. Or maybe it suggests they occupy too much space near the rim, so much so that their team’s interior scores don’t have the space they need to score inside.
So, while every metric requires context to be fully understood, measurement metrics are different in that they don’t just require context to be understood, they require context to hold any value with respect to the game being played. If Player A makes five 3PT shots per game, that means that Player A scores 15 points per game from 3PT shots, which contributes directly to their team’s score in that game. Being ten feet from the hoop does not directly have any effect on the game as specified in the rules of the game. Measurement metrics offer insight into a situation or relationship, but rarely offer the entire story. They need to be collected along with other information in order to put together a more complete picture.
Acknowledging that an analyst has to perform some process in order to give a measurement metric or a metric based on measurements any usefulness in the context of the game being played is essential in using metrics to better understand the game. Knowing what you can and can’t draw conclusions from is instrumental in sports analytics.
Points
Points aren’t a category of metric as much as they’re one single metric that exists slightly differently in every sport. The reason they need to be defined as one of the three fundamental types of metrics is that they cannot be treated or thought of as either a count or measurement metric. Points are, as one could assume, the number of points scored by either a team or an individual. This is not the number of times a team or an individual scores — which would be a count metric — but it is the actual number of points scored. So, if a field goal kicker in American football makes 2 field goals, the number of field goals made would be 2 but the points scored would be 6.
At first, it may seem trivial to make this distinction, but it’s important to consider the major conceptual differences between count metrics and points metrics. One such difference is that there is no inverse to a points metric. You can count the number of times a kicker makes a field goal and the number of times a kicker misses a field goal, but you can’t really count the number of points they didn’t score. This comes into play in analysis when thinking about efficiency. A soccer/football player may score more goals than anyone else (total goals being a points metric) but if they achieve that feat by taking a huge number of reckless, low-quality shots, that context would not be captured by the points metric alone.
Another important conceptual difference is that points metrics, depending on the sport, can be counted in increments other than 1. By definition, counts metrics go up by 1 whenever the given action is recorded, but points are counted at the liberty of the rules of the game. This becomes especially important conceptually when you consider metrics that are based on or utilize points metrics in their calculations A common example being points-per-game-like metrics, which exist in some form in essentially every sport. A basketball player can average 30 points per game by accumulating points from any amount of three-pointers, two-pointers, or free throws. All three of those actions correspond to different point values.
Co-founder of First Line Sports Analytics