There are a number of different well-known types of APM metrics, each of which uses slightly different techniques to reach their final output.
Let’s look at a simple example of how this sort of thing can be useful even for the casual observer: Quantifying a player’s defensive abilities has always been one of the toughest areas within analytics.Anyone with a keen eye and experience can get a good rough idea by watching players, and one can use NBA.com’s advanced on/off court logs to determine that a team tends to suffer defensively when one particular player steps on or off the court.A simple look at RPM rankings, or those of any other APM derivative, is not a surefire way of determining which players are “better” than others in any sort of concrete sense.These metrics have error margins, and even their creators would acknowledge that there will always be parts of the game they struggle to pick up entirely.The statistician can control for such elements as coaching, opponent, time in between games and more.
They can also use longer stretches of prior games to “inform” the model (the process of setting a concrete statistical baseline with which to compare subsequent players).
Last year’s prime example here would be Milwaukee’s Khris Middleton, who finished 10th overall for RPM despite coming into the year as a relatively unknown player.
A large factor was the way his impact on the court for the Bucks was consistently felt even when he wasn’t putting up traditional numbers.
Carmelo Anthony is another, finishing just 400th of 474 total players last season mainly due to a big negative figure defensively – confirming to many the prevalent opinion that he’s a one-way player and short of a true superstar for that reason.
A word of caution, one that could be applied to any single-number metric out there: This is not an end-all statistic.
No stat can track how well a guy contests a shot, or whether he pushes himself to 100 percent to get around every single screen set against him.