Is per worth the hype

STEPHEN SHEA
Summer 2017

Imagine that at the conclusion of an annual wellness visit with your physician, the doc proclaimed, “There is something wrong with you,” never to continue the conversation again. That would be health care’s equivalent to presenting a coach with his guard’s poor Player Efficiency Rating (PER).

PER usually refers to a specific formula created by John Hollinger and popularized by ESPN.com (while John wrote for them), but there are variations employed by coaches around the globe. While the specifics may differ, all versions of PER are similar. They combine mostly box score data into a single number that is meant to approximate a player’s performance.

Trying to evaluate a player based on the box score alone can feel like choosing a home based only on its square footage. The box score is limited, but if we flashback more than a decade to when PER was introduced, we find a sports world where the box score was nearly all statisticians had to work with, thus the motivation to extract as much as possible out of it.

The analytics revolution has championed two principles: the importance of efficiency and context.

For too long, players were judged based on quantity of performance. The best players were the ones that scored the most points, grabbed the most rebounds or dished the most assists. Early analytics acknowledged that quantity was important, but argued that it must be judged in conjunction with efficiency. It’s much more informative (and exciting) to know Rudy Gay scored 32 on 12-for-18 shooting than to know he scored 29 points going 11 for 37.

PER was intelligent in that it weighed both quantity and efficiency in its formula. Players scored the highest by doing the most and maintaining efficiency in spite of a larger workload. By including efficiency, PER provided a different perspective for coaches and fans. PER can function as part of a coach’s quick first glance at a player’s performance to see if anything jumps out that doesn’t jive with his or her review of the other stats. Here’s an example:

In 2016-17, JaVale McGee had a better PER than Steph Curry, John Wall, Blake Griffin, DeAndre Jordan, Marc Gasol, and Paul George (among many, many others). Does this mean JaVale is more valuable to Golden State than Steph? Should the Clippers rush to try and swap Jordan for McGee? Of course not. PER has its role, but its not a complete measure.

This brings us to the second major principle of modern analytics, the idea that production cannot be judged independent of context.

Examples of production include a player’s points scored or FG%. Examples of context would be how often the player is open when he shot or the types of shots he took. McGee scored well in PER, in part, because he shot 65% from the field. But a closer look at his context reveals that McGee only shoots when he has layups or dunks, and he only gets those opportunities because his teammates, such as Thompson, Curry and Durant, draw his defender for help when attacking.

The analytics community has always been aware that context influences performance, but it wasn’t until recently that it was able to assess a player’s situation. The key has been more-detailed data led by spatial-tracking efforts. PER doesn’t know what percent of a player’s shots are open. It doesn’t know how often a player draws a double team when driving to the hoop. It doesn’t know if a player is guarded by Kawhi Leonard or Kyle Korver. PER doesn’t incorporate the type of information that can now be gathered with spatial-tracking tools, and so, PER doesn’t account for context.

PER falls short in another manner. Suppose that you have won the choice of one among 10 boxes, each containing a mystery amount of cash. The game’s host offers to provide you with some more information on the boxes. He will either tell you how much money is in each box, or he will tell you the total sum of the money in all the boxes. Which would you prefer to know?

Arguably, PER is worse than getting the total cash rolled into one number because it’s output in a nonsensical unit. It’s as if the game’s host took the total in each box, ran it through a complicated formula and then told you there is a total of 27 goodforyas in the boxes. What’s a goodforya? He doesn’t know, but he’s sure 27 is better than 26. Well…it’s better most of the time.

PER’s formula was intelligent in that it included efficiency with production, but in doing so, it created a number so abstract that it’s hard to interpret.

Players must be evaluated on their performance in the context of their function. If a player’s role is to spread to the corner and keep their defender from helping in the lane on drives, then evaluate the player in that regard. He or she could help keep the lane open all game, improving the team’s efficiency without ever taking a shot. In that scenario, the player would provide a lot of value on offense while posting a terrible PER.

How do teams below the professional level get access to the data they need to properly evaluate their players? Take a look at ShotTracker TEAM, a new system that automatically captures complete stats for an entire team during practices and games. Not only does ShotTracker TEAM give you box score metrics, but also important measures for context like contested vs. uncontested shots, all in real-time.

PER is good for a quick glance. Perusing PER should be used in the same fashion as skimming the back cover of a novel to see if it might be an enjoyable read. The process isn’t perfect, but it’s better than judging the book on its title alone. It won’t be until you dig into the pages that you’ll discover if you’ve made a good choice or not. A player’s value isn’t found in a catch-all metric. It’s in the details of their performance and the success with which they function in their lineups and with their team.