Outside the Box Score: How We Count Affects How We Think

No matter how hard we try to avoid it, the way stats are counted naturally drives our thinking about the game of basketball. Nowhere is this more apparent than in the consistent split between how offensive and defensive contributions are evaluated and compensated in the NBA. This phenomenon is pretty easily explainable. Most things that are measured, counted and tracked focus on the ball.

Shots, assists, rebounds, blocks and steals are the bricks and mortar of much of basketball analysis, and they are all about people doing things with the rock. Even as we move beyond box score-based metrics, those labels still drive our understanding.

Here at Nylon Calc, Krishna Narsu has been working on perfecting his measure of shot difficulty. A matter of continual internal discussion is whether to use a single model to account for the difficulty of all shots or to split it up into separate algorithms for various kinds of shots. A stumbling block has been that last word: shots.

Since they are all shots – field goal attempts. Shouldn’t they be considered the same thing? But a dunk and a three point jump shot are lumped together more by conventions of labeling than real similarity in the activities. The absurdity of the category is demonstrated by the fact that the most efficient ‘shooter’ of all time can’t actually shoot. It might well be that the factors affect a layup the same way they impact a midrange pull up and/or a catch and shoot three, but that’s an empirical question, and the argument they all should be treated similarly simply because “they are all shots” is an exercise, albeit an understandable one, in question begging.

One of the tougher problems in basketball analysis is moving from the team to the individual level. On a team level, a defensive rebound is a very good thing, and positive in a way that is easily[1. Relatively speaking, of course.] quantifiable. However, all of the value from that defensive rebound is not created by the rebounder. Someone contested the shot, someone boxed out, and maybe the ball just bounced right to a guy. But that value is still largely credited in our minds’ eye to the rebounder because he got the stat. Similarly, blocks and steals are assigned value rather than things like shot contests or deflections and loose ball recoveries because those are the stats that were chosen to be recorded.

Even moving into more modern charted and tracked metrics, how we count and categorize things affects perceptions of value and importance. The quest for efficiency is definitely a good thing, but if we choose the wrong endpoints for analysis, it can go horribly wrong. “Isolations” produce inefficient shots, because creating shots against set defense is hard. But an “isolation play”[2. As tracked by Synergy.] is only one result of what we would more colloquially refer to as an “iso.”

Maybe LeBron isos on the left wing[3. Picture any Finals possession…]. If he ends up pounding the air out of the ball and shooting a jumper, that’s probably a bad possession. But what if he hits an open shooter for a spot up J, or Timo Mozgov on a cut under the rim after drawing help defenders? “Just run that play where they don’t guard the tall guy under the basket again” doesn’t seem like particularly useful advice, so you have to take the good with the bad when the play is examined as a whole.

Let’s take for a moment a made up and simplified, but ultimately realistic example. Say the Pelicans are running a spread pick-and-roll attack with Tyreke Evans working off a ball screen from Anthony Davis, with Eric Gordon spotting up. For purposes of this example, the only three possible outcomes are Tyreke barreling to the basket, him hitting Davis on the roll, or finding Eric Gordon for the spot up. I’ll further assume that when this play is run, 60% of the time it ends with Tyrke finishing the play[1. Either shooting, getting fouled in the act or turning it over.], 25% of the time he drops it off to Davis, and 15% of the time the ball finds Gordon on the kick out.[2. This is actually slightly conservative, as 31.4% of plays ending via pick-and-roll saw the roll man shoot as opposed to 29.4% in this example.] We’re ignoring the other two players and things like offensive rebounds, since Synergy “plays” don’t include possible offensive rebounds

According to Synergy, Evans scored .8 points per play  as a ball-handler in a pick-and-roll. While this is slightly above average for P&R guards[4. Leaguewide average was .79 points per play in 2014/15.], it’s below average for all halfcourt plays[4. .83 points per play for the league on non-offensive rebound plays.] Why would you run such a negative play? The answer, as should be obvious from the set up, is that those other good outcomes can arise from the Evans pick and roll. As the graphic below demonstrates, the play in it’s entirety is actually better than average.

Even with the low efficiency Evans shot taking up the majority of the outcomes, the chance of one of the other positive plays is enough that taken as a whole, the Evans/Davis pick-and-roll suddenly becomes a great offensive option! But if we just focused on the smallest detail – the times Evans shoots, it would look like the worst offense in the league. This isn’t to say you wouldn’t hope Evans was better, either at finishing himself or, more likely, by being better at finding teammates. Still, by only counting the plays he himself finished when measuring his contributions, we’d be missing the forest for the individual metric trees.