Putting Back the “Kobe Assist”

Last week Kirk Goldberry published his latest research on Grantland featuring a new metric he has named the “Kobe Assist”. The idea is that some players are good at missing shots in a way that falls into the hands of their teammates more often than their opponents, as if on purpose; as if this were a hidden skill only visible through some new data analysis.

In contrast to his fascinating insights on shot location published in the New York Times before the NBA Finals, this “Kobe Assist” research is composed of over 3,000 words with only two images and one video anecdote. While his presentation was disappointing, what really saddened me were the two wrong turns Goldberry took when he was oh-so-close to one of the most powerful new visualizations in all of basketball: showing the relationship between where a player shoots and misses and where the rebound happens. Now that would help you win games!

Wrong Turn #1 – Premature Conclusion from Scant Evidence

Instead of continuing down the road of his sound data science methods from before, Goldberry veers off to cherrypick anecdotal evidence from a small dataset and jump to a conclusion about Kobe. He does not offer systematic evidence to show that some NBA players possess a skill for missing shots that create quick putbacks for teammates. How consistent is this skill over time? More importantly, what other factors (like teammate offensive rebounding ability) need to be accounted for to know it is not them but the “Kobe Assist” at work? Without that systematic evidence, there is no science yet supporting the claim for a “Kobe Assist.” Goldberry could merely be finding evidence that Kobe misses a lot of shots on a team with big men like Bynum, Gasol, Odom, and Howard who are exceptionally talented at rebounding them.

For instance, systematic analysis has shown that 90% of rebounding can be explained by how well the players rebounded the year before (see Stumbling on Wins). While Kirk does cite things like two year league-wide averages and that Kobe leads the league over that period, he doesn’t make a systematic breakdown of it. Looking over a dataset, there are usually outliers (Lamar Odom in 2012 on Dallas for instance) but an outlier neither proves or disprove a dataset. And unfortunately, we don’t see Kirk’s dataset or any systematic approach, which makes the Kobe Assist harder to swallow.

Wrong Turn #2 – Seduced by “It’s Complicated”

Goldberry’s second wrong turn was falling for the “beautiful complexity” argument of basketball. This common argument says that basketball is a dynamic team game in constant motion, and therefore the boxscore statistics recording discrete events for each player (e.g. a rebound, an assist) do a poor job of modeling what really happens and why. In contrast to baseball which is ‘basically an individual sport,’ Goldberry and others argue that too much gets missed in the NBA to trust those boxscore numbers.

This neat rhetorical slight-of-hand allows Goldberry to shift some of the credit for the points Dwight Howard scored in that video clip back to Kobe Bryant when we aren’t looking. I understand the desire to view basketball as an art that takes deep knowledge to understand, but I also think the “Kobe Assist” is very sloppy thinking and really bad science.

As it turns out, NBA basketball is not that complicated. Unlike global weather patterns or chaos theory–both of which Goldberry alludes to–the effort to explain basketball by modeling it scientifically has two enormous legs up on the natural sciences: the rules dictating how basketball works are written down in a book (the rule book!) and detailed observations of events are recorded in an extensive log (game statistics!).

Armed with these two datasets and solid understanding of economic methods for simulating controlled scientific experiments, we now know winning basketball games boils down to a team’s skill for:

  1. Turning possessions into points as efficiently as possible — defined as offensive efficiency which boils down primarily to high percentage shooting and avoiding turnovers
  2. Helping their opponent waste their possessions as much as possible — defined as defensive efficiency; primarily low percentage opponent shooting and forcing turnovers
  3. Getting the ball when it is up for grabs — primarily rebounding missed shots, because it is awfully hard to score points when you do not have the ball!

To Goldberry and others this kind of statement seems to be a heresy by researchers following old school economics methods that reduce a beautifully nuanced and complicated game of mano-a-mano into a sterile spreadsheet. While we agree there are better methods for presenting these insights than a boring old table, these insights above come from a scientific model capable of explaining what wins basketball games (at least 95% of it!).

Let’s look again at Goldberry’s primary anecdote of one possession and think it through again:

  1. Bad: Kobe Bryant spends the Lakers’ possession without scoring any points.
  2. Good: Dwight Howard regains possession of the ball.
  3. Good: Dwight Howard quickly spends this possession by scoring points.

The credit for that bucket was rightly recorded to Dwight Howard in the boxscore for that game. The argument that Kobe Bryant deserves some sliver of the credit is still possible, but totally unconvincing given Goldeberry’s evidence so far. Was it something about the timing of Kobe’s shot? Does he exhibit a pattern his teammates are able to cue in on to position themselves better?

Even if a systematic model eventually demonstrates there is such a thing as a skill for the “Kobe Assist,” the amount of credit it can take from Howard and pass to Bryant is extremely small. And indeed Kirk mentions that a majority of Kobe’s misses turn into defensive rebounds, which is good for the opponent. So Kobe’ misses are still much more bad than good.

Goldberry’s Near Miss

I mentioned at the top that part of what frustrated me so much about this research was how Goldberry’s two wrong turns caused him to narrowly miss a powerful visualization that could explain something valuable about basketball.

The relationship between where a player shoots and misses and where the rebounds happens could reveal to teams not only which shots go in the most, but which shots are most likely to be kept and put back for points. Goldberry already presents shooting data and rebounding data by precise location. Why not combine these?

Rather than chase the “Kobe Assist” we would plead with Kirk Goldberry to take the chocolate and the peanut butter from his research and combine them into something amazing. If we know both where a player shoots and which missed shots his team mates are most likely to turn into easy buckets or new possessions, we would know exactly where he should shoot, where he shouldn’t, and where his teammates should position themselves to secure more rebounds. That kind of insight could be coached as an effective strategy to help a team win a lot more games.

The ability to analyze statistics, discover legitimate scientific insights, and then present them understandably is hard. Kirk Goldberry has done an amazing job presenting analysis around shot and rebound location. While I think he airballed on the “Kobe Assist,” I also think the path his prior work was on could produce insights capable of transforming the game!

Dre (with several Steve Nash assists from Jeremy)

Comments are closed.