Meet the Mapmakers Who Are Changing the NBA

When it comes to data analytics and sports

baseball has led the way.

The heart of the game is the one-on-one confrontation

between a batter and a pitcher,

and there's a clear start and end to every play.

Now think of basketball.

Players switch from offense to defense in a moment

and move freely all over the court.

To understand basketball, you need to understand space.

It's a mapping problem,

and that's why cartographer Kirk Goldsberry is one

of the most exciting researchers in the game.

Basketball is clearly a spatial sport.

What I mean by that is the court space is a character

in the play, and for the most part basketball analytics,

until very recently, had just ignored that fact.

Every year in the NBA, NBA players take

about 200,000 field goal attempts, and each

of those field goal attempts is accompanied

by an xy coordinate.

That is the key, key ingredient to the court vision product.

[Mark] Goldsberry's method starts with dividing the court

into 1,284 one square foot areas.

By tracking the shots taken and made by every player

in the league, he can establish a baseline expectation

for the value of a shot in a certain spot,

and then compare individual players to those baselines.

Sport View is this crazy, crazy system.

It's essentially a tracking system

for every moment of a basketball game.

It works by hanging six cameras

in the rafters of NBA arenas,

and then tracking the player locations 24 times per second

and the location of the basketball 24 times per second.

It's giving these guys' games sort of an MRI and exposing

where their real strengths and weaknesses are.

The data is so valuable that the NBA paid

for the installation of the cameras in every arena

in the league before the 2013-2014 season,

but now that all that data is available the trick is

to know what to do with it.

When I first got the data the first thing I did was,

sort of, look for people that could help me,

and so I targeted a few people

in the stats department at Harvard

that I knew were working on relatively similar projects.

The most exciting part of this research for me has been

being able to see a data set of this quality

for something that is this huge.

There are about 1,000 games in the season and 10 players

on the court, and each player has two xy coordinates

that are collected at 25 frames per second, so the number

of space time data points ends up in the billions.

The data comes in basically a big text file,

and then we read the data from that data base using one

of these programming languages, say R or Python,

and then I can work with that data interactively

and say pull up a possession and look at the positions

of all the players at a particular moment

in a particular game.

A lot of the time when people come across

a giant data set like this they think

that the insights are just going to sort of jump out

at them automatically.

In fact, most of the time you have to choose

what sort of angle you're going to take

to figure out how you're going to turn this data

into some kind of insight.

Expectant possession value takes a particular moment

in a basketball game and assigns it a point value based

on how many points we expect the offense to score

before they give the ball to the other team.

If you take a look at LeBron James,

he is one of the most effective scorers overall in the NBA,

but if you just look at his overall numbers it's deceptive.

Two years ago he led the league in both points

and field goal percentage in the paint, which is incredible.

Outside the paint he's more average.

He's not bad, but he's just average.

Whereas somebody like Kevin Durant

is really good everywhere, but he lacks

that really dominant aspect that LeBron has near the basket.

One of the things I think that's coolest

about this player tracking data set is

that it really is challenging us as scientists

to ask the bigger questions about movement.

Whether it's traffic, or whether it's the movement

of people in cities.

A lot of the concepts we're working on will inform

our future work in non-basketball domain.

Goldsberry's analysis opens up a new way

of evaluating everything that a player does on the court.

Just call it what Goldsberry and his team do, databall.