NBA HD: Visualizing Shot Selection by Position

The positional revolution has gained a full head of steam over the past month.  Although talk of tearing down the walls of traditional positions has been going on for years, Drew Cannon’s brilliant article at Basketball Prospectus blew the discussion wide open and sparked a slew of articles from the game’s brightest writers and analysts.

Definition is the root of the issue.  What is a point guard? Besides height, what differentiates a power forward from a center?  Why do we call a player who can’t shoot a lick a shooting guard?

Here’s an attempt at quantifying those definitions from the shot selection standpoint. Using Hoopdata.com’s player shot location data, I’ve calculated the average shot location shares of each position (the positional designations on Hoopdata come from dougstats.com).

We want to outgrow the conventions of traditional positions but let’s see if we can observe divisions in the first place.  Hoopdata breaks down shot types into 5 buckets: at the rim (layups and dunks), <10 feet, 10-15 feet, 16-23 feet, and 3-point shots.  Here’s how the five positions look, in terms of percentage of shots in each location.  So what does a point guard’s shot makeup look like compared to a shooting guard? Where do we see the biggest disparities between positions?

Here we see that the typical point guard attacks the basket more than the typical shooting guard and then the basket attack trends upward with the following positions.  Most point guards work out of the pick-and-roll which lends itself to penetration to the rack or dishes to the rolling big.  They’re getting almost all of their at rim baskets on penetration as opposed to bigs who can get layups/dunks from offensive rebounds.

Looking further, we see that the mid-range jumper is the least populated area for shots but there isn’t much distinction between positions in the mid-range.  What’s also interesting is that point guards, shooting guards, small forwards, and power forwards all shoot the long two in similar doses, with centers only taking about 18 percent of their overall game from here.

From the 3-point line, it makes sense that shooting guards launch the most from deep and the centers the least.  Nothing too ground-breaking there.

Perhaps what’s most interesting is how similar point guards and small forwards are in their shot palette.  The blue and green bars are nearly identical with each other from zone to zone.  Below is the graph in table form along with the assisted percentages and field goal percentages from each shot location, courtesy of Hoopdata.

There’s plenty of good stuff in the table above but for now, let’s dig deeper and see which players get classified in a particular position but shoot nothing like their traditional brethren.  To get there, I calculated each player’s z-score (which, in simple terms, calculates the magnitude of deviation from the norm) compared to the positional mean from each shot location.  Then, I took the absolute value of those z-scores and summed each location together for the Zsum to get the final aggregated score.  Note: I only looked at players who averaged 20 MPG and played 20 games last season.

In the first table below, we find that Miami Heat point guard Carlos Arroyo deviates the most from the shot selection of a traditional point guard.  In particular, 65.3 percent of his shots come from long twos and he barely attacks the basket or launches from downtown.  His z-scores total to 8.19 which is the highest sum of the point guard bunch.  Perhaps is good that he doesn’t attack the basket, as he only converts on 47.8 percent of his tries which is far below new Charlotte Bobcat Shaun Livingston’s 71.4 percent success rate.

The first table displays the “Least Alike” players in the group and the next table shows the “Most Alike” which tells us who are the most protypical point guards in their shot selection.  Orlando point guard Jameer Nelson tops that list.  I’ll save the commentary for a later date but I found this to be a pretty interesting exercise.  Which players are positional contrarians? Find out below. (My apologies for the blurriness).

SHOOTING GUARDS

SMALL FORWARDS

POWER FORWARDS

CENTERS