Basketball is a game of matchups. The offense is constantly moving around and looking to force switches that can create matchups they can exploit. If a defensive liability like JJ Redick is on the floor, the opposing offense will certainly look to attack him as often as possible. Meanwhile, shifty point guards like Chris Paul will hope to draw slow big men onto them on the perimeter where they can abuse their speed advantage.

Here’s what happens when the offense forces the switch and creates a mismatch:

DeAndre Jordan is a good defender against other big men, but he isn’t versatile. Certainly not enough to be able to guard Stephen Curry on the perimeter. As you’d expect, it was an easy blow-by for Curry and Jordan was simply forced to intentionally foul him.

This is why defensive versatility is so important. It is far more difficult for the opposing offense to find a mismatch if the defensive players are able to guard multiple positions. Traditional defensive metrics (like field goal percentage allowed) don’t properly represent the diverse skill set of these players even though it’s a huge part of their value on defense. Let’s try to tackle that problem.

The NBA Stats website offers comprehensive head-to-head match up data for the past three seasons. The question: how should we actually use this data to quantify the versatility of defenders?

People in the past have attempted to go about this using the number of possessions a player guarded each position. For example, let’s say a player defended guards for 30 possessions, forwards for 60 possessions, and centers for 10 possessions. We could use the following chi-squared goodness of fit formula to measure their versatility:

In this case, the expected value is 1/3 because there are three options (guards, forwards, centers). If a player was perfectly versatile and defended every position for an equal amount of possessions, their versatility index would be 0. If a player spent every possession guarding a single position, their versatility index would be 2. The lower the versatility index, the more versatile the defender is.

I went through this calculation for over 300 players from this season and noticed that there was a clear trend in the results.

Using this methodology, the most versatile defenders are just the tallest ones. The least versatile defenders are the shortest ones. While I think that this isn’t completely off base from the truth, the correlation does seem a bit extreme. According to the coefficient of determination for this relationship, 70% of the variance in defensive versatility can be predicted purely from a player’s height. There must be more to it, though, right? Is Clint Capela *really* the third-most versatile defender in the NBA?

Let’s try something else. I used the regression line to determine each player’s versatility versus their expected versatility based on their height. For example, Marquese Chrisss had the lowest versatility index in the league at 0.0439. Based on his height of 81 inches, he had an expected versatility index of 0.3156. The difference between the two is 0.2717, which we’ll try to use to represent his versatility. In this case, a positive value indicates a player who is more versatile than expectation while a negative value represents a less versatile player, so a higher number is better.

In other words, we’re basically creating the residual plot of the above graph:

The top of the graph has some names that you’d actually expect to see. Chris Paul, Kyle Lowry, and Patrick Beverley don’t really have the size to guard bigs, but they’re certainly defensive bulldogs who are frequently matched up against forwards. Some other plots that I didn’t label near the top of the graph include Draymond Green, James Harden, and Marcus Smart, who are all also versatile defenders.

While this residual plot seems to be an improvement over the original methodology, there are still some glaring issues. Isaiah Thomas is arguably the biggest defensive liability in basketball, largely due to the fact that he’s shorter than the average American adult male. The idea that his defensive versatility is well above average is absurd. Likewise, it’s peculiar that Giannis Antetokounmpo and Jonathan Isaac, two of the league’s best wing defenders, are also the league’s two least versatile defenders according to this methodology.

So, this strategy is clearly quite imperfect. I would’ve expected Giannis to be one of the league’s most versatile defenders and Isaiah Thomas to be one of the least. Why is the opposite true according to our current technique? Remember that we’re looking at the number of possessions a player guards each position — not how well they perform in those possessions. Just the total volume.

Isaiah Thomas is the type of player who opposing offenses consistently attempt to get their best players to switch onto in order to create favorable mismatches. In other words, maybe the actual reason he guards other positions for more possessions is because he’s *not* versatile. Also, our use of a residual plot in the case of Isaiah Thomas may be ineffective because he’s the shortest player in the dataset by a fair margin, so the linear extrapolation may not be very accurate.

Meanwhile, Giannis Antetokounmpo plays alongside two NBA All-Defensive candidates in guard Eric Bledsoe and center Brook Lopez. He’s probably not asked to guard opposing positions as much even if he’s capable of it.

I think this demonstrates the problem with the common strategy of measuring versatility based on how often a player guards different positions. This can tell us about strategy and a player roles, but it isn’t a good indicator of actual ability. In the next article on this subject, I’ll try to use a metric for evaluating actual defensive ability in conjunction with the positional matchup data.

## Comments