Defense is half of the game of basketball, but it has been difficult to gather information about the defensive capabilities of all the players in the league until very recent seasons. Due to the paucity of data about defensive performance and the limitations imposed by television broadcasting contracts, it was practically impossible to truly know who the best defenders were, night in and night out.
Offense, by contrast, is well-documented. The box score provides a good deal of useful information on individual offensive output, and even more granular data has been available throughout the 21st century by virtue of play-by-play data.
Ball Don’t Lie
In the words of the great Sheed: men and women lie, numbers can lie, but ball don’t lie. Even though it has proven difficult to evaluate individual defense, there is a truth worth seeking. Individual defensive performance impacts which team wins and loses each game. If we have any interest in understanding what happens in the NBA and providing actionable information for teams (as well as entertainment for fans), we must pursue any avenue possible to understand and model this effect. The box score may lie, but ball don’t lie.
Rasheed_Wallace
Some basketball fans still think of defense in terms of the plays recorded in the box score: steals, blocks and defensive rebounds. As I’ve written previously, steals are unreliable indicators of defensive performance. In addition, steals and blocks constitute a very small percentage of the possession in which a player participates (no more than 6-8% for even stellar defenders). I have also found that a large portion of defensive rebounds are uncontested and add minimal value on the player level.
These more familiar measures lie by inclusion and by omission. They tell half-truths by implying that players who make a few plays in a game are good defenders. But ball don’t lie. The point of defense is not to get a steal and a block each game; the goal of defense is to prevent the opposition from scoring.
Who Puts in Work?
What a defensive statistic needs to tell us, in order to be valuable, is: who prevents the other team from scoring? The first step toward answering that query involves making another: who prevents the player they’re guarding from scoring?
Some have used the field goal percentage allowed by a player as a proxy for this latter question, but doing so is fraught with obstacles. Evaluating only the percentage of shots made “against” the defender does not take account of the type of shot or the shooter, both of which are at least as important to the outcome of the shot as the identity of the defender.
In order to determine who does a good job or a bad job of preventing their opponent from scoring, it is necessary to evaluate the individual matchup data from nba.com in light of the defensive load each player carries. It is vastly more difficult to prevent LeBron James from scoring on a drive than it is to prevent Ben Simmons from scoring on a jump shot. I measure Defensive Load as way of recognize the differences between playing suffocating defense on a premiere scorer, as in the first clip, and helping off of a non-scorer to contribute, as in the second clip.
Compare this with the work of an elite help defender, Draymond Green, on a team with plenty of defensive depth:
As it turns out, the majority of defensive impact is made by players who carry a heavy Defensive Load. The combination chart below measures the amount of players at each level of Defensive Load (the histogram in orange). Each bar’s color represents the 50th percentile of per-minute performance of defenders in that group, compared with the other groups. The line overlaying the bars keeps a running count of the percentage of total Points Saved by all the players in the previous groups. As you can easily see, the players on the right side of the histogram (the high-load defenders) do the better part of the damage.
Load Management
What’s more, it is clear that the players who are appointed as their teams’ primary “stoppers” are much, much more successful in preventing their opponents from scoring than are other players. The chart below compares each player’s points allowed with expected points allowed. Players are clustered by Defensive Load, and the field is separated into “positive” and “negative” performers to enable us to see at a glance whether a given player allowed more points or fewer points than expected. There is a visible relationship between Defensive Load and effectiveness, and the effect is most noticeable for the cluster that is farthest to the right – the stoppers.
Calculating each player’s Defensive Load and comparing it against his team’s defensive depth and his effectiveness in preventing his opponent from scoring gives us a good baseline of how good a defender he is. As outlined in my previous work, we can use load-adjusted effectiveness to credit each player with an appropriate amount of points saved for his team by preventing his opponent from scoring. I call the result “Shooting Defense.”
Steals are NOT the Same as Good Defense
One of the quickest ways to prevent the other team from scoring is, of course, to create a turnover before they can get up a shot. In order to properly credit a player for forcing turnovers, we have to overcome another hurdle. Steals are often recorded for a player who didn’t truly force the turnover. One player pressures an opposing ballhandler and knocks the ball loose or forces him to deliver an errant pass. The ball is recovered or intercepted by his teammate, who gets credit for a steal.
None of which even mentions turnovers that are the result of offensive fouls; including offensive fouls gives us more information to utilize, but makes it more difficult to use steals as a unit of measure for defensive performance. In order to properly assign credit for forcing turnovers, we need to evaluate the turnovers committed by each player’s opponents, relative to that specific opponent’s propensity to turn the ball over and relative to the defensive team’s ability to force turnovers. The result is Non-Shooting Defense.
Non-Shooting Defense has three major components: Solo Steals (steals that are credited to the correct defender), Charges Drawn, and “Other Turnovers.” The chart below compares the total amount of turnovers a player forced with the number of those turnovers which were Solo Steals. Players with tall green bars are good pickpockets, while players with tall blue bars but short green bars are good at creating turnovers in other ways.
We can also benefit by examining the relationship between Solo Steals and Charges Drawn at the player level. While the chart below can be filtered to include all players or only those meeting certain criteria, perhaps its most interesting use is to identify the outliers who produce far more of one type of turnover than other.
The Best Defenders in the League
But what can you do with all that? What do the results tell us? First of all, it allows us to evaluate a player’s defense based on how effective he is at preventing his opponent from scoring, rather than evaluating the player on the basis of three plays per game (using the “stocks” method). In the table below, you can sort each season horizontally by effectiveness. Defensive Effectiveness is defined here as expected points allowed (based on the quality of the player whom the defender is guarding) divided by actual points allowed. Very few players are able to play significant minutes and record values far above 1.0, but those who do are rightly recognized as among the games’ best defenders.
As Childish Major says in “NoEyeInTeam,” defense is “not an I thing, it’s a we thing, it’s a team thing.” In order for a defensive metric to have value, it needs to be able to operate in the context of a team by explaining the strengths and weaknesses of the team. Even more important than telling us who the best defenders in the league are (which we will get to soon enough), Matchup-Based Defense can help us understand what makes a team good or bad on the defensive end.
Front offices, coaches, color commentators, and fans all want to explain why a team is performing the way it is, and we can do that by examining the data at the team level. The area chart tool will display as many selections as you like, but I recommend selecting one team in one season for the most accurate representation. The size of each player’s area represents the total amount of points he saved the team that season, which obviously depends on playing time as well as defensive ability. In order to represent that duality, the blocks are color-coded according to Defensive Load.
Team with large gray or red blocks tend to struggle on defense because they are relying too heavily on defenders who cannot stop quality opponents. Team with a lot of green blocks have good defensive depth, and some of their defenders may not look as strong on the leaderboard as expected, since they don’t have to do as much. Teams with only one or two green blocks may also struggle due to overburdening their key defenders. These trends and more will shine through quite clearly with even a cursory view of the data.
And Now It’s Time for the Main Event
Look, let’s be honest. You know what you came here for: you came here to see who this model says is the best defender in the league. You probably wanted to either a) win an argument on social media, b) find a better statistic for measuring defense so that you can win arguments on social media, or c) find a goofy leaderboard that fails the smell test so that you can hold analytics up to ridicule … in order to win arguments on social media.
I know it.
You know it.
So, without further adieu, here’s the list of the best defenders in the league since 2013-14, when the NBA began recording defensive matchups.
Well … kind of. This is the “hands-off” leaderboard, sorted by Total Points Saved per 100 Possessions. As you can see, the top 100 player-seasons includes a lot of centers, but also features one or more campaigns from Giannis Antetokounmpo, young Anthony Davis, Serge Ibaka, Paul Millsap, Jerami Grant, Andre Roberson, Kevin Durant, Danny Green, Michael Carter-Williams, Jrue Holiday, Robert Covington, James Johnson, Kawhi Leonard, Michael Kidd-Gilchrist, Josh Smith, and Patrick Beverley.
Why are there so many centers at the top? This happens when you don’t filter defensive data. Since they tend to get more blocks and have greater impact on opponent scoring, big men can dominate defensive rankings quite easily. Look a little closer at the “Relative Load” column, though, and you’ll find that most of the centers fall below 1.0. A Relative Load below 1.0 means that a player has a lighter Defensive Load than average for his team. He doesn’t have to guard scoring threats as often as his teammates do.
That’s not really what we mean when we ask “who is the best defender in the league?” though, is it? What we really want is an answer that considers only high-load defenders – pure stoppers. By selecting only players whose Defensive Load is at least 5% higher than average for their team, we find a new appreciation for the centers who remain (like Draymond Green) while also highlighting defenders across the defensive spectrum. In addition to the players listed above, we now find the following non-centers in the top 50 player-seasons alone: DeAndre’ Bembry, Jeff Teague, Josh Richardson, Mario Chalmers, Eric Bledsoe, Victor Oladipo, Wesley Johnson, John Wall, Maurice Harkless, Elfrid Payton, Marcus Smart.
Who Should Have Been Defensive Player of the Year?
If we further stipulate that a player must both carry a heavy Defensive Load and do it effectively (with a Load-Adjusted Effectiveness greater than 1.0), then the best defender in the league last year was Rudy Gobert. In the 2017-18, the data supports Jrue Holiday as the rightful DPOY, contrary to the actual award going to Gobert. In 2016-17, the “real-life” DPOY Draymond Green was in a dead heat with Andre Roberson in Total Points Saved, though Roberson was better per-possession.
Kawhi Leonard was the runaway Defensive Player of the Year in 2015-16, both in reality and in the model. Kawhi’s first award in 2014-15, by contrast, should have gone to Draymond Green or Kawhi’s teammate, Danny Green. I guess the voters weren’t seeing green that year …
A bad joke, I know, but it makes for a nice segue into the final award to give: 2013-14 Defensive Player of the Year.
Do you remember the 2013-14 season? A lot has changed since then, and I wouldn’t blame you if you struggled to put together a coherent image of the what the league was like at that time. The best high-load defender in the league in that season was Josh Smith – yes, that Josh Smith. The one you remember as a past-his-prime chucker.
Back in 2013, though, Josh Smith was a dangerous weakside shot-blocker and a versatile combo forward. John Wall and Mario Chalmers also have solid arguments for this season, but I have to wonder if this is a case where the filter misleads us. If we reduce the requirement from a minimum Relative Load of 1.05 to a minimum of 1.0, we see that Serge Ibaka, Anthony Davis, and Dwight Howard all rank above Smith for the season. The award went to Joakim Noah, who was a terrific rim protector but whose Load-Adjusted effectiveness was below 1.0.
Here is the filtered leaderboard, for those who are interested in determining the best defensive seasons since 2013-14.
How Is My Favorite Player Doing?
Say that you want to evaluate the defensive performance of a certain player, instead of just looking at the best defenders. Fans care about the players on their team. Coaches and scouts primarily want to evaluate their own players and upcoming opponents. In fact, while NBA front offices are always in search of the best players possible, making good roster decisions requires detailed information about every player – good, bad, or in between. I’ve released a tool on Google Sheets for precisely this purpose. For any player-season between 2013-14 and 2018-19, you can select the player and season to instantly view their defensive performance measured by Matchup-Based Defense. Here is the link.
The tool also provides several different percentiles for each player. Using some NTILE (100) OVER (PARTITION BY … ORDER BY …) clauses in SQL Server, I was able to determine each player’s per-possession percentile when compared against 1) all players in all seasons, 2) players who play the same position in all season, 3) players who play the same position in the same season, and 4) players at all positions in the same season. This functionality comes in handy if you want to know how good or bad a player was for his context.
Conclusions
There is a way to determine which players provide the most defensive value. In this post and its predecessors, I have described the type of analysis required to answer the types of questions we want to answer. The next article in this series will provide an update on the defensive performance of every player in the league this season. Later on, I will also do some data dives in previous season’s data to identify the players who should have been named to All-Defensive teams, players who were underrated/overrated, and players whose defensive play significantly impacted their teams’ fortunes. Until next time, make sure to leave your responses in the comments!
Great article Greg! Although overall I found the graphs a little complicated. You should try to produce a glossary or something (if this already not exists). Nevertheless, great job! Keep it going!