Defensive rating doesn't measure defense
Defensive rating — the number of points a player or team allows per 100 possessions — is the cleanest-looking advanced defensive stat in basketball. It's printed on Basketball Reference, cited on broadcasts, and quoted in MVP and Defensive Player of the Year debates as if it were a verdict on how well an individual player guards the floor. It is not. The player-level defensive rating is a team statistic with a single player's name attached, and the gap between what the stat is doing and what people think it's doing has been distorting defensive evaluation for two decades.
What the stat actually computes
A player's defensive rating is calculated by taking the team's defensive points-per-possession allowed while the player is on the floor, then making a small adjustment for individual events like steals, blocks, and defensive rebounds. The "small adjustment" is the operative phrase. The overwhelming majority of the number is determined by what the other four defenders on the floor did, plus what the offense happened to attempt, plus how the team's rotations and coverages were set up by the coaching staff. The player whose name is on the stat is one input among many, and usually not the biggest one.
This is why two players on the same defense usually have defensive ratings within one or two points of each other. The five-man unit is the actual defensive entity. Splitting the credit for the unit's results into five equal-ish slices and printing those slices as if they were individual stats is a format convention, not a measurement. The defensive rating column on a stat page is, mathematically, the team's defensive rating with a small individual nudge attached.
The lineup-correlation problem
A useful test of any "player" defensive stat is to ask whether it moves when the player changes teams. Player defensive rating fails this test badly. A defender who posts a 105.3 on a top-five defense will, after a trade to a bottom-ten defense, typically post a 113.8 the next year despite no observable change in his defensive play. The reverse is also true: a defender on a bad team who gets traded to a contender will look like he discovered defense in the offseason. He didn't. The team around him got better, and the stat that bears his name moved accordingly.
The honest interpretation of this pattern is that the defensive rating column is mostly capturing the quality of the team, with a thin layer of individual signal on top. The thin layer is real, and it can be teased out with enough careful adjustment, but the raw number on a stat page does not perform that adjustment for you. The all-in-one DPOY-style metrics built on top of defensive rating inherit the same problem and then add their own adjustment layer, which is why their year-to-year stability for the same player is worse than for offensive stats of similar construction.
The shot-quality confound
Defensive rating credits a player for shots the offense missed. Many of those misses had nothing to do with the player or his teammates. Across an 82-game season, the shooting variance of opposing offenses against a given five-man unit accounts for roughly 30-40% of the unit's defensive rating swings. A unit that allows the same shots in the same locations to the same players will give up a different points-per-possession total depending on how well the offense happens to shoot in any particular sample.
This is why "defensive rating against" can change five points in either direction across consecutive months for the same lineup playing the same schemes. The defense didn't get better or worse. The opponent's threes started or stopped going down. The stat is downstream of finishing variance the defense doesn't control, and the player whose name appears on it gets credit or blame that belongs to opponents he didn't even guard.
The minutes-played selection bias
The players who post the best defensive ratings in any given season are usually role players on top defensive teams who play primarily against second-unit offenses. The starting backcourt of a contending team posts a worse defensive rating than the eleventh man on the same team, not because the eleventh man is the better defender, but because he plays his minutes against opposing benches while the starters draw the opposing starters. Opponent quality cuts the same way for defensive rating as it does for any other on-court stat, and the standard reporting format does not adjust for it.
This is the source of most of the famous "advanced stats love him" moments in NBA discourse. A defensive specialist who plays deep bench minutes against the dregs of the league posts a 102.4 defensive rating, the highlight chart goes around, and the argument gets made that he's a top-five defender in basketball. He isn't. He's a passable defender who plays against the wrong people. The stat doesn't see the difference.
What does measure defense, approximately
Three categories of defensive stat do better than raw defensive rating at isolating the individual contribution. The first is defensive on-off, which compares team defensive performance with the player on the floor versus off it. This still has lineup- correlation problems but at least controls for who's on the court. The second is matchup data — what specific opponents shoot when this player is the primary defender on them. Tracking data has made this available for over a decade. The third is the film-derived stats around shot contests, deflections per possession, screen navigation, and recovery rate, all of which capture defensive actions that an individual player demonstrably performed.
None of these is a clean one-number summary. They are noisier, less stable, and harder to put on a broadcast graphic. They are also closer to actually measuring defense than the rating on the back of the player's card. The teams that win on defense in May and June are usually the ones whose front offices have stopped treating defensive rating as the primary input and started treating it as one piece of a stack.
How to read it
A player's defensive rating is fine as a hygiene check. A starting wing on a contending defense should post something roughly in line with his team's overall defensive rating. A wide gap in either direction is worth investigating, but the gap itself doesn't tell you whether the player is the cause. It just tells you that there's something to investigate.
The mistake the discourse keeps making is treating defensive rating as if it were the basketball equivalent of ERA — a clean, individual measure of how much the player gave up. ERA is already a noisy team-influenced stat, and at least pitchers face one batter at a time and personally produce most of the events the stat counts. A basketball defender is one-fifth of a unit that is being attacked by an offense whose decisions and finishing have far more to do with the points-allowed total than the defender's individual play. Treating the on-court number as if the defender produced it is the central confusion that defensive rating creates and that the more honest defensive stats have been trying to clean up.