On a Tuesday night in late January 2023, Russell Westbrook posted 12 points, 6 rebounds, and 11 assists for the Lakers. The headlines, dutifully, called it a “near triple-double.” The same night, Mikal Bridges scored 38 on 17 shots in Brooklyn. One of those box scores looked Hall-of-Fame impressive. The other looked like a quiet wing’s good evening. The analytics, gently, asked us to pay attention to the second one.
This is the unsolved problem of modern NBA coverage. The box score, invented in the 1850s for baseball and grafted onto basketball later, was designed to record events. It was never designed to measure efficiency. You can have a stat line that looks like a season and a true shooting percentage that looks like a streetball pickup at the YMCA. Advanced stats exist because, somewhere between 2002 and 2014, a generation of analysts got tired of pretending those were the same thing.
If you have read our general guide to sports analytics, this is the basketball-specific follow-up. The metrics below are the ones that survive five years of arguments. The order matters.
True shooting and effective field goal percentage
The two simplest, and the two you cannot skip.
Effective field goal percentage (eFG%) adjusts standard field goal percentage to reflect the fact that a three is worth more than a two. The math is straightforward: (FG + 0.5 * 3P) / FGA. A player shooting 40% from three is generating the same point yield as a player shooting 60% from two. The eFG% reflects that. Field goal percentage does not.
True shooting percentage (TS%) goes one step further and folds in free throws. The formula uses 2 * (FGA + 0.44 * FTA) as the denominator, which is a rough estimate of total scoring possessions. The output is a single number that captures how efficient a player was, given every shot and every foul drawn. League average TS% in 2024-25 hovered around .576. Elite scorers — Curry, Durant, Jokic — live north of .620 in good years. A volume scorer at .540 is, in plain terms, hurting his team’s offense relative to a competent role player at .580.
The reason both matter, and not just one, is that TS% includes free throws, which can mask poor shooting from the field. James Harden’s late-Houston years are the canonical example. Eye-popping foul rates kept TS% high while eFG% was telling a quieter story. Both numbers, read together, are honest.
Usage rate
Usage rate (USG%) estimates the percentage of team possessions a player “uses” while on the floor, where a use means ending a possession with a shot, a turnover, or shooting free throws. The league mean is 20%, by definition, since five players are on the floor at once.
The reason usage is so useful is that it gives you the denominator for everything else. A player putting up 25 points on 28% usage is doing it on a manageable shot diet. A player putting up 25 on 35% usage is consuming a third of his team’s possessions. The same scoring line carries different implications for ceiling, sustainability, and playoff scaling.
The pattern to look for is what coaches and front offices call “efficiency at usage.” A player who maintains a .600 TS% at 32% usage is doing something almost nobody can do. Most players’ efficiency degrades as their usage rises — the marginal shot is harder than the average shot, and defenses adjust. Players who do not degrade are the ones whose teams contend.
Net rating, on/off, and lineup data
This is where the conversation gets unstable and where the temptation to overclaim is highest.
Net rating is points scored minus points allowed per 100 possessions. A team net rating of +6 is title-contender territory. A team net rating of -4 is lottery territory. Easy enough.
On/off splits are the same calculation, but restricted to the minutes a specific player is on or off the floor. If the Knicks are +9 net with Jalen Brunson on the court and -2 when he sits, the gap of 11 points is the on/off split. The number is intuitive. It is also, in any given month, capable of lying to your face.
The honest version of on/off is that the data needs thousands of possessions to stabilize, and even then, it is partly a function of the bench unit a player happens to play alongside. A reserve point guard whose on-court minutes are spent with three starters will show a much better on/off split than a starter whose off-court minutes are spent with three reserves. Read the lineup. Then read the number.
Five-man lineup data is the next layer down, and the most informative when used carefully. The 2024-25 Celtics’ starting five posted a net rating that, in regular season minutes, looked like a historical outlier. The 2024-25 Pistons had a five-man lineup that posted +18 over 280 minutes and won them basically nothing in April. Lineup data is a starting point, not a verdict.
BPM, RAPTOR, EPM, and the all-in-one stat fight
At some point in any analytics conversation, someone will say “well, his BPM is…” and you will need to know what they mean. Box Plus/Minus (BPM), RAPTOR (FiveThirtyEight’s old model), EPM, LEBRON, DARKO — these are all attempts to produce a single number that tells you how much a player contributes per 100 possessions, accounting for context.
The way to think about them, without pretending one is the answer, is as panel data. If a player’s BPM is +4, his RAPTOR is +3.5, his EPM is +4.2, and his on/off is +6, you have a coherent picture. If his BPM is +4 but his EPM is -1 and his on/off is -3, you have a stat fight, not a player evaluation, and you have to go look at the film.
The dirty secret of advanced stats coverage is that the all-in-ones disagree more than they agree on the borderline cases. Nikola Jokic looks like a top-three player in every model. The 12th and 13th best players in the league, in any given year, are a different question depending on which model you trust. That is fine. It is also the part of the conversation most likely to be smuggled past you.
Where this gets weird
A list of the places NBA advanced stats consistently break.
Garbage time pollution. A meaningful chunk of late-third and fourth-quarter minutes are played at a different intensity. Some sites filter them out. Some do not. The same player can have meaningfully different splits depending on the cutoff used. Always check.
Defense is undermeasured. Offensive efficiency is solved. Defensive impact, individually, is not. Steals and blocks are unreliable proxies. Defensive RAPTOR, EPM-D, and similar metrics are improvements but still miss communication, switching IQ, and the gravitational effect of a great rim protector on opponents’ shot selection. Beat writers who actually watch film are still ahead of the public defensive data here.
Playoffs are a different game. Regular season net ratings do not always survive seven games against a coordinated opponent. A team that ran a +7 net in the regular season can look ordinary when its third option is being scouted, double-teamed, and forced into uncomfortable touches. Some teams scale up in the postseason. Some scale down. The advanced stats rarely predict which.
Coaching changes nuke historical comparisons. A player’s career numbers reflect every system he has played in. Comparing 2018 Houston Harden to 2024 Brooklyn Harden is comparing two different jobs. The stat sheets do not warn you.
How to actually use this on a Tuesday night
You are watching a Lakers-Knicks game. You want to talk about it the next morning without sounding like the third caller on a sports radio show. A short workflow.
- Open the box score. Note who used the most possessions.
- Pull up Basketball Reference or NBA Stats. Check TS% for the night’s top scorers. A 30-point night on .500 TS is a different night than a 30-point night on .640.
- Check the lineup splits, particularly the closing five. Who was on the floor in the fourth-quarter run? What was their net rating in 5-minute samples this season?
- Read one beat writer who watched the game. They will tell you what the stats cannot — who was being defended, who was setting cross-screens, what the coverage was on pick-and-roll.
- Form an opinion. Hold it loosely. Update on Thursday’s game.
That is the practice. The metric is not the story. The metric is one of the three or four ingredients you use to tell a story you would not have noticed otherwise.
The one number I still trust most
If you put a gun to my head and asked for one advanced stat that has, across a decade of using these, lied to me least often, it would be this: true shooting percentage in playoff series of five games or more. Volume scorers who maintain their TS% as the defenses tighten and the rotations shrink are, in my experience, the players who win series. Volume scorers whose TS% drops six points in the playoffs are, in my experience, the ones who get eliminated.
It is not a complicated metric. It does not need a proprietary model. You can calculate it on a napkin if someone hands you the box scores. The fanciest analytical mind I know in this sport keeps a spreadsheet of exactly that, and nothing else, for the postseason. He has been right more often than the models. I have learned not to argue.
