January 14, 2024. Boston Garden, second quarter. The Celtics put on the floor a starting five — Jrue Holiday, Derrick White, Jaylen Brown, Jayson Tatum, Kristaps Porziņģis — and outscore the Indiana Pacers 27-12 across the next nine minutes of play. That five-man unit, on the season to that point, will have posted a net rating of +18.4 across roughly 850 minutes, one of the most dominant five-man combinations of the modern NBA. The next-most-used Celtics lineup will be substantially less dominant. The lineup after that, less so still. Boston’s entire season-long success — and ultimately their championship run — is, in a real sense, the story of that one specific five-man combination doing things on the floor that the sum of its individual parts does not predict. The math has a name for it. The math is called lineup synergy. Most public NBA coverage doesn’t use the term, doesn’t name the math, and frequently treats the Boston starting five’s dominance as some combination of “talent” and “chemistry” without specifying what specifically is being measured.
Lineup synergy — the question of why some five-man basketball units consistently outperform what their individual players’ on/off data would predict — is, in 2026, one of the more interesting unsolved problems in public NBA analytics. The vocabulary exists. The data infrastructure exists. The careful versions of the analysis, inside team analytics departments, have been informing roster construction for at least a decade. But the public-facing version of the conversation is, in my opinion, still in its early phase. Most NBA writing about “lineup chemistry” or “team fit” is intuition with statistical decoration. The actual math of why certain five-man combinations work — and why their performance can’t be predicted by adding up the individual ratings — is more interesting and more constrained than the typical coverage suggests.
I have been writing about basketball analytics since 2014, and the analytical frame that has changed my own reading of the modern NBA most measurably is the one this article is going to unpack. Lineup synergy — what it actually measures, where it comes from, how the public data captures it, and the math of why some five-man units click while others don’t, is the subject of this article.
The origin: where lineup analysis came from
The systematic study of NBA lineup data became possible with the public release of play-by-play with substitution information, which Basketball-Reference began publishing in cleaner form in the late 2000s. Dean Oliver’s earlier work on team possession analysis included some lineup-aware components, but the modern era of five-man lineup analysis emerged primarily through Cleaning the Glass and Basketball-Reference’s lineups tool in the 2010s.
The first serious public-facing five-man lineup work came from John Hollinger at ESPN and Kevin Pelton at Basketball Prospectus / ESPN in the late 2000s. Both writers consistently surfaced specific five-man lineup net ratings in their season coverage, particularly for unusual combinations whose performance differed substantially from the players’ individual on/off data. The pattern they were identifying — that lineup data was its own analytical layer, not just an aggregation of individual player numbers — became standard in serious NBA analysis by the early 2010s.
The deeper conceptual work — explaining why some lineups outperform their individual components — has been a slower-moving project. The most-cited public work probably belongs to Justin Jacobs at his “Squared 2020” newsletter and various academic researchers working on “complementarity” and “fit” measurements. The proprietary versions inside team analytics departments are more sophisticated; the public versions are good enough for serious editorial work but rarely venture beyond the descriptive layer (this lineup is good) into the explanatory (why is it good).
By 2026, the public-facing infrastructure includes Basketball-Reference’s lineups search (cleanest historical archive), NBA.com Stats five-man combinations (current season with deep filtering), and Cleaning the Glass’s garbage-time-filtered lineup data (most analytically rigorous). The conversation has matured. The “why” of lineup synergy is still being built.
How lineup synergy works: in plain language
The basic measurement is straightforward. A five-man lineup’s net rating is the team’s point differential per 100 possessions while that specific five-player combination is on the floor. The math is mechanical: total points scored minus total points allowed, divided by total possessions, multiplied by 100.
The lineup’s expected net rating, if the five players were operating independently, would be roughly the sum of their individual on/off splits (with adjustments for the bench-unit context they’re each compared against). A lineup that posts a net rating substantially higher than the sum-of-parts prediction is showing what analysts call positive synergy. A lineup that underperforms its sum-of-parts is showing negative synergy.
The synergy effect comes from a few identifiable mechanisms. Spacing complementarity: when all five players can credibly shoot from three, the floor is wider and the driving lanes are larger. The collective effect is bigger than any individual player’s spacing contribution would predict. Defensive versatility: when all five players can switch onto multiple positions, the defense can match up against more offensive sets without forcing mismatches. Two-way creators: when the lineup has multiple players who can run pick-and-roll as primary or secondary action, the offensive options compound.
The negative-synergy mechanisms are the mirror images. Non-shooter overlap: two players who can’t shoot three together create a shrunken floor that the defense exploits. Position redundancy: a lineup with three small guards is structurally weaker against opposing rebounders than the individual ratings would suggest. Defensive holes: a single player who can’t guard their position is partially shielded by a strong defender behind them; the same player in a lineup without that defender is exposed.
The critical component: the difference between sum and combination
The single most important conceptual insight in lineup analysis is that basketball is not additive at the team level. A team built from five players who are each individually +3 in on/off does not produce a +15 net rating; it might produce +10, or +6, or +18, depending on how the players’ specific skills interact.
The non-additivity is most visible at the extremes. A lineup with one elite creator and four floor-spacing shooters produces an offense that’s much better than the sum of the individual on/off splits — the creator’s gravitational effect makes everyone else’s shooting more impactful. A lineup with two elite ball-dominant scorers can produce an offense that’s worse than the sum, because the two players don’t fit together well in the half-court.
The 2023-24 Boston Celtics starting five is the cleanest recent example of an additive expectation being wildly exceeded. Each of the five starters individually had a strong on/off split. The combined lineup’s net rating was so much higher than the sum-of-parts prediction would suggest that it produced one of the more historically anomalous results in modern lineup data. The team’s championship was, in part, the championship of that one specific combination.
Lineup synergy vs the alternatives: a comparison
The major frameworks for measuring lineup-level performance:
| Framework | What it measures | Strengths | Weaknesses |
|---|---|---|---|
| Five-man lineup net rating | Points per 100 with specific 5 on floor | Direct observation; widely available | Small samples; noisy at lineup level |
| On/off splits | Individual player’s team net rating impact | More stable than lineup data | Mixes player effect with bench context |
| Adjusted plus-minus (RAPM) | Individual player impact, isolated from teammates | Cleanest individual-player measure | Requires regression; high variance per player |
| Synergy index (sum-vs-actual) | Lineup minus predicted-from-individuals | Identifies fit effects | Custom calculation; less public availability |
| Lineup minute volume | Total minutes played by specific 5 | Reveals coaching trust patterns | Doesn’t directly measure quality |
The full lineup analysis uses three or four of these together. The five-man net rating shows what happened. The individual on/off splits show what was expected. The synergy index shows where the gap is. The minute volume shows how much the team has invested in the combination.
What the data needs: inputs
Lineup analysis requires play-by-play with substitution data, possession counts for each lineup configuration, and individual player on/off data for the sum-of-parts comparison. The standard public sources (Basketball-Reference, NBA.com/stats, Cleaning the Glass) provide all of this. The proprietary versions inside teams use deeper inputs: actual shot quality faced by each lineup, defensive coverage tagging, possession outcomes by play type.
The minimum sample size for stable lineup analysis is roughly 200 minutes. Below that, the noise dominates. 500-800 minutes produces reasonably stable estimates. The Celtics starting five’s 850 minutes in 2023-24 was, by lineup-analysis standards, a robust sample.
The harder input — and the one that’s mostly limited to proprietary versions — is quality-of-shot data filtered by lineup. The shot diet a lineup generates (corner threes vs mid-range vs at-the-rim) is much more predictive of future success than the raw made-or-miss numbers. Public coverage usually can’t access this depth.
Building the analysis: a working framework
- Pull the five-man lineup data for the team in question, filtered to lineups with 200+ minutes of play.
- Calculate the sum-of-parts net rating by adding the individual on/off splits.
- Identify the synergy gap. Lineups exceeding their predicted net rating by 4+ points have measurable positive synergy. Lineups underperforming by 4+ points have negative synergy.
- Look at the shot-quality and defensive coverage data for high-synergy lineups when available. Identify the specific mechanisms (spacing, switching, complementarity).
- Cross-reference with playoff translation history. High-synergy regular-season lineups don’t always translate to playoff success — some synergies depend on specific opposing-team weaknesses that varied playoff opponents may not have.
Where this gets weird: common mistakes
Small-sample overconfidence. A lineup with 80 minutes of play and a +20 net rating is impressive on the leaderboard and almost meaningless as evidence. The signal in 80 minutes is dominated by variance. 200+ minutes is the floor for meaningful analysis.
Bench-unit comparison failures. A starting five’s on/off data is calculated against the bench units the team uses. Comparing two teams’ on/off splits requires accounting for the different bench units they replace. The sum-of-parts comparison can be misleading if the bench-unit baseline differs.
Garbage-time pollution. Lineups that play heavy minutes in blowouts post inflated or deflated net ratings that don’t reflect competitive-game performance. Cleaning the Glass’s garbage-time-filtered versions are the better source for serious analysis.
The injury confound. A starting five’s strong net rating may partly reflect that they’ve all been healthy together; an alternative lineup that’s posted weaker numbers may have been used primarily when one of those starters was injured and the team was operating in a depleted state. Filtering for healthy-roster context fixes part of this.
Coach attribution. A coach who finds a strong lineup and rides it for 800 minutes can produce great synergy data without doing much beyond identifying the right combination. A coach who runs ten different five-man combinations with limited minutes each can’t produce the same lineup-level signal regardless of the underlying quality. The coaching’s contribution to “discovering” the right lineup is real but hard to measure.
When lineup synergy analysis shines: use cases
Closing lineup identification. A team’s most-used five-man combination in close-and-late minutes is, by design, the lineup the coach trusts most. The synergy analysis of that specific lineup is more predictive of playoff success than the team’s overall net rating.
Trade evaluation. A trade that adds a player to a team should be evaluated on how the player fits into the existing rotation, not just their individual on/off. The synergy analysis can predict whether the addition will produce better-than-sum results.
Free-agency targeting. A team building around a specific star player and core should target free agents whose skills create positive synergy with the existing pieces. A team with three poor shooters in its core needs floor-spacers; a team without a primary creator needs creation. The lineup math can frame these decisions.
Coaching evaluation. A head coach who consistently identifies and rides high-synergy lineups across multiple seasons and multiple rosters is doing measurable work that the win-loss record alone doesn’t capture. Erik Spoelstra (Heat), Steve Kerr (Warriors), and Joe Mazzulla (recent Celtics) all have track records of effective lineup management.
A working example: the 2024 Celtics starting five
The 2023-24 Boston Celtics starting five (Holiday, White, Brown, Tatum, Porziņģis) is one of the cleaner lineup-synergy case studies of the modern era. Across the regular season, the unit played 850 minutes — a robust sample by any standard — and posted a net rating of +18.4. The sum-of-parts prediction, based on each player’s individual on/off splits, was approximately +12.5. The synergy gap of roughly 6 points per 100 possessions was historically unusual.
The mechanisms were identifiable. The lineup had elite spacing (all five could credibly shoot from three), elite defensive switchability (all five could guard at least two positions adequately, four could guard three or more), and two distinct primary creators (Tatum and Brown) operating against defenses that couldn’t hide a single weak defender. The specific complementarity between Holiday’s point-of-attack defense and Porziņģis’s rim protection made the team’s ceiling considerably higher than a sum-of-parts prediction would suggest.
The playoff translation followed the regular-season profile. The Celtics’ starting five posted strong-but-not-historic net ratings in the playoffs (the 16-3 championship run produced numerous closeout games but with somewhat smaller margins than the regular season suggested), partly because varied playoff opponents found different specific weaknesses to attack. The synergy held; the absolute net rating compressed slightly under postseason intensity. The framework had captured both halves of the prediction accurately.
The limits: what lineup synergy cannot tell you
Lineup synergy cannot predict outcomes of individual games. The framework is a season-level structural analysis.
Lineup synergy cannot fully isolate the coach’s contribution. A coach who identifies a great lineup is doing useful work; the credit allocation between coaching and player composition is unsolved.
Lineup synergy cannot capture the impact of injuries that change the available rotation. A team whose starting five is highly synergistic but rarely available because of injuries is, in practical terms, less effective than the lineup data alone suggests.
Lineup synergy cannot replace film study for tactical understanding. The numbers can show which lineups outperform expectations. The film reveals what specific actions and matchups produce the gap.
One additional limit: synergy analysis depends on having stable lineup composition to study. The NBA’s shift toward more lineup variation (more positions defended by smaller players, more pace, more bench-driven rotations) means that the average team uses more lineups in shorter samples than was the case ten years ago. The framework still works but with smaller per-lineup sample sizes, which reduces the certainty of the inferences.
Frequently asked questions
How many minutes are needed for stable lineup analysis?
Minimum 200 minutes for any signal. 500-800 minutes produces reasonably stable estimates. Below 200, the noise dominates and the lineup’s net rating is largely random.
Is the synergy effect mostly offensive or defensive?
The literature suggests both, but the offensive synergies are easier to identify and measure. Spacing complementarity, two-way creators, and rim-running fit produce visible offensive synergy. Defensive synergies — particularly around switching and rotation — are real but harder to isolate from individual defensive ability.
Can a coach create synergy or just identify it?
Both, in practice. Some synergies are designed (a coach deliberately pairing players whose skills compound), some are discovered (a coach trying a combination and observing the result). The best lineup-management coaches in recent NBA history have done both.
Where can I see lineup data?
NBA.com/stats five-man combinations is the most accessible public source with current-season data. Basketball-Reference’s lineups search provides historical depth. Cleaning the Glass’s lineup data is garbage-time-filtered and is, in my opinion, the most analytically rigorous source available to the public.
Sources and further reading
- Cleaning the Glass — the canonical source for garbage-time-filtered lineup data.
- Basketball-Reference — historical lineup data going back to the early 2000s.
- NBA.com/stats — official current-season five-man combinations with deep filtering.
- Zach Lowe at The Ringer — analytical NBA writing that consistently integrates lineup data into prose.
- Justin Jacobs’ Squared 2020 — public analytical writing on lineup complementarity and fit.
The Celtics starting five that opened this article — Holiday, White, Brown, Tatum, Porziņģis, 850 minutes, +18.4 net rating, championship — is the kind of lineup that lineup-synergy analysis was built to describe. The math says the combination was substantially better than its parts. The film shows you why. The trophy at the end of the season is the validation. For the broader frame on reading modern NBA roster construction through analytics, our guide to pace and space is the natural companion piece.
