Lineup data and five-man units: reading a rotation
Five-man lineup numbers show which combinations of players work, but they rest on small samples. Learn how to read them, how big the noise is, and how to use pairs and trios instead.
A basketball team is not a collection of individuals. It is a series of five-player combinations, and the combination on the floor at any moment matters as much as the names on the roster. Two players can be excellent apart and awkward together. A bench group of modest players can outplay a lineup of stars because their skills fit. Coaches know this instinctively, and they spend much of a game managing who is on the floor with whom. Lineup data is the statistical record of that management: for each group of five, how many possessions it played, how many points it scored and allowed, and how the team fared while it was on the floor.
At first glance, lineup tables look like gold. They say, for example, that a particular unit outscored opponents by 15 points per 100 possessions, while another was outscored by 12. It is tempting to conclude that the first should play more and the second less. Sometimes that is right. Often it is a mistake, because lineup numbers are among the noisiest statistics in the sport, built from samples far smaller than the headline figures suggest.
This article explains how to read a lineup table, why the samples are so small, what the noise looks like in practice, and how to extract real information anyway. It builds on the ideas in net rating and on our discussion of plus-minus and on/off numbers, and it connects to the design questions in lineup construction and rotation management.
What lineup data contains
A lineup table lists every distinct group of five players who shared the floor for a team, together with some or all of the following:
- Minutes or possessions played together. This is the key column.
- Points scored and allowed, usually converted into offensive rating, defensive rating and net rating per 100 possessions.
- Shooting and rebounding splits, which indicate how the unit produced its results.
- Games played, which tells you whether the group appeared steadily or in a concentrated burst.
The same information can be sorted by any column. Sorted by minutes, a table shows the coach's actual preferences: the starters, the usual closers, the reserve group. Sorted by net rating, it shows the units with the best and worst results, but the top and bottom of the list are almost always dominated by groups with tiny samples.
Alongside the five-man tables, many sites offer data on two-man and three-man combinations, and on the effect of single players on or off the floor. These smaller groupings are more stable, for reasons explained below, and often more useful.
The sample-size problem
Here is the heart of the matter. A typical NBA team uses dozens of different five-man lineups over a season. A few get a lot of playing time: the starting unit might play more than a thousand minutes together if the team is healthy, which works out to a couple of thousand possessions. But most units play only a few minutes in total, and many appear once and never again. A lineup that played 20 minutes together has had around 40 possessions at each end. A result drawn from forty possessions is closer to a coin flip than to a verdict.
To put numbers on it, consider the randomness in basketball scoring. A single possession produces zero, one, two or three points, usually, and the spread of outcomes is wide. As a rough approximation, the uncertainty in a lineup's net rating, measured per 100 possessions, is about 140 divided by the square root of the number of possessions it played. That is a rule of thumb, not an exact law, but it is accurate enough to be instructive.
| Possessions together | Rough uncertainty in net rating |
|---|---|
| 100 | about 14 points per 100 |
| 250 | about 9 points per 100 |
| 1,000 | about 4 to 5 points per 100 |
| 2,500 | about 3 points per 100 |
A lineup that has played 250 possessions and posted a net rating of plus 10 could easily be a plus 1 lineup that was lucky or a plus 19 lineup that was a little unlucky. An outcome at the extremes of the table is almost by definition an outlier, and outliers in small samples regress toward the middle. This is the same effect described in small samples in basketball and regression to the mean: the units at the top of the table are, on average, less good than they look, and the units at the bottom are, on average, less bad.
The noise is made worse by three-point shooting. Both teams' shooting percentages on threes fluctuate a great deal over short stretches, and a lineup that happens to catch a hot night from deep, or an opponent's cold one, can look dominant over a few dozen possessions. Because the total effect is built from so many random shots, the advice is to read a unit's rebounding, turnover and free-throw numbers alongside its shooting results, and to be suspicious of a lineup whose success depends on opponents missing open shots.
What a lineup table can legitimately tell you
None of this makes lineup data useless. The information in a table is strong in several ways.
Who plays together
Minutes tell you how the coach actually runs the team. The starting group and the usual closing group are visible at a glance. You can see whether a coach staggers stars so that one is always on the floor, or plays them together for long stretches. You can see which reserves are paired with which starters. This is descriptive, not evaluative, and it needs no statistical caution at all.
The big, stable units
For the largest samples, usually the starting five and the top few lineups, the results are meaningful. A unit with 1,500 possessions has uncertainty on the order of three or four points per 100, small enough that a result of plus 10 or minus 8 is informative. These are the units that define a team's season, and their numbers are usually close to the team's overall net rating.
Patterns across many small units
Even if individual lineups are noisy, patterns across several of them are informative. If every lineup with a certain player at center performs well, and every lineup without him performs poorly, that pattern is more convincing than any single number. Grouping lineups by shared characteristics, such as those with two big men versus one, or with three guards versus two, produces larger samples and clearer answers. This kind of aggregation is how analysts study ideas like small-ball lineups and five-out basketball with some confidence.
Why combinations matter
The deeper insight of lineup data is that basketball is a game of fit. The value of a player depends on who he plays with. A rim-running center is much more valuable alongside a creative guard who can deliver lobs and a roster of shooters who pull defenders out of the paint. The same center with non-shooting teammates may find his space crowded. A defender who excels at switching is more valuable in a lineup of players who can also switch. Lineup numbers, even when noisy, make this interdependence visible in a way that individual statistics cannot.
Reading a rotation
Used with care, lineup data helps you read a coach's rotation. Here is a method.
- Sort by possessions and look at the top group. That is the core unit. Note its net rating, then note how much the team's overall net rating differs from it.
- Look at the closing lineups. Find the lineups that played the most in the last five minutes of close games, if the source offers it. The coach's closers reveal whom he trusts most.
- Identify the stagger. Check how often the best players are on the floor together versus separately. A team that wants to keep an offensive engine on the floor at all times will stagger minutes; a team that wants to maximize a big-three combination will play them together.
- Compare units by type. Group lineups by how many big men or guards they feature. Look at the team's performance in each group.
- Check the bench. Reserve units often play a large share of a game. If the second unit consistently loses ground, the coach's rotation may be giving back what the starters win.
- Be careful with the hidden gems. A lineup with a strong net rating that plays few minutes might be a hidden asset, or a result of luck in a small sample. The way to tell is to ask whether its success has a basketball reason: a skill fit, a stylistic edge, a consistently favorable matchup.
A famous example of a combination that mattered is the small lineup Golden State used in the mid-2010s, with Draymond Green at center surrounded by shooters and ball-handlers. It was often called the "death lineup," and it did more than produce good numbers in a sample; it changed the way teams thought about positions and the value of speed, spacing and switching. That sort of lineup earns attention not just because its net rating was high but because there was a clear, repeatable mechanism behind the results. The mechanism, more than the number, is what lets you trust it.
Why coaches cannot simply play the best lineup
If lineup tables identify a best unit, why not play it all game? The answer is that a lineup is not a free-standing object. It is built from five people who tire, foul, get injured and face opponents who adapt.
Fatigue is the first limit. Even a superb group cannot play 48 minutes, and the pace of the modern game makes long stretches costly. Substitution patterns therefore have to balance the quality of the lineup on the floor with the energy of the players in it. A unit that looks outstanding in the 8 or 10 minutes it plays at a time may look very ordinary if forced to play 20.
Foul trouble is the second. A star with three fouls in the first half may need to sit, and the best unit is suddenly unavailable. How coaches handle those moments is covered in foul trouble: how coaches manage minutes when stars pick up fouls. The result is that the best lineup is a resource to be spent at the right moments, not a default.
Matchups are the third. A lineup that dominates one opponent might be exposed by another. A small, fast group can overwhelm a slow team and be punished by a big, physical one. The best coaches change their lineups in response to what the opponent does, and the numbers over a season average across many different opponents, which can hide the fact that a unit is excellent in some matchups and poor in others.
Finally, there is the simple fact that the best five players do not always make the best five-man unit. A team might own five strong scorers who all need the ball, and the lineup of those five can have too little spacing and too little defense. A lineup that includes a pass-first guard, a rim protector and three flexible wings might outperform it. Fit is the point of lineup analysis, and it is why the tables so often surprise us.
A worked example in miniature
Suppose a team has a starting unit that has played 1,800 possessions together with a net rating of plus 6, and a bench unit that has played 900 possessions with a net rating of minus 3. The rest of the team's possessions are spread among a handful of mixed units that average roughly zero.
The starters are the foundation: their plus 6, with uncertainty of about three points, is a trustworthy figure. The bench unit's minus 3, resting on 900 possessions, carries an uncertainty of roughly 4 to 5 points, so it is entirely possible that the true value is around zero or even slightly positive. A coach who looked at the table and decided that the bench unit was a failure would be making a judgment on thin evidence. A coach who looked at its component numbers, such as the turnovers, the rebounding and the shot quality, might find a specific problem, for instance that the unit has no one who can create off the dribble, and fix that instead.
Now suppose a mixed unit that played only 60 possessions shows a net rating of plus 20. That is the sort of number that gets noticed, and it should be viewed with a raised eyebrow. With 60 possessions, the uncertainty is nearly twenty points. The sensible response is to ask why it might work, to watch the film and to give it more minutes if there is a plausible reason, not to promote it on the strength of the number alone.
Pairs and trios: the better compromise
Because five-man units are so sparse, analysts often turn to smaller groups. Two-man and three-man combinations appear in many more possessions, so their numbers are less noisy. They also answer more practical questions. Does the team do better when two particular players are on the floor together? How does a given guard-big pairing fare in the pick and roll? Does a certain wing help the offense when he plays with the second unit?
Pair statistics have their own traps, mostly because the other three players on the floor are not controlled. If the pair spends most of its time with strong teammates, it will look better than it is. The standard remedy is to examine the pair against a baseline: how the team performs with each player alone, and how it performs with neither. That comparison shows whether the combination does something more than add two good players together.
Another approach is to use regression-based methods, which estimate each player's contribution while holding the others constant. Methods of this type, used by analysts and teams, are better in principle for sorting out who deserves credit, but they need a large amount of data and still carry uncertainty, and the results are usually shown as a range rather than a single number. For a fan, the main benefit of understanding them is to appreciate why a single on-off number should be held loosely.
Context that changes the meaning of a lineup number
A few contextual factors can change how a lineup's results should be read.
- Opponents. Starters generally face opposing starters; second units face second units. A reserve group that outscores other benches is doing something worthwhile, but it is not the same as beating strong starters.
- Game state. Lineups that play in blowouts, either leading or trailing, are surrounded by garbage-time minutes in which the effort and the quality of play are distorted. Some analysts exclude those possessions for cleaner comparisons.
- Injuries and availability. A lineup that played together for ten games because one starter was out cannot be evaluated the same way as one that played through the full season.
- Pace and style. A unit that plays faster has more possessions per minute. Reading performance in per-100 terms, not per-minute terms, keeps the comparison fair.
- Opponent shooting luck. A lineup that allowed few points might have been fortunate in that opponents missed good shots. Looking at shot quality, not just shooting percentage, offers a check.
What to take away
Lineup data is the closest statistics comes to measuring how a basketball team actually functions, but it is also one of the least reliable categories because of small samples. Use it to understand who plays with whom and how the coach structures a game. Trust the large-sample units, usually the starters and the most common closers. Treat anything under a few hundred possessions as a story to check, not a fact. Prefer pairs, trios and groupings by type to individual five-man lines, and always ask whether a result has a basketball explanation. The numbers can point you toward a question; the game itself, watched carefully, will usually supply the answer. You can look at team and player information on the teams and players pages.