How the NBA has changed since 1996: a statistical analysis

The NBA of the last 30 years did not simply become a league that shoots more threes. The data describes a much deeper transformation: teams play faster, lose the ball less often, generate more shots per possession, the mid-range has lost most of its weight and — above all — two-point shooting has become dramatically more efficient. This analysis covers 30 seasons, from 1996-97 to 2025-26, using a public dataset of team box scores.

Between 1996-97 and 2025-26, points per team per game go from 96.9 to 115.6, a 19.3% increase. But only about half of that growth comes from playing more possessions. The other half comes from scoring more efficiently within each possession.

The most spectacular number in the dataset is the three-point attempt: more than doubled, from 16.8 to 37.0 per team per game, while the conversion rate is practically identical — 36.0% in 1996-97 and 36.0% in 2025-26. The NBA revolutionised which shots it takes, far more than how well it shoots them.

The key numbers: NBA 1996-97 vs NBA 2025-26

The picture is unambiguous: more pace, more shots, many more threes, fewer turnovers and a completely different shot diet. But the most interesting findings only appear once you go past the surface values.

Why teams score almost 19 more points per game

In 1996-97 a team produced 96.9 points on average; in 2025-26 it produces 115.6. The difference is 18.7 points per team per game.

Only two variables can push scoring up: the number of possessions and the points produced per possession. Average possessions go from about 92.3 to 100.7 per team, while efficiency goes from 105.0 to 114.8 points per 100 possessions.

Decomposing the scoring increase mathematically gives an almost perfectly symmetrical result: about 9.3 of the 18.7 extra points are associated with the rise in pace, and about 9.4 with the rise in efficiency — roughly 49.8% pace and 50.2% efficiency.

That distinction matters. Saying the modern NBA scores more simply because it plays faster is not consistent with these numbers. Over the long run, speed and efficiency contributed almost exactly the same amount.

The three-point explosion is mostly a volume story

In 1996-97 a team attempted 16.8 threes per game. In 2025-26 the figure is 37.0 — an increase of 120%. The share of field goal attempts taken from behind the arc moves from 21.2% to 41.5% over the same period: almost one shot in two is now a three.

The conversion rate tells a completely different story: 36.01% in 1996-97 and 35.96% in 2025-26. Practically no change. In fact the best single-season 3P% in the data is the 36.68% of 2008-09, higher than the 2025-26 figure.

So the rise of the three-pointer was not driven primarily by better shooting. The real revolution was the willingness to take far more threes while holding a similar efficiency at much higher volume.

The path is not linear either. The three-point share of attempts drops from 21.2% in 1996-97 to 15.9% in 1997-98, then starts a long-term climb. The dataset lets us observe that discontinuity, but not explain it. A few thresholds show how fast the acceleration was: the three-point rate passes 25% in 2013-14, 30% in 2016-17, 35% in 2018-19 and 40% for the first time in 2024-25.

It was not a handful of teams: the entire league moved

The change becomes even clearer when you look at individual teams. In 1996-97 the team with the highest three-point rate took 29.9% of its field goal attempts from behind the arc; the lowest was at 14.5%, and the league median was 20.8%. In 2025-26 the range is completely different: from 33.9% to 49.7%, with a median of 42.1%.

The key fact is this: the team that shoots the fewest threes in 2025-26 still shoots them more often than the team that shot the most in 1996-97. The modern minimum, 33.9%, beats the 1996-97 maximum, 29.9%, by about four percentage points.

This is not a transformation driven by a few extreme teams. The entire distribution of the NBA has shifted. Curiously, the standard deviation of three-point rate across teams is similar in both eras: about 4.1 percentage points in 1996-97 and 4.2 in 2025-26. Teams still differentiate themselves from one another — but they do so inside a new paradigm in which everyone shoots many more threes.

The real casualty is the mid-range

The clearest shot-selection shift shows up in where the points come from.

In 1996-97: 41.9% of points came from the paint, 20.2% from the mid-range, 18.6% from three and 19.3% from the free-throw line. In 2025-26: 43.2% from the paint, just 6.5% from the mid-range, 34.4% from three and 15.9% from the line.

The share of points in the paint is remarkably stable: +1.3 percentage points in thirty years. The mid-range, on the other hand, loses 13.7 percentage points — a 67.8% contraction of its relative share — while threes gain 15.8 points.

The shot diet can therefore be summarised like this: the paint stays central, and most of the space once occupied by the mid-range is absorbed by the three-pointer. Free throws also lose relative weight, from 19.3% to 15.9% of total points. The mid-range bottoms out in 2024-25 at 6.4% of points, essentially the same as the 6.5% of 2025-26.

The counterintuitive finding: two-point shooting improved the most

If three-point volume is the most visible transformation, the rise in two-point percentage is probably the most important one for efficiency. 2P% goes from 48.0% in 1996-97 to 55.0% in 2025-26 — almost seven percentage points. Over the same period, 3P% barely moves.

That helps explain why effective field goal percentage rises from 49.3% to 54.6% and True Shooting from 53.6% to 58.1%.

A mathematical decomposition of the eFG% increase between 1996-97 and 2025-26 is particularly revealing. Of the roughly 5.27 percentage points of improvement, a decomposition that also distributes the interaction effects attributes about 4.79 points to the growth in 2P%, about 0.50 points to the change in the two-versus-three shot mix, and essentially nothing to the change in 3P%.

In purely accounting terms, about 91% of the eFG% improvement between the two endpoints is associated with better two-point conversion. This is not a causal demonstration: the dataset cannot establish why 2P% went up. But the association with the simultaneous collapse of the mid-range is very strong — the share of points from mid-range falls sharply exactly while two-point efficiency climbs.

In 2025-26, an average two is worth more than an average three

Another counterintuitive result appears when you simply convert shooting percentages into points produced per attempt. In 2025-26 a two-pointer at 55.0% produces about 1.100 points per attempt, while a three at 36.0% produces about 1.079.

Average two-point shooting has become so efficient that, in four of the last five seasons in the dataset, raw points per attempt on twos exceeded those on threes. In 1996-97 the situation was completely different: a two was worth about 0.961 points, a three about 1.080.

This does not mean the two shot types are interchangeable, nor does it say which shot a team should take. Observed attempts are the outcome of the selection teams made: they only shoot the twos they consider good enough. What it does show is how much the average quality of the twos actually taken has risen.

In accounting terms, all of the scoring growth comes from threes

There is a second way to look at the offensive transformation: separating points scored on twos, on threes and at the line.

In 1996-97 a team produced about 60.0 points on two-point field goals, 18.1 on threes and 18.7 at the line. In 2025-26 it produces about 57.3 from twos, 39.9 from threes and 18.4 from the line. Compared with 1996-97, that means +21.8 points per game from three, −2.7 from twos and −0.3 from the line.

Total scoring rises by 18.7 points, but points from three grow by almost 21.8. In accounting terms, the extra points from three more than explain the entire net increase, absorbing a small decline in points from twos and free throws.

It is important to keep this separate from the previous section. The growth in points from three comes mostly from volume. The growth in overall efficiency is strongly associated with better two-point shooting.

More possessions — and more shots per possession

Teams do not just get more possessions. Field goal attempts go from 79.3 to 89.1 per game while possessions grow from about 92.3 to 100.7. Normalising for pace, field goal attempts rise from about 86.0 to 88.4 per 100 possessions, and three-point attempts from 18.2 to 36.7 per 100 possessions.

Part of this extra shot-generating capacity is consistent with another very clear trend: teams lose the ball far less often.

Turnovers are down almost 20% at equal possessions

In 1996-97 the data records 15.7 turnovers per game, about 17.0 per 100 possessions. In 2025-26 it is 13.8 per game and 13.7 per 100 possessions — a normalised drop of 19.4%.

At the same time assists rise from 22.0 to 26.7 per game, +21.3%. The assist-to-turnover ratio therefore goes from about 1.41 to 1.94, an improvement of nearly 38%.

These numbers describe an offence that generates more shots and wastes fewer possessions, without necessarily showing an equally large rise in the share of assisted baskets: assisted field goals are about 61.0% of makes in 1996-97 and 63.7% in 2025-26. The growth in assists per game therefore also reflects more possessions, more shots and more makes.

Twos and threes are created in two different ways

The share of assisted two-point baskets goes from 57.0% to 53.5%. The share of assisted threes goes the other way, from about 80.0% to 85.4%.

In 2025-26, in other words, three-pointers are far more often the end of a pass, while a larger share of two-point baskets is created without an assist. The data suggests a growing division of labour between the two shot types — although, again, it describes the outcome of possessions rather than identifying the tactical mechanisms that produce it.

Fewer free throws relative to shot volume

The free throw attempt rate, measured as FTA/FGA, goes from 32.0% to 26.4% — a relative decline of 17.5%. The share of points coming from the line falls too, from 19.3% to 15.9%.

In absolute terms the drop is less visible, because faster play offsets part of the relative reduction: free throw attempts go from 25.3 to 23.5 per game. Personal fouls also fall, from 22.1 to 19.9 per team per game.

The trend is not perfectly linear: the free throw rate bottoms out in 2024-25 at 24.3%, before rebounding to 26.4% in 2025-26.

Offensive rebounding fell — and is now coming back

Offensive rebounds go from 12.7 per team per game in 1996-97 to 11.4 in 2025-26. Normalised for possessions the decline is sharper: from 13.7 to 11.3 per 100 possessions, −17.7%.

The trajectory, however, is not linear. The low point comes in 2020-21, at about 9.8 offensive rebounds per 100 possessions. From there the value climbs back to 11.3 in 2025-26, recovering roughly 15% from the bottom. The long run shows a reduced emphasis on offensive rebounding; the most recent seasons show a partial reversal.

The highest-scoring season is not the fastest one

Pace peaks in 2019-20, at about 100.8. In 2025-26 pace is slightly lower, 100.2, yet average scoring is higher: 115.6 against 111.8.

Between 2019-20 and 2025-26 scoring rises by about 3.8 points despite a small reduction in possessions. The decomposition gives a pace effect of about −0.9 points and an efficiency effect of about +4.7.

The most recent phase of the NBA's evolution is therefore different from the thirty-year trend: once pace reached an already very high level, further scoring growth came mainly from efficiency. Consistently, 2025-26 posts the dataset's maximum for points per 100 possessions, 2P% and True Shooting % at the same time.

More threes does not automatically mean more wins

The dataset also allows a comparison between team characteristics and win percentage. Aggregating at team-season level, the correlation between three-point rate and win percentage is positive but not particularly strong. In the first ten-season block, 1996-97 to 2005-06, it is about 0.21. In 2006-07 to 2015-16 it rises to 0.29. In the last decade, 2016-17 to 2025-26, it falls to just 0.14.

By contrast, the effective field goal percentage differential between a team and its opponents correlates with win percentage at 0.86, 0.88 and 0.86 across the three blocks. Even in the last decade, actual three-point accuracy is far more associated with winning than raw frequency: 3P% correlates about 0.59 with win percentage, against 0.14 for volume.

These are descriptive correlations and do not prove causality. But the statistical message is clear: once high three-point volume becomes a league-wide standard, taking a lot of threes stops being a differentiator. What matters far more is how efficiently a team converts possessions relative to its opponent.

The same dynamic appears when comparing teams in the top and bottom quartile for three-point volume. In the 2006-07 to 2015-16 decade the top quartile averaged about 13 percentage points more win percentage than the bottom quartile. In the last decade the average gap falls to about five points. Three-point adoption looks like it has reached maturity: what once strongly separated strategies has increasingly become a shared requirement.

It is the same logic that governs tactics inside a manager game — a system is not strong in the abstract, it is strong against what the other team does. Our basketball tactics guide works through that idea with sixteen concrete systems.

Home-court advantage has shrunk

The data also shows a progressive reduction in home-court advantage. In 1996-97 the home team wins 57.5% of games, with an average margin of +2.6 points. In 2025-26 home win percentage is 55.4% and the average margin +1.7.

The trend is even clearer when grouping seasons into ten-year blocks. In 1996-97 to 2005-06 home teams win 60.4% of games with a +3.27 margin. In 2006-07 to 2015-16 it falls to 59.4% and +2.93. In 2016-17 to 2025-26 it falls further, to 56.2% and +2.09.

The dataset's seasonal maximum is the 62.8% of 2002-03; the minimum is the 54.3% of 2023-24. Home advantage still exists, but in recent data it is markedly smaller than in the first two decades observed.

Games end with bigger margins — but there is a scale effect

In 1996-97 the average absolute margin of a game is about 11.0 points. In 2025-26 it reaches 13.3. Games decided by 20 or more points go from 14.0% to 22.8%, while games inside five points fall from 28.4% to 24.5%.

At first sight that looks like a sharp rise in imbalance. But when the margin is measured against the total number of points scored in the game, the picture changes: the absolute margin averages 5.74% of the combined score in 1996-97 and 5.79% in 2025-26. Practically the same value.

In nominal terms the gaps are larger because more points are scored. In relative terms, the average distance between the two teams is remarkably stable. It is a good example of why comparing eras through raw statistics alone can produce misleading conclusions.

Three different NBAs in thirty years

The first decade is defined by low pace, plenty of mid-range points and a relatively small share of threes. In the second decade the change is still gradual: threes and efficiency rise, but pace stays close to previous levels. The third decade brings the clearest break: threes climb past a third of all attempts, the mid-range halves again, pace increases and offensive efficiency accelerates.

The 10 most important findings in the data

1. Scoring rose 19.3%, from 96.9 to 115.6 points per team.
2. That increase splits almost exactly in half between more possessions and better efficiency.
3. Three-point attempts grew 120%, while 3P% is practically unchanged.
4. The mid-range lost about two thirds of its weight, from 20.2% to 6.5% of points.
5. The share of points in the paint stayed remarkably stable, around 42–43%.
6. 2P% rose seven percentage points, far more than 3P%.
7. Turnovers per 100 possessions fell almost 20%, and the assist-to-turnover ratio went from 1.41 to 1.94.
8. The three-pointer is now structural: the team that shoots the fewest threes today shoots more of them than the team that shot the most in 1996-97.
9. In the last decade three-point volume has a relatively weak relationship with win percentage; the shooting efficiency differential is far more closely associated with results.
10. Home-court advantage declined, while the growth in nominal game margins is almost entirely absorbed once you account for higher total scoring.

The real evolution of the NBA: not one revolution, but four

Read together, the data describes four simultaneous transformations.

• A shot-selection revolution: far more threes, far less mid-range and a surprisingly stable share of points in the paint.
• A two-point efficiency revolution: the 55.0% of 2025-26 is enormously better than the 48.0% of thirty years earlier.
• A possession-management revolution: fewer turnovers, more shots per 100 possessions and a much better assist-to-turnover ratio.
• A pace revolution, which increased the number of offensive opportunities but on its own cannot explain the scoring explosion.

The result is a game in which every possession produces more value on average, and in which the statistical organisation of offence is radically different.

How the Swish & Dunk engine models this evolution

All of the above is the reason the Swish & Dunk match engine does not decide games with a single roll on team Overall. It simulates possession by possession, and its structure maps almost one-to-one onto the mechanics the data describes.

Shot selection is a choice, not a fixed rate

On every possession the engine picks between a close shot, a mid-range and a three, weighting three things: the shooter's ratings, their position and the offensive tactic. The weight of a three is the player's three-point rating multiplied by a positional factor — 1.3 for a shooting guard, 1.1 for a point guard, 1.05 for a small forward, 0.7 for a power forward, 0.35 for a centre — and by the tactic's three-point tendency, which ranges from 0.85 for the Triangle to 1.35 for Pace and Space. That is exactly the league-wide spread the data describes: within one system a club can play a 1996-style shot diet, within another a 2026-style one.

The base rates match reality

Before ratings are applied, the engine starts from base odds of 60% on close shots, 42% from mid-range, 36% from three and 75% at the line. Those numbers are not arbitrary: 36% is almost exactly the NBA three-point average across all thirty seasons analysed here, and it deliberately stays stable no matter how many threes a team takes. Volume is a decision; accuracy belongs to the player.

Why the mid-range is the weakest shot in the game too

Put the base rates into points per attempt and you get the same hierarchy as the modern NBA: a close shot is worth about 1.20 points, a three about 1.08 and a mid-range only 0.84. A manager who fills the roster with mid-range specialists and runs a low three-point tendency is reproducing the 1990s — and, exactly as in the data, pays for it in efficiency. The tactics guide lists which systems push shots towards the arc and which push them into the paint.

Pace is a lever, not a constant

Roughly 15% of possessions start as a fast break, scaled by the tactic's transition tendency — from 0.75 for the Princeton to 1.50 for Run and Gun, capped at 85%. Since a transition shot gets a bonus of up to 12 percentage points against a defence that does not get back, pace in Swish & Dunk does what it did in the real NBA: it adds possessions and slightly raises the value of each one, but it never explains scoring on its own.

Turnovers and assists behave like the modern league

Every possession runs a turnover check from a 13% base rate, moved by ball security, basketball IQ and decision making against defensive pressure — close to the 13.7 turnovers per 100 possessions of today's NBA rather than the 17.0 of 1996-97. And the assist model reproduces the split described above: a made three is more likely to be credited as assisted (a bonus of 8 percentage points), a close shot less likely (−3), matching the 85.4% of assisted threes against 53.5% of assisted twos in the real data.

Home court, fouls and margins

Home advantage is deliberately small — a 1.2% boost on every rating of the home team — because it is applied to dozens of duels per possession. It produces a home win rate in line with the recent NBA, not with the 60%+ of the early 2000s. Shooting fouls start from an 8% base rate, non-shooting fouls from 5%, with the bonus from the fifth team foul in a quarter: the same mechanics that make free throws a real but decreasing share of scoring.

Why this matters for your club

Because the engine models the causes rather than the outcomes, the statistical evolution of basketball is something you can choose in the game. You can build a team of shooters and a system that lives beyond the arc, or a paint-heavy roster that makes its living at 60% base odds close to the rim. What the thirty-year dataset says applies inside the game too: volume from three, by itself, does not win games — the efficiency differential does. If you want to see how a whole club is built around those choices, start from what Swish & Dunk is.

Method, data and what these numbers cannot tell us

This analysis describes precisely what changed, but on its own it cannot establish why it changed. Without introducing external sources, the dataset cannot attribute these transformations to rule changes, tactical philosophies, player characteristics, technological innovation, changes in officiating or other historical factors. In the same way, the correlations between statistics and winning should not be read as causal relationships.

The source is the public Kaggle dataset Historical NBA Data and Player Box Scores, and specifically the team box score file TeamStatisticsExtended.csv. No external source was used for the figures in this article.

The file structure matters. The main analysis uses regular season games identified through the gameId pattern: all rows explicitly labelled “Regular Season” share the same prefix, but 3,652 rows in that same group have an empty gameType field. Using the prefix avoids discarding a meaningful part of the regular season.

The sample comprises 71,082 team-game rows, equivalent to 35,541 games, across 30 seasons from 1996-97 to 2025-26. The main metrics used — points, FGA, 3PA, FTA, assists, turnovers, rebounds, possessions, pace, eFG% and True Shooting — have complete coverage within the sample. Some more specific absolute statistics have incomplete coverage in the earliest seasons; where possible the corresponding percentage metrics, available across the whole interval, were used instead. The number of games also varies between some seasons, which is why the analysis favours per-game values, per-possession values and percentages over raw totals.

Conclusion: what 30 years of NBA statistics really say

The transformation of the NBA from 1996-97 to 2025-26 can be summarised in one sentence: the league replaced a huge share of the mid-range with the three-pointer, while simultaneously increasing pace, the quality of its two-point shots and its ability to keep possessions alive.

Three-point volume is the most visible sign of the change, but not necessarily the most important one for explaining efficiency. Three-point percentage is practically the same as thirty years ago. What changed radically is the number of threes attempted, the weight of the mid-range, 2P%, turnovers and the number of possessions.

And that is exactly where the data undercuts the simplest story about the modern NBA. Teams do not score more only because they shoot more threes. They do not score more only because they run more. They score more because the offensive system as a whole produces more possessions, wastes fewer of them, and converts those possessions with significantly higher efficiency.

Frequently asked questions

How much has NBA scoring gone up in the last 30 years?

In the dataset, points per team go from 96.9 in 1996-97 to 115.6 in 2025-26: 18.7 more points per game, a 19.3% increase.

Does the NBA score more only because it plays faster?

No. Possessions rise about 9.5%, but points per 100 possessions also rise about 9.3%. Decomposing the total scoring increase attributes roughly half of it to pace and half to efficiency.

How much has three-point shooting grown in the NBA?

Three-point attempts go from 16.8 to 37.0 per team per game, +120%. The share of field goal attempts taken from three goes from 21.2% to 41.5%.

Do NBA teams shoot better from three than 30 years ago?

Not at league-average level. Three-point percentage is 36.01% in 1996-97 and 35.96% in 2025-26. The change is almost entirely about volume.

What happened to the mid-range shot?

The share of points coming from the mid-range falls from 20.2% to 6.5%, a drop of about 68%. It is one of the deepest transformations in the whole dataset.

Have two-point shots become more efficient?

Yes. Two-point percentage goes from 48.0% to 55.0%. It is the single factor most closely associated with the rise in effective field goal percentage.

Does shooting more threes mean winning more?

The relationship exists but is weak, especially in the last decade. Between 2016-17 and 2025-26 the correlation between three-point rate and win percentage is about 0.14, while the effective field goal percentage differential against the opponent correlates about 0.86 with winning.

Do NBA teams turn the ball over less?

Yes. Turnovers fall from about 17.0 to 13.7 per 100 possessions, a 19.4% reduction, while the assist-to-turnover ratio improves from 1.41 to 1.94.

Has home-court advantage shrunk in the NBA?

In this dataset, yes. Home win percentage goes from 57.5% in 1996-97 to 55.4% in 2025-26, and the ten-season averages fall from 60.4% in the first block to 56.2% in the last.

What is the most surprising finding about the NBA's evolution?

One of the most striking is that in 2025-26 even the team with the lowest three-point rate, 33.9%, shoots threes more often than the team with the highest rate in 1996-97, which was at 29.9%. The three-point revolution moved the entire distribution of teams, not just a few outliers.

We used these same NBA numbers to recalibrate the Swish & Dunk engine: read how the new match engine plays like the NBA.

Last updated: October 2, 2026