Basketball Data Science All studies Scouting one-pagers Code on GitHub
46 seasons of rosters

The players didn't shrink. The jersey numbers did.

Ask a basketball fan what happened to NBA height over the last twenty years and you will hear a confident story: the game spread out, centers gave way to wings, and the league got smaller. Forty-six seasons of roster data (18,947 player-seasons, 1980-81 through 2025-26) say the confident story is about three-quarters measurement artifact.

Terms used in this analysis
The 2019-20 height cliff
−0.61 in
Largest year-over-year move in the 45-season record
Number-height correlation
.44 → .12
The jersey number used to tell you the height. It barely does now.
Median jersey number
25 → 14
1980-81 to 2025-26. The low numbers won.
Sample
18,947
Player-seasons across 46 years of rosters
The height story

Not a slope. A cliff.

Mean listed height did rise for two decades, peaking at 79.4 inches in 2002-03, and it has drifted down since. The question is how it came down, and that is exactly the kind of question a summary statistic hides and a time series answers. The figure below is the study's centerpiece for one reason. The decline is not a slope, it is a cliff: a single-season drop of 0.61 inches between 2018-19 and 2019-20, the largest year-over-year move in the entire 45-season record and 2.3 times the next-largest drop. Trends produced by basketball evolution do not move like that. Rule changes do.

Mean listed height by season. The reason this chart exists: the 2019-20 cliff is visible to the naked eye, and no smooth "the game got smaller" story survives contact with it. Hover any season for the exact value.
The height decline split into a slow drift, one step, and a riseMean listed height, 2002-03 peak to 2025-2678.57979.5inches79.37peak2002-03−0.35slow drift16 seasons−0.61rule changeone offseason+0.15since6 seasons78.57now2025-2675%of the net −0.80 indecline happened inthe one offseasonwhen the leaguebegan requiringheights measuredwithout shoes.The regime model fitsthat step directly:−0.62 in95% CI −0.79 to −0.46
The height decline split into a slow drift, one step, and a riseMean listed height, 2002-03 peak to 2025-2678.57979.5inches79.37peak2002-03−0.35slowdrift16seasons−0.61rulechangeoneoffseason+0.15since6seasons78.57now2025-2675%of the net −0.80 in decline happened in theone offseason when the league beganrequiring heights measured without shoes.The regime model fits that step directly:−0.62 in95% CI −0.79 to −0.46
The decline from the 2002-03 peak, split into its three parts. The slow drift before 2018-19 is small, the single offseason of the rule change is most of it, and heights have risen since. The box fits that step with the regime model instead of reading it off two seasons.

2019-20 is precisely when the league began requiring measured heights without shoes. Two checks pin the attribution down. A regime-aware model (piecewise trends with an explicit level-shift term at 2019) estimates the rule-change step at −0.62 inches (95% CI [−0.79, −0.46]), and fitting the step explicitly keeps it from contaminating the trend slopes on either side. And among the 388 players rostered in both 2018-19 and 2019-20, the same drop appears within the same human beings. So roster turnover, the one mundane alternative explanation, cannot account for it. Since the rule change, the league has actually been getting slightly taller.

The same players, listed before and after the 2019-20 ruleThe same 388 players, listed in 2018-19 and again in 2019-20listed shorter: 215unchanged: 152listed taller: 21Mean change per player: −0.57 inSame people, so a change in whoplays cannot explain the drop.Returning players listed shorter, each offseason2018-19 to 2019-20: 55%median of the other 44: 0%1980-81 to 1981-822024-25 to 2025-26
The same players, listed before and after the 2019-20 ruleThe same 388 players, listed in2018-19 and again in 2019-20listed shorter: 215unchanged: 152listed taller: 21Mean change per player: −0.57 inSame people, so a change in who plays cannotexplain the drop.Returning players listed shorter, eachoffseason2018-19 to 2019-20: 55%median of the other 44: 0%1980-81 to 1981-822024-25 to 2025-26
The within-player check. Each dot is one player rostered in both seasons, colored by how his listed height changed across the rule change: more than half were listed shorter. The bars show the same share for every offseason. In the typical one almost no returning player is listed shorter, and the smaller bars after the rule change are consistent with later re-measurements. A change in who plays cannot produce a drop inside the same people.
Positions

Convergence is real. It just isn't the story.

Height by position tells the second half of the story. Positional convergence is real. But it is gradual and modest next to the step, which is why the two need to be separated before either can be read.

Height by listed position over time. Necessary because "the league got smaller" and "positions converged" are different claims. This figure shows the second is real but gradual, which is what isolates the step in the first figure as a measurement event.
The cultural signal

Big men no longer wear big numbers

The study's original question, jersey numbers, turns out to carry the cultural signal. Number choice used to encode position: big men wore big numbers. That convention has dissolved.

Jersey-number conventions by era. The cultural companion to the height story: which numbers players choose, and how the old big-man-big-number convention faded.

The correlation in the next chart is one number per season, and the diagram below shows what it summarizes, for the first season and the latest.

Each player as a dot: jersey number against listed heightEach dot is one player: his jersey number across, his listed height upcentersguards and forwardsbest-fit line1980-81r = +0.446'0"6'6"7'0"0255075100jersey number2025-26r = +0.176'0"6'6"7'0"0255075100jersey numberThe more the cloud climbs from low numbers to high, the larger r: the number tellsyou more about the height.
Each player as a dot: jersey number against listed heightEach dot is one player: his jersey numberacross, his listed height upcentersguards and forwardsbest-fit line1980-81r = +0.446'0"6'6"7'0"0255075100jersey number2025-26r = +0.176'0"6'6"7'0"0255075100jersey numberThe more the cloud climbs from low numbers tohigh, the larger r: the number tells you moreabout the height.
What the correlation measures. In the first season the centers (blue) mostly wear high numbers and the cloud climbs from left to right; in the latest season most centers wear low numbers too, and the best-fit line is nearly flat. Each point on the next chart is one season's version of this cloud.
The correlation between jersey number and height, by season: from +0.44 to +0.12. This is the single chart that summarizes the study's answer: the number on the jersey used to tell you how tall its wearer was, and now it barely does.

The correlation between a player's number and his height fell from +0.44 to +0.17 across the sample (its 2024-25 floor was +0.12): a quiet, steady cultural change that, unlike the height "decline," is exactly what it appears to be.

46 seasons of number choices

The low numbers won

The numbers themselves deserve a season-by-season look, not just a correlation. Two explorable views below (0 and 00 are pooled into a single bucket throughout: distinct uniforms, same cultural choice; the pooling is noted on each figure). The median tells the story in one line: the typical jersey number nearly halved, from 25 in 1980-81 to 14 in 2025-26 (13 the season before). The low numbers won.

The median jersey number by season, with the middle-50% band. Necessary because the correlation chart above says the number-height link dissolved but not what players chose instead: the entire distribution slid steadily toward low numbers, with the median nearly halving over 46 seasons. Hover any season for the 0/00 and 50+ shares.
The full jersey-number distribution, one season at a time: drag the slider or press play to walk 46 seasons, and switch the bin size (1, 5, or 10 numbers per bar) to trade detail for shape. This is the exploratory view the medians compress away: any bar with ten or fewer wearers lists them on hover (a player traded mid-season shows once, with his final team, because the roster archive is end-of-season). Watch the high-number right tail thin out and the 0/00 bar (first bar, pooled) grow from 1.1% of the league in 1980-81 to its 6.8% all-time peak in 2024-25 (5.6% in 2025-26). Numbers 50 and up went from 7.7% of players to 4.8%.
The WNBA

The WNBA never had the convention to lose

The WNBA started in 1997, when the NBA's big-man-big-number habit was already fading, and its rosters show it. The jersey/height correlation was +0.20 across the WNBA's first five seasons and +0.19 across its last five: a change of about zero (95% CI [-0.10, +0.09]), and an interval that rules out a drop anywhere near the NBA's. Numbers above 55 were never common there at all.

WNBA number conventions by season: the share of players in numbers above 55, in 0 or 00, and in numbers containing a digit from six through nine. Necessary because it shows the high-number habit the NBA abandoned was never present in the WNBA to begin with.
The WNBA's jersey/height correlation by season. The flat line is the finding: the WNBA began about where the NBA has fallen to and stayed there.

Lining the two leagues up by league age rather than calendar year sharpens the contrast. At the same ages, the NBA's correlation was building, from +0.07 at league ages 8 to 12 to +0.33 at ages 25 to 29, while the WNBA's held at +0.21 and +0.19. The NBA grew the convention as it aged and has since let it go; the WNBA, born later, never built it.

Jersey/height correlation by league age, NBA and WNBA. Necessary because calendar years compare a league in its fourth decade with one in its ninth; league age compares like with like.
Share of players in numbers above 55 by league age. Rare in both leagues at every league age the two share.
Share of players wearing a number with a digit from six through nine, by league age. The NBA's share fell as it aged; the WNBA's did not.
A natural experiment

College rules and the numbers rookies pick

For decades the NCAA barred the digits six through nine on basketball uniforms, so almost every WNBA rookie arrived having never been allowed to wear one. The NCAA dropped the restriction for the 2023-24 season, which makes the 2024 to 2026 rookie classes the first exposed to the new rule. Among NCAA rookies, the share wearing a digit from six to nine went from 8.5% in the ten classes before to 13.0% in the exposed classes, a change of +4.5 points with a 95% interval of [-2.8, +12.8]. In the NBA, which drafts from the same college system, it went from 22.9% to 21.3%. Neither change clears zero, and with 92 exposed WNBA rookies the interval is wide enough to hide a real shift of a few points. This is a question to rerun as each new class arrives.

WNBA rookies wearing a number with a digit from six to nine, by class, NCAA alumni against everyone else. The first three classes after the rule change sit inside the earlier range.
The same rookie measure for the NBA, as a cross-check: men's college basketball had the same rule and the same change.
The lesson

A rehearsal for every "the metric moved" argument

Why does any of this matter beyond trivia? Because it is a low-stakes rehearsal of a high-stakes failure mode. A headline metric moved, and the obvious narrative was ready. The narrative was wrong. And the error only shows up when you decompose the change, model the break explicitly, and check it within the same players. Every study in this family runs on that discipline; this one just happens to make the lesson visible in a single picture.

Method, code, and verification

The full technical write-up (data provenance, model specification, validation gates, and stated limitations) is the README. The analysis is implemented independently in R and Python and reconciled to numeric tolerance; run_checks.sh replays every gate.