The NBA didn't get shorter. The ruler changed.

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 seventy percent measurement artifact.

Terms used in this analysis

CI: Confidence interval.

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 larger than the runner-up. 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.

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 −0.64 inches (95% CI [−0.80, −0.47]), 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.

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 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 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.

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%.

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.