Did NBA offenses become solar systems?

"Heliocentrism" is the idea that modern NBA offenses orbit one star who creates most of the scoring, while the WNBA stayed more egalitarian. This study tests both halves of that idea on every regular-season game the play-by-play covers: 35,492 NBA games from 1996-97 to 2025-26 and 6,469 WNBA games from 1997 to 2026, with 1,838,338 assisted baskets. The short answer: the typical NBA team spreads its scoring slightly more than it did thirty years ago, not less; the WNBA has not changed measurably; and 2026 is the only one of the 30 seasons compared in which the WNBA was clearly the more concentrated league.

Crediting every basket to the player who created it

The play-by-play records one thing about who set up a basket: on an assisted make, the passer. So every made field goal is credited to one creator, the passer when it was assisted and the shooter when it was not, weighted by the shot's points. For each team the study then measures the top creator's share of the points created, the top two creators' share, and the effective number of creators (1 divided by the sum of squared shares, which weighs the large shares and barely moves for a bench player with two assists).

The primary measures are computed within each game and averaged over a team's games. That choice came from the first pass: on season totals, a star who misses games lowers the share even if the offense runs through them every night they play. Both versions are reported. Eras are compared by the first five seasons against the last five, with 95% Welch intervals for the change; season means carry 95% t intervals across teams.

Passers are matched to players by name within the shooter's team and game, in three passes, each accepted only when it finds exactly one player. At least 99.8% of assisted baskets resolve in every season, and 993 (0.05%) are left out. Parsed team assists equal the league's game logs in at least 99.7% of team-games in every season except WNBA 1997 (92.1%) and WNBA 1998 (94.2%), and made field goals match in at least 99.9%. Every table was computed in Python and again in R from the same events, and the two agree on every cell.

The typical NBA offense spread out slightly

Over 1996-97 to 2000-01, the average NBA team's top creator accounted for 31.5% of its created points in a game. Over 2021-22 to 2025-26 it was 30.5%, a change of -1.0 point with a 95% interval of -1.9 to -0.1. The effective number of creators per game rose from 5.32 to 5.53 (+0.21, interval +0.08 to +0.34). The most recent season, 2025-26, was the least concentrated NBA season of the 30, at 29.1% (27.8% to 30.5%).

A hinge fit on the season means puts the NBA's turn in 2008, but its approximate 95% set runs from 2005 to 2022. The data date a slow drift, not a moment when the league changed.

Counting by season overstates the decline nearly four times over

On season totals the NBA's top creator share fell from 25.9% to 22.0%, a change of -3.9 points (95% interval -5.3 to -2.5), and the effective number of creators rose from 7.32 to 8.87. That reads like a league moving away from stars. Per game, the change is -1.0 point. The difference is what a rise in games missed by top players would produce, though games played were not measured here, so that cause is unverified. Any usage or concentration claim built on season totals mixes availability into it.

The same gap runs through every season, not only the two eras compared.

The WNBA is now the more concentrated league

Season by season, the NBA's per-game top creator share was higher than the WNBA's, with an interval excluding zero, in 5 of 30 seasons (WNBA years 2001, 2006, 2007, 2009, and 2016). The WNBA's was higher in one: 2026, where the gap was -3.0 points (-5.3 to -0.6), the NBA's 29.1% against the WNBA's 32.1%. In most seasons the two leagues cannot be told apart.

Both leagues assist more of their baskets than they did, which spreads creation by how it is counted. The NBA's assisted share rose from 60.9% to 62.3% (+1.4 points, interval +0.4 to +2.3), and the WNBA's from 62.4% to 67.4% (+4.9, interval +3.5 to +6.4). Part of the NBA's per-game fall is this counting effect, not only a change in who has the ball.

Heliocentric teams exist; a heliocentric league does not

Individual team-seasons do orbit one player. On season totals, the 2008-09 Hornets ran 47.9% of their created points through Chris Paul, and the 2016-17 Thunder 46.8% through Russell Westbrook. Against their own league and season, the standouts include Trae Young's 2021-22 Hawks (39.1%, 2.84 standard deviations above that season's teams) and Luka Doncic's 2023-24 Mavericks (37.4%, 2.62). In the WNBA, Courtney Vandersloot's 2020 Sky (39.6%) and Caitlin Clark's 2024 Fever (39.5%) top the list. These teams are outliers in their seasons. The league averages show the typical NBA team did not follow them.

Set against every team-season, those standouts sit far above their leagues in every era, while the league means stay flat or drift down.

What this does not show

An assist credits only the last pass. A drive and kick credits the player who kicked, and the pass before it is invisible, so creation by primary ball handlers is undercounted, and a rise in assists spreads creation by construction. Assists are also a scorekeeper's judgment that varies by arena and era, which this study does not adjust for. Free throws carry no assist in the feed and are left out, so a star who draws fouls creates more than these shares show. The study covers the regular season only, and team-seasons from one franchise inside a five-season block are treated as independent, so the era intervals are somewhat narrow. Passers are matched by name text, and a confident wrong match is possible where two teammates share it, though the matcher refuses any name that fits more than one player.

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.