RAPM — Regularized Adjusted Plus-Minus — is the gold standard for measuring individual player impact in basketball. It controls for the quality of teammates and opponents, measuring the scoring margin a player generates per 100 possessions. The problem: it requires play-by-play stint data, which the NBA did not systematically record until the late 1990s.
That left an entire era — the Bad Boys, Showtime, Jordan’s ascent, the Dream Team generation — analytically dark. No lineup data. No on/off splits. No way to disentangle individual contributions from team performance using the tools modern analysts take for granted.
This project exists to change that. Every game in this database was manually reconstructed from video: every lineup change logged, every possession counted. Nine seasons are currently published, spanning 1969–70 through 1995–96, with the 1985–96 window as the primary focus of ongoing reconstruction. It is the only dataset of its kind for this era, and it is still growing.
Each game in this database was reconstructed manually from video. Every substitution is logged, creating a complete record of which five players were on the court for every possession. Points scored and allowed are recorded for each stint, enabling the calculation of offensive rating, defensive rating, and net rating for every lineup combination.
RAPM is then estimated using ridge regression across the full leaguewide sample — the same mathematical framework used for modern RAPM, applied to the reconstructed stint data. Confidence intervals are reported alongside all estimates. At current sample sizes, intervals span approximately ±9–10 points per 100 possessions for most players. RAPM estimates should be read as directional signals, not precise measurements.
Playoff data is logged separately and currently covers approximately 70% of playoff games in the 1985–1996 window. Regular season coverage is approximately 16% and ongoing.
For a full technical treatment of the RAPM methodology, see the RAPM series in the Analytics Analysis section.
The Data Is Available for Purchase
Four tiers of data access are available, from cleaned season summaries to full possession-level stint files and pre-computed RAPM models. Custom consulting engagements are also available for organizations with specific research needs.
Justin Jacobs holds a PhD in Statistics from UMBC and is a recipient of the Presidential Early Career Award in Science and Engineering — the highest honor the U.S. government awards to early-career scientists. He spent nearly a decade as a research statistician at the Department of Defense and Sandia National Laboratories, with over 30 peer-reviewed publications and a patent.
Since the 2012–13 season he has worked with multiple NBA front offices on analytic development, player valuation, and machine learning, including a role as Senior Basketball Researcher for the Orlando Magic. The Historical RAPM Project is an independent research effort to bring modern analytical tools to bear on an era of basketball history that has, until now, been impossible to study at the possession level.
Squared Statistics
This analysis draws on the Historical RAPM Project — the only possession-level lineup database for the pre-play-by-play NBA era, reconstructed game by game from video since 2021.
View the Project → Data & Consulting →
Data available for the 1984–85 through 1995–96 NBA seasons · squared2020.com
