Investigate / Basketball / Prediction
What Recruiting Rank Predicts in the NBA
The Recruiting Services Consensus Index (RSCI) combines rankings from major recruiting services into a consensus top 100. Nearly every top-10 recruit reached the NBA within five seasons. Among the recruits who made it, rank had almost no relationship with four-season NBA value.
Result
Final rank was most useful for estimating whether a recruit would reach the NBA. The top 10 had a 92.6% attainment rate. Among NBA entrants, rank retained a modest relationship with minutes and starts, while its relationship with box-score value was close to zero.
- 92.6%
- of RSCI top-10 recruits appeared in an NBA game inside five seasons.
- 17.0%
- of recruits ranked 51–100 did the same.
- 291
- NBA entrants in the four-season quality analysis.
First threshold
Top-10 recruits almost always reached the NBA.
The full sample includes every nationally ranked recruit in the study. Here, “reached the NBA” means at least one regular-season game during the five basketball seasons after a recruit’s class. Draft picks, contracts, and later arrivals do not count.
Within that window, 93% of the top 10 reached the league. The rates were 64% for ranks 11–25, 36% for ranks 26–50, and 17% for ranks 51–100.
Lines show 95% confidence intervals, the range of attainment rates reasonably compatible with each observed group under this interval procedure. Tied final ranks are retained.
Log-rank logistic model, averaged across classes; band from class-and-player resampling.
After entry
Recruiting rank tracked minutes more closely than NBA value.
Among the 291 recruits who reached the NBA on time, better-ranked players generally logged more minutes and starts. The rank correlations were −0.251 for minutes and −0.281 for starts.
Value Over Replacement Player (VORP) estimates a player’s total box-score contribution above a replacement-level player. Its four-season correlation with recruiting rank was −0.067. Box Plus/Minus (BPM), a per-100-possession box-score estimate, was also close to zero after weighting each season by minutes.
Medians keep a few very high-VORP stars from determining the group comparison. The individual-player rank correlations below use every entrant’s ordering, not these four summaries.
Points are Spearman ρ estimates; lines are 95% class-aware bootstrap intervals.
All recruits
Top-10 recruits produced the most total VORP.
With zero VORP assigned to non-entrants, mean first-four-season VORP was 1.54 for the top 10 and 0.08 for ranks 51–100. Much of the difference across all recruits comes from the top group’s 92.6% NBA attainment rate.
Observed group means. VORP can be negative; class-aware intervals are widest in the top 10 because a few stars contribute heavily.
Exact rank and broad groups
Four broad groups retained most of the attainment signal.
The exact-rank model had the best held-out result. A version using 1–10, 11–25, 26–50, and 51–100 was close behind. Its log loss was 0.010 higher, with a 95% interval from -0.009 to +0.026.
Log loss grades the predicted probabilities and penalizes confident mistakes, so lower is better. The small held-out difference means the exact slot added little forecasting accuracy once the model kept the top 10 separate.
Both lines use the approved attainment specification and average over recruiting classes. The shaded band is the exact-rank model’s 95% class-and-player bootstrap interval. In held-out classes, exact rank scored 0.504 log loss and 0.794 AUC; the four groups scored 0.514 and 0.790.
AUC measures how often a model ranks an eventual NBA entrant above a non-entrant. AUC 0.5 is chance ordering and 1.0 is perfect ordering; it does not test whether the probabilities themselves are well calibrated. The 25- and 50-player band models performed worse because they grouped the top 10 with lower ranks.
Who ranked best?
The four ranking systems performed similarly.
ESPN, Rivals, and 247Sports are present in the historical RSCI record for the same 456 recruits from 2013–2018. All four systems were within about one AUC point and 0.014 log-loss points of one another.
RSCI had the highest AUC. ESPN had the lowest log loss by 0.0001. The pairwise uncertainty intervals included zero. The same close grouping appeared for overall VORP, entrant VORP, and minutes. This sample showed no measurable forecasting advantage for the consensus.
Points show the held-out estimate; lines show 95% class-and-player bootstrap intervals. AUC grades the ordering of entrants and non-entrants. Log loss grades the probabilities and gives larger penalties to confident mistakes.
Adding ESPN/Rivals/247 disagreement changed held-out attainment log loss by about 0.001 and left the value forecast unchanged.
Largest residuals
These players finished far from the rank-only forecast.
These are the largest gaps between leave-one-class-out expected VORP and actual first-four-season VORP. They are labeled by residual direction because the model does not test why a player exceeded or missed the forecast.
The right column is actual minus forecast VORP. Forecasts are out of sample by recruiting class. “Above” and “below” describe model residuals, not player labels.
Sample and scope
Limits
This sample only includes nationally ranked RSCI recruits. It excludes most high-school players and Division I recruits. This is a predictive analysis within that group, with no causal estimate.
Measurement details
Final RSCI consensus rank is the primary exposure. ESPN, Rivals, and 247Sports ranks are the component values preserved in the same historical RSCI record. Ties remain tied. Repeat listings caused by reclassification or a prep year are reviewed and counted once; same-name players at different colleges remain separate.
Attainment covers the five basketball seasons immediately after the recruiting class. Quality covers four consecutive seasons from NBA debut for on-time entrants. A season with no appearance contributes zero. BPM is minutes-weighted; VORP is summed. Drafted and reached NBA remain separate fields.
The attainment model uses log rank plus a linear class-year adjustment. Every predictive metric is leave-one-recruiting-class-out. Intervals use 1,000 resamples of classes and players within classes. VORP and BPM are box-score measures, not complete descriptions of defense, role, health, or development.
Sources
Recruiting: final RSCI rankings, including the individual-service rankings preserved in the historical RSCI record. NBA appearances and minutes: NBA Stats. Starts, identity, BPM, VORP, awards, and draft tables: Basketball Reference. NBA outcomes were built independently from the cited player-season and draft tables.