Can a team depend less on its stars when the stakes rise? Basketball offers a way to explore that question. This study starts with a narrower test: whether substitutes who play more during the regular season also score more in the playoffs.
The calculations below use the 48 values embedded in the original website. They reveal a modest association, but also inconsistencies in the published results and team list. These are checks of the chart data, not independently validated NBA findings. The original data export and notebook are needed to resolve them.
See the data audit and reproducible calculations ↗01 /The question, without the jargon
Every basketball team has five players on the court. The players who begin a game are its starters; the others enter as substitutes, collectively called the bench. The regular season determines which teams qualify for the playoffs, the elimination tournament that decides the champion.
In this article, bench production means scoring by substitutes. It does not include every contribution they make, such as defending or creating opportunities for teammates. Depth is the broader idea of having capable players available beyond the main lineup. A large roster and a capable supporting group are different things.
The organizational question is familiar: how much responsibility should go to established top performers, and how much should go to the people who support them? This first analysis asks whether greater regular-season use of substitutes is associated with their later scoring. It does not yet test whether depth helps a team win.
02 /What we actually measure
The original study describes NBA Stats data for 2023–24, 2024–25 and 2025–26, collected through nba_api using LeagueDashPlayerStats and starter_bench_nullable='Bench'. This revision can inspect the website’s team-level chart values; it cannot verify the original extraction or aggregation without the source files.
| Concept | Meaning | Measured here? |
|---|---|---|
| Usage | How much substitutes play | Yes, using the reported average minutes per bench player per game |
| Quality | How effectively the supporting players contribute | Only partly: playoff scoring omits defense and does not adjust for minutes |
| Star dependence | How concentrated the team’s contribution is among a few players | No; player-level contribution and lineup data are needed |
Each point represents one team in one season. The horizontal axis is its reported average regular-season minutes per bench player; the vertical axis is its reported average playoff points per bench player. Both are per-player averages, not total team bench minutes or points.
Measurement matters. Averaging player averages does not automatically control for roster size. A player with a brief appearance may receive the same weight as a regular contributor. Trades, games played, and changes between starting and substitute roles can also affect the result. More bench minutes could reflect development opportunities, injuries to starters, or game circumstances; this dataset cannot distinguish them.
03 /The evidence—and how stable it is
Recalculating the 48 chart values gives Pearson r = 0.329, a modest positive linear association, with a conventional two-sided p = 0.023 and an approximate 95% confidence interval of 0.049 to 0.561. The squared correlation is 0.108: the simple fitted relationship describes about 10.8% of the observed variation in scoring. This is not the percentage of playoff success caused by bench usage.
Hover or tap a point for its team abbreviation and values. Select a season in the legend to show or hide it. The original chart values are retained for auditability.
A small p-value describes how unusual the statistic would be under the test’s zero-correlation model and assumptions. It does not establish causation or practical importance. Teams recur across seasons, so treating all 48 observations as independent can misstate uncertainty. All p-values and intervals here are exploratory and do not account for dependence within teams.
| Check | Result | Interpretation |
|---|---|---|
| Compare teams within the same season | Partial r = 0.249 p = 0.096 | After removing season means, the association is weaker and does not clear 0.05 under the conventional test. |
| Compare ranks instead of raw values | Spearman ρ = 0.308 p = 0.033 | The direction is similar when comparing relative order; this does not establish causation. |
| Omit one season at a time | r = 0.287–0.389 p = 0.028–0.111 | The positive direction persists, but crossing the significance threshold depends on the sample. |
| Omit one observation at a time | r = 0.278–0.368 | No single plotted record reverses the sign. This does not validate the underlying records. |
The season-specific correlations are 0.015, 0.315 and 0.295, respectively; their p-values are 0.957, 0.235 and 0.268. None clears 0.05 individually. With only 16 records per season, uncertainty is substantial. The near-zero first-season result also contradicts the original claim of a consistently positive trend of similar strength.
04 /Cases that raise better questions
The following records lie above or below the pooled fitted line. These are descriptive examples, not formally established exceptional cases. Scoring differences alone cannot tell us whether coaching, efficiency, injuries or teamwork caused the result.
Was the higher scoring associated with more playoff playing time, better shooting, or different player availability? Minutes and efficiency data could separate those explanations.
The chart gives 13.80 minutes; the original card incorrectly used 10.6. The “very low usage” interpretation therefore does not follow from this record. Role clarity is a hypothesis, not a measured explanation.
Low scoring could reflect fewer opportunities, poor shooting, or both. It does not measure defensive contribution or prove excessive dependence on stars.
The observed scoring is below the fitted line. Explaining why requires checking who played, their minutes, and the opponents they faced.
05 /What the evidence means
Playing time alone is an incomplete indicator of usable depth. The provisional association is modest, varies by season, and weakens when season differences are removed. Even if the source data confirms it, teams with greater bench usage may differ in overall strength, player availability or other ways that also influence playoff scoring.
The practical implication is to evaluate opportunity alongside performance. A coach considering more minutes for substitutes would need to know whether those players maintain team effectiveness and whether the benefit outweighs the immediate cost of resting starters. The present analysis cannot estimate that tradeoff or recommend an optimal allocation of minutes.
A stronger study should define success before choosing predictors. There is no single “ultimate team”: winning now, sustaining performance across a season, and developing future talent are distinct objectives. The next analysis below makes playoff team performance the primary outcome.
06 /Next test: does depth help teams succeed?
The current chart lacks team performance, player-level contributions, lineup information and roster continuity. The design below is a concrete next analysis, not an additional empirical finding.
Test 1 · Supporting quality and playoff performance
Hypothesis: among similarly strong playoff teams, those whose supporting units perform better during the regular season also perform better in the playoffs.
Use playoff team net rating as the primary outcome: points scored minus points allowed per 100 possessions. Normalizing by possessions helps compare teams playing at different speeds. See the NBA statistics glossary.
Define supporting-unit quality in advance as regular-season team net rating during possessions with at least two non-starters on the court, identifying starters separately for each game. This requires lineup or play-by-play data and still reflects teammates and opponents, rather than an isolated bench effect. Specify a minimum possession threshold before examining results and report which teams it excludes.
Compare a baseline model using regular-season team net rating and season indicators with a model that also includes supporting-unit quality. In a separately labeled sensitivity model, account for playoff opponent strength using opponents’ regular-season net ratings weighted by possessions faced. Report the quality coefficient, uncertainty and improvement in prediction on a held-out season. Avoid interpreting prediction as proof of a development effect.
Test 2 · Star concentration and how depth is built
Measure star concentration as the share of a team’s regular-season points scored by its top three scorers, selected using regular-season data only. This captures scoring concentration, not complete dependence. With a larger historical sample, test whether supporting-unit quality matters differently for teams with more concentrated scoring. Do not divide teams into “star” and “collective” groups using a threshold selected to maximize the result.
Then examine one possible route to depth: roster continuity, defined as the share of current regular-season minutes played by players who were on that team in the previous season. Test its association with supporting-unit quality while accounting for prior team strength and available payroll data. This can identify patterns consistent with continuity’s value; successful teams may also retain players, so it cannot by itself show that retention builds depth.
Money, scouting, coaching and development are plausible additional explanations. Fewer injuries, creativity and teamwork are separate outcomes or mechanisms that require their own measures. Age and trainer quality should not be treated as demonstrated explanations without relevant data.
First rebuild and validate the sample from source data, excluding play-in games. Document extraction dates, aggregation weights and missing records. Expand beyond three seasons if comparable data are available; with 48 records, keep models small. Specify the primary hypothesis and model before fitting them, account for repeated teams, and retain weak or negative results. A larger correlation is not the research objective.
07 /Beyond basketball: capacity that can step in
For a company, the useful parallel is whether work can continue effectively when a key contributor is unavailable or demand increases. Basketball makes that allocation problem visible, but this study contains no corporate data. The following are management questions and possible tests, not business effects established by the NBA analysis.
| Basketball question | Organizational translation | What to measure |
|---|---|---|
| Who can contribute beyond the starters? | Are critical tasks covered by more than one capable person? | Share of critical tasks with a trained, assessed backup |
| Does playing time develop useful capacity? | Does supervised responsibility improve independent delivery? | Quality and completion time before and after a development rotation |
| How dependent is performance on a few stars? | Does delivery deteriorate when a key contributor is absent? | Changes in output, errors and delays during planned absences, accounting for workload |
| Does continuity help? | Does retaining shared experience improve coordination? | Handoff failures and recovery time, alongside staffing stability |
A practical pilot: choose a recurring workflow, identify tasks with only one capable owner, and give a backup supervised responsibility. Define acceptable quality and turnaround time before the pilot. Compare results with a similar workflow over the same period, accounting for demand and task complexity, and include training time in the cost. A randomized or staggered rollout, where feasible, would strengthen the evaluation.
The outcome should fit the industry: incident recovery and release quality in software, account continuity in professional services, or throughput and defects in operations. Company size and manager-to-staff ratios could be contextual variables in a later study; neither is a direct measure of supporting capability. Cross-industry comparisons need comparable definitions and separate evidence.
Conclusion for decision-makers
Ask whether the organization can perform beyond its usual top contributors—and measure that capability directly. Headcount and opportunities are inputs; reliable performance when others step in is the outcome that matters.
The basketball reanalysis provides a modest, provisional association between substitute usage and later scoring. It does not establish that giving substitutes more minutes improves team success. Its value is in sharpening the next decision: test whether supporting talent can maintain quality under pressure, then evaluate which investments develop that capacity at an acceptable cost.
08 /Data, audit & reproducibility
This revision uses the numbers embedded in the original public page, retrieved September 16, 2026. The public repository contained website files but no source notebook or raw NBA export at the time of review. The original extraction dates, game coverage and weighting cannot yet be verified.
- Statistical mismatch: the original article reported pooled r = 0.300 and p = 0.038; its chart values give r = 0.329 and p = 0.023. The seasonal bar values also differed from the calculations.
- Team-list mismatch: the 2023–24 array includes Golden State and omits Phoenix. Golden State lost in the play-in; Phoenix played in the first round. See the official NBA schedule and results. The row is retained here to audit the original display, not relabeled or treated as a verified playoff record.
- Card mismatch: Minnesota’s 2025–26 minutes are 13.80 in the plotted data, rather than the 10.6 previously shown in the text.
- Interpretation: claims about efficiency, injuries, role clarity, coaching and later team success have been replaced with questions that require additional evidence.
Pearson tests and Fisher-transformed 95% confidence intervals use SciPy. The season-adjusted calculation correlates deviations from each season’s mean, with two season indicators accounted for in the test’s degrees of freedom (44). Sensitivity checks preserve the original records. No check resolves source-data validity or dependence between repeated teams.
CONTINUE EXPLORINGBack to the portfolio ↗