NBA Last-Five Hit Rates: How Much Can Five Games Tell You?

A player clearing a line in four of the last five games has an 80% historical hit rate for that window. The percentage is correct, but its apparent precision can hide how little evidence five appearances provide. Read recent form alongside a stable event definition, a broader comparison, and the reason the player's circumstances may have changed.

Start with the count

Keep the fraction beside the percentage before interpreting a hot run. Brown and colleagues' work on binomial intervals shows that small samples can leave substantial uncertainty around an observed proportion. Brown et al. (2001) In a fictional record, four hits in five appearances and 16 hits in 20 both equal 80%. Under independent, constant-probability teaching assumptions, their nominal 95% Wilson intervals are approximately 37.6% to 96.4% and 58.4% to 91.9%. These ranges are illustrations of sampling precision, not tonight's NBA probabilities. The same displayed percentage can therefore represent very different amounts of information, and the denominator belongs in your first reading of the result.

Understand what an interval says

An uncertainty interval is useful only when its assumptions and interpretation travel with it. Greenland and colleagues explain that confidence levels concern the repeated behavior of an interval procedure under an appropriate model. Greenland et al. (2016) A calculated 95% interval does not assign a 95% probability that tonight's player's chance lies inside those endpoints. Nor can it correct for missing appearances, a changing role, or a threshold selected after inspecting the data. Treat the interval as a check on how much a simplified count can establish, then examine whether the games actually answer your basketball question.

Compare windows without double-counting them

Last-five and last-ten figures are overlapping descriptions, not two independent confirmations of a trend. Gelman and Stern show why a difference in statistical labels is not itself evidence of a difference between estimates. Gelman and Stern (2006) The related practical lesson is to compare the observations and uncertainty directly instead of treating one window as convincing because its label looks stronger. The last five are already included in the last ten, and both may sit inside the season record. Show a single sequence of games with the relevant role changes marked. That reveals which observations actually changed the conclusion.

Allow a hot run to cool without inventing a story

An extreme recent stretch can move closer to a longer-run level without a new basketball cause. Barnett and colleagues explain regression to the mean, especially when observations are selected because they are unusually high or low. Barnett et al. (2005) This does not say a particular NBA player must miss the next over, and it does not rule out a real improvement in opportunity. It means that selecting the hottest five games makes ordinary variation part of the explanation you must consider. Check minutes, shot attempts, and lineup changes before replacing the baseline with the recent rate.

Choose your window before choosing your conclusion

Use a recent window because it matches a documented research question, not because it produces the most appealing result. Cawley and Talbot demonstrate that selecting against noisy evaluation results can overstate performance. Cawley and Talbot (2010) If a player became a starter five games ago, the new role gives the window a clear purpose; it still leaves only five observations. Record the broader comparison and assess the proposed rule on later appearances. A useful note might say, “Four of five since the rotation change, with limited evidence about the size or persistence of the change.” Add it to the player-prop evidence note so the attractive percentage remains attached to its actual meaning.

References

• Brown, L. D., Cai, T. T., and DasGupta, A. (2001). Interval Estimation for a Binomial Proportion. Statistical Science, 16(2), 101-133, including discussion. DOI: 10.1214/ss/1009213286. Author-hosted published full text; interval-method discussion and Section 3.1.1, equation 4.
• Greenland, S., Senn, S. J., Rothman, K. J., Carlin, J. B., Poole, C., Goodman, S. N., and Altman, D. G. (2016). Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. European Journal of Epidemiology, 31, 337-350. DOI: 10.1007/s10654-016-0149-3. Publisher full text; statistical models and confidence-interval interpretation.
• Gelman, A., and Stern, H. (2006). The Difference Between “Significant” and “Not Significant” is not Itself Statistically Significant. The American Statistician, 60(4), 328-331. DOI: 10.1198/000313006X152649. Author-hosted full text; introduction and Section 2.
• Barnett, A. G., van der Pols, J. C., and Dobson, A. J. (2005). Regression to the mean: what it is and how to deal with it. International Journal of Epidemiology, 34(1), 215-220. DOI: 10.1093/ije/dyh299. University-hosted published full text; pages 215-216.
• Cawley, G. C., and Talbot, N. L. C. (2010). On Over-fitting in Model Selection and Subsequent Selection Bias in Performance Evaluation. Journal of Machine Learning Research, 11, 2079-2107. Publisher abstract refreshed; full-text introductory verification inherited from the research bank.

Prepared with AI assistance and source checks. Published by StatChecker.