nist-stat-handbook

Route by group count: one-sample t/z or exact binomial for one process, two-sample or paired t and F-test for two, and ANOVA followed by Tukey, Scheffé, or Bonferroni multiple comparisons for three or more. The cheatsheet's comparison-test selector maps each case, and chapter 7 explains test-interval duality.

List every component as a standard deviation, group by Type A (statistical) or Type B (other means), combine by root-sum-squares with sensitivity coefficients, and expand with U = k·u where k≈2 gives roughly 95% coverage. Use Welch-Satterthwaite when combined degrees of freedom are unknown.

Use CUSUM or EWMA charts for small sustained drifts, since Shewhart charts respond mainly to large abrupt shifts. Tune the EWMA memory depth λ or the CUSUM V-mask, compare candidates by Average Run Length, and define an OCAP before the chart signals.

No. The pack synthesizes concepts and method-selection logic in original words and deliberately does not reproduce the per-distribution formula galleries, worked case studies, datasets, plot images, or Dataplot/R code from the web Handbook. Those are described, not copied.

It is thin on modern machine learning, Bayesian methods beyond conjugate reliability priors, time-series analysis and forecasting, Bayesian or optimal experimental design, and survey or causal-inference methodology. It focuses on applied frequentist statistics for measurement and quality.

Run the 4-plot at the start of any univariate analysis, before fitting or testing. Its run-sequence, lag, histogram, and normal-probability panels check the four assumptions of fixed location, fixed scale, randomness, and fixed distribution; passing all four means the process is in statistical control.