Statistics and SPC Calculators API
Hypothesis tests, ANOVA, regression, normality and power calculations, plus control charts, process capability, gage R&R, designed experiments, Weibull reliability and time-series decomposition, computed from the numbers you send.
One-sample, two-sample (Welch or pooled) and paired t-tests with the estimate, statistic, degrees of freedom, p-value, confidence interval and Cohen's d from caller-supplied numbers. Pure computation; no data is stored or fetched.
One-way and two-way ANOVA with interaction (Type II or III sums of squares), R-squared, Tukey HSD post-hoc comparisons and the Brown-Forsythe (median-centred Levene) test of equal variances. Sums of squares use centred two-pass accumulation and match NIST StRD SiRstv.
Chi-square test of independence (with optional Yates correction), goodness of fit, Fisher's exact test for 2x2 tables and Cramer's V, with expected counts and a low-expected-cell warning.
Mann-Whitney U, Wilcoxon signed-rank, Kruskal-Wallis, Friedman and sign tests with average-rank ties, exact p-values for samples up to 50 and tie-corrected normal approximations.
Ordinary least squares by Householder QR with coefficient standard errors, t-tests and intervals, R-squared, F, VIF, Durbin-Watson, Breusch-Pagan and confidence and prediction intervals. Matches NIST StRD Norris and Longley.
Binary logistic regression by iteratively reweighted least squares with odds ratios and intervals, deviance, AIC, Hosmer-Lemeshow goodness of fit, a confusion matrix at a chosen threshold and complete-separation detection.
Anderson-Darling, Shapiro-Wilk (Royston AS R94) and Lilliefors tests of normality with optional normal probability plot coordinates. Shapiro-Wilk and Anderson-Darling agree with SciPy and statsmodels.
Power or required sample size for a two-sample t-test (noncentral t), one-way ANOVA (noncentral F), two proportions (Cohen's h) and a correlation (Fisher z). Exactly one of n or power is solved.
Control limits and Western Electric or Nelson rule violations for I-MR, X-bar R, X-bar S, p, np, c and u charts from caller-supplied measurements or counts, with optional fixed Phase II limits.
Cp, Cpk, Pp, Ppk, Cpm and expected parts per million out of specification from individual values or subgroups, with R-bar, S-bar or pooled within-subgroup sigma, an optional Box-Cox transform and a Bissell interval for Cpk.
Crossed gage repeatability and reproducibility by the ANOVA method: variance components, percent contribution, percent study variation, percent tolerance and number of distinct categories for a balanced parts by operators by replicates study.
Full and fractional factorial (with defining relation, resolution and alias structure), Plackett-Burman and central composite designs in standard order, optionally randomised with a seeded xorshift128+ shuffle that is reproducible across languages.
Effects, coefficients and standardised effects for a two-level factorial, with Lenth's pseudo standard error and margin of error for unreplicated designs or a pure-error ANOVA when runs are replicated, sorted as a Pareto of effects.
Maximum-likelihood Weibull, lognormal or exponential life fits to right-censored failure data with Fisher-matrix confidence bounds, B-lives, reliability at chosen times and mean time to failure.
Classical moving-average additive or multiplicative decomposition into trend, seasonal indices and remainder, with Wang-Smith-Hyndman strength of seasonality and trend measures. STL is not provided.