@sipemu/anofox-statistics

A comprehensive statistical hypothesis testing library compiled to WebAssembly for JavaScript/TypeScript applications. All tests are validated against R.

Installation

npm install @sipemu/anofox-statistics

Usage

Web/Browser

import init, {
  tTest,
  shapiroWilk,
  mannWhitneyU,
  oneWayAnova,
  JsTTestKind,
  JsAlternative,
  JsAnovaKind
} from '@sipemu/anofox-statistics';

// Initialize the WASM module
await init();

// Two-sample Welch's t-test
const result = tTest(
  new Float64Array([1.2, 2.3, 3.1, 4.5, 5.2]),
  new Float64Array([2.1, 3.4, 4.2, 5.6, 6.1]),
  JsTTestKind.Welch,
  JsAlternative.TwoSided,
  0.0,  // null hypothesis value
  0.95  // confidence level
);

console.log(result);
// { statistic: -2.34, df: 7.89, p_value: 0.047, mean_x: 3.26, mean_y: 4.28, ... }

Node.js

const { tTest, JsTTestKind, JsAlternative } = require('@sipemu/anofox-statistics');

const result = tTest(
  new Float64Array([1.2, 2.3, 3.1]),
  new Float64Array([2.1, 3.4, 4.2]),
  JsTTestKind.Welch,
  JsAlternative.TwoSided
);

Available Tests

Parametric Tests

| Function | Description | |----------|-------------| | tTest | Student's, Welch's, or paired t-test | | yuenTest | Yuen's test for trimmed means (robust) | | oneWayAnova | One-way ANOVA (Fisher or Welch) | | twoWayAnova | Two-way ANOVA with interaction | | repeatedMeasuresAnova | One-way RM-ANOVA with sphericity test and GG/HF corrections | | brownForsythe | Brown-Forsythe test for homogeneity of variances |

Nonparametric Tests

| Function | Description | |----------|-------------| | mannWhitneyU | Mann-Whitney U test (Wilcoxon rank-sum) | | wilcoxonSignedRank | Wilcoxon signed-rank test | | kruskalWallis | Kruskal-Wallis H test | | brunnerMunzel | Brunner-Munzel test | | rank | Rank a sample, with average rank assigned to ties |

Distributional Tests

| Function | Description | |----------|-------------| | shapiroWilk | Shapiro-Wilk normality test | | dagostinoKSquared | D'Agostino's K² omnibus normality test |

Categorical Tests

| Function | Description | |----------|-------------| | chiSquareTest | Chi-square test of independence (with optional Yates correction) | | chiSquareGoodnessOfFit | Chi-square goodness-of-fit test | | gTest | G-test (likelihood-ratio chi-square) | | fisherExact | Fisher's exact test (2×2 tables) | | binomTest | Exact binomial test | | mcnemarTest | McNemar's test for paired proportions | | mcnemarExact | Exact McNemar test for small samples | | propTestOne | One-sample proportion test | | propTestTwo | Two-sample proportion test | | cohenKappa | Cohen's kappa for inter-rater agreement | | cramersV | Cramér's V effect size | | phiCoefficient | Phi coefficient (2×2 tables) | | contingencyCoefficient | Pearson's contingency coefficient |

Correlation Tests

| Function | Description | |----------|-------------| | pearsonCorrelation | Pearson product-moment correlation | | spearmanCorrelation | Spearman rank correlation | | kendallCorrelation | Kendall's tau correlation | | partialCorrelation | Partial correlation (controlling for covariates) | | semiPartialCorrelation | Semi-partial (part) correlation | | intraclassCorrelation | Intraclass correlation coefficient | | distanceCorrelation | Distance correlation (Székely-Rizzo) | | distanceCorrelationTest | Distance correlation independence test |

Equivalence Tests (TOST)

| Function | Description | |----------|-------------| | tostTTestOneSample | One-sample TOST | | tostTTestTwoSample | Two-sample TOST (Welch / Student) | | tostTTestPaired | Paired TOST | | tostCorrelation | Correlation equivalence (Pearson / Spearman) | | tostPropOne | One-proportion TOST | | tostPropTwo | Two-proportion TOST | | tostWilcoxonPaired | Paired Wilcoxon TOST | | tostWilcoxonTwoSample | Two-sample Wilcoxon (Mann-Whitney) TOST | | tostYuen | Trimmed-mean (Yuen) TOST | | tostBootstrap | Bootstrap TOST with percentile CI |

Modern / Kernel Tests

| Function | Description | |----------|-------------| | energyDistanceTest1d | Energy distance test (1D) | | energyDistanceTest | Energy distance test (multivariate) | | mmdTest1d | Maximum Mean Discrepancy test (1D) | | mmdTest | MMD test (multivariate) |

Forecast Comparison Tests

| Function | Description | |----------|-------------| | dieboldMariano | Diebold-Mariano test | | clarkWest | Clark-West test for nested models | | spaTest | Superior Predictive Ability test | | mspeAdjustedSpa | MSPE-adjusted SPA test | | modelConfidenceSet | Model Confidence Set procedure |

Resampling

| Function | Description | |----------|-------------| | permutationTTest | Permutation t-test |

Enums

// T-test variants
enum JsTTestKind { Welch, Student, Paired }

// Alternative hypotheses
enum JsAlternative { TwoSided, Less, Greater }

// ANOVA variants
enum JsAnovaKind { Fisher, Welch }

// Kendall correlation variants
enum JsKendallVariant { TauA, TauB, TauC }

// Intraclass correlation forms
enum JsICCType { ICC1, ICC2, ICC3, ICC1k, ICC2k, ICC3k }

// Method for correlation TOST
enum JsCorrelationTostMethod { Pearson, Spearman }

// Kernel types for MMD
enum JsKernel { Gaussian, Laplacian, Linear }

// Loss functions for forecast comparison
enum JsLossFunction { SquaredError, AbsoluteError }

// Variance estimators (Diebold-Mariano)
enum JsVarEstimator { Acf, Bartlett }

// MCS statistics
enum JsMCSStatistic { Max, Range }

Examples

Normality Testing

import init, { shapiroWilk, dagostinoKSquared } from '@sipemu/anofox-statistics';

await init();

const data = new Float64Array([2.3, 3.1, 2.8, 3.5, 2.9, 3.2, 2.7]);

const sw = shapiroWilk(data);
console.log(`Shapiro-Wilk: W=${sw.statistic.toFixed(4)}, p=${sw.p_value.toFixed(4)}`);

const dk = dagostinoKSquared(data);
console.log(`D'Agostino K²: K²=${dk.statistic.toFixed(4)}, p=${dk.p_value.toFixed(4)}`);

One-Way ANOVA

import init, { oneWayAnova, JsAnovaKind } from '@sipemu/anofox-statistics';

await init();

const groups = [
  new Float64Array([23, 25, 28, 31, 27]),
  new Float64Array([31, 33, 35, 37, 34]),
  new Float64Array([41, 43, 45, 47, 44])
];

const result = oneWayAnova(groups, JsAnovaKind.Fisher);
console.log(`F(${result.df_between}, ${result.df_within}) = ${result.statistic.toFixed(2)}`);
console.log(`p-value: ${result.p_value.toFixed(6)}`);

Two-Way ANOVA

import init, { twoWayAnova } from '@sipemu/anofox-statistics';

await init();

// 2 (factor A) × 3 (factor B) design, 2 reps per cell
const values   = new Float64Array([10, 12,  20, 22,  30, 32,  15, 17,  25, 27,  35, 37]);
const factorA  = new Uint32Array([   0,  0,   0,  0,   0,  0,   1,  1,   1,  1,   1,  1]);
const factorB  = new Uint32Array([   0,  0,   1,  1,   2,  2,   0,  0,   1,  1,   2,  2]);

const result = twoWayAnova(values, factorA, factorB);
console.log(`Factor A: F=${result.factor_a.f_statistic.toFixed(2)}, p=${result.factor_a.p_value.toFixed(4)}`);
console.log(`Factor B: F=${result.factor_b.f_statistic.toFixed(2)}, p=${result.factor_b.p_value.toFixed(4)}`);
console.log(`A×B:      F=${result.interaction.f_statistic.toFixed(2)}, p=${result.interaction.p_value.toFixed(4)}`);

Repeated-Measures ANOVA

import init, { repeatedMeasuresAnova } from '@sipemu/anofox-statistics';

await init();

// Rows = subjects, columns = conditions
const data = [
  new Float64Array([10, 12, 15]),
  new Float64Array([11, 14, 17]),
  new Float64Array([ 9, 13, 16]),
  new Float64Array([12, 15, 18]),
];

const result = repeatedMeasuresAnova(data, /* computeSphericity */ true);
console.log(`Within-subjects: F=${result.within_subjects.f_statistic.toFixed(2)}, p=${result.within_subjects.p_value.toFixed(4)}`);
if (result.sphericity) {
  console.log(`Mauchly W=${result.sphericity.w.toFixed(4)}, p=${result.sphericity.p_value.toFixed(4)}`);
  console.log(`Greenhouse-Geisser p=${result.greenhouse_geisser.p_value.toFixed(4)}`);
}

Equivalence Testing (TOST)

import init, { tostTTestTwoSample, JsTTestKind } from '@sipemu/anofox-statistics';

await init();

const treatment = new Float64Array([10.2, 11.1, 9.8, 10.5, 10.9]);
const control   = new Float64Array([10.0, 10.8, 9.9, 10.3, 10.7]);

// Test equivalence within ±1.0 of zero difference
const result = tostTTestTwoSample(treatment, control, -1.0, 1.0, JsTTestKind.Welch, 0.05);
console.log(`Equivalent: ${result.equivalent}`);
console.log(`TOST p-value: ${result.tost_p_value.toFixed(4)}`);

Forecast Model Comparison

import init, {
  dieboldMariano,
  JsLossFunction,
  JsAlternative,
  JsVarEstimator
} from '@sipemu/anofox-statistics';

await init();

const errors1 = new Float64Array([0.5, -0.3, 0.2, -0.1, 0.4]);
const errors2 = new Float64Array([0.3, -0.5, 0.1, 0.2, -0.2]);

const dm = dieboldMariano(
  errors1,
  errors2,
  JsLossFunction.SquaredError,
  1,                       // horizon
  JsAlternative.TwoSided,
  JsVarEstimator.Bartlett
);
console.log(`DM statistic: ${dm.statistic.toFixed(4)}, p-value: ${dm.p_value.toFixed(4)}`);

TypeScript Support

Full TypeScript definitions are included. All functions and enums are properly typed:

import init, {
  tTest,
  JsTTestKind,
  JsAlternative
} from '@sipemu/anofox-statistics';

await init();

const result = tTest(
  new Float64Array([1, 2, 3]),
  new Float64Array([4, 5, 6]),
  JsTTestKind.Welch,
  JsAlternative.TwoSided,
  0.0,
  0.95
);

// TypeScript knows result has: statistic, df, p_value, mean_x, mean_y, conf_int, etc.
console.log(result.p_value);

Validation

All statistical tests are validated against R's implementation using extensive test suites. Results match R's output within numerical precision (typically 10⁻¹⁰).

Changelog

0.4.2

• Fixed: p-values smaller than ~1.4×10⁻¹⁴ no longer underflow to 0.0. The underlying Rust library now uses each distribution's survival function (sf) instead of 1 - cdf(x) everywhere a tail probability is computed. Affects all tests that use a normal, Student's t, chi-square, or F approximation (Mann-Whitney, Wilcoxon, t/Welch/paired, Yuen, every ANOVA variant, Levene, Brunner-Munzel, Kruskal-Wallis, chi-square, McNemar, Shapiro-Wilk, D'Agostino, Pearson/Spearman/Kendall/partial correlation, ICC, proportion tests, Cohen's kappa, Diebold-Mariano, Clark-West, and all TOST variants).
• Added: rank, twoWayAnova, repeatedMeasuresAnova bindings — the JS package now exposes the same statistical surface as the Rust crate.

License

MIT License — see LICENSE for details.

Links

• GitHub Repository
• Rust crate on crates.io
• API Documentation