ab-testing-c

Statistical A/B testing in pure C — chi-squared tests, Welch's t-test, and confidence intervals. All math from scratch, zero external dependencies beyond the C standard library.

What This Gives You

• Chi-squared test for comparing two proportions (with Yates correction)
• Welch's t-test for comparing two independent samples with unequal variances
• Confidence intervals for proportions (normal approximation) and means (t-distribution)
• Utility functions: sample mean, standard deviation, gamma function, erf, normal CDF — all from scratch
• Runs anywhere C99 compiles — embedded systems, edge computing, kernels

Quick Start

#include "ab_testing.h"

/* Chi-squared: compare conversion rates */
TestResult r = chi_squared_test(150, 80, 1000, 1000);
if (r.significant) {
    printf("Significant difference (p=%.4f)\n", r.p_value);
}

/* Welch's t-test: compare continuous metrics */
double control[] = {1.0, 2.0, 3.0, 4.0, 5.0};
double treatment[] = {10.0, 11.0, 12.0, 13.0, 14.0};
TestResult t = welch_t_test(control, 5, treatment, 5);

/* Confidence interval for a proportion */
ConfidenceInterval ci = proportion_ci(500, 1000, 0.95);
printf("Rate: %.1f%% [%.1f%%, %.1f%%]\n", ci.mean * 100, ci.lower * 100, ci.upper * 100);

API Reference

Test Functions

Function	Parameters	Returns
chi_squared_test(a, b, n1, n2)	Successes in each group, total counts	TestResult
welch_t_test(s1, n1, s2, n2)	Sample arrays and sizes	TestResult

Confidence Intervals

Function	Parameters	Returns
proportion_ci(successes, n, confidence)	Success count, total, confidence level	ConfidenceInterval
mean_ci(sample, n, confidence)	Sample array, size, confidence level	ConfidenceInterval

Result Types

typedef struct {
    double statistic;    /* chi-squared or t statistic */
    double p_value;
    double df;           /* degrees of freedom */
    int significant;     /* 1 if p < alpha */
} TestResult;

typedef struct {
    double lower;
    double upper;
    double mean;
    double confidence;
} ConfidenceInterval;

Utility Functions

Function	Description
sample_mean(data, n)	Arithmetic mean
sample_std(data, n)	Bessel-corrected standard deviation
gamma_func(x)	Lanczos approximation (g=7)
erf_func(x)	Abramowitz-Stegun approximation
normal_cdf(x)	Standard normal CDF

How It Fits

C port of ab-testing-rs. Part of the SuperInstance OpenConstruct ecosystem. Use this when you need statistical testing in environments where you can't pull in scipy or a Rust runtime.

Testing

31 assertions across tests for chi-squared, Welch's t-test, proportion CI, mean CI, and utility functions.

Installation

# Build and test
make test

# Just build
make

# Clean
make clean

Requires a C99 compiler and math.h. No other dependencies.