Simple nonparametric confidence intervals

This JavaScript library calculates conservative nonparametric confidence intervals of the expected value (also known as mean or average) based on observations. It works great to calculate confidence intervals for the true average product rating based on known ratings, for example.

It uses the Dvoretzky–Kiefer–Wolfowitz inequality to calculate confidence intervals without making assumptions about the underlying probability distribution.

Install by copying into your project for now. See unit tests for detailed usage.

Example usage:

const confidence = require('./confidence');

const ratings = {
  1: 0, 2: 3, 3: 9, 4: 53, 5: 144,
};

const [lower, upper] = confidence.calculateConfidenceInterval(ratings, 0.95);

expect(lower).toBeCloseTo(4.24);
expect(upper).toBeCloseTo(4.78);

Tests and ESLint can be run through npm run test or using e.g. the Jest and ESLint plugins for VS Code.

See LICENSE.txt for the license (Apache 2.0).