confidence.js

Confidence

Confidence.js is a light-weight JavaScript library to help you make sense of your A/B test results.

Getting started

Download confidence.js

Include confidence.js on your page.

<script src="path/to/confidence.js"></script>

Usage

Initialization

var myConfidence = new Confidence();

Confidence helps you compare the variants in your A/B test. Variants in Confidence.js look like this:

variant = {
    id: 'A',                // short identifier
    name: 'Variant A',      // descriptive identifier
    conversionCount: 50,    // number of events that successfully converted
    eventCount: 300         // total number of events tracked
}

addVariant(variant)

Adds a variant to your A/B test. You can add and compare as many variants as you'd like.

Parameters:

• variant: the variant object you'd like to add to this A/B test

// first, create some variants
variantA = {
    id: 'A',
    name: 'Alluring Alligators',
    conversionCount: 1500,
    eventCount: 3000
}

variantB = {
    id: 'B',
    name: 'Belligerent Bumblebees',
    conversionCount: 2500,
    eventCount: 3000
}

// then add the variants to your A/B test
myConfidence.addVariant(variantA);
myConfidence.addVariant(variantB);

getResult()

Evaluates the variants in your A/B test and determines which is the winning variant, if there is one.

Returns an object containing:

• hasWinner: true if a winner could be calculated, false otherwise
• hasEnoughData: true if there is enough data to calculate a statistically significant result, false otherwise
• winnerID: the ID of the winning variant, or null if there isn't one
• winnerName: the name of the winning variant or null if there isn't one
• confidenceInterval: the confidence interval, or null if there is no winner.
• ex: { min: 0.154, max: 0.187 }
• readable: human readable result.
• ex: There is not enough data to determine a winner.

Examples

Case 1: There is not enough data to determine a result.

// create some variants
variantC = {
    id: 'C',
    name: 'Cranky Capybaras',
    conversionCount: 5,
    eventCount: 50
};

variantD = {
    id: 'D',
    name: 'Diligent Ducklings',
    conversionCount: 60,
    eventCount: 200
};

variantE = {
    id: 'E',
    name: 'Effervescent Elephants',
    conversionCount: 30,
    eventCount: 40
};


// add the variants to your A/B test
myConfidence.addVariant(variantC);
myConfidence.addVariant(variantD);
myConfidence.addVariant(variantE);


// evaluate the variants to get the result
result = myConfidence.getResult();

/*
{
    hasWinner: false,
    hasEnoughData: false,
    winnerID: null,
    winnerName: null,
    confidenceInterval: null,
    readable: 'There is not enough data to determine
        a conclusive result.'
}
*/

Case 2: There is enough data, but there is no clear winner.

// create some variants
variantF = {
    id: 'F',
    name: 'Freaky Flamingos',
    conversionCount: 1501,
    eventCount: 3000
};

variantG = {
    id: 'G',
    name: 'Gregarious Gorillas',
    conversionCount: 1500,
    eventCount: 3000
};

// add the variants to your A/B test
myConfidence.addVariant(variantF);
myConfidence.addVariant(variantG);

// evaluate the variants to get the result
result = myConfidence.getResult();

/*
{
    hasWinner: false,
    hasEnoughData: true,
    winnerID: null,
    winnerName: null,
    confidenceInterval: null,
    readable: 'We have enough data to say we cannot
        predict a winner with 95% certainty.'
}
*/

Case 3: There is enough data and there is a clear winner.

// create some variants
variantH = {
    id: 'H',
    name: 'Hungry Hippopotami',
    conversionCount: 2500,
    eventCount: 3000
};

variantI = {
    id: 'I',
    name: 'Irritable Iguanas',
    conversionCount: 1500,
    eventCount: 3000
};

// add the variants to your A/B test
myConfidence.addVariant(variantH);
myConfidence.addVariant(variantI);

// evaluate the variants to get the result
result = myConfidence.getResult();

/*
{
    hasWinner: true,
    hasEnoughData: true,
    winnerID: 'H',
    winnerName: 'Hungry Hippopotami',
    confidenceInterval: { min: 82, max: 84.67 },
    readable: 'In a hypothetical experiment that
        is repeated infinite times, the average
        rate of the "Hungry Hippopotami" variant
        will fall between 82% and 84.67%, 95%
        of the time'
}
*/

Run Tests

npm install
npm test

TODO

• Variant name parameter optional
• requires changes to addVariant, and errors.js
• add "not provided" default name if left blank
• add removeVariant function
• zscore table lookup to provide more accurate results if 95% confidence is not available

Issues and Questions

Found a bug? Create an issue here on GitHub!

For general questions, tweet me @jessicaraygun.

Author

Developed and maintained by Jessica Thomas, Data Scientist @ sendwithus.com