Calculate Correlation

What is this

A simple node package that calculates the Pearson correlation coefficient of two arrays. No dependencies needed.

Installation

As with any other node package:

npm install calculate-correlation

Usage

The package exports a function.

The function takes two required arguments: two arrays of numbers, with equal length.

And the function returns the correlation coefficient of the two arrays of numbers.

The function also accepts a third optional argument, a config object. More info about it below.

Examples

Use it as follows:

// import the fn
const calculateCorrelation = require("calculate-correlation");

// given 4 points: (2,3), (5,3), (4,6) and (1,7)
const x = [2, 5, 4, 1];
const y = [3, 3, 6, 7];

const correlation = calculateCorrelation(x, y);

console.log(correlation); // logs -0.442807443
console.log(typeof correlation); // logs number

Example with config options:

// import the fn
const calculateCorrelation = require("calculate-correlation");

// given 4 points: (2,3), (5,3), (4,6) and (1,7)
const x = [2, 5, 4, 1];
const y = [3, 3, 6, 7];
const config = {
  string: true,
  decimals: 5,
};

const correlation = calculateCorrelation(x, y, config);

console.log(correlation); // logs -0.44281
console.log(typeof correlation); // logs string

Be careful

• Passing less than 2 or more than three arguments will return an error.
• Passing two arrays with non-numerical values in them will return an error.
• Passign two arrays with different length will also return an error.

If the input is correct, the default result will be a Number between 1 and -1, rounded up to 9 decimal positions.

Config options

The function accepts an object as the third argument. This object can setup two options:

More details

As of now, the formula used by the package to calculate correlation is the one for samples.

Future

Some kind of possible roadmap:

• Limit result number to n decimal positions.
• Add a config option to specify the decimal positions in result.
• Add a config option to specify if the result is preferred as string or number.
• Add a config option to calculate correlation of a population instead of a sample.
• Throw different error messages depending on what was the problem with the input.
• Return other statistic values like average, standard deviation, variance, covariance, etc.