pymccorrelation 0.2.1

Compute correlation coefficients with uncertainties

pymccorrelation

A tool to calculate correlation coefficients for data, using bootstrapping and/or perturbation to estimate the uncertainties on the correlation coefficient. This was initially a python implementation of the Curran (2014) method for calculating uncertainties on Spearman's Rank Correlation Coefficient, but has since been expanded. Curran's original C implementation is MCSpearman (ASCL entry).

Currently the following correlation coefficients can be calculated (with bootstrapping and/or perturbation):

• Pearson's r
• Spearman's rho
• Kendall's tau

Kendall's tau can also calculated when some of the data are left/right censored, following the method described by Isobe+1986.

Requirements

• python3
• scipy
• numpy

Usage

pymccorrelation exports a single function to the user (also pymccorrelation).

from pymccorrelation import pymccorrelation

[... load your data ...]

The correlation coefficient can be one of pearsonr, spearmanr, or kendallt.

For example, to compute the Pearson's r for a sample, using 1000 bootstrapping iterations to estimate the uncertainties:

res = pymccorrelation(data['x'], data['y]',
                      coeff='pearsonr',
                      Nboot=1000)

The output, res is a tuple of length 2, and the two elements are:

• numpy array with the correlation coefficient (Pearson's r, in this case) percentiles (by default 16%, 50%, and 84%)
• numpy array with the p-value percentiles (by default 16%, 50%, and 84%)

The percentile ranges can be adjusted using the percentiles keyword argument.

Additionally, if the full posterior distribution is desired, that can be obtained by setting the return_dist keyword argument to True. In that case, res becomes a tuple of length four:

• numpy array with the correlation coefficient (Pearson's r, in this case) percentiles (by default 16%, 50%, and 84%)
• numpy array with the p-value percentiles (by default 16%, 50%, and 84%)
• numpy array with full set of correlation coefficient values from the bootstrapping
• numpy array with the full set of p-values computed from the bootstrapping

Please see the docstring for the full set of arguments and information including measurement uncertainties (necessary for point perturbation) and for marking censored data.