benfordslaw-analysis 1.1.0

Use this package to analyse your data with Benford's law

Benford's law analysis

Benford's law is a digit-law, which states that digits from numbers follow a specific frequency. This specific frequency has been observed in many numerical datasets, as discovered by Simon Newcomb and Frank Benford. You can find on wikipedia more information about this mysterious law.

Benford's law is a helpful tool to detect fraud, do science, or just investigate the quality of data. You can also read my blog on Towards Data Science with a summary of Benford's law and this paper, where I used Benford's law to study digit patterns of the distances between stars in our Milky Way.

Installation

With pip install benfordslaw-analysis you can install the package.

Usage

After installing, you can run in Python from benfordslaw_analysis.analysis import Analysis. This imports the class Analysis. With this class you can verify if Benford's law is hidden in your own data.

For example, make a plot with Benford's law versus random data with:

from benfordslaw_analysis.analysis import Analysis
from random import uniform
random_data = [uniform(-10, 10) for i in range(0,1000)]
bl = Analysis(random_data)
bl.plot_first_digit('Random stuff')

Note that we use the Euclidean distance between the digit frequency from Benford's law and your own data as a measure and that we use Poisson error bars (based on the number of data points).

Euclidean distance

The normalized Euclidean distance is a quick way to test whether your data follows Benford law. This value is situated between 0 and 1, the closer to 0 the better. However, it is not a formal statistic because it is sample size independent. In the literature there are several other measures (Chi-square, Kolmogorov-Smirnov, ..) that are used. However, I noticed in my own research that size dependency is a limitation in bigger datasets and classifies all bigger datasets as non-Benford, even though they are Benford by eye.

Citing

If you find benfordslaw_analysis a useful tool for your own research, please cite in the following way:

@misc{benford_py,
      author = {Jurjen, de Jong},
      title = {{benfordslaw_analysis: a Python Implementation of Benford's Law analysis}},
      year = {2021},
      howpublished = {\url{https://github.com/jurjen93/Benfords_law}},
}