A Mathematical Analysis of Benford's Law and its Generalization

TL;DR

This paper explores Kossovsky's generalization of Benford's law, modeling data distributions to explain digit patterns and suggesting applications in density estimation for data science.

Contribution

It provides a mathematical explanation for Kossovsky's generalization of Benford's law and demonstrates its potential for improving density estimation in data analysis.

Findings

Models show higher compliance than natural data.

Compliance depends on the constancy of periodized density functions.

Generalized law can enhance statistical pattern recognition.

Abstract

We explain Kossovsky's generalization of Benford's law which is a formula that approximates the distribution of leftmost digits in finite sequences of natural data and apply it to six sequences of data including populations of US cities and towns and times between earthquakes. We model the natural logarithms of these two data sequences as samples of random variables having normal and reflected Gumbel densities respectively. We show that compliance with the general law depends on how nearly constant the periodized density functions are and that the models are generally more compliant than the natural data. This surprising result suggests that the generalized law might be used to improve density estimation which is the basis of statistical pattern recognition, machine learning and data science.