Distribution of the C statistic with applications to the sample mean of Poisson data

TL;DR

This paper investigates the distribution of the C statistic for Poisson data, especially in low-count regimes, providing insights for model testing and parameter estimation in astronomy.

Contribution

It characterizes the distribution of the C statistic and its minimum and delta forms, improving understanding for low-count Poisson data analysis.

Findings

In high-count limit, C statistic behaves like chi-squared.

In low-count regime, C statistic's expectation is lower than chi-squared.

Application to X-ray data reveals biases in chi-squared use for Poisson data.

Abstract

The C statistics, also known as the Cash statistic, is often used in astronomy for the analysis of low-count Poisson data. One of the challenges of the C statistic is that its probability distribution, under the null hypothesis that the data follow a parent model, is not known exactly. Such distribution is needed for model testing, namely to determine the acceptability of models and then to determine confidence intervals of model parameters. This is in contrast with the accurate knowledge, for Gaussian data, of the $\chi^2$ statistic for any number of free parameters in the parent model. This paper presents an effort towards improving our understanding of the C statistic by studying (a) the distribution of for a fully specified model, (b) the distribution of $C_{min}$ resulting from a maximum-likelihood fit to a simple one-parameter constant model, i.e., a model that represents the sample mean of $N$ Poisson measurements, and (c) the distribution of the associated $\Delta C$ statistic that is used for parameter estimation. The results confirm the expectation that, in the high-count limit, both statistics have the same mean and variance as a $\chi^2$ statistic with same number of degrees of freedom. It is also found that, in the low-count regime, the expectation of the C statistic can be substantially lower than for a $\chi^2$ distribution. These result have implications for hypothesis testing in the low-count Poisson regime that are also discussed in the paper. The paper makes use of recent X-ray observations of the astronomical source PG 1116+215 to illustrate the application of the C statistic to Poisson data. These measurements are also used to identify biases in the use of the $\chi^2$ statistic for Poisson data, especially in the low-count regime.