Is the Gompertz family a good fit to your data?

TL;DR

This paper introduces new statistical tests based on characterisation and Stein's method to assess whether data follow a Gompertz distribution, crucial for accurate modeling in biology and survival analysis.

Contribution

It develops a novel family of goodness-of-fit tests for the Gompertz distribution using characterisation and weighted L^2 methods, including a bootstrap approach for unknown parameters.

Findings

The tests effectively identify Gompertz distribution fit in simulations.

Application to real data demonstrates practical utility.

The method shows high power and consistency.

Abstract

That data follow a Gompertz distribution is a widely used assumption in diverse fields of applied sciences, e.g., in biology or when analysing survival times. Since misspecified models may lead to false conclusions, assessing the fit of the data to an underlying model is of central importance. We propose a new family of characterisation-based weighted $L^2$-type tests of fit to the family of Gompertz distributions, hence tests for the composite hypothesis when the parameters are unknown. The characterisation is motivated by distributional transforms connected to Stein's method of distributional approximation. We provide the limit null distribution of the test statistics in a Hilbert space setting and, since the limit distribution depends on the unknown parameters, we propose a parametric bootstrap procedure. Consistency of the testing procedure is shown. An extensive simulation study as well as applications to real data examples show practical benefits of the procedures: the first data set we analyse consists of lifetimes of fruitflies, the second has been synthetically generated from life tables for women born in Germany in 1948.