Simple sufficient condition for inadmissibility of Moran's single-split test

TL;DR

This paper provides a simple sufficient condition under which Moran's single-split test for a parameter is inadmissible, and proves the inadmissibility of its multi-dimensional Gaussian version, advancing the understanding of test optimality.

Contribution

It introduces a new notion of regular admissibility and derives a simple criterion for the inadmissibility of Moran's test, including a proof for the Gaussian case.

Findings

Moran's test is inadmissible under certain conditions.

The paper establishes a simple criterion for inadmissibility.

It proves the Gaussian multi-dimensional Moran's test is inadmissible.

Abstract

Suppose that a statistician observes two independent variates $X_1$ and $X_2$ having densities $f_i(\cdot;\theta)\equiv f_i(\cdot-\theta)\ ,\ i=1,2$ , $\theta\in\mathbb{R}$. His purpose is to conduct a test for \begin{equation*} H:\theta=0 \ \ \text{vs.}\ \ K:\theta\in\mathbb{R}\setminus\{0\} \end{equation*} with a pre-defined significance level $\alpha\in(0,1)$. Moran (1973) suggested a test which is based on a single split of the data, \textit{i.e.,} to use $X_2$ in order to conduct a one-sided test in the direction of $X_1$. Specifically, if $b_1$ and $b_2$ are the $(1-\alpha)$'th and $\alpha$'th quantiles associated with the distribution of $X_2$ under $H$, then Moran's test has a rejection zone \begin{equation*} (a,\infty)\times(b_1,\infty)\cup(-\infty,a)\times(-\infty,b_2) \end{equation*} where $a\in\mathbb{R}$ is a design parameter. Motivated by this issue, the current work includes an analysis of a new notion, \textit{regular admissibility} of tests. It turns out that the theory regarding this kind of admissibility leads to a simple sufficient condition on $f_1(\cdot)$ and $f_2(\cdot)$ under which Moran's test is inadmissible. Furthermore, the same approach leads to a formal proof for the conjecture of DiCiccio (2018) addressing that the multi-dimensional version of Moran's test is inadmissible when the observations are $d$-dimensional Gaussians.