A statistical formalism for alignment analysis

TL;DR

This paper introduces two analytical methods for detecting and quantifying anisotropic alignment signals in vector fields, providing more statistically robust tools than existing Monte Carlo-based approaches.

Contribution

It presents two novel, analytical techniques for alignment analysis that do not depend on Monte Carlo simulations, improving statistical robustness and sensitivity.

Findings

The methods outperform existing techniques in detecting alignment signals.

They provide analytical expressions for statistical significance.

The approaches are validated through extensive Monte Carlo simulations.

Abstract

The detection of anisotropies with respect to a given direction in a vector field is a common problem in astronomy. Several methods have been proposed that rely on the distribution of the acute angles between the data and a reference direction. Different approaches use Monte Carlo methods to quantify the statistical significance of a signal, although often lacking an analytical framework. Here we present two methods to detect and quantify alignment signals and test their statistical robustness. The first method considers the deviance of the relative fraction of vector components in the plane perpendicular to a reference direction with respect to an isotropic distribution. We also derive the statistical properties and stability of the resulting estimator, and therefore does not rely on Monte Carlo simulations to assess its statistical significance. The second method is based on a fit over the residuals of the empirical cumulative distribution function with respect to that expected for a uniform distribution, using a small set of harmonic orthogonal functions, which does not rely on any binning scheme. We compare these methods with others commonly used in the literature, using Monte Carlo simulations, finding that the proposed statistics allow the detection of alignment signals with greater significance.