Overview
Research addressed the problem of testing whether a multivariate distribution exhibits axial symmetry around an unspecified direction. The developed testing procedure relies on a Kolmogorov-Smirnov-type statistic, constructed using projected data and sample splitting. The approach is applicable under a simple-spectrum assumption regarding the covariance matrix.
Research Context
The problem investigated involves determining if a multivariate distribution possesses axial symmetry, specifically when the axis of symmetry is not predefined or known. This constitutes a challenge in statistical inference, as typical symmetry tests often require a pre-specified direction or focal point for symmetry. The proposed method specifically targets cases where this direction is an unknown parameter of the distribution.
Approach
The research posited that, under a simple-spectrum assumption on the covariance matrix, any potential symmetry axis must align with an eigenvector of that covariance matrix. This assumption simplifies the problem by converting the search for an unknown direction into a test against a finite set of candidate directions, specifically the eigenvectors of the covariance matrix.
For each candidate direction identified, a Kolmogorov-Smirnov-type statistic was constructed. This construction involved two primary steps: utilizing projected data and employing sample splitting. The use of projected data likely transformed the multivariate observations into a form more amenable to testing along each candidate axis, while sample splitting is a technique often used in statistical inference to develop test statistics with known asymptotic properties or to validate procedures.
The researchers derived the asymptotic distribution of this statistic within a triangular-array framework. Establishing the asymptotic distribution is crucial for understanding the statistical properties of the test statistic under large sample sizes, enabling the calculation of p-values and critical regions for hypothesis testing. Furthermore, the validity of a bootstrap procedure was established under suitable regularity conditions. Bootstrap validity provides a practical method for approximating the sampling distribution of the statistic, particularly when its theoretical asymptotic distribution might be complex or when sample sizes are not large enough for asymptotic approximations to be fully accurate.
Findings
- Under a simple-spectrum assumption on the covariance matrix, any axial symmetry direction must coincide with an eigenvector of the covariance matrix.
- The problem of testing axial symmetry around an unknown direction can be reduced to testing a finite set of candidate directions, which are the eigenvectors of the covariance matrix.
- A Kolmogorov-Smirnov-type statistic can be constructed for each candidate direction using projected data and sample splitting.
- The asymptotic distribution of this Kolmogorov-Smirnov-type statistic was derived under a triangular-array framework.
- Bootstrap validity was established for the statistic under suitable regularity conditions.
- These developments lead to a feasible testing procedure for axial symmetry when the symmetry direction is unspecified.
Why This Matters
The development of a feasible testing procedure for axial symmetry without requiring a pre-specified symmetry direction addresses a specific challenge in multivariate statistical analysis. The reliance on the simple-spectrum assumption for the covariance matrix and the subsequent reduction to a finite set of candidate directions simplifies what otherwise would be a more complex inference problem.