Multivariate symmetry models (Q4355972)

The authors develop methods of inference for multivariate symmetry models, which means \(P(A) = P(\gamma A)\) for all Borel sets \(A\) and \(\gamma\in \Gamma_0\), a compact subgroup of the group \(\Gamma\) of orthogonal transformations on \(R^d\). Methods of inference are based on the notion of ``probabilities over half spaces'', defined by \(\{x\in R^d: s'x\leq t\}\). where \(s'x\) denotes the inner product. Several notions of ``symmetric distributions'' on \(R^d\), \(d\geq 2\), are developed in terms of orthogonal subgroups and half spaces. A goodness of fit test is presented after preparatory sections on symmetrization of measures, efficiency, symmetric shift models and identifiability. The proposal requires definition and discussion of stochastic norm, stochastic search, and of computable, and asymptotically valid, critical values, obtained by bootstrap methods. Then \(n^{1/2}\) consistency is proved (Theorem 1), and the asymptotic form of the test is given (Theorem 2). For the entire collection see [Zbl 0861.00032].