Comparing Normal Linear Experiments and Transformation of Observations *
Consider random experiments \({\mathcal N}_1={\mathcal N}(A\beta, \sigma V)\) and \({\mathcal N}_2={\mathcal N}(B\beta,\sigma W)\) realized by random vectors \(X\in \mathbb{R}^n\) and \(Y\in \mathbb{R}^m\), respectively, where \(A\), \(V\), \(B\), \(W\) are known matrices, while \(\beta\in \mathbb{R}^p\) and \(\sigma>0\) are unknown parameters. A new ordering induced by unbiased estimation of \(\beta\) and \(\sigma\) is proposed. The vector \(X\) is called ``at least as good as \(Y\) if for every parametric function \(\psi= k'\beta+c\sigma\), where \(k\in \mathbb{R}^p\) and \(c\in \mathbb{R}\), and for each unbiased estimator \(\widehat{\psi}(Y)\), whenever such exists, there exists an unbiased estimator \(\widehat{\psi}(X)\), s.t. \(\text{var} [\widehat{\psi}(X)]\leq \text{var} [\widehat{\psi}(Y)]\). This concept is an extension of the ordering of linear experiments w.r.t. unbiased estimation of \(\beta\) only, introduced by the author [Ann. Inst. Stat. Math. 35, 461-473 (1983; Zbl 0553.62065)]. The main result of the present paper states that the vector \(X\) is as good as \(Y\) iff there exists an \(m\times n\) matrix \(F\) and a normal vector \(U\in \mathbb{R}\) independent on \(X\) with expectation zero and covariance matrix \(\sigma\cdot G\) s.t. \(R(G)\subseteq R(V+AA')\) and \(F(X+U)\) has the same distribution as \(Y\). Here \(A'\) and \(R(M)\) denote the transposition of \(A\) and the range of a matrix \(M\), respectively. This theorem shows the relation of the above ordering and the ordering of the experiments based on statistical information, see \textit{E. L. Lehmann}, Ann. Stat. 16, No. 2, 521-533 (1988; Zbl 0672.62008). There are some connections between the above ordering and the concepts of linear and quadratic sufficiency considered by several authors. These connections are indicated in Section 4.
- Matrix loss in comparison of linear experiments
- Non-existence of unbiased estimators of ordered parameters
- Order relations for linear models: a survey on recent developments
- Comparison of linear experiments with known covariances
- Quotient of information matrices in comparison of linear experiments for quadratic estimation
- Stochastic ordering and Schur-convex functions in comparison of linear experiments
- Unbiased estimation of \(P(X>Y)\) for two-parameter exponential populations using order statistics
- scientific article; zbMATH DE number 806888
- scientific article; zbMATH DE number 3940433
- A complete sufficient statistic for the linear model under normality and a singular covariance matrix
- Comparing location experiments
- Comparison of linear experiments with known covariances
- scientific article; zbMATH DE number 49097 (Why is no real title available?)
- Linear Statistical Inference and its Applications
- Linear transformations preserving best linear unbiased estimators in a general Gauss-Markoff model
- Optimal allocation of units in experimental designs with hierarchical and cross classification
- Ordering of nonnegative definite matrices with application to comparison of linear models
- Orderings of linear models
- Reduction problems in comparison of linear models
- Stochastic ordering and Schur-convex functions in comparison of linear experiments
- Sufficiency and completeness in the linear model
- Bemerkungen zum Vergleich linearer normaler Experimente. (Remarks on the comparison of linear normal experiments)
- Comparison of normal linear experiments by quadratic forms
- Sandwich theorem in comparison of multivariate normal experiments
- On a matrix result in comparison of linear experiments
- Quotient of information matrices in comparison of linear experiments for quadratic estimation
- Estimation and experiments comparison with respect to the matrix risk
- Matrix equations with restraints and their statistical applications
- A surprising result in comparing orthogonal and nonorthogonal linear experiments
- Stochastic ordering and Schur-convex functions in comparison of linear experiments
- Comparison of linear experiments with known covariances
- Order relations for linear models: a survey on recent developments
- Perfect linear models and perfect parametric functions
This page was built for publication: Comparing Normal Linear Experiments and Transformation of Observations *
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4379702)
