scientific article; zbMATH DE number 1304769

From MaRDI portal
Publication:4251607





We consider the linear model (1) \(y=X\beta+\varepsilon\), where \(y\) is an \(n\times 1\) observable random vector with expectation vector \(E(y)=X \beta\) and covariance matrix \(\text{cov}(y)=\sigma^2V\). Equivalently, the unobservable error vector \(\varepsilon\) has expectation \(E(\varepsilon)=0\) and \(\text{cov}(\varepsilon)=\sigma^2V\). The \(n\times p\) full column rank model matrix \(X\) and the \(n\times n\) positive definite matrix \(V\) are known, \(\beta\) is a \(p\times 1\) vector of unknown parameters and \(\sigma^2\) is an unknown positive scalar. For convenience, we may briefly denote the model (1) as (2) \({\mathcal M}=\{y,X\beta,\sigma^2V\}\). While estimating the parameters of the model \({\mathcal M}\), it is natural to consider the influence on the estimates of some changes or perturbations in the data: regression diagnostics measure such influence. A fundamental perturbation is the omission of at least one of the observations from the model. It is of interest, therefore, to have good matrix methods available to check this influence. It is on such ``handy tricks that our interest focuses in this paper. For the entire collection see [Zbl 0912.00017].











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4251607)