The radial process for confidence sets

From MaRDI portal





Suppose that the sample \(x_ n\) has distribution \(P_{\theta,n}\), where \(\theta \in \Theta\) is unknown and may be either finite or infinite dimensional. Of interest is the value \(\tau = T(\theta)\), where \(T\) is a given function. This paper treats the problem of constructing a good confidence set \(C_ n\) for \(\tau\). What does `good' mean for a confidence set? The most basic requirement is that \(C_ n\) have the intended coverage probability: \[ P_{\theta,n}[\tau \in C_ n] = \alpha \quad \forall \theta \in \Theta,\tag{1} \] either exactly or at least asymptotically. Further requirements on \(C_ n\) can be either geometrical or probabilistic. Viewing \(C_ n\) as a set-valued estimator of \(\tau\) suggests other criteria for good confidence sets. This paper will pursue a geometrical good -- minimizing the geometrical extent of \(C_ n\) -- subject to (1) and the condition that the \(C_ n\) center on \(\tau\) in a certain probabilistic sense. When \(\tau\) is vector-valued or function-valued, the centering requirement turns out to determine the shape and placement of the confidence set \(C_ n\). For the entire collection see [Zbl 0773.00018].











This page was built for publication: The radial process for confidence sets

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3139513)