Spherical regression (Q5903270)

Let \(S^ p\) be the unit radius sphere in p-dimensional Euclidean space. Consider the following problem in ''spherical regression''. Let \(u_ 1,...,u_ r\) be fixed points in \(S^ p\) and let \(\nu_ 1,...,\nu_ r\) be random points on \(S^ p\), independently distributed such that the density of \(\nu_ i\) is of the form \(g(\nu_ i'Au_ i)\) with respect to a uniform measure on \(S^ p\), where A is a \(p\times p\) unknown orthogonal matrix \((A'A=I)\). In this paper a statistical procedure is developed for estimating and testing the unknown parameter A. In the case of the circle \((p=2)\) the matrix A represents a counter-clockwise rotation by an angle \(\theta\). If \(\theta_ i\) is the angle from \(u_ i\) to \(\nu_ i\) then \(\theta_ 1,...,\theta_ n\) are independent and identically distributed with a density of the form \(g(\theta_ i- \theta)\).