A simple alternative derivation of a useful theorem in linear ''errors-in- variables'' regression models together with some clarifications (Q1090033)

A study of the linear errors-in-variables model: \[ X_ j=\xi_ j+\epsilon_ j,\quad j=1,2,...,N,\quad \sum^{N}_{j=1}\gamma_ j\xi_ j+\gamma_ 0=0, \] where X is an observable random variable; \(\xi\) is an unobservable r.v.; \(\epsilon\) are measurement errors (independent r.v., \(E(\epsilon_ j)=0\); \(Var(\epsilon_ j)=\lambda_ j\); \(Cov(\xi,\epsilon)=0)\) and \(\gamma\) is a (structural) vector parameter. The purpose is to derive a result of \textit{O. Reiersøl} [Confluence analysis by means of lag moments and other methods of confluence analysis. Econometrica 9, 1-24 (1941)] using only elementary matrix algebra: under some assumptions concerning M (covariance matrix of the X's) and L (diagonal covariance matrix of the \(\lambda\) 's), \(\gamma\) is contained in a simplex whose vertices are obtained from the different regressions \(X_ j\) in terms of \(X_ 1,...,X_ n\). A (counter)example in the trivariate case is also analyzed.