Linear Algebra and Its Applications · Problem 3.3.30

Orthogonality — Gilbert Strang

Mathematics · statistics-variance-reduction

Problem

A doctor takes four readings of your heart rate. The best solution to x=b1,…,x=b4x = b_1, \ldots, x = b_4 is the average x^\hat{x} of b1,…,b4b_1, \ldots, b_4. The matrix AA is a column of 1s. Problem 29 gives the expected error (x^−x)2(\hat{x} - x)^2 as σ2(ATA)−1=\sigma^2 (A^T A)^{-1} = \_\_\_\_. By averaging, the variance drops from σ2\sigma^2 to σ2/4\sigma^2/4.

Related theory

Browse the problem library