Foundations for the Theory of Least Squares
Bruce M. Hill · Journal of the Royal Statistical Society Series B (Statistical Methodology) · 1969
Summary The Bayesian theory of least squares is founded upon a weaker and more tangible form of prior knowledge than the conventional assumption of normality. The underlying assumption is a form of conditional uniformity on spheres for the “actual errors” in the experiment. This provides a unified theory appropriate for randomization models in the analysis of variance as well as for classical least-squares analysis.