Determination of hypothesis testability in linear statistical models

Abstract

Analysts conducting experiments must frequently deal with situations in which data is incomplete or missing. This creates problems that can seriously affect classical hypothesis testing by introducing extraneous terras into the hypothesis in a complicated way. A technique exists that allows an analyst to determine precisely which experimental terms are actually present in a proposed hypothesis and what that hypothesis would actually be testing if employed. This paper examines the mathematics underlying the technique and applies the theory to a widely used data analysis computer package. A computer program is presented to facilitate implementation of the method.