Some computing methods for unbalanced analysis of variance and covariance

Abstract

Many computing algorithms are used for analysis of variance and covariance. Choice of an ideal algorithm is dependent on many factors including balance, type of model (fixed, random or mixed), number of dependent variables, covariates, sample size, required auxiliary output, numerical conditioning, and the statistical hypotheses to be tested. In BMD and BMDP, several algorithms are customized to particular classes of problems. In this paper, attention is focused on the general linear hypothesis (general linear model), repeated measures, and efficient computation for multiple observations per cell. Discussion is primarily directed towards testing fixed effects.

Keywords:

• general linear hypothesis
• repeated measures
• analysis of variance
• covariates
• unbalanced
• unequal cell size
• contrast
• missing values