Abstract
Maximum likelihood analysis of repeated measures. Biased tests and non-exact F distributions are deficiencies in the classical ANOVA approach to repeated measures in the case of non-spherical variance covariance matrices or unbalanced data. Individual differences in developmental patterns have also been typically ignored by being shoved into the error term. This paper discusses two case studies that illustrate analysis opportunities in the solution of these problems by using maximum likelihood methods. Evidence suggests that the latter allow for more precise inference by using more accurate models of the covariance matrix. They also provide estimators with known and favorable asymptotic properties in the case of unbalanced data. Individual differences in growth can be quantified by inclusion of new covariance parameters, which provide some answers to Barlow and Hensen critical remarks on classical experimental psychology. Some potential problems in the use of these methods and characteristics of available software are finally discussed.
