ABSTRACT
Noncentrality-based confidence interval estimates provide a superior alternative to significance testing for assessing model fit in most standard areas of behavioral statistics, from the t test through multiple regression and analysis of variance to the analysis of covariance structures. These confidence intervals provide all the information inherent in a significance test, and more, and deal with situations more traditional interval estimates cannot handle. For example, in the analysis of variance, noncentrality interval estimation allows computation of exact confidence intervals for (a) standardized measures of effect size and (b) statistical power. In multiple regression, one can compute an exact confidence interval on the squared multiple correlation. Because of computational complexities, noncentrality-based confidence intervals seldom have been computed, except in the analysis of covariance structures. Most of the reasons for not using these interval estimates are no longer relevant in the microcomputer age. In this chapter, we review some of the standard techniques, and provide computational examples.
