A Design-Sensitive Adjustment to the Parsimony Ratio for Evaluating Omnibus Fit of Structural Equation Models

Abstract

In response to Marsh and Hau's (1996) recent article on the potential for inferential errors when parsimony is rewarded in the evaluation of overall fit of structural equation models, Hoyle proposes a design-sensitive adjustment to the standard parsimony ratio. The design-sensitive parsimony ratio distinguishes be tween free parameters in a model that are discretionary and those that are required to reflect the design of the research. Hoyle argues that in parsimony adjustments to normed indices of omnibus fit, parameters dictated by research design should not contribute to the downward adjustment to fit indices effected by the parsimony ratio. A reconsideration of Marsh and Hau's example, a simplex model, showed that the design-sensitive parsimony ratio renders a more reasonable upper bound for parsimony indices than does the standard parsimony ratio. A brief description of 4 classes of research design for which the design-sensitive parsimony ratio should be used routinely is presented.