Some Factors Affecting the Success of Specification Searches in Covariance Structure Modeling

Applications of Covariance Structure Modeling often include modifications on the initially specified model. The aim of these modifications is to obtain a model which fits better and/or is more parsimonious than the original model. This process, known as a specification search, is not always straightforward and requires good judgement. Therefore, research which focuses on this practice is essential in order to identify the methods which would most likely lead an investigator to the "true" model. This study provides an extension to the work of MacCallum (1986), and examines the effects of several search strategies when applied to initial models which vary in their degree of "correctness". Results indicate that restricting modifications to those which can be justified on the basis of prior theoretical knowledge, greatly improves the success of a specification search. Assigning priority to the measurement model during such a search appears to offer no advantage. In addition, results point to the importance of a carefully formulated initial model which is based on sound prior theory. When little theoretical knowledge is available, however, a full initial model which includes all permissible parameters in the B and F matrices appears to be better than an ill-constructed one.