Modeling Strategies: In Search of the Holy Grail

Modeling strategies are subject to debate for virtually all statistical procedures. Witness the sharp disagreements over stepwise regression, the interpretation of clusters in cluster analysis, or the identification of outliers and influential points. The largely objective basis of statistical algorithms does not remove the need for human judgment in their implementation. So it is not surprising that the use of structural equation models is subject to disputes over the best way to formulate and test models. Though I must admit considerable scepticism about whether it is possible to have a single generic strategy that would prove optimal over all substantive areas and types of structural equation models, articles and discussions such as those of Hayduk and Glaser (2000) and Mulaik (1998) are very helpful in that they bring out the merits and limits of the alternative procedures. The options that are the focus of their discussions are (a) the one-step procedure and (b) the four-step approach. In a sense, we could see the one-step procedure as there at the birth of contemporary structural equation models. One of the attractive features of structural equation models was the ability to simultaneously model the latent variable and the measurement models in one step.1 Defenders of this original practice include Hayduk (1987, 1996) and Fornell and Yi (1992). More recently, Mulaik (1998) has advocated a four-step method that is the subject of the Hayduk and Glaser (2000) article. STRUCTURAL EQUATION MODELING, 7(1), 74–81 Copyright © 2000, Lawrence Erlbaum Associates, Inc.