Structural Equation Modeling (SEM) basically estimates a series of
separate, but interdependent multiple regression equations simultaneously. The specification of
the model is heavily influenced by theory. SEM has the ability to incorporate latent variables
(similar to factor analysis) that can be approximated by the more common standard measured
variables seen in typical regression equations. Common uses of SEM are in applications
to predict customer satisfaction. Typically, special software has been designed and marketed
especially for this type of analysis. Perhaps the most widely used software is called LISREL. The
development of SEM models can be tedious and requires a significant amount of experience in
working with the software as well as the data. Seven steps are usually involved in developing a
Structural Equation Model:
(1). Developing a theoretically based model
(2). Constructing a path diagram of causal relationships
(3). Converting the path diagram to a set of structural and measurement equations
(4). Choosing the input matrix type and estimating the proposed model
(5). Assessing the identification of the model equations
(6). Evaluating Goodness of Fit
(7). Making the modifications to the model if theoretically justified.
