Modeling Development and Testing

Structural equation modeling (SEM) provides scholars the opportunity to articulate, and then test, a system of relationships between variables. This statistical approach is useful in advancing health communication research because scholars have the opportunity to statistically test the overarching picture outlined by the theoretical structure of a set of variables. Certainly there are other statistical tests that may be more appropriate to the type of questions asked by scholars or the types of variables that are measured; indeed, SEM is not infallible and there is much room for the misuse of this technique, particularly as software becomes increasingly intuitive to use. Nevertheless, SEM generally has two advantages not available from other statistical tests: (1) the simultaneous analysis of an entire theoretical model that articulates a set of relationships between constructs; and (2) the use of latent (error-free) variables.

Many argue that, at multiple levels, communication is not only affected by internal and external factors, but that communication itself influences outcomes. Consequently, scholars who study health communication have an opportunity to examine a broader network of theoretical variables in order to understand the complexity of the relationships between them. In turn, this allows scholars and others to gain a greater understanding of communication within this process as well as within the broader nomological network.

Viewing communication as a process suggests that one might fully test theoretical models rather than analyzing small or discrete sections of the model individually. In many circumstances, it may be necessary and thus reasonable to do both. Nonetheless, in order to understand how all the theoretical variables work together, a test of the full theoretical model is necessary. This is where SEM is useful. Take, for example, the extended parallel process model (EPPM), which is a fear-appeal theory used routinely in health communication research. The theory suggests that responses to a fear appeal message depend upon the relationship between perceived threat and perceived efficacy; that a fearful emotional response is a mediating variable; and that ultimately, the attitudinal or behavioral response is fully conditional upon the relationships between threat, efficacy, and fear. SEM allows one to analyze the entire system of variables (in this example, threat, efficacy, fear, attitudes, and intentions) simultaneously.

Although there are complex issues that cannot be fully captured or described in a general review of SEM, the entire SEM process is typically represented in three steps: model specification, model estimation, and model evaluating.