Basic issues

The gold standard for studying causal relations is experimentation. As Fisher (1956) labored so hard to demonstrate, experimental manipulations have the ability to disentangle factors in a way that is usually not possible with nonexperimental data. By creating independence among causes, experimentation can lead to a great reduction in ambiguity about effects. There is little doubt for most scientists that well designed and properly analyzed experiments provide the most powerful way of assessing the importance of processes, when appropriate and relevant experiments are possible.

In this chapter I address a topic that generally receives little attention in discussions of SEM, its applicability to experimental studies. I hope to deal with two common misconceptions in this chapter, (1) that multivariate analysis is only for use on nonexperimental data, and (2) that when experiments are possible, there is no need for SEM. In fact, I would go one step further and say that the value of studying systems using SEM applies equally well to experimental and nonexperimental investigations.

There are several reasons why one might want to combine the techniques of SEM with experimentation. First, using experiments to evaluate multivariate relationships provides inherently more information about the responses of a system to manipulation. It is often difficult and sometimes impossible to exert independent control over all the variables of interest in a system. Examination of how the various pathways among variables respond to experimental treatment can yield important insights into system function and regulation.