ABSTRACT
Social scientists have been turning with increasing frequency to the analysis of nonrecursive linear models to study complex causal relationships among a set of variables. In this chapter, the authors review and examine critically a fourth option for achieving identification proposed by Greenberg and Logan. The method differs from more conventional strategies in two ways. First, it requires observations for at least three time points. Second, it achieves identification by making assumptions about the consistency of parameter values rather than about the fixed values of any individual coefficients or else about the ratios of different coefficients. To illustrate the consistency approach, they consider a three-wave generalization of the model. The authors deals with a “true” structural model, generated the observed correlation matrix among the variables in the model over a very long period of time, and then attempted to recover the underlying structure from this matrix at different points as the system approached equilibrium.
