ABSTRACT
Latent variables provide a way to infer properties of constructs that cannot be directly measured. We build statistical models, such as confirmatory factor models, to identify latent constructs and test theories about them. This methodology has been extraordinarily successful and has become extremely popular. Increasingly, the methods and models for identifying and estimating properties of latent variables are becoming reified. That is to say, it is increasingly common for an experiment to be designed around the methodology of latent variable modeling and perhaps around a particular type of model. This is not necessarily a bad thing. One would hope that an expensive data collection initiative would be likely to provide convincing evidence about a thesis or its antithesis. But the well–oiled gears of latent variable modeling may be so attractive that we do not consider alternate theories that do not fit into this smoothly functioning machine.
