ABSTRACT
Statistical models with latent variables are very popular in the social and behavioral sciences. Much of this popularity is explained by the contribution these models make to the solution of the severe measurement problems that have plagued these sciences. Although theoretical developments have lead to some improvement in the quality of the measurement procedures used in these sciences, a lot is still “measurement by fiat” (Torgerson, 1958). Researchers in these fields collect responses to sets or scales of indicator variables that are assumed to be related to the underlying theoretical construct, and use a sub ject’s scale score as a proxy for the unobservable latent score. Although most measurement scales are used after meticulous item selection and test con struction procedures with the aim to enhance the reliability and the validity of the scale, the problem still remains that the link between the unobserved latent score and the observed scale score is an indirect one.
