ABSTRACT
In introducing unmeasured intervening variables into a stimulus-response setup, one is attempting to avoid the extreme operational orientation that, in many practical instances, has characterized much empirical research in sociology and other social sciences. The causal connections between unmeasured variables and their indicators should also be made explicit so that implications for tests and estimating procedures can be noted. In models involving additional unmeasured variables, a similar strategy of locating both causal indicators and effect indicators would also apply. In general, the more complex and numerous the linkages among the unmeasured variables, the simpler the required assumptions about the indicators and the more numerous causal indicators must be in order to avoid identification problems. It would be helpful to develop necessary and sufficient conditions for the general model involving unmeasured as well as measured variables, but in the absence of such general criteria, one can readily investigate the properties of specific models that seem most appropriate in a given context.
