ABSTRACT
This chapter deals with a situation in which the multiple-indicator approach is advantageous. The models to be tested involve variables which are almost certainly highly related; hence, the epistemic correlations and structural parameters are very strong. Alvin L. Jacobson and N. M. Lalu have compared three measurement strategies, the single-indicator approach, the index approach, and the multiple-indicator approach. The chapter examines more specifically with criteria for the selection of indicators, given several from which to choose. Herbert L. Costner has proposed a criterion to determine whether measurement error is random or nonrandom, provided one has multiple indicators of each theoretical construct and can express the correlations among indicators in terms of path coefficients. Model specification and parameter estimation should yield similar results regardless of the particular set of indicators used, unless there are major differences in the nature of their measurement error.
