Theory, Deduction, and Rules of Correspondence

Causal models are strictly untestable with a single indicator for each variable, unless one assumes very slight measurement error. With multiple indicators for each variable, incorporated as an extension of the causal model, estimates of the path coefficients may be derived that are subject to sampling variability but are not distorted in the same manner as estimates based on single indicators. Under certain specified conditions, the coefficients thus derived will be inconsistent, thus providing a clue to the existence of nonrandom measurement error of specific kinds. In this paper, procedures for two-indicator and three-indicator models are explored.