On how not to make inferences about measurement error

The growing interest in causal models with unmeasured variables & in using information about measured variables to infer relationships involving unmeasured variables ignores fundamental reasons for considering this approach untenable. Specific attention is given to an example that illustrates the existence of certain untenable inferences. It is assumed in this example, drawn from path analytic methods, that certain other variables are measured perfectly & that they have or do not have certain empirical relationships. Further, it is conceptually inappropriate to consider 'measurement error,' a concept referring to incorrect measurement of one variable, to exist when one variable is used to assign a value to another that is independently defined. Further, regularities cannot be varied for an infinite population, & infinite populations are involved in any universal theory. If observed & predicted variables differ, it should never be assumed that the theory is error-free. 1 Figure. W. H. Stoddard.