Difficulties with using correlations to determine the relative strength of effects of latent variables

Abstract

It is common to use correlation data to estimate path coefficients in structural equation models. In this article we argue that it might be problematic to make inferences from path coefficients based upon correlations to the relative strength of exogenous model variables’ effects because correlations only express the degree of linearity in the relation between variables but not their slopes. We use a data example to illustrate possible misinterpretations. Additionally we analyze whether using covariance data input provides information about relative strength of effects. Our data example shows that neither correlation nor covariance data based estimations of LISREL path coefficients necessarily need to reflect relative strength of the effects of exogenous variables.