To study the dynamics among latent variables, researchers often use structural equation modeling (SEM). The real interest of SEM is often in the structural part, which explains the relationships among the latent variables. The indicators are peripheral; they exist because it is impossible to obtain the latent variables’ scores. Perhaps because the indicators are peripheral in developing substantive theories, how to choose indicators for SEM has received little attention in the literature. The attention has been largely drawn to model estimation and evaluation given a set of indicators, but the importance of indicators for the quality of model estimation has been overlooked. In this paper, we study the trade-offs among the variable types, quantity, and quality of indicators with respect to the quality of estimating structural-level model parameters. Better selections of indicators can substantially improve estimation quality without increasing the sample size. Using many lower-loading indicators is not necessarily inadvisable.
Read full abstract