Modeling Unmeasured Third Variables in Longitudinal Studies

This article discusses techniques to account for unmeasured third variables in longitudinal designs. As an extension of the synchronous common factor (SCF) model, which is the basis of the cross-lagged panel correlation technique, a series of less restrictive synchronous common factor (LRSCF) models are introduced. LRSCF models can be tested by using structural equation modeling. Although both SCF and LRSCF models rest on a 2-wave design, the SCF model requires only 2 variables measured at each wave, whereas LRSCF models require 3 variables. This enables several restrictions of the SCF model to be relaxed. Among others, a particular advantage is that the strength of lagged relations among the variables can be estimated. It is recommended that LRSCF models be applied routinely whenever possible third variables might have been omitted.