Against the ubiquity of the random intercept cross-lagged panel model

Researchers in psychology often use longitudinal data to estimate cross-lagged effects—that is, how a variable at one time point (e.g., X_(t-1)) influences another at a later time point (e.g., Y_t). In a recent critique, Lucas (2023) argued that the traditional cross-lagged panel model is “almost never the right choice” because it fails to account for stable trait factors. As an alternative, he recommends models incorporating latent variables, such as the random-intercept cross-lagged panel model. In this comment, we challenge the view that including random intercepts is inherently superior or necessary for estimating cross-lagged effects. We distinguish between two strategies for addressing confounding: (1) adjusting for observed covariates, and (2) modeling latent variables to capture unmeasured, time-invariant influences. We argue that blanket recommendations to always include random intercepts are premature and potentially misleading. In many research contexts, adjusting for observed covariates remains a reasonable and defensible approach.