Are Cross-Lagged Effect Estimates Larger in the Traditional Cross-Lagged Panel Model Than in the Random Intercept Cross-Lagged Panel Model?

Cross-lagged panel models are widely used to examine whether one psychological construct predicts change in another over time, yet alternative specifications can yield meaningfully different conclusions. This article compares the traditional cross-lagged panel model (CLPM) with the random-intercept cross-lagged panel model (RI-CLPM), which partitions stable person-level differences from within-person fluctuations over time. We make two contributions. First, we derive analytical results that specify when CLPM cross-lagged estimates are expected to be larger than the corresponding RI-CLPM estimates—most notably when the constructs are strongly associated at the stable between-person level and when the RI-CLPM cross-lagged effects are small. The derivations further indicate that RI-CLPM cross-lagged effects are typically estimated less precisely (i.e., with greater sampling variability) than CLPM effects. Second, we illustrate these implications in two case studies that systematically compare CLPM and RI-CLPM estimates for (a) reciprocal relations between self-esteem and depression and (b) reciprocal relations between achievement and academic self-concept, using 5 and 14 longitudinal studies, respectively. Across both applications, RI-CLPM cross-lagged effects were smaller and accompanied by larger standard errors than CLPM estimates. We conclude by discussing implications for model choice and for interpreting cross-lagged effects in psychological research.