You have measured two variables across three waves and you want to know which predicts which. The classical cross-lagged panel model is the first method that comes to mind — and the most criticised method of the past decade.

Longitudinal and dyadic data analysis
Choosing a model for longitudinal data depends on whether your question is within-person or between-person.

What the classical CLPM does

The cross-lagged panel model (CLPM) takes each variable’s previous measurement, and the other variable’s previous measurement, as predictors. If the cross-lagged paths are significant, the conclusion drawn is “X predicts Y over time”.

The problem: that interpretation assumes stable differences between individuals have been controlled in the model. CLPM does not control them.

The hidden assumption: does everyone return to the same mean?

In CLPM what is modelled is each participant’s deviation from the overall sample mean, not from their own stable level. The model therefore implicitly assumes that all individuals share the same long-run level.

In real data this is almost never true. Some people are chronically more anxious, others chronically less so. If those stable individual differences do not enter the model, the cross-lagged paths become confounded with their residue.

The meaning of the result changes: because individual differences are uncontrolled, cross-lagged paths from a CLPM may reflect a between-person pattern rather than a within-person process. You want to say “when a person’s anxiety rises, their sleep worsens”; the model is actually saying “anxious people sleep worse”.

What RI-CLPM changes

The random-intercept cross-lagged panel model (RI-CLPM) defines a stable individual level (a random intercept) for each participant and estimates the cross-lagged paths on deviations from that level.

The model thereby separates out the fixed between-person differences, leaving within-person fluctuation. If your question is “when a person’s X increases, does their Y increase too?”, this is the model you want.

CLPMRI-CLPM
What it modelsDeviation from the sample meanDeviation from the person’s own level
Stable individual differencesNot controlledModelled as a separate layer
Level of interpretationBetween-personWithin-person
Minimum waves23

Which should you choose?

  1. Clarify your questionAre you asking about within-person change or about differences between groups? The answer determines the model.
  2. Check the number of wavesRI-CLPM requires at least three measurements. With two waves it cannot be specified.
  3. Test measurement invarianceDo not compare waves without showing that you are measuring the same construct over time.
  4. Report bothFitting both models and discussing the difference is a strong approach in peer review.
Note: RI-CLPM is not always the “more correct” model. If your research question really is about a between-person pattern, CLPM may be appropriate. What is wrong is moving to interpretation without having chosen the model deliberately.

In short

Choosing a model for longitudinal data is a theoretical decision, not a technical preference. What matters is less which model you use than being able to explain why you used it. That justification belongs explicitly in the method section.

Longitudinal and Dyadic Analysis course

CLPM, RI-CLPM, latent growth models and APIM for dyadic data — all hands-on, working with your own data structure.

Course details →

References

  1. Hamaker, E. L., Kuiper, R. M., & Grasman, R. P. P. P. (2015). A critique of the cross-lagged panel model. Psychological Methods, 20(1), 102–116.
  2. Mulder, J. D., & Hamaker, E. L. (2021). Three extensions of the random intercept cross-lagged panel model. Structural Equation Modeling, 28(4), 638–648.