Causal relationships in longitudinal observational data: An integrative modeling approach

Area of Science:

• Psychology
• Statistics
• Computational Science

Background:

• Observational studies in psychology often lack causal interpretability.
• Advances in statistics and computational science offer methods to infer causality from correlational data.

Purpose of the Study:

• To design and empirically test a novel approach for identifying potential causal factors in longitudinal correlational data.
• To integrate interventional and time-restrained notions of causality.

Main Methods:

• Developed and tested a new causal discovery approach.
• Utilized principled simulations to evaluate the method.
• Applied the approach to a large cohort study on early-life determinants of cognitive development.

Main Results:

• Simulation results demonstrated both the potential and limitations of discovering causal factors in observational data.
• Identified plausible early-life determinants of cognitive abilities in 5-year-old children in the illustrative application.

Conclusions:

• Exploratory causal discovery holds promise for psychological research using observational data.
• Discussed the limitations, potential misuses, and misinterpretations of these methods.