Using Instrumental Variables to Measure Causation over Time in Cross-Lagged Panel Models

Area of Science:

• Social Sciences
• Psychology
• Biostatistics

Background:

• Cross-lagged panel models (CLPMs) are standard for analyzing causal relationships in longitudinal data.
• CLPMs' ability to detect lagged effects diminishes with longer assessment intervals.
• Existing methods struggle with inferring causality when time intervals are extensive.

Purpose of the Study:

• To enhance causal inference in CLPMs by incorporating instrumental variables (IVs).
• To develop a modeling strategy for estimating both distal and proximal effects.
• To address the limitation of time intervals affecting the detectability of lagged effects.

Main Methods:

• Integration of instrumental variables (IVs) into a two-wave, two-variable CLPM framework.
• Estimation of both Granger-causal (distal) and contemporaneous (proximal) effects.
• Utilizing simulations and an empirical application to validate the proposed model.

Main Results:

• The proposed IV-CLPM approach allows for the estimation of distal effects (decaying with time) and proximal effects (accruing over time).
• Proximal effects provide crucial insights into causality when distal effects become undetectable due to long time intervals.
• Demonstrated the impact of time intervals on causal inference and presented strategies to overcome these limitations.

Conclusions:

• The IV-CLPM offers a robust method for causal inference in panel studies, irrespective of the time interval.
• Significant proximal effects with negligible distal effects indicate that the time interval may be too long for standard CLPM lagged effect estimation.
• Highlights the utility and constraints of using genetic variables as IVs in large-scale panel data analysis.