Inference with cross-lagged effects-Problems in time.

Area of Science:

• Quantitative Psychology
• Statistical Modeling
• Causal Inference

Background:

• Vector autoregressive (VAR) models are widely used to infer causality from cross-effects.
• Interpreting these cross-effects can be problematic when continuous processes are modeled discretely.

Purpose of the Study:

• To highlight the interpretive issues of discrete-time models for continuous processes.
• To propose and demonstrate continuous-time modeling as a solution for accurate causal inference.

Main Methods:

• Simulations were used to demonstrate problems with discrete-time models.
• Stochastic differential equations (SDEs) were parameterized for continuous-time inference.
• An empirical example with intensive longitudinal data was analyzed.

Main Results:

• Discrete-time models can inaccurately represent continuous processes, leading to misinterpretations of causal effects.
• Continuous-time models reveal denser effect matrices than expected from discrete models.
• Switching to continuous-time modeling requires careful consideration of regularization, time lag, and model order.

Conclusions:

• Continuous-time modeling provides a more accurate framework for inferring causality from dynamic systems.
• Proper model specification and parameter interpretation are crucial when applying continuous-time approaches to real-world data.