Conditional Direction Dependence Analysis: Evaluating the Causal Direction of Effects in Linear Models with...

Area of Science:

• Statistics
• Causal Inference
• Econometrics

Background:

• Direction dependence analysis (DDA) uses higher-order moments to infer causal direction (x → y or y → x).
• Existing DDA methods are limited when moderators affect variable relationships.
• Moderation complicates the assessment of causal direction in statistical models.

Purpose of the Study:

• Introduce a conditional direction dependence analysis (CDDA) framework.
• Enable evaluation of causal direction in conditional regression effects.
• Assess CDDA performance under moderation and model misspecification.

Main Methods:

• Developed a conditional direction dependence analysis (CDDA) framework.
• Employed Monte-Carlo simulations to evaluate CDDA performance.
• Tested CDDA in scenarios with moderators affecting effect strength and causal direction, and with model misspecifications.

Main Results:

• CDDA significance tests are suitable for moderation scenarios.
• CDDA effectively discerns regions where causal direction is identifiable based on moderator values.
• The framework demonstrates robustness against functional model misspecifications.

Conclusions:

• CDDA extends DDA to effectively analyze causal directions in moderated relationships.
• The proposed framework enhances causal inference capabilities in the presence of moderators.
• CDDA provides a valuable tool for researchers investigating complex causal mechanisms.