The search for causality: A comparison of different techniques for causal inference graphs

Area of Science:

• Psychology
• Causal Inference
• Data Science

Background:

• Estimating causal relations is crucial for understanding psychological phenomena.
• Observational data alone is insufficient for complete causal inference.
• Combining observational and experimental data can improve causal estimation.

Purpose of the Study:

• To evaluate the performance of various causal discovery algorithms.
• To identify optimal algorithms for estimating causal relations in psychological research.

Main Methods:

• Simulation study comparing Peter and Clark, Downward Ranking, Transitive Reduction, Invariant Causal Prediction (ICP), and Hidden Invariant Causal Prediction (HICP) algorithms.
• Application of algorithms to an empirical dataset to assess real-world performance.

Main Results:

• ICP and HICP algorithms demonstrated superior performance across most simulation conditions.
• Empirical application highlighted similarities and differences among the evaluated algorithms.

Conclusions:

• Invariant Causal Prediction (ICP) and Hidden Invariant Causal Prediction (HICP) are effective for causal discovery.
• The combined use of ICP and HICP algorithms is recommended for future psychological research.