MERLiN: Mixture Effect Recovery in Linear Networks

Causal inference concerns the identification of cause-effect relationships between variables. However, often only a linear combination of variables constitutes a meaningful causal variable. We propose to construct causal variables from non-causal variables such that the resulting statistical properties guarantee meaningful cause-effect relationships. Exploiting this novel idea, MERLiN is able to recover a causal variable from an observed linear mixture that is an effect of another given variable. We illustrate how to adapt the algorithm to a particular domain and how to incorporate a priori knowledge. Evaluation on both synthetic and experimental EEG data indicates MERLiN's power to infer cause-effect relationships.