Causal Inference About Good and Bad Outcomes.

Area of Science:

• Cognitive psychology
• Decision-making
• Causal inference

Background:

• Learning asymmetries, where individuals learn differently from positive versus negative outcomes, are well-documented.
• Existing models often attribute these asymmetries to stable outcome distributions.
• The role of beliefs about the underlying causal structure has been less explored.

Purpose of the Study:

• To investigate how beliefs about hidden causal structures influence valence-dependent learning asymmetries.
• To propose and test a Bayesian model explaining how causal attributions shape learning from good and bad outcomes.

Main Methods:

• Two experiments were conducted with sample sizes of 80 and 255 participants.
• Participants' beliefs about hidden causal structures were explicitly manipulated.
• Behavioral data were analyzed in conjunction with a novel Bayesian computational model.

Main Results:

• The study demonstrated learning asymmetries consistent with the proposed Bayesian model.
• Evidence suggests that attributions to hidden causes, rather than stable outcome distributions, drive learning biases.
• Specific patterns of learning less from negative outcomes (when attributed to hidden causes) and more from positive outcomes were observed.

Conclusions:

• Causal attributions play a mechanistic role in generating biased learning from valenced outcomes.
• Understanding the perceived causal structure of an environment is crucial for explaining learning differences.
• This framework offers new insights into decision-making and belief updating under uncertainty.