Non-bayesian inference: causal structure trumps correlation.

Area of Science:

• Cognitive psychology
• Decision science
• Causal inference

Background:

• Conditional probability judgments are fundamental to decision-making.
• Previous research often assumes statistical information directly drives these judgments.
• The role of underlying causal structures in probability estimation is less understood.

Purpose of the Study:

• To investigate whether causal links, independent of statistical relations, affect conditional probability judgments.
• To compare the impact of different causal structures (chain, common cause, predictive, diagnostic, direct, indirect) on probability estimations.

Main Methods:

• Three experiments were conducted, manipulating the causal structure between three variables.
• Participants' conditional probability judgments were recorded under varying causal conditions.
• A Bayesian learning model was used to assess its explanatory power.

Main Results:

• Causal chains led to higher probability judgments than common cause structures.
• Predictive causal chains (evidence as cause) yielded higher judgments than diagnostic chains (evidence as effect).
• Direct causal chains resulted in higher judgments than indirect chains.

Conclusions:

• Causal structure significantly influences conditional probability judgments, beyond statistical information.
• A Bayesian learning model failed to account for the observed effects.
• An explanation-based hypothesis, emphasizing belief updates about causal structure, better explains the findings.