Reasoning about causal relationships: Inferences on causal networks

Area of Science:

• Cognitive Psychology
• Decision Science
• Artificial Intelligence

Background:

• Bayesian Probabilistic Causal Networks are a dominant framework for causal inference in psychology.
• These networks provide a normative model for calculating probabilities based on causal relationships.

Purpose of the Study:

• To provide a tutorial on normative calculations within Bayesian Probabilistic Causal Networks.
• To systematically review and compare human judgments against these normative models.

Main Methods:

• Review of normative calculation methods for Bayesian Probabilistic Causal Networks.
• Systematic analysis of behavioral studies comparing human judgments to normative predictions across various network structures and inference types.

Main Results:

• Human judgments generally align directionally with normative calculations (increasing or decreasing as predicted).
• Two systematic deviations were observed: violation of the Markov assumption and reduced sensitivity to network parameters and structure.

Conclusions:

• Human causal inference shows systematic deviations from normative Bayesian models.
• Future research should explore these deviations, particularly the violation of the Markov assumption and parameter insensitivity.