A dependence detection heuristic in causal induction to handle nonbinary variables.

Area of Science:

• Cognitive Science
• Psychology
• Artificial Intelligence

Background:

• Human causality estimation is a key cognitive science question.
• Previous models, like pARIs, described binary causal estimation.
• A gap existed in modeling multi-valued causal induction.

Purpose of the Study:

• To develop and validate a new descriptive model for human causal induction with multi-valued variables.
• To extend the pARIs model's applicability beyond binary variables.
• To analyze the properties of the new model through simulations and experiments.

Main Methods:

• Developed the pARIs_mean model, extending the pARIs framework to multi-valued variables.
• Conducted a causal induction experiment with human participants to gather response data.
• Performed computer simulations to analyze model properties and performance with limited data.

Main Results:

• The pARIs_mean model showed a high correlation (r=0.976) with human causal induction estimates.
• Computer simulations indicated the model effectively estimates population mutual information with sparse data.
• Model performance was robust under conditions of near-equal and small probabilities for cause and effect.

Conclusions:

• The pARIs_mean model is a valid and highly descriptive tool for human causal induction with multi-valued variables.
• The model offers insights into human causal estimation tendencies and performs well in data-limited scenarios.
• This research advances computational cognitive science by providing a more versatile model for causal reasoning.