Feature inference with uncertain categorization: Re-assessing Anderson's rational model

Area of Science:

• Cognitive Psychology
• Decision Making
• Computational Modeling

Background:

• Categories aid predictions about unobserved features.
• Human categorization often involves uncertainty.
• Anderson's (1991) rational model assumes conditional independence of features within categories.

Purpose of the Study:

• To evaluate Anderson's model under conditions of conditional independence.
• To compare Anderson's model against alternative inference models.
• To investigate the influence of multiple categories on feature inferences.

Main Methods:

• Five experiments were conducted, utilizing a novel paradigm in four and an existing paradigm in one.
• The study assessed Anderson's model's predictive accuracy against competing models.
• Participant inferences were analyzed under task environments adhering to conditional independence.

Main Results:

• Anderson's model demonstrated superior predictive performance compared to competing models.
• The findings indicate that inferences are influenced by both the most likely and other candidate categories.
• The model's success suggests that people do not solely rely on the most probable category.

Conclusions:

• Anderson's model provides a strong fit for human inferences in settings with conditional independence.
• Inferences are influenced by a broader consideration of categories than previously assumed.
• Relaxing the conditional independence assumption in Anderson's model may improve its performance in environments with within-category feature correlations.