Accurate knowledge about feature diagnosticities leads to less preference for unidimensional strategy

Area of Science:

• Cognitive Psychology
• Artificial Intelligence
• Machine Learning

Background:

• Classification learning often shows a preference for unidimensional categorization.
• Participants favor perfectly diagnostic dimensions over partially diagnostic ones.
• This unidimensional bias is absent in array-based classification tasks.

Purpose of the Study:

• To investigate the influence of learning feature diagnosticity on categorization preferences.
• To replicate previous findings on unidimensional bias in classification learning.
• To explore the role of accurate knowledge in reducing unidimensional categorization.

Main Methods:

• Replication of prior classification learning experiments (Experiment 1).
• Introduction of repeated testing to enhance learning of partially diagnostic features (Experiment 2).
• Application of Bayesian modeling to assess participants' knowledge of feature diagnosticity.

Main Results:

• Experiment 1 confirmed the preference for unidimensional categorization with perfectly diagnostic dimensions.
• Experiment 2 demonstrated a decrease in unidimensional categorization after learning partially diagnostic features.
• Accurate knowledge (≥75% diagnosticity accuracy) was crucial for using a dimension in categorization.

Conclusions:

• Accurate knowledge of feature diagnosticity reduces the preference for unidimensional categorization.
• This finding offers an explanation for why simple categorization is often preferred in learning.
• Understanding diagnosticity is key to flexible and accurate classification strategies.