Rule following and rule use in the balance-scale task

Area of Science:

• Cognitive Psychology
• Developmental Psychology
• Computational Neuroscience

Background:

• Connectionist models are used to explain cognitive development.
• Latent Class Analysis (LCA) has been applied to critique these models, specifically regarding the balance-scale task.
• Quinlan et al. argue LCA reveals flaws in a connectionist model, suggesting it fails to capture the torque rule and attributes non-existent rules to children.

Purpose of the Study:

• To address criticisms of connectionist models of cognitive development based on Latent Class Analysis (LCA).
• To highlight the tendency of LCA to identify small, unreliable, and difficult-to-interpret classes.
• To recommend cautious application of LCA in psychological research and developmental rule analysis.

Main Methods:

• Rejoinder to Quinlan et al. (this issue).
• Analysis of the tendency of LCA to find small and unreliable classes.
• Documentation of this tendency through network simulations, synthetic data, and psychological research.
• Discussion of statistical reasons for unreliable class identification.

Main Results:

• Latent Class Analysis (LCA) exhibits a tendency to find small, unreliable, and difficult-to-interpret classes.
• This tendency is empirically supported by simulations and psychological research.
• A focus on rule-diagnosis via LCA may overlook crucial developmental phenomena not explained by simple rules.

Conclusions:

• Latent Class Analysis (LCA) should be used with caution in developmental psychology.
• Small and unreliable classes identified by LCA should be discounted.
• Simple extensions to existing network models may be sufficient to achieve torque-rule performance in balance-scale tasks.