Structural parameter interdependencies in computational models of cognition.

Area of Science:

• Cognitive science
• Computational psychology
• Psychometrics

Background:

• Computational models of cognition estimate latent psychological variables using adjustable parameters.
• Strong intercorrelations between model parameters can impair parameter estimation and interpretation.
• Parameter interdependencies are common in models combining subjective value functions and probabilistic choice rules.

Purpose of the Study:

• To identify and trace structural parameter interdependencies in prominent computational models of cognition.
• To investigate the impact of parameter intercorrelations on parameter estimation accuracy.
• To propose and evaluate methods for alleviating parameter interdependencies in cognitive models.

Main Methods:

• Analysis of parameter interdependencies in established models (cumulative prospect theory, generalized context model, SIMPLE model).
• Simulation studies using a generic choice model to assess the effects of parameter intercorrelations.
• Application of alternative stochasticity specifications (parameter stochasticity, constant error term) to empirical data.

Main Results:

• High parameter intercorrelations were found to hamper parameter estimation accuracy, particularly for detecting group differences and external variable associations.
• Alternative specifications of model stochasticity, such as parameter stochasticity or a constant error term, can alleviate these interdependencies.
• Alleviating parameter interdependencies led to different conclusions in empirical analyses of risky choice data.

Conclusions:

• Parameter interdependencies represent a significant challenge in computational cognitive modeling.
• Adjusting model stochasticity offers a viable solution to improve parameter estimation and interpretation.
• These findings provide practical guidance for designing and applying more robust computational models in cognitive research.