Assessing the informational value of parameter estimates in cognitive models

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Mathematical Psychology

Background:

• Mathematical models of cognition frequently involve parameters estimated from data.
• The informativeness of these parameter estimates is a critical but often neglected aspect.
• Standard errors, typically within a maximum likelihood framework, measure parameter estimate informativeness.

Purpose of the Study:

• To explore the informativeness of parameter estimates in cognitive models.
• To demonstrate the utility of standard error calculations for assessing parameter informativeness beyond the maximum likelihood framework.
• To apply these methods to specific cognitive models, revealing novel insights into their estimation properties.

Main Methods:

• Interpreting standard error as the standard deviation of parameter estimates across multiple samples.
• Extending standard error calculation for interpretable intervals outside the strict maximum likelihood framework.
• Combining standard error analysis with graphical methods for model evaluation.

Main Results:

• Standard error calculations provide valuable, interpretable information about parameter estimates, even when maximum likelihood assumptions are not fully met.
• Application to the ALCOVE and exemplar-based random walk models revealed previously unknown characteristics of these models.
• The analysis yielded a combination of favorable and unfavorable findings regarding the estimation of these specific cognitive models.

Conclusions:

• Standard error analysis offers a robust and broadly applicable method for assessing the informativeness of parameter estimates in cognitive models.
• This approach enhances understanding of model identifiability and reliability, crucial for scientific progress.
• The findings encourage wider adoption of these methods in cognitive science for more rigorous model evaluation.