Parameter recovery, bias and standard errors in the linear ballistic accumulator model

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Psychometrics

Background:

• The linear ballistic accumulator (LBA) model is widely used for analyzing response time data in cognitive experiments.
• Accurate parameter estimation and inference are crucial for the reliable application of the LBA model.

Purpose of the Study:

• To validate the `glba` R package for fitting the LBA model using maximum likelihood estimation.
• To compare methods for computing parameter standard errors within the LBA framework.
• To demonstrate the application of LBA model fitting and inference in analyzing implicit learning data.

Main Methods:

• Parameter recovery simulation study to assess the accuracy of the `glba` package.
• Comparison of Hessian-based and bootstrap methods for calculating parameter standard errors.
• Application of the LBA model to empirical data from an implicit learning experiment.

Main Results:

• The `glba` package demonstrates good parameter recovery at sufficient sample sizes, though bias can occur with smaller samples.
• The Hessian-based method for computing standard errors is found to be adequate and significantly faster than the bootstrap method.
• The LBA model effectively captures typical implicit learning effects through its parameters.

Conclusions:

• The `glba` R package is a validated tool for LBA model fitting, with performance dependent on sample size.
• The Hessian-based method offers an efficient and reliable approach for estimating parameter standard errors in LBA models.
• LBA model analysis provides insights into the mechanisms underlying implicit learning.