Comparing Eight Parameter Estimation Methods for the Ratcliff Diffusion Model Using Free Software.

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Psychometrics

Background:

• The Ratcliff Diffusion Model is crucial for analyzing psychological experiment data.
• Multiple software tools exist for estimating diffusion model parameters.
• Focus on comparing freely available (open-source) estimation routines is needed.

Purpose of the Study:

• To compare the performance of widely used diffusion model parameter estimation tools.
• To evaluate the accuracy and bias of different estimation approaches, particularly open-source options.
• To provide guidance on selecting appropriate methods for real-world data analysis.

Main Methods:

• Computer simulations were used to assess parameter recovery.
• Various estimation algorithms were compared, including Bayesian, Maximum Likelihood, Kolmogorov-Smirnov, Chi-squared, and EZ.
• Performance was evaluated based on the accuracy of recovered parameters like drift rate, starting point, and non-decision time.

Main Results:

• Starting point and non-decision time were recovered more reliably than drift rate.
• Bayesian approaches demonstrated superior performance with a low number of trials.
• Kolmogorov-Smirnov and Chi-squared methods exhibited greater bias compared to Bayesian and Maximum Likelihood methods.
• The EZ method produced significant bias in threshold separation, non-decision time, and drift rate when the starting point was not centered (z ≠ a/2).

Conclusions:

• The choice of diffusion model parameter estimation method significantly impacts results.
• Bayesian methods are recommended for low-trial datasets.
• The EZ method should be used cautiously, especially when biased starting points are suspected, due to potential for deviant estimates.