Diffusion versus linear ballistic accumulation: different models for response time with different conclusions about...

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Decision Science

Background:

• Two main classes of evidence-accumulation models, diffusion and racing accumulator pairs, dominate rapid binary choice research.
• The Ratcliff diffusion (RD) and linear ballistic accumulator (LBA) models are the least similar within their respective classes.
• Previous research showed model mimicry when only evidence accumulation rates differed.

Purpose of the Study:

• To investigate potential divergent inferences between RD and LBA models when response caution and nondecision time parameters vary.
• To examine the fit of RD and LBA models to a dataset with a practice manipulation not previously surveyed.
• To compare the explanatory power of LBA and RD models for practice effects.

Main Methods:

• Simulations comparing RD and LBA model performance under varying parameter conditions.
• Analysis of a dataset from Dutilh et al. (2009) using a practice manipulation.
• Model fitting and comparison of RD and LBA to the experimental data.

Main Results:

• Simulations revealed trade-offs between response caution and nondecision time parameters, potentially leading to divergent inferences.
• RD model fits to the Dutilh et al. data indicated practice affected all parameters.
• The LBA model provided a simpler and alternative account of practice effects in this dataset.

Conclusions:

• While RD and LBA models can yield equivalent inferences under certain conditions, differences in response caution and nondecision time can lead to divergent conclusions.
• The LBA model offers a more parsimonious explanation for practice effects observed in the examined dataset.
• Findings highlight the importance of considering model assumptions and parameter trade-offs in evidence accumulation modeling.