Accumulator models debate

Lens adapted to this topic: What established models explain and assume

The mainstream modeling view treats evidence accumulation as a useful framework for explaining joint choice and response-time data. It emphasizes established model families—especially diffusion, race, and linear ballistic accumulator models—and compares them by their assumptions about noise, competition, evidence, and thresholds. These models are valued for tractability, interpretable parameters, and links to behavioral and neural dynamics.

• Diffusion and race models are prominent approaches for modeling dynamic decisions through evidence accumulation.
• Linear ballistic accumulator models provide computationally tractable race models for simple-choice behavior and can fit choices and response times.
• Different accumulator architectures vary in whether they use one or multiple accumulators, relative or absolute evidence, competition, and ballistic or diffusive noise.
• Accumulator parameters have been studied in relation to neural dynamics, connecting computational models with brain processes.
• Accumulator models can be extended to account for persistence and disengagement in test-taking, illustrating their application beyond basic choice tasks.

Lens adapted to this topic: Broader models and challenges to standard distinctions

A dissenting or broader research perspective argues that familiar labels such as diffusion model and accumulator model may impose an overly narrow division. Generalized accumulation frameworks allow options to correspond to directions in an accumulation space, while newer models extend accumulation to continuous option spaces and biologically motivated competition. This perspective focuses on whether experiments can genuinely identify the underlying architecture rather than merely fit behavior with interchangeable models.

• Diffusion and racing accumulators may be only special cases of a more general evidence-accumulation process, so model labels alone may not identify the underlying mechanism.
• Discriminating models requires targeted manipulations, including tests of whether near-zero evidence is filtered and whether drift-rate components can be separated.
• Accumulator frameworks can be generalized to continuous option spaces rather than only a fixed set of discrete alternatives.
• A radial-basis leaky competing accumulator model is proposed as a biologically plausible framework for decisions involving continuous options.
• The broader debate concerns identifiability: similar behavioral predictions can arise from different assumptions about evidence, competition, and noise.