A Cautionary Note on Evidence-Accumulation Models of Response Inhibition in the Stop-Signal Paradigm

Matzke, Dóra; Logan, Gordon D.; Heathcote, Andrew · 2020 · Computational Brain & Behavior

DOI: 10.1007/s42113-020-00075-x

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper investigates the validity of using evidence-accumulation models to characterize response inhibition within the stop-signal paradigm. While the stop-signal race model is a standard framework for studying the ability to withhold ongoing responses, previous approaches often relied on descriptive distributions (like the ex-Gaussian) that lack direct psychological interpretation. Recent proposals, such as the racing Wald model and a novel lognormal race model, attempt to model the "runners" in the race as evidence-accumulation processes to provide deeper insights into cognitive mechanisms like decision thresholds and accumulation rates. The authors question whether these models function as true "measurement models," defined as having a one-to-one mapping between data-generating parameters and their estimates, which is essential for quantifying individual differences and experimental effects. To evaluate this, the authors conducted a series of simulation studies and fits to empirical data. They examined two specific architectures: the racing Wald model, which separates evidence-accumulation rates from decision thresholds, and the stop-signal lognormal race model, which identifies only the ratio of thresholds to rates. Using Bayesian parameter estimation via Differential Evolution Markov Chain Monte Carlo, they tested parameter recovery across four scenarios with varying trial counts: 100 stop trials (representative of clinical studies), 200 and 400 stop trials (typical experimental designs), and 3,200 stop trials (a near-asymptotic case). The simulations generated synthetic data with known true parameters to assess bias and uncertainty in the recovered estimates. The results demonstrate that neither model serves as a reliable measurement model under realistic experimental conditions. While go-process parameters were recovered with high accuracy and appropriate uncertainty estimates, stop-process parameters exhibited severe bias. Specifically, the stop accumulation rate was consistently underestimated, while the scale parameter and non-decision time were overestimated. This bias persisted even in the largest simulation condition with 3,200 stop trials, where confidence intervals failed to cover the true values. Furthermore, in smaller samples, the models sometimes produced narrow posterior distributions for biased parameters, creating a false sense of precision. The authors found that the structural limitations of these models, particularly regarding parameter-dependent lower bounds and the partial observability of stop trials, prevent the recovery of psychologically meaningful parameters. The significance of these findings is a cautionary warning against interpreting stop-signal evidence-accumulation parameters as direct measures of underlying cognitive processes. Because the models cannot reliably recover the true data-generating parameters, inferences about individual differences or group deficits (such as those in schizophrenia) based on these parameters are likely invalid. The authors conclude that researchers should remain skeptical of claims derived from these models and suggest that descriptive approaches, which do not claim to isolate specific cognitive components, may currently offer more robust characterizations of stop-signal performance.

Key finding

Evidence-accumulation models for the stop-signal paradigm are not measurement models because their data-generating parameters cannot be recovered in realistic experimental designs.

Methodology

simulation_modeling

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via author_sweep_intake on 2026-05-28.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• signal detection theory