Response time distributions and the Stroop task: A test of the Cohen, Dunbar, and McClelland (1990) model.

Mewhort, D. J.; Braun, Jacqueline G.; Heathcote, Andrew; Heathcote, Andrew · 1992 · Journal of Experimental Psychology Human Perception & Performance

DOI: 10.1037//0096-1523.18.3.872

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 evaluates the validity of the Cohen, Dunbar, and McClelland (1990) connectionist model in explaining performance on Stroop-like interference tasks. The Stroop effect, characterized by slower response times when naming the color of a conflicting color word compared to neutral or congruent stimuli, is widely attributed to the automaticity of reading versus the controlled nature of color naming. Cohen et al. proposed a model where evidence for potential responses accumulates in parallel, with interference arising from stronger pathways associated with reading experience. While the model successfully predicted changes in mean response latency (MRT), Mewhort, Braun, and Heathcote sought to determine whether it accurately accounted for the underlying distribution of response times, arguing that MRT changes are driven by shifts in distribution shape. The researchers tested the model by simulating 10,000 trials for congruent, neutral, and incongruent conditions across 135 combinations of noise and processing parameters. The model architecture consists of a 12-node encoder network with input, hidden, and output layers, trained via back-propagation to establish asymmetric weights favoring word reading over color naming. Response latency was modeled as a linear function of the cycles required for evidence accumulators to reach a threshold, with Gaussian noise added to simulate human variability. The simulated latency distributions were characterized using the ex-Gaussian distribution, which separates the leading edge of the distribution (parameter $\mu$) from the skew (parameter $\tau$). The results demonstrated that while the model correctly predicted the ordering of mean response latencies across conditions, it failed to replicate the shape of the latency distributions observed in human subjects. Specifically, the model’s predictions for the skew and variance of the distributions did not align with empirical data, even though these distributional changes are the primary drivers of the MRT differences. The study found that the model’s ability to predict MRT was robust across various noise parameters, but the mechanism underlying this prediction was flawed. The model predicted MRT successfully but for the wrong reason, as it did not capture the specific ways in which interference alters the tail and skew of response time distributions. The authors conclude that the Cohen, Dunbar, and McClelland model is not an adequate account of human performance in the Stroop task. Although the model captures the general trend of increased latency under interference, its failure to model the distributional properties of response times suggests that its theoretical assumptions about parallel evidence accumulation and pathway strength are insufficient. This finding implies that future models of cognitive interference must account for the detailed shape of response time distributions, not just mean values, to provide a valid explanation of the Stroop effect.

Key finding

The Cohen, Dunbar, and McClelland model correctly predicts mean response latency changes in the Stroop task but fails to account for the underlying changes in the shape of response time distributions, rendering it an inadequate explanation of human performance.

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 (2 acquisition events logged).

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.

• dual task multitasking