Bayesian analyses of cognitive architecture.

Houpt, Joseph W.; Heathcote, Andrew; Eidels, Ami · 2017 · Psychological Methods

DOI: 10.1037/met0000117

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 addresses the challenge of empirically determining cognitive architecture—specifically, how humans temporally organize the processing of multiple information sources (e.g., parallel vs. serial processing). While Systems Factorial Technology (SFT) and its key statistic, the Survivor Interaction Contrast (SIC), provide a rigorous framework for identifying these architectures, historical inference relied on visual assessment or null-hypothesis significance testing (NHST). The authors argue that NHST has limitations, including an inability to provide evidence for null hypotheses and assumptions of independence between SIC components. Consequently, this study develops and evaluates two Bayesian inference methods—a parametric test and a nonparametric test—to improve the statistical toolkit for SFT. The parametric Bayesian test assumes that channel completion times follow an inverse Gaussian distribution, derived from a Brownian motion process with drift. This approach models specific architectures (Parallel-OR, Parallel-AND, Serial-OR, Serial-AND, and Coactive) by combining these distributions according to their respective stopping rules. The authors implemented this using Hamiltonian Monte Carlo sampling in STAN. The nonparametric Bayesian test utilizes a Dirichlet process prior to model uncertainty in survivor functions without assuming specific distributional forms. To address the low prior probability of specific SIC shapes (particularly the flat SIC indicative of Serial-OR), the authors employ a Region of Practical Equivalence (ROPE) and calculate Bayes factors using an encompassing prior approach. The authors compare these new Bayesian methods against the existing NHST approach through simulation studies and empirical data analysis. The simulations explore scenarios where parametric assumptions hold true and where they do not, assessing the accuracy of model selection. The results demonstrate that Bayesian approaches offer distinct advantages, such as the ability to quantify evidence for specific architectural models rather than merely rejecting nulls, and the capacity to incorporate hierarchical structures for group-level inferences. The nonparametric test, in particular, provides a flexible method for identifying architecture without relying on potentially incorrect assumptions about the underlying stochastic processes of information accumulation. The significance of this work lies in providing a more robust and informative statistical framework for cognitive psychology. By enabling researchers to directly compare competing models of cognitive architecture and stopping rules, these Bayesian methods facilitate more precise conclusions about human information processing. This advancement allows for better resolution of long-standing debates regarding whether cognitive systems process information in parallel or series, offering tools that are less prone to Type I error inflation and capable of handling complex experimental designs.

Key finding

The proposed parametric and nonparametric Bayesian tests provide a robust and flexible statistical framework for identifying cognitive architecture and stopping rules, offering advantages over traditional null hypothesis significance testing.

Methodology

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.

• dual task multitasking