A tutorial on cue combination and Signal Detection Theory: Using changes in sensitivity to evaluate how observers integrate sensory information

Jones, Pete R. · 2016 · OpenAlex-citations

DOI: 10.1016/j.jmp.2016.04.006

archive: archived pipeline: cataloged verified

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

Summary

This tutorial paper addresses the methodological challenges in evaluating how observers integrate multiple sensory cues, specifically focusing on the limitations of using sensitivity metrics alone to infer decision-making strategies. The authors note that while Signal Detection Theory (SDT) is widely used to measure perceptual sensitivity (e.g., $d'$), the field lacks clarity regarding the diversity of plausible ideal-observer models and the specific assumptions underlying them. Consequently, different studies may define "ideal" performance differently, leading to inconsistent conclusions about whether human observers combine cues optimally. The paper aims to clarify what can and cannot be inferred from behavioral estimates of sensitivity by systematically comparing various cue-combination models. The study employs a theoretical and mathematical approach rather than empirical experimentation. The authors formulate and compare fourteen distinct algorithms for cue combination, ranging from simple strategies like "1-look" (attending to only one cue) and "2-look" (probability summation) to more complex linear summation models. Using SDT, the paper derives quantitative predictions for each model, detailing the assumptions regarding internal noise (e.g., Gaussian, additive, independent) and the formation of the decision variable. The authors provide MATLAB code to simulate these models and calculate expected levels of measurement error. The analysis focuses on simple detection tasks with two cues, deriving formulas to predict changes in sensitivity when moving from single-cue to multi-cue conditions. The primary finding is that predicted sensitivity improvements often differ surprisingly little between qualitatively distinct models of cue combination. For instance, a specific increase in $d'$ can be consistent with multiple different decision strategies, making it difficult to distinguish between them based on sensitivity data alone. The authors demonstrate that the differences between competing models are frequently smaller than the measurement error expected in typical experiments with standard trial counts. For example, distinguishing between certain strategies with 95% confidence would require significantly more trials (e.g., at least 400) than are commonly used. This indicates that sensitivity alone is insufficient for determining the specific efficiency or mechanism of sensory integration. The significance of this work lies in its caution against overinterpreting sensitivity data in cue-combination research. The authors conclude that because sensitivity metrics cannot uniquely identify the underlying decision process, researchers must be aware of the implicit assumptions in their chosen models. The paper provides a comprehensive taxonomy of models and tools to quantify prediction errors, urging the field to adopt more rigorous experimental designs or alternative analytical approaches to accurately assess how observers integrate sensory information. This tutorial serves as a critical guide for interpreting psychophysical data and avoiding erroneous conclusions about optimal sensory processing.

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.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-25; verification: verified.

Topics

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

• signal detection theory
• multimodal feedback