Signal Detection Analysis (Claude Skill)

Separates sensitivity from response bias in two-alternative behavioral data, with the formulas, edge-case corrections, and reporting conventions that signal detection theory requires.

   
Type Claude Skill
Supplier Awesome Cognitive and Neuroscience Skills (community OSS, MIT)
Availability GA — one of ~40 research skills in the collection (last upstream push 2026-07-13)
Pricing Free / OSS (MIT)
Capabilities Read/Write — methodology guidance; Claude writes and runs the computation locally
Verified works · 2026-08-06
Security cleared · 2026-08-06 — legitimate NeuroAIHub repo transfer confirmed; osv-advisory flag traced to old/withdrawn scipy GHSA IDs, not applicable to the pinned 1.18.0

How to install

  • Claude Code — plugin marketplace (installs all skills in the collection): 
    /plugin marketplace add HaoxuanLiTHUAI/awesome_cognitive_and_neuroscience_skills
/plugin install awesome-cognitive-and-neuroscience-skills@awesome-cognitive-and-neuroscience-skills

    Restart Claude Code afterwards. The skills are description-activated — there is no slash command; ask for a d-prime analysis and Claude loads the skill.

  • Claude Code — single-skill alternative. This skill declares a required dependency on the collection’s research-literacy skill, so copy both: 
    git clone https://github.com/HaoxuanLiTHUAI/awesome_cognitive_and_neuroscience_skills
cp -r awesome_cognitive_and_neuroscience_skills/skills/signal-detection-analysis ~/.claude/skills/
cp -r awesome_cognitive_and_neuroscience_skills/skills/research-literacy ~/.claude/skills/

    (Project-scoped alternative: copy into .claude/skills/ instead.)

  • Underlying software — none specific. The computation needs only a normal-quantile function, e.g. scipy.stats.norm.ppf from pip install scipy.

What it does

Steps through the analysis in order:

  1. Research planning — state the question and justify SDT over raw percent-correct before touching the data.
  2. Response matrix — classify trials into hits, misses, false alarms and correct rejections.
  3. Rate computation — hit rate and false-alarm rate per condition per subject.
  4. Extreme-value correction — apply the log-linear rule (add 0.5 to every cell) when a rate hits 0 or 1, rather than silently clipping, which would otherwise make d′ infinite.
  5. Sensitivity — d′ = z(H) − z(FA) for the equal-variance case; d_a or A_z when the ROC slope departs from 1 or when distributional assumptions do not hold.
  6. Bias — criterion c = −0.5 × [z(H) + z(FA)], likelihood ratio ln(β) = d′ × c, or the normalized relative criterion c′ = c / d′.
  7. Reporting — document the correction applied, the sensitivity measure chosen, and the tests run.

Primary use cases: recognition-memory and perceptual-discrimination experiments, yes/no vs. forced-choice paradigm comparison, disentangling a bias shift from a genuine sensitivity change in clinical or pharmacological studies.

Notes

AI-generated content — verify before use. The collection’s README states that all skills “were generated by AI from academic papers and other publicly available sources” and that “the content has not been individually verified by human domain experts.” This skill’s front-matter carries review_status: ai-generated. It cites Green & Swets 1966, Macmillan & Creelman 2005, Stanislaw & Todorov 1999 and Hautus 1995 — check the formulas and the correction rule against those references, particularly the choice between the log-linear and 1/(2N) corrections, on which the sources differ.

Related catalogued skill in the same collection: Drift-Diffusion Model, for when reaction times as well as accuracy are available.

Sources

Installed this tool?

Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.