Statistical Power & Sample Size

What it does

• A priori power analysis: give an effect size, α, and target power (typically 0.80/0.90) and get the required n per group and total.
• Closed-form solvers: independent/paired t-tests, one-way ANOVA, one/two proportions, correlation, chi-square (effect size w), and regression R² increments, all through one interface in scripts/power.py (sample_size, power, mde, power_curve).
• Simulation-based power: for designs with no formula — logistic/Poisson regression, mixed-effects models, cluster-randomized trials with an ICC, survival analysis, mediation, interactions — via simulate → fit the planned model → repeat (≥1,000 reps) with a reported Monte Carlo CI.
• Real-world adjustments: multiplicity (α/m or FDR by simulation), attrition (n_enroll = n/(1−dropout)), clustering design effect (DEFF = 1+(m−1)·ICC), and unequal allocation ratios.
• Effect-size guidance and reporting: prefers SESOI over shrunken pilot estimates over Cohen conventions, provides d/f/r/η²/OR/h/w conversions, insists on sensitivity analysis rather than a single number, and explicitly warns against circular post-hoc "observed" power.

Who it's for

• Graduate students and researchers who must justify a sample size in a thesis or paper
• Anyone writing an IRB protocol, grant application, or pre-registration
• Clinical and epidemiological researchers needing enrollment numbers that account for dropout and clustering
• Data analysts sizing A/B tests or computing MDE for a fixed traffic budget

Examples

1. "How many per group to detect d = 0.5 at 80% power, α = 0.05?" → returns n per group and total, inflates for 20% attrition, and drafts the reporting paragraph.
2. "I can only recruit 30 per group — what can I detect?" → computes the MDE at that n and generates a power-vs-effect-size curve figure for the grant.
3. "20 clinics randomized, 30 patients each, ICC = 0.05 — what's my power?" → writes a cluster-simulation script, runs 5,000 replicates, and reports power with a Monte Carlo 95% CI.

Install the skill I found on Claude Skill Mart.
Copy the skills/statistical-power folder from the GitHub repo K-Dense-AI/scientific-agent-skills into my ~/.claude/skills/statistical-power/.
When it's done, tell me in one line what this skill can do.