statistical-analysis-advisor

name	statistical-analysis-advisor
description	Advises on appropriate statistical tests, validates assumptions, performs analysis with effect sizes and confidence intervals. Outputs a statistical analysis plan or results interpretation with test justification and assumption checks.
allowed-tools	Read, Write, Bash
effort	medium

Statistical Analysis Advisor

When to activate

When choosing statistical tests, interpreting results, checking assumptions, or when a reviewer questions your analysis. Use before running any analysis to ensure correct test selection.

When NOT to use

Skip for simple descriptive statistics (mean, SD), data visualization without inference, or when the statistical approach is already validated and approved.

Instructions

1. Characterize the data:

   • Outcome type: Continuous, categorical, ordinal, time-to-event, count
   • Groups: Number of groups, paired vs. independent
   • Distribution: Normal (Shapiro-Wilk), equal variance (Levene's)
   • Sample size per group

2. Test selection decision tree:

   • 2 independent groups, normal → Independent t-test
   • 2 independent groups, non-normal → Mann-Whitney U
   • 2 paired groups, normal → Paired t-test
   • 3+ groups → One-way ANOVA (Kruskal-Wallis if non-normal)
   • Repeated measures → RM-ANOVA or Friedman
   • Time-to-event → Log-rank test, Cox regression
   • Correlation → Pearson (normal) or Spearman (non-normal)

3. Assumption checks:

   • Normality: Shapiro-Wilk (n<50) or Kolmogorov-Smirnov (n≥50)
   • Homogeneity of variance: Levene's test
   • Sphericity: Mauchly's test (for RM-ANOVA)
   • If violated: Report the violation and use the non-parametric alternative

4. Report results with:

   • Test statistic and degrees of freedom
   • Exact p-value (not just <0.05)
   • Effect size (Cohen's d, η², r, OR as appropriate)
   • 95% confidence interval
   • Sample size

5. Multiple comparisons: If >2 comparisons, specify correction:

   • Bonferroni (conservative)
   • Holm-Bonferroni (step-down)
   • Benjamini-Hochberg FDR (exploratory)

Output Format

STATISTICAL ANALYSIS: [Research question]

DATA CHARACTERISTICS:
  Outcome: [type] | Groups: [X] | Distribution: [normal/non-normal]
  Normality: Shapiro-Wilk W=[X], p=[X]
  Variance: Levene's F=[X], p=[X]

TEST SELECTED: [Test name] — [justification]

RESULTS:
  [Test]([df]) = [statistic], p = [exact p]
  Effect size: [Cohen's d / η² / r] = [value] ([small/medium/large])
  95% CI: [[lower], [upper]]
  n = [X] per group

INTERPRETATION: [Plain language conclusion]
LIMITATIONS: [Assumptions, power, generalizability]

Example

STATISTICAL ANALYSIS: Does treatment X reduce blood glucose vs. vehicle?

DATA CHARACTERISTICS:
  Outcome: Continuous (mg/dL) | Groups: 2 (treatment vs. vehicle) | Distribution: Normal
  Normality: Shapiro-Wilk W=0.97, p=0.34
  Variance: Levene's F=0.12, p=0.73

TEST SELECTED: Independent t-test — normal distribution, equal variance, 2 independent groups

RESULTS:
  t(46) = 3.82, p = 0.0004
  Effect size: Cohen's d = 1.12 (large)
  95% CI: [18.4, 59.2] mg/dL difference
  n = 24 per group

INTERPRETATION: Treatment X significantly reduced fasting blood glucose by 38.8 mg/dL
(compared to vehicle), a large and clinically meaningful effect.
LIMITATIONS: Single timepoint, single strain, not adjusted for body weight.