Skill: Two-Group Statistical Comparison

Use When

• User wants to test whether a continuous measurement differs between two experimental groups (e.g., treated vs. control, knockout vs. wild-type)
• User needs a p-value, effect size, and confidence interval for a two-group comparison
• User is unsure which statistical test is appropriate and wants automatic method selection

Inputs

• Required:
  • Data table (CSV or TSV) with a numeric value column and a group column (exactly two group levels)
• Optional:
  • Group variable column name (default: auto-detected if only one non-numeric column)
  • Value variable column name (default: auto-detected if only one numeric column)
  • Paired comparison: true or false (default: false)
  • Alternative hypothesis: two-sided (default), greater, or less
  • Output directory (default: ./compare_output)

Workflow

1. Read data; validate that group column has exactly two levels.
2. Run normality test (Shapiro-Wilk) on each group separately.
3. If both groups are normal: run Levene's test for equal variance.
   • Equal variance (p > 0.05): Student's t-test.
   • Unequal variance (p ≤ 0.05): Welch's t-test.
4. If either group is non-normal: Mann-Whitney U test (unpaired) or Wilcoxon signed-rank test (paired).
5. Compute effect size: Cohen's d (parametric) or rank-biserial correlation r (non-parametric).
6. Compute 95% confidence interval for the difference in means (parametric) or Hodges-Lehmann estimator (non-parametric).
7. Generate a box plot with individual data points and significance annotation.
8. Write a structured result and plot (PDF) to output directory.
9. Report method used, test statistic, p-value, effect size, confidence interval, and interpretation.
10. Save the executable analysis script (analysis_code.py or analysis_code.R) and software version manifest (session_info.txt) used to produce results.

Output Contract

• Result (TSV): group1, group2, method_used, statistic, p_value, effect_size, effect_size_type, ci_lower, ci_upper, n_group1, n_group2
• Box plot with data points (PDF)
• Decision log explaining method selection (printed to stdout)
• Reproducibility artifacts: analysis_code.py or analysis_code.R, plus session_info.txt

Limits

• Requires exactly two groups; use stat-compare-multiple-groups for three or more groups.
• Minimum recommended sample size: n ≥ 5 per group (results are unreliable below this).
• Paired comparison requires equal sample sizes in both groups.
• Common failure: group column contains more than two levels — filter data before running.