data-analysis

Data Analysis is a research-oriented skill that generates rigorous statistical analysis code and subjects it to a structured 4-round review. It helps select appropriate statistical tests, interpret results, and produce analysis reports complete with p-values, effect sizes, and confidence intervals — aimed at analyzing experimental data for a paper. It solves the problem of ad-hoc, error-prone statistical scripting by imposing a consistent code structure and a multi-pass review that catches mathematical, data-handling, and reporting flaws.

The skill structures analysis code into labeled sections (import, load data, dataset preparation, descriptive statistics, preprocessing, analysis, and saving results) and then runs four review rounds: fundamental code and statistical flaws, data handling (missing values, units, preprocessing, test choice and assumptions), per-table sensibility and uncertainty measures, and cross-table completeness and consistency. It ships two helper scripts: `stat_summary.py`, which detects data types, recommends tests, runs comparisons, and outputs effect sizes and significance stars (requiring numpy and scipy), and `format_pvalue.py`, which formats p-values as stars, LaTeX, or plain text using only the standard library. A test-selection table maps data types to tests (t-test, Mann-Whitney U, ANOVA, Kruskal-Wallis, chi-square, Fisher's exact, Pearson/Spearman, OLS/logistic/mixed effects), and rules require reported uncertainty for every value, appropriate tests, and no hallucinated results. The scripts operate on local user-supplied data files with common scientific packages.

The target users are researchers, data scientists, and graduate students preparing quantitative results for publication. Use cases include comparing methods across metrics, choosing and validating a statistical test, formatting p-values for a LaTeX table, and reviewing analysis code for statistical soundness before submission.