Generate Statistical Analysis Scripts in Python and R

This skill writes robust, reproducible statistical analyses in Python and R, grounded in core principles: always check assumptions (normality, homogeneity of variance, independence) before applying a test, report effect sizes alongside p-values for practical significance, apply corrections for multiple comparisons, provide and interpret confidence intervals in context, document methodology, and handle missing data transparently.

In Python it builds a StatisticalAnalysis class (using pandas, numpy, scipy.stats, statsmodels) with methods for descriptive statistics (including skewness and kurtosis), normality checking via both Shapiro-Wilk and D'Agostino's tests, and automatic group comparison that selects the right test based on assumption checks: an independent t-test (with Levene's test deciding equal vs. unequal variance) when both groups are normal, or a Mann-Whitney U test with rank-biserial correlation as effect size when they're not. The R template (tidyverse, psych, effsize, car, ggplot2, corrplot) mirrors this logic - descriptive stats grouped by the independent variable, Shapiro and Levene assumption tests, then branching to an equal-variance t-test, Welch's t-test, or a Wilcox/Mann-Whitney test with Cohen's d or Cliff's delta as the effect size.

def interpret_effect_size(d, test_type='cohens_d'):
    """Provide interpretation of effect sizes"""
    if test_type == 'cohens_d':
        if abs(d) < 0.2:
            return "negligible"
        elif abs(d) < 0.5:
            return "small"
        elif abs(d) < 0.8:
            return "medium"
        else:
            return "large"

For visualization it builds publication-ready, statistically annotated plots - a box plot with individual points and a p-value/effect-size annotation, paired with a histogram comparing group distributions. For power analysis it calculates required sample size per group from a target effect size, alpha, and desired power using statsmodels' ttest_power. For reporting it generates an APA-style statistical report string including descriptive statistics per group, the test used, the test statistic and p-value, Cohen's d with its interpretation label, and a significance conclusion.

Use it when running a statistical group comparison or hypothesis test in Python or R that needs correct assumption checking, automatic test selection (parametric vs. non-parametric), effect size reporting, power/sample-size analysis, or a formatted APA-style write-up.

Key recommendations to follow: always run exploratory data analysis before formal testing, use robust methods when assumptions are violated (as the automatic test-selection logic already does), report confidence intervals alongside point estimates, weigh practical significance alongside statistical significance, validate findings via cross-validation or replication, document all preprocessing decisions, and version-control analysis scripts for reproducibility.

Input is a dataset with a dependent variable and a grouping/independent variable, plus an alpha level (default 0.05). Output is a full analysis: descriptive statistics, assumption test results (normality, variance homogeneity), the automatically selected test with its statistic and p-value, an effect size with interpretation, an annotated comparison plot, and optionally a power/sample-size calculation or an APA-formatted report.

Researchers, analysts, and data scientists running group-comparison statistical analyses in Python or R who need correct, assumption-checked test selection, effect sizes, and reproducible, publication-ready reporting rather than defaulting to a single test regardless of data shape.