statistics_skill

Overview

This skill provides a practical toolkit for statistical analysis tailored to data analytics workflows. It covers descriptive statistics, probability, inferential tests, and regression methods so you can summarize data, test hypotheses, and build predictive models. The focus is on actionable techniques and common pitfalls to support robust, repeatable analysis.

How this skill works

The skill inspects datasets to compute core descriptive metrics (mean, median, variance, percentiles) and evaluates distribution shape and dispersion. It guides sampling considerations, constructs confidence intervals, and runs hypothesis tests (t-tests, chi-square, ANOVA) while reporting p-values and assumptions. For modeling, it fits linear, multiple, and logistic regressions, performs validation and diagnostics, and flags issues like multicollinearity and outliers.

When to use it

• Summarize central tendency and spread before modeling or reporting
• Determine sample size and evaluate sampling strategy for studies
• Test hypotheses to support business or scientific decisions
• Build and validate regression models for prediction or inference
• Diagnose model problems like multicollinearity, heteroscedasticity, or influential points

Best practices

• Always visualize distributions before choosing tests or models
• Check and state test assumptions; use non-parametric alternatives when violated
• Pre-register analysis plans or correct for multiple tests to avoid p-hacking
• Perform diagnostics: residual analysis, VIF for multicollinearity, and leverage for outliers
• Prefer confidence intervals and effect sizes alongside p-values for interpretation

Example use cases

• Compare conversion rates between variants using chi-square or t-tests
• Estimate customer lifetime value distribution and summarize percentiles
• Build a regression model to predict churn and validate with holdout or cross-validation
• Calculate required sample size for A/B testing and plan enrollment
• Detect and mitigate multicollinearity when adding correlated predictors to a marketing mix model