Academic Statistical Analysis

Problem

Research analyses often fail because the workflow, not the model, is wrong: the wrong test is chosen, assumptions are skipped, p-values are over-read, effect sizes are missing, or the APA report is incomplete. statistical-analysis turns this into a repeatable path for experimental, observational, and hypothesis-testing work.

How It Works

• Test selection: maps sample design, variable type, and normality to parametric, nonparametric, or Bayesian routes, with multiple-comparison correction.
• Assumption checks: uses assumption_checks.py to inspect normality, variance homogeneity, linearity, and outliers, with Q-Q plots, residual plots, and boxplots.
• Core analysis: covers t-tests, ANOVA, chi-square tests, linear and logistic regression, correlation, and Bayesian alternatives such as Bayesian t-tests, ANOVA, and regression.
• Interpretation and reporting: computes Cohen's d, η²_p, r, R², Cramér's V, confidence intervals, and APA-style result text.

Boundaries

It is intended for rigorous research reporting, not ad-hoc business metrics. Effect-size benchmarks are heuristics; p-values are not posterior probabilities; non-significant results do not prove the null. Hierarchical models, missing-data mechanisms, and sequential analysis still require domain-specific judgment.