ReproStat: Reproducibility Diagnostics for Statistical Modeling

Tools for diagnosing the reproducibility of statistical model outputs under data perturbations. The package implements bootstrap, subsampling, and noise-based perturbation schemes and computes coefficient stability, p-value stability, selection stability, prediction stability, and a composite reproducibility index on a 0–100 scale. Cross-validation ranking stability for model comparison and visualization utilities are also provided.

Typical workflow

1. Call run_diagnostics with your formula and data.

2. Inspect individual metrics with coef_stability, pvalue_stability, selection_stability, and prediction_stability.

3. Summarise with reproducibility_index.

4. Visualise with plot_stability.

5. Compare competing models with cv_ranking_stability and plot_cv_stability.

See also

Useful links:

• https://ntiGideon.github.io/ReproStat

• https://github.com/ntiGideon/ReproStat

• Report bugs at https://github.com/ntiGideon/ReproStat/issues

Author

Maintainer: Gideon Nti Boateng gidiboateng200@gmail.com