RobustiPy: An efficient next generation multiversal library with model selection, averaging, resampling, and explainable artificial intelligence

The "multiverse" of research choices is a major hidden problem: Scientists make many defensible but different modeling decisions (which variables to include, how to process data, etc.) that can produce wildly different results—sometimes varying as much as the actual phenomena being studied. RobustiPy helps researchers systematically explore these choices rather than hiding them.

Transparency through specification curves: The library generates "specification curves" that show how results change across different analytical choices, making it visible when findings are robust versus fragile. This addresses a key reproducibility concern where the same dataset can support contradictory conclusions depending on modeling decisions.

Practical validation across multiple fields: The paper demonstrates RobustiPy's effectiveness across economics, sociology, psychology, and medicine with ten empirical case studies, including re-analyses of "widely cited findings with documented discrepancies," suggesting the tool can catch and correct problematic research in real-world applications.

Computational efficiency enables scale: By benchmarking on approximately 672 million simulated regressions, RobustiPy achieves "state-of-the-art computational efficiency," making rigorous robustness analysis practical rather than prohibitively time-consuming—removing a major barrier to responsible research practices.

Explainability built into the framework: The library integrates explainable AI methods that quantify "the marginal contribution of each covariate," helping researchers and audiences understand which specific analytical choices drive results, supporting more honest communication about research limitations.