Specification search
Explore admissible outcome constructions and candidate-control combinations, with support for multiple focal estimands.
Published in Patterns · 14 August 2026
See how conclusions change across the analytical multiverse.
RobustiPy is an open-source Python library for multiverse analysis and model uncertainty assessment. It brings specification search, resampling, model comparison, validation, and interpretation into one reproducible workflow.
5
Controlled simulationsreported in the paper
10
Empirical replicationsacross four research domains
≈672m
Regression-equivalent fitsin the reported time-profiling benchmark
One coherent toolkit
Assess robustness from more than one angle.
RobustiPy helps researchers make defensible modelling choices explicit, estimate their consequences, and inspect uncertainty across specifications.
Bootstrap inference
Quantify sampling uncertainty across specifications with reproducible bootstrap-based routines.
Model selection and averaging
Compare specifications using fit criteria and summarise results with weighted or unweighted estimates.
Out-of-sample validation
Use cross-validation to assess whether model performance holds beyond the estimation sample.
Joint inference
Evaluate evidence across related estimates rather than relying on isolated significance tests.
Feature-level explanations
Inspect variable influence in the full-specification predictive model with explainable-AI tools.
A transparent workflow
From modelling choices to interpretable evidence.
Frame the choice space
Define the outcomes, predictors, controls, estimators, and resampling strategy that are defensible for the question.
Estimate the multiverse
Run combinations systematically, with sampling and parallelisation options when the specification space is large.
Interpret results together
Compare estimates, uncertainty, fit, predictive performance, and feature influence across a shared set of outputs.
Patterns · Open access
RobustiPy: An efficient next-generation multiversal library with model selection, averaging, resampling, and explainable AI
The paper introduces the library and demonstrates its use across simulations and empirical replications in economics, sociology, psychology, and medicine.
Learn and apply
Everything needed to get started.
Begin with a practical walkthrough, consult the API documentation, or work through complete examples in the source repository.
Documentation
API reference and package documentation on Read the Docs.
Getting started
A concise path from installation to fitting and plotting results.
Interpret the figures
Read each panel in the standard RobustiPy results output.
Empirical examples
Reproducible applications across several substantive domains.
Tutorials and talks
Introductions, demonstrations, and recorded project material.
Software archive
Versioned RobustiPy releases preserved on Zenodo.
Creators
Built by an interdisciplinary team.
RobustiPy was developed at the University of Oxford by researchers working across computational social science, demography, and data science.
Daniel Valdenegro
Co-author · Software development
GitHub profile ↗Jiani Yan
Co-author · Software development
Personal website ↗Duiyi Dai
Co-author · Code review
Personal website ↗Charles Rahal
Software initiator · Lead contact
Personal website ↗Open by design
Research software that can be inspected and improved.
RobustiPy is released under the GNU General Public License v3.0. Contributions, bug reports, feature requests, and reproducible examples are welcome through GitHub.
Acknowledgements
Research support
The authors are grateful for funding from the ESRC (grant ES/W002302/1), the Leverhulme Trust (grant RC-2018-003) for the Leverhulme Centre for Demographic Science and Nuffield College, and a Grand Union DTP ESRC studentship.
