Integer Programming for Generalized Causal Bootstrap Designs
Authors: Jennifer Rogers Brennan, Sebastien Lahaie, Adel Javanmard, Nick Doudchenko, Jean Pouget-Abadie
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the refined confidence intervals achieved through simulations of small geographical experiments. In Section 5, the paper includes a dedicated section titled 'Simulations on real data' and discusses 'Empirical Results', 'Dataset', 'Settings', 'Baselines', and 'Analysis'. |
| Researcher Affiliation | Collaboration | The authors list affiliations as '1Google Research 2University of Southern California 3Meta'. Adel Javanmard is affiliated with both '1' (Google Research, industry) and '2' (University of Southern California, academia). Since there is a mix of academic and industry affiliations among the authors, the paper is classified as a collaboration. |
| Pseudocode | Yes | The paper includes 'Algorithm 1 Our Causal Bootstrap Procedure' in Appendix A.9, which provides a structured, step-by-step procedure. |
| Open Source Code | No | The paper states: 'Integer programs to compute the optimal causal bootstrap imputation were solved using the CP-SAT solver (https://developers.google.com/optimization/cp/cp_solver).' This refers to a third-party solver used by the authors, not the release of their own implementation code. There is no explicit statement or link indicating that the authors have made their specific methodology's source code available. |
| Open Datasets | Yes | To work with a realistic dataset, we look to the International Monetary Fund s publicly-available Gross Domestic Product (GDP) country-level report for the years 2017 2019, restricted to the top 50 countries by GDP (IMF, 2024). ... IMF. World economic outlook database, 2024. https://www.imf.org/en/Publications/ WEO/weo-database/2024/April. |
| Dataset Splits | Yes | To showcase the validity and usefulness of our method, we compare two assignment strategies: the standard complete randomization design ( Compl. R. ) treating exactly half of the countries, and a matched-pairs design ( Matched Pairs ). The matched-pairs design is constructed by matching the country with the largest GDP in 2018 with the country with the second largest GDP for that year, the third largest with the fourth largest, and so on. |
| Hardware Specification | No | The paper mentions: 'Experiments were parallelized on a cloud CPU cluster using 100 workers.' This describes a general computing environment ('cloud CPU cluster') but does not provide specific hardware details such as CPU models, GPU types, or memory specifications. |
| Software Dependencies | No | The paper states: 'Integer programs to compute the optimal causal bootstrap imputation were solved using the CP-SAT solver (https://developers.google.com/optimization/cp/cp_solver).' While a specific solver is named, no version number for the CP-SAT solver is provided, which is required for a reproducible description of software dependencies. |
| Experiment Setup | Yes | The paper describes specific experimental settings: 'we compare two assignment strategies: the standard complete randomization design ( Compl. R. ) treating exactly half of the countries, and a matched-pairs design ( Matched Pairs ).' It also details the construction of the matched-pairs design and the estimators used: 'We consider two estimators: first, the standard nocovariates difference-in-means estimator, and second, the doubly-robust estimator from Section 3.2...' Additionally, it states: 'All reported confidence intervals have 95% nominal coverage' and specifies 'resampling the assignment according to each design 500 times'. |