Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Does Representation Guarantee Welfare?
Authors: Jakob de Raaij, Ariel D Procaccia, Alexandros Psomas
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, in Section 5 we experimentally show that it is possible to select 2-representative panels in real-world instances. We analyze data from four different citizens assemblies: one from a Western European country and three from Australian states. Our experiments show that (approximate) 2representative panels exist with the current panel sizes for two of the datasets, and would exist by picking a slightly smaller panel for a third dataset. |
| Researcher Affiliation | Academia | Jakob de Raaij Harvard University EMAIL Ariel D. Procaccia Harvard University EMAIL Alexandros Psomas Purdue University EMAIL |
| Pseudocode | No | The paper describes the methodology for checking m-representative panels using an integer linear program and the Gurobi solver but does not provide pseudocode or an algorithm block for this process. |
| Open Source Code | No | We do not release our code since it is very straightforward, cleaning data as described in Appendix A.6 and solving a linear program using external software as described in Section 5. |
| Open Datasets | Yes | For EUR1, we used the European Social Survey [19], which collected data on a wide variety of features and their intersections, including those used in EUR1. ... For AUS1, AUS2, and AUS3, we used data from the 2021 Australian Census [1]. Through their online data tools, we obtained data on the intersections of individuals features. |
| Dataset Splits | No | The paper does not include experiments in the machine learning sense that would require training/test/validation splits. The empirical analysis focuses on the feasibility of selecting representative panels from a pool of volunteers based on real-world citizens assembly data. |
| Hardware Specification | Yes | We used Gurobi, run on a 14-inch Mac Book Pro (2023) with Apple M3 Pro chip, to check whether the program is feasible, i.e., whether an m-representative panel exists. |
| Software Dependencies | No | The paper mentions using "Gurobi" as a solver but does not provide a specific version number. |
| Experiment Setup | Yes | Experiments. For each m between 1 and the number of features considered in the panel, |F|, we found the largest size of an m-representative panel up to rounding using individuals from the pool of volunteers. Furthermore, we found for each m between 1 and |F| and panel size k between 1 and the pool size the smallest ε for which an ε-approximately m-representative panel exists. For a given m and desired panel size k, we checked for the existence of an m-representative panel of size k using an integer linear program. Each person in the pool of volunteers corresponds to one binary variable, encoding whether this person is in the panel or not. The constraints on the variables are the size of the panel, k, and having to be m-representative (up to rounding or ε-approximately) to the underlying population data. |