Research map / Graph learning / Graph algorithms and clustering
Cluster Randomized Designs for One-Sided Bipartite Experiments
Jennifer Rogers Brennan, Vahab Mirrokni, Jean Pouget-Abadie
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use bias-minimizing clustering H(C) instead of direct bipartite clustering (ablation) | cluster-randomized design · Bipartite SBM p=0.005 | relative bias: 9.10 -> 3.90 | Improved |
| use H(C) clustering instead of unit-level randomization | cluster-randomized design · Bipartite SBM p=0.0 | relative bias: 12.44 -> 0.02 | Improved |
| H(C) vs true clustering | cluster-randomized design · Bipartite SBM | relative bias: 3.88 -> 3.90 | No clear effect |
| replace H(C) objective with Tr(Var(d)) objective (ablation) | cluster-randomized design · Bipartite SBM | relative bias: 3.90 -> 3.84 | No clear effect |
| increase cross-cluster edge probability p from 0 to 0.5 | H(C) design · Bipartite SBM | relative bias: 0.02 -> 12.98 | Hurt |
| widen pure-exposure neighborhood Delta from 0.1 to 0.5 | H(C) design · Nonlinear outcomes | relative bias: 1.000 -> 0.001 | Improved |
| use H(C) instead of direct clustering at Delta=0.5 (ablation) | cluster-randomized design · Nonlinear outcomes | relative bias: 0.600 -> 0.001 | Improved |
| use H(C) clustering instead of EXPOSURE-DESIGN at Delta=0.3 (ablation) | cluster-randomized design · Nonlinear outcomes | relative bias: 0.509 -> 0.457 | Improved |
| use H(C) instead of unit-level randomization on power-law graphs (strong latent structure) | cluster-randomized design · Power-law graph | relative bias: 4.9 -> 1.9 | Improved |
| use H(C) instead of true clusters on power-law graphs (weak structure) (ablation) | cluster-randomized design · Power-law graph | relative bias: 5.2 -> 3.5 | Improved |
| use H(C) vs direct clustering on power-law graphs (ablation) | cluster-randomized design · Power-law graph | relative bias: 3.3 -> 3.5 | No clear effect |
| use H(C) instead of direct clustering in Airbnb misspecified-exposure simulation (ablation) | cluster-randomized design · Airbnb marketplace simulation | RMSE | Improved |
Closest papers by meaning
- Independent-Set Design of Experiments for Estimating Treatment and Spillover Effects under Network Interference (ICLR 2024 · similarity 0.88)
- Online Experimental Design With Estimation-Regret Trade-off Under Network Interference (NeurIPS 2025 · similarity 0.87)
- Probabilistic Factorial Experimental Design for Combinatorial Interventions (ICML 2025 · similarity 0.85)
- Higher-Order Causal Message Passing for Experimentation with Complex Interference (NeurIPS 2024 · similarity 0.85)
- Balancing Interference and Correlation in Spatial Experimental Designs: A Causal Graph Cut Approach (ICML 2025 · similarity 0.84)
- Near-Optimal Experiment Design in Linear non-Gaussian Cyclic Models (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
