Research map / Probabilistic and scientific modeling / Treatment effects
Spatial Deconfounder: Interference-Aware Deconfounding for Spatial Causal Inference
Ayush Khot, Miruna Oprescu, Maresa Schröder, Ai Kagawa, Xihaier Luo
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 |
|---|---|---|---|
| add deconfounder substitute confounder (C-VAE) to spatial+ head | spatial+ · Local confounding, PM2.5->m (r_d=1), hidden rho_pop | direct-effect bias: 0.05 ± 0.02 -> 0.02 ± 0.01 | Improved |
| add C-VAE substitute confounder to unet head | unet · Spatial confounding, PM2.5->m (r_d=1), hidden rho_pop | direct-effect bias: 0.06 ± 0.01 -> 0.04 ± 0.01 | Improved |
| add C-VAE substitute confounder to unet head | unet · Spatial confounding, SO4->PM2.5 (r_d=1), hidden OC | direct-effect bias: 0.07 ± 0.02 -> 0.06 ± 0.01 | Improved |
| add C-VAE substitute confounder to unet head | unet · Spatial confounding, SO4->PM2.5 (r_d=1), hidden OC | spillover bias: 0.05 ± 0.02 -> 0.07 ± 0.03 | Hurt |
| use deconfounder neighborhood radius R=1 instead of R=2 (ablation) | C-VAE-spatial+ · Local confounding, PM2.5->m (r_d=1), hidden rho_pop | direct-effect bias: 0.04 ± 0.01 -> 0.02 ± 0.01 | Improved |
| use deconfounder radius R=2 instead of R=1 (ablation) | C-VAE-unet · Spatial confounding, PM2.5->m (r_d=1), hidden rho_pop | direct-effect bias: 0.06 ± 0.02 -> 0.04 ± 0.01 | Improved |
| use spatial+ head instead of durbin | C-VAE-spatial+ vs durbin · Local confounding, SO4->PM2.5 (r_d=2), hidden NH4 | direct-effect bias: 0.10 ± 0.00 -> 0.02 ± 0.00 | Improved |
| compare C-VAE-spatial+ (R=1) to durbin spillover estimate | C-VAE-spatial+ vs durbin · Local confounding, PM2.5->m (r_d=1), hidden q_summer | spillover bias: 0.02 ± 0.00 -> 0.06 ± 0.00 | Improved |
| sparsity sweep 10% to 70% (ablation) | Spatial Deconfounder · stress tests | bias / predictive p-value | Hurt |
| radius and latent-dimension sweeps (ablation) | Spatial Deconfounder · stress tests | bias | No clear effect |
| inject localized single-cause unobserved confounders (ablation) | Spatial Deconfounder · stress tests | bias | Hurt |
Closest papers by meaning
- SpaCE: The Spatial Confounding Environment (ICLR 2024 · similarity 0.92)
- GST-UNet: A Neural Framework for Spatiotemporal Causal Inference with Time-Varying Confounding (NeurIPS 2025 · similarity 0.91)
- Smooth Sailing: Lipschitz-Driven Uncertainty Quantification for Spatial Associations (NeurIPS 2025 · similarity 0.89)
- Discovering Latent Causal Graphs from Spatiotemporal Data (ICML 2025 · similarity 0.87)
- Generalizing Causal Effects from Randomized Controlled Trials to Target Populations across Diverse Environments (ICML 2025 · similarity 0.86)
- Balancing Interference and Correlation in Spatial Experimental Designs: A Causal Graph Cut Approach (ICML 2025 · similarity 0.85)
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.
