CLAM: Causal Spatial Disaggregation to Infer Local Effects From Coarse Data

Learning spatially fine-grained patterns from coarse-resolution data is inherently difficult; doing so in a causal setting - estimating high-resolution effects from coarse interventional data - adds an extra layer of complexity.We introduce CLAM, a method for estimating fine-grained causal effects when only coarse-resolution data on interventions and outcomes is available. We assume high-resolution contextual covariates exist that modulate these effects and can be exploited to infer localised causal effects, support counterfactual reasoning, and enable disaggregation of the outcome. Through simulation studies, we demonstrate that CLAM can recover spatially varying causal impacts under diverse conditions.This has important implications for domains such as public health and environmental policy, where decisions are made at broad scales but causal pathways vary locally. Code is available https://anonymous.4open.science/r/clam-DE2C/Causaldisaggregationshared.ipynb