Generalized propensity score approach to causal inference with spatial interference.

Area of Science:

• Spatial statistics
• Causal inference
• Environmental science

Background:

• Spatial phenomena often exhibit interference, where exposures at one location impact outcomes at others.
• Standard causal inference methods fail due to violations of the stable unit treatment value assumption.

Purpose of the Study:

• To develop a novel causal framework for estimating direct and spillover effects in the presence of spatial interference.
• To account for the varying influence of exposures based on spatial proximity.

Main Methods:

• Proposed a generalized propensity score to address confounding in spatial interference settings.
• Developed a Bayesian spline-based regression model to manage dimensionality.
• Assumed no unmeasured confounding for valid inference.

Main Results:

• The generalized propensity score effectively removes measured confounding.
• The Bayesian spline model demonstrates accuracy and appropriate coverage properties in simulations.
• The method was applied to estimate wildland fire impacts on Western US air pollution (2005-2018).

Conclusions:

• The new framework enables robust causal inference for spatial phenomena with interference.
• The approach provides a method to disentangle direct and spillover effects.
• This research offers valuable insights into the spatial relationships between environmental events and pollution.