ATD: Collaborative Research: Causal Inference with Spatio-Temporal Data on Human Dynamics in Conflict Settings

How do airstrikes affect the subsequent location and nature of insurgent violence in violent settings like Afghanistan and Iraq? How does insurgent violence change in response to civilian casualties? And how do large-scale aid programs affect the frequency, type, and location of insurgent violence in these settings?

These questions are examples of important issues that face policy-makers on a daily basis. Yet while the study of human dynamics of violence has grown enormously over the past decade, we still lack methods for conducting causal inference in these challenging settings. Indeed, scholars have continued to rely on approaches that do not fully account for the spatial and temporal dynamics that characterize fast-moving interactions between governments, insurgents, and civilians.

Existing methods typically aggregate fine-grained geo-spatial data at much coarser temporal and spatial units, throwing away the advantages of current micro-level data and leaving scholars unable to capitalize on future improvements to high-frequency, high-resolution geo-spatial data. Taken together, existing frameworks risk mistaken causal inferences about the efficacy of both violent and non-violent interventions in these settings, leaving policy-makers blind to possible unintended consequences and negative externalities of proposed policies.

This project will develop a comprehensive spatio-temporal causal inference framework that avoids data aggregation and imposes no structural assumptions on spatial spillover and temporal carryover effects. It defines causal quantities of interest under stochastic interventions, which represent counterfactual treatment assignment strategies. The estimation strategy employs inverse probability weighting based on an estimated propensity score surface.

Among others, this project will develop mediation analysis, effect modification, and sensitivity analysis for these complex spatio-temporal settings. It will also investigate other important problems such as optimal treatment allocations, the spatial range of treatment spillover, and causal inference in the presence of persistent treatments. This project team will use airstrikes and economic assistance in fragile settings as empirical examples.

The goal is to publish articles in leading general science, statistical, and political science journals. The PIs will also publish software packages to automate and to extend the spatio-temporal framework to other issues and settings. The project will provide research training opportunities at a graduate level.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.