Primary Submission Category: Causal Inference in Networks
Causal Inference when Intervention Units and Outcome Units Differ
Authors: Fabrizia Mealli, Georgia Papadogeorgou, Guido Imbens,
Presenting Author: Fabrizia Mealli*
We study inference for causal effects in settings characterized by interference stemming from having two distinct sets of units: units to which the intervention is applied and units on which the outcomes are measured. We call this bipartite interference: treatments applied to one intervention unit can affect multiple outcome units, and the outcome of a unit may depend on the treatments applied to multiple intervention units.
Examples of this setting can be found across many disciplines. In air pollution epidemiology, the interventional units could be pollution emitters such as power plants (intervention units) which may install a filter on their smokestack or not, and the outcome units could be members of the population residing within specific geographical areas. Similarly, in the economics of housing, housing prices at different locations (outcome units) may be affected by whether appropriate cleaning has taken place in nearby contaminated hazardous-waste disposal sites (intervention units).
We consider both usual and new causal estimands, highlithing similarities and differences with more common settings of causal inference with unit-to-unit interference. Estimators for these quantities in bipartite settings will be introduced and their performance evaluated from a design-based perspective. For inference, finite sample and asymptotic properties will be investigated. Optimal designs for such effects will be discussed that exploit the topology of the bipartite graph.