Primary Submission Category: Causal Inference in Networks
Average and Conditional Inward and Outward Spillovers of One Unit’s Treatment under Network Interference
Authors: Fei Fang, Laura Forastiere,
Presenting Author: Fei Fang*
In a connected social network, users may have varying levels of influence on others when they themselves receive interventions. For example, giving an advertisement to a more influential person can have on average a greater impact on others’ purchase decisions. Understanding and evaluating these effects can provide valuable insights for various applications such as targeting strategies in marketing and behavioral interventions in public health. Under a partial interference assumption, we define influence effects in two ways: i) the inward average spillover effect on a unit’s outcome of a neighbor’s treatment, and ii) the outward average spillover of a unit’s treatment on their neighbors’ outcomes. We investigate the comparison between the two causal effects in directed networks with different properties, including the conditions under which they are equivalent. Additionally, we develop Horvitz-Thompson estimators for assessing both effects, on average and conditioning on categorical covariates, as well as weighted least square estimators for these effects conditioning on continuous covariates. We derive design-based variance estimators and establish the consistency and asymptotic normality. Through simulations, we verify the empirical performance of our proposed estimators. Finally, we employ our approach to investigate inward and outward average and conditional spillover effects of an information session on the adoption of weather insurance among rice farmers in China.