Statistical Theory for Causal Inference with Interference and Spillover Effects in Network-Connected Nigerian Communities

Standard causal inference methods assume no interference between units, an assumption violated when programme effects spill over from treated to untreated individuals through social networks, as is common in Nigerian community-level health, agriculture, and education interventions. This dissertation develops statistical theory for causal inference with interference and spillover effects in network-connected Nigerian communities, making original contributions to potential outcomes theory under partial interference. New identification conditions for direct and indirect causal effects are derived under stratified interference and partial interference assumptions appropriate for Nigerian community cluster structures. Asymptotic theory for inverse probability weighted estimators under network-correlated outcomes is established, extending existing results to non-i.i.d. network data. A novel network-adaptive doubly robust estimator that remains consistent under either propensity score or outcome model misspecification is introduced with theoretical efficiency bounds. Simulation experiments calibrated to social network data from 30 Nigerian villages across three geopolitical zones demonstrate that the proposed estimator reduces bias by 34 percent relative to naive ignoring-interference estimators. The theory was applied to re-estimate spillover effects of a community health worker programme in Kaduna State, revealing that indirect vaccine coverage effects on untreated households represented 41 percent of total programme impact, previously unattributed by standard analysis. The dissertation contributes causal inference theory applicable to the growing literature on community-based interventions in African development contexts. Keywords: causal inference, interference, spillover effects, potential outcomes, network statistics