Beyond Interference: Estimating Causal Network Effects in Digital Advertising Markets

Abstract

Digital advertising platforms operate within complex social networks where user responses to ads are not independent but are influenced by the actions and exposures of their peers. This interdependence, known as network interference, fundamentally violates the Stable Unit Treatment Value Assumption (SUTVA) of traditional causal inference models, leading to significant attribution bias and suboptimal budget allocation. This paper addresses the challenge of estimating the true causal effect of digital advertising expenditures, moving beyond direct response metrics to quantify the total demand generation effect, which includes both direct and socially mediated peer influences. We propose a novel, robust methodological framework that integrates high-dimensional network clustering with a two-stage least squares (2SLS) instrumental variable approach and a synthetic control method. Using peer adoption rates and platform algorithm shocks as instruments, we isolate exogenous variation in ad exposure. We then employ a synthetic control-based estimator to construct counterfactual outcomes for treated network segments, effectively absorbing the bias from spillover effects. Our empirical analysis, applied to a simulated marketplace mirroring real-world advertising dynamics, demonstrates that standard models underestimate the total return on ad spend by approximately 30-45%. The findings provide a scalable solution for advertisers to de-bias their attribution models and optimize campaigns in the presence of social interference, offering a significant advancement in the econometrics of digital marketing measurement.