BEYOND CORRELATION : Decoding Local Impact Through, Counterfactual Policy Evaluation

This paper serves as a pedagogical synthesis of causal evaluation specifically tailored for territorial applications. It clarifies how modern public policy evaluation relies heavily on counterfactual logic and causal attribution to move beyond mere correlations and establish true programmatic impacts. Rooted primarily in the post-positivist research paradigm, this quantitative approach emphasizes objective measurement, rigorous empirical observation, and the systematic isolation of variables to uncover reliable causal mechanisms. As governments increasingly decentralize decision-making, the need for such rigorous territorial policy evaluation has never been more critical. The paper synthesizes the theoretical framework of the potential outcomes model (Rubin, 1974), detailing the core parameters of interest (ATE, ATT, ATU) and the necessary identification conditions for robust causal inference. It systematically examines quasi-experimental methods designed to overcome selection bias in observational data. A particular focus is placed on the instrumental variables (IV) approach, heavily illustrated by the seminal study of Angrist and Krueger (1991) on the economic returns to education, demonstrating how exogenous variations can identify causal effects even in the presence of unobservable confounders. Furthermore, this study deeply discusses the specific, complex challenges inherent to territorial evaluation—such as spatial spillovers, small unit sample sizes, policy contagion, and highly fragmented data structures. To address these geographical constraints, it highlights recent methodological advancements, notably the synthetic control method (Abadie, 2021), which offers a transparent framework for comparative case studies. Finally, the paper explores the contemporary frontier of causal machine learning, demonstrating how advanced predictive algorithms and spatial big data are fundamentally reshaping local public decision-making and resource allocation.