Balancing score methods for continuous treatment evaluation: measuring the dose-response function, evidence from the Common Agricultural Policy

This dissertation develops a unified applied framework for causal inference with continuous treatments based on balancing-score methods, combining a methodological synthesis with an empirical application. The first chapter reviews and systematises the literature on balancing score methods for continuous treatments, organising a wide range of approaches within a common framework built around the generalized propensity score. It structures these methods along three dimensions—how the balancing score is constructed (from parametric treatment models to direct covariate-balancing and machine-learning estimators), how it enters the outcome stage (regression-based, weighting-based, and design-based estimators), and whether estimation is global or local—and reviews the associated diagnostics for overlap and covariate balance, together with recent extensions to doubly robust, multivariate, and heterogeneous-effect settings. A recurring theme is that no single estimator dominates: credible estimation depends on aligning the estimation strategy with the structure of the data, and on diagnostic assessment of balance and common support. The second chapter applies this framework, estimating the dose–response function of Common Agricultural Policy (CAP) Pillar I support on regional economic and employment outcomes across European NUTS-3 and OECD-equivalent regions, with support measured as its intensity relative to the regional agricultural economy. The estimated effects are non-linear and differ across outcomes. Higher support intensity is associated with significantly stronger growth in total gross value added, up to a maximum beyond which returns diminish, and with significantly higher total employment growth that levels off as support rises. Within agriculture, support significantly cushions the decline of agricultural employment at low intensities, while its effect on agricultural value-added growth is statistically indistinguishable from zero across the support range—consistent with Pillar I being decoupled income support rather than a production subsidy. A common thread across the significant effects is the diminishing marginal effectiveness of support as its intensity rises. Together, the findings demonstrate the value of continuous-treatment evaluation and point to a dual economic role for CAP Pillar I: supporting regional labour markets and output at moderate intensities while cushioning agricultural labour adjustment, rather than expanding agricultural production.