Interpretable Causal Machine Learning Evidence On The Impact Of Renewable Energy On Co₂ Emissions

This study examines the impact of renewable energy share on per capita CO₂ emissions using a combination of machine learning-based causal inference and explainable artificial intelligence methods. The relationship between renewable energy and carbon emissions has mostly been addressed in the literature using correlation-based approaches. However, the magnitude, direction, and inter-country variability of the causal effect between renewable energy and carbon emissions remain largely unclear. This study aims to fill this gap. Analyses were conducted using a global panel dataset covering the post-1995 period. In this study, the causal effect of renewable energy share on CO₂ emissions was estimated using the Causal Forest method within the Double Machine Learning framework. Furthermore, the mechanisms behind the obtained heterogeneous treatment effects were interpreted using SHapley Additive exPlanations-based explainability analysis. The findings show that an increase in the share of renewable energy significantly and causally reduces per capita CO₂ emissions on average. The negative conditional mean of treatment effects for all observations reveals that renewable energy transition does not lead to an increase in emissions under any economic or structural conditions. However, the magnitude of the effect differs significantly between countries. Explainable causality analysis shows that energy intensity is the most dominant determinant of this heterogeneity; per capita income and industrial structure play nonlinear and context-sensitive roles. The analysis conducted for Turkey reveals that structural constraints limit the effectiveness of renewable energy transition in middle-income and energy-intensive economies. Overall, this study demonstrates the causal effect of renewable energy policies on emission reduction not only at the average level but also in a heterogeneous and explainable manner. By combining causal inference with explainable machine learning, the study offers a new and powerful empirical framework for evaluating energy and climate policies.