Causal analysis of energy retrofit for real-life cases using Average Treatment Effect

Energy retrofit of existing buildings is one of the key strategies for achieving building decarbonization. However, its actual impact on energy reduction remains underexplored due to unobserved confounding factors. in other words, most existing studies have relied on correlational (association) data or simulation-based analyses without providing causally interpretable estimates for real-life cases. This study employs a partial identification approach to estimate the Average Treatment Effect (ATE), considering unobserved confounding factors. For this purpose, monthly energy use data were collected from 210 post- retrofit public day-care buildings in South Korea. Retrofit data were used as treatment variables, and causal graphs were developed based on domain knowledge. Results show that envelope insulation retrofit and glazing replacement reduce energy use during the heating period, while having minimal impact on cooling energy. Replacement of high-efficiency heat pumps with old boilers and new installation of heat recovery ventilators exhibit a wide uncertainty in the estimated ATEs. This approach addresses limitations of association-based methods and offers interpretable evidence of the retrofit effect. The findings of this study are expected to support policymakers and building managers in prioritizing retrofit measures. This work shows how post-retrofit evaluation can move beyond association toward uncertainty-aware causal estimation.