Traditional data-driven approaches emphasize input–output correlations and neglect dependencies among inputs, risking missed insights into key drivers of energy performance. Consequently, approaches that transcend correlation-centric analysis are warranted. Within this context, causal inference, which accounts for both statistical associations and temporal cause–effect relations, constitutes a promising direction. However, researchers cannot feasibly specify all causal relations relying solely on domain knowledge. Causal discovery is a data-driven methodology for analyzing causal relationships among variables, providing not only measures of association but also information on causal directionality. The authors employ two causal discovery algorithms—PC (Peter-Clark) and FCI (Fast Causal Inference)—on weather data. The discovered causal structures are compared, and two validation approaches are introduced to evaluate their statistical reliability; the authors also build on the identified causal structure to analyze the resulting causal pathways. The results show that both algorithms provide insights into causal relationships among variables, and the proposed validation approaches help establish the statistical reliability of the discovered structures. Moreover, the analysis of causal pathways indicates that causal effects can be identified and estimated with reliability.

Building energy models often include multiple, strongly interdependent weather variables, yet whether these variables provide distinct and interpretable information beyond outdoor air temperature (OAT) remains unclear. This issue is important for weather normalization, benchmarking, and policy-oriented performance evaluation, where model transparency and coefficient stability matter alongside predictive accuracy. This study applies causal discovery using the Peter–Clark (PC) algorithm to monthly heating-season weather and energy data from four childcare facilities in Seoul, South Korea. A synthetic negative-control case, in which energy use was generated independently of weather, was also included. A synthetic negative-control case, in which energy use was independent of weather, was included to check whether the method avoids spurious weather–energy links; no such links were identified. Graph-based total-effect estimates were compared with coefficients from simple and multiple linear regression models. Across the four real buildings, OAT was the only weather variable with a direct edge to heating energy use in all estimated graphs. Relative humidity, vapor pressure, evaporation, and global horizontal irradiance were connected to heating energy use mainly through pathways involving OAT. Because the estimated graphs contained only a direct OAT-to-energy path, the total-effect estimate of OAT coincided with its simple-regression coefficient. By contrast, jointly entering the weather variables into multiple regression produced sign reversals for several variables and substantial changes in coefficient magnitude, whereas the identifiable graph-based total-effect estimates remained directionally consistent and physically interpretable across the four real buildings. These findings support the use of an OAT-centered parsimonious model as a transparent baseline for the operationally homogeneous buildings examined. Additional weather variables may be warranted when they provide reproducible predictive gains or represent building-specific physical mechanisms.