Abstract
•Fixes virtual projection bias in two-stage network DEA models.•indirect and direct approaches for bias correction in DEA models.•Decomposes efficiency into stage- and system-level measures.•Validated via banking case study and sensitivity analysis.
Two-stage network Data Envelopment Analysis (2SN DEA) is widely used to evaluate decision-making units (DMUs) with internal production structures. However, conventional 2SN DEA models may produce biased efficiency estimates when the generated virtual projection point lies outside the feasible input–output production frontier. From a managerial perspective, existing models also have important limitations. They usually explain inefficiency as a stage-level production problem, while paying insufficient attention to system-level managerial inefficiency and ripple effects between connected stages. More importantly, many conventional 2SN DEA models implicitly assume that supply-chain stages can be freely rearranged across DMUs. To address these issues, we propose a unified non-radial analysis framework that imposes additional restrictions on the reconfiguration of conventional network production technology. These restrictions ensure that the generated projection point remains feasible within the input–output production frontier. The proposed model is evaluated against existing approaches using an empirical case study from the supply chain of motor manufacturing industry, along with a sensitivity analysis based on alternative reference sets. The results, further supported by the proposed theorems, demonstrate that the new approach effectively relocates previously identified exterior projection points onto the boundary of the input–output production frontier. Through the corresponding dual programming model, we further introduce a novel mechanism to account for stage-level and system-level failures, as well as the ripple effect. The proposed model enables a decomposition of the overall efficiency of each DMU into stage efficiency and system efficiency, offering deeper managerial insights.