A single two-stage network DEA model ensuring feasibility and interpretable efficiency measurement
提出一个单阶段两阶段网络DEA模型，同时评估阶段和整体效率，确保投影结果径向可解释且在生产可能集内可行，避免现有模型的无解或偏差问题。
Data envelopment analysis (DEA) is a well-known data-enabled technique to evaluate the relative efficiency of empirical production technology of decision-making units (DMUs) that transform inputs (resources) into outputs (products). Such technologies may consist of two or more internal stages. In a two-stage network (2SN) DEA, the outputs of the first stage serve as inputs for the second stage and are referred to as intermediates. DEA evaluates relative efficiencies based on observed DMUs under a very few assumptions. However, conventional 2SN DEA models may indicate that none of the observed DMUs are overall efficient. To address this issue, a few approaches have been proposed in the literature to ensure that at least one efficient DMU is identified through sequential or iterative processes. In this study, we propose a single model for simultaneously evaluating stage and overall efficiencies. Unlike other existing models, our model ensures that the resulting projections are radially interpretable, remain within the feasible production possibility set, and prevents infeasibility or bias in evaluating stage efficiencies. We demonstrate the model using a real-world numerical example and compare its results to those of existing models.