The Potential for Endogeneity Bias in Data Envelopment Analysis
本文通过蒙特卡洛模拟证明，当投入资源受产出反馈影响（即内生性）时，数据包络分析（DEA）的效率估计会产生偏倚，导致使用低水平内生资源的低效单位面临更严苛的效率目标，尤其在小样本下。
Data envelopment analysis has become an important technique for modelling the relationship between inputs and outputs in the production process, particularly in the public sector. However, whenever measures of the output of public sector activity receive public attention, there is a strong possibility that there will be a feedback from the achieved output to the resources devoted to the activity. In other words, the level of resources is endogenous. The implications of such endogeneity for standard econometric estimation techniques are well known, and methods exist to deal with the problem. Most commentators have assumed that endogeneity poses no analogous problems for DEA because the technique merely places an envelope around feasible production possibilities. Using Monte Carlo simulation techniques, however, this paper shows that the efficiency estimates generated by DEA in the presence of endogeneity can be subject to bias, in the sense that inefficient units using low levels of the endogenous resource may be set tougher efficiency targets than equally inefficient units using more of the resource, particularly when sample sizes are small. The paper concludes that, in such circumstances, great caution should be exercised when comparing efficiency measures for units using different levels of the endogenous input.