Abstract
In data envelopment analysis (DEA), the concept of efficiency is examined in either Farrell (DEA) or Pareto senses. In either of these senses, the efficiency status of a decision making unit (DMU) is classified as either weak or strong. It is well established that the strong DEA efficiency is both necessary and sufficient for achieving the Pareto efficiency. For the weak Pareto efficiency, however, the weak DEA efficiency is only sufficient, but not necessary in general. Therefore, a DEA-inefficient DMU can be either weakly Pareto efficient or Pareto inefficient. Motivated by this fact, we propose a new classification of DMUs in terms of both DEA and Pareto efficiencies. To make this classification, we first demonstrate that the Farrell efficiency is based on the notion of FGL dominance. Based on the concept of dominance, we then propose and substantiate an alternative single-stage method. Our method is computationally efficient since (1) it involves solving a unique single-stage model for each DMU, and (2) it accomplishes the classification of DMUs in both input and output orientations simultaneously. Finally, we present a numerical example to illustrate our proposed method.
Similar content being viewed by others
Notes
Throughout our paper, we define the concept of Farrell efficiency with respect to strongly disposable technology and refer to it as the Farrell measure of efficiency. However, as per Färe et al. (1985), it is called as the weak measure of efficiency, as they define the Farrell measure of efficiency with respect to weakly disposable technology.
Though Färe et al. (1985, pp. 28 and 46) employ this notion to characterize the weakly efficient subsets of the input and output correspondences and the graph of the technology, they do not call it as a dominance concept.
Bertsimas and Tsitsiklis (1997) presented an LP problem to find out a feasible solution with the maximum number of positive components to a homogeneous system of equations (Exercise 3.27, p. 136).
While dealing with the estimation of a piecewise log-linear technology, one may encounter negative data since the log transformation of values less than 1 are always negative (Zarepisheh et al. 2010; Mehdiloozad et al. 2014). One may also refer to, e.g., Pastor and Ruiz (2007), Sahoo and Tone (2009) and Sahoo et al. (2012), among others, for several examples of applications with negative data.
References
Ali, A. I., & Seiford, L. M. (1993). Computational accuracy and infinitesimals in data envelopment analysis. INFOR, 31, 290–297.
Andersen, P., & Petersen, N. C. (1993). A procedure for ranking efficient units in data envelopment analysis. Management Science, 39, 1261–1264.
Banker, R. D., Charnes, A., & Cooper, W. W. (1984). Some models for estimating technical and scale inefficiencies in data envelopment analysis. Management Sciences, 30, 1078–1092.
Bertsimas, D., & Tsitsiklis, J. N. (1997). Introduction to linear optimization. Belmont: Athena Scientific.
Chang, K. P., & Guh, Y. Y. (1991). Linear production functions and the data envelopment analysis. European Journal of Operational Research, 52, 215–223.
Charnes, A., & Lemke, C. E. (1954). Computational theory of linear programming; I: The bounded variables problem. United States, Office of Naval Research, O.N.R. research memorandum: Graduate School of Industrial Administration, Carnegie Institute of Technology.
Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring the efficiency of decision making units. European Journal of Operational Research, 2, 429–441.
Charnes, A., Cooper, W. W., & Thrall, R. M. (1986). Classifying and characterizing efficiencies and inefficiencies in data envelopment analysis. Operations Research Letters, 5, 105–110.
Charnes, A., Cooper, W. W., & Thrall, R. M. (1991). A Structure for classifying and characterizing efficiency and inefficiency in data envelopment analysis. Journal of Productivity Analysis, 2, 197–237.
Cooper, W. W., Seiford, L. M., & Tone, K. (2007). Data envelopment analysis: A comprehensive text with models, applications, references and DEA-solver software. Boston: Kluwer Academic Publishers.
Dantzig, G. B. (1955). Upper bounds, secondary constraints, and block triangularity in linear programming. Econometrica, 23, 174–183.
Debreu, G. (1951). The coefficient of resource utilization. Econometrica, 19, 273–292.
Dulá, J. H., & Hickman, B. L. (1997). Effects of excluding the column being scored from the DEA envelopment LP technology matrix. Journal of the Operational Research Society, 48, 1001–1012.
Färe, R., Grosskopf, S., & Lovell, C. A. K. (1985). The measurement of efficiency of production. Boston: Kluwer Nijhoff.
Farrell, M. J. (1957). The measurement of productive efficiency. Journal of the Royal Statistical Society, Series A, 129, 253–351.
Jahanshahloo, G. R., Hosseinzadeh Lotfi, F., Shoja, N., Tohidi, G., & Razavyan, S. (2005). A one-model approach to classification and sensitivity analysis in DEA. Applied Mathematics and Computation, 169, 887–896.
Koopmans, T. (1951). Activity analysis of production and allocation. New York: Wiley.
Krivonozhko, V. E., Utkin, O. B., Volodin, A. V., & Sablin, I. A. (2005). About the structure of boundary points in DEA. Journal of the Operational Research Society, 56, 1373–1378.
Mehdiloozad, M., Sahoo, B. K., & Roshdi, I. (2014). A generalized multiplicative directional distance function for efficiency measurement in DEA. European Journal of Operational Research, 214, 679–688.
Pareto, V. (1909). Manuel d’economie politique. Paris: Giars & Briere.
Pastor, J., & Ruiz, J. (2007). Variables with negative values in DEA. In J. Zhu & W. Cook (Eds.), Modeling data irregularities and structural complexities in data envelopment analysis (pp. 63–84). Berlin: Springer.
Ray, S. C. (2008). The directional distance function and measurement of super-efficiency: An application to airlines data. Journal of the Operational Research Society, 59, 788–797.
Sahoo, B. K., Kerstens, K., & Tone, K. (2012). Returns to growth in a non-parametric DEA approach. International Transactions in Operational Research, 19, 463–486.
Sahoo, B. K., & Tone, K. (2009). Radial and non-radial decompositions of profit change: With an application to Indian banking. European Journal of Operational Research, 196, 1130–1146.
Thrall, R. M. (1996). Duality, classification and slacks in DEA. Annals of Operations Research, 66, 109–138.
Winston, W. L. (2003). Operations research: Applications and algorithms. Boston: Duxbury Press.
Zarepisheh, M., Khorram, E., & Jahanshahloo, G. R. (2010). Returns to scale in multiplicative models in data envelopment analysis. Annals of Operations Research, 173, 195–206.
Acknowledgments
We wish to thank Victor V Podinovski and Rajiv D Banker (Guest Editors) and two anonymous referees of the journal, whose invaluable inputs and comments considerably improved the contents of our article. The usual caveat applies.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
Proof of Lemma 3.1
Since part (ii) is an immediate consequence of part (i), we prove only part (i).
Let \(\left( {\mathbf{x}_o ,\mathbf{y}_o } \right) \in NE_I \). By Definition 2.1, there exists \(\hat{{\theta }}<1\) such that \(\left( {\hat{{\theta }}\mathbf{x}_o ,\mathbf{y}_o } \right) \in T^{DEA}\). Since \(\hat{{\theta }}<1\), we have \(\mathbf{x}_o >^{+}\hat{{\theta }}\mathbf{x}_o \). Hence, by Definition 3.1, it follows that \(\left( {\hat{{\theta }}\mathbf{x}_o ,\mathbf{y}_o } \right) {\succeq }_I \left( {\mathbf{x}_o ,\mathbf{y}_o } \right) \).
Conversely, let \(\left( {{{\hat{\mathbf{x}}}},{{\hat{\mathbf{y}}}}} \right) \in T^{DEA}\) dominates \(\left( {\mathbf{x}_o ,\mathbf{y}_o } \right) \) in the FGL sense. By Definition 3.1, we have \(\mathbf{x}_o >^{+}{{\hat{\mathbf{x}}}}\), and \({{\hat{\mathbf{y}}}}{\ge }\mathbf{y}\). We define \(\hat{{\theta }}:=\max \left\{ {\frac{\hat{{x}}_i }{x_{io} }\left| {\hbox { }x_{io} >0} \right. } \right\} \). Then, \(\hat{{\theta }}<1\) and \(\hat{{\theta }}\mathbf{x}_o \,{\geqq }\,{{\hat{\mathbf{x}}}}\). Since \(\left( {{{\hat{\mathbf{x}}}},{{\hat{\mathbf{y}}}}} \right) \in T^{DEA}\), as per the strong disposability assumption, \(\left( {\hat{{\theta }}\mathbf{x}_o ,\mathbf{y}_o } \right) \in T^{DEA}\). Hence, the optimal value of \(\theta ^{*}\) in model (5) is less than one, indicating \(\left( {\mathbf{x}_o ,\mathbf{y}_o } \right) \in NE_I \).
Proof of Proposition 3.1
Note that parts (iii), (v) and (vii) are immediate consequences of parts (iv), (vi) and (viii), respectively. Moreover, the proof of part (vi) is similar to that of part (iv). Therefore, we are left with in proving only parts (i), (ii), (iv) and (viii).
Part (i) Let \(\left( {\mathbf{x}_o ,\mathbf{y}_o } \right) \in E\) and assume by contradiction that \(\left( {\begin{array}{l} \mathbf{x}_o -{{\hat{\mathbf{s}}}}^{-} \\ \mathbf{y}_o +{{\hat{\mathbf{s}}}}^{+} \\ \end{array}} \right) \in \Omega _o \). Let \({{\hat{{\varvec{\upmu }}}}}\) be the intensity variable corresponding to this element in (11). For any direction vector \(\mathbf{g}\), \(\left( {{\beta }^{\prime },{\varvec{\upmu }^{\prime }},\mathbf{s}^{{-}^{\prime }},\mathbf{s}^{{+}^{\prime } }} \right) :=\left( {0,{{\hat{{\varvec{\upmu }} }}},{{\hat{\mathbf{s}}}}^{-},{{\hat{\mathbf{s}}}}^{+}} \right) \) is then a feasible solution to model (8). Since the first phase of model (8) is a minimization problem, the optimal value of \(\beta \) is non-positive, which is a contradiction by Proposition 2.4.
In a similar way, the converse can be proved by the way of contradiction and with the help of Proposition 2.4.
Part (ii) Let \(\left( {\mathbf{x}_o ,\mathbf{y}_o } \right) \in {E}^{\prime }\). Then, part (ii) of Proposition 2.4 follows that \(\left\{ {\left( {\begin{array}{l} \mathbf{x}_o \\ \mathbf{y}_o \\ \end{array}} \right) } \right\} \subseteq \Omega _o\). To prove the equality, assume by contradiction that \(\left( {\begin{array}{l} \mathbf{x}_o -{{\hat{\mathbf{s}}}}^{-} \\ \mathbf{y}_o +{{\hat{\mathbf{s}}}}^{+} \\ \end{array}} \right) \in \Omega _o\) such that \(\left( {\begin{array}{l} {{\hat{\mathbf{s}}}}^{-} \\ {{\hat{\mathbf{s}}}}^{+} \\ \end{array}} \right) \ge \mathbf{0}_{m+s}\). Let \({{\hat{{\varvec{\upmu }} }}}\) be the intensity variable corresponding to this element in (11). For any direction vector \(\mathbf{g}\), \(\left( {{\beta }^{\prime },{\varvec{\upmu }^{\prime }},\mathbf{s}^{{-}^{\prime }},\mathbf{s}^{{+}^{\prime } }} \right) :=\left( {0,{{\hat{{\varvec{\upmu }} }}},{{\hat{\mathbf{s}}}}^{-},{{\hat{\mathbf{s}}}}^{+}} \right) \) is then a feasible solution to model (8). Since the second phase of model (8) is a maximization problem, the optimal sum of slacks is positive, which is a contradiction by Proposition 2.4.
In a similar way, the converse can be proved by the way of contradiction and with the help of Proposition 2.4.
Part (iv) By Lemma 3.1, \(\left( {\mathbf{x}_o ,\mathbf{y}_o } \right) \in NE_I \) if and only if there exists \(\left( {{{\hat{\mathbf{x}}}},{{\hat{\mathbf{y}}}}} \right) \in T^{DEA}\) such that \(\mathbf{x}_o >^{+}{{\hat{\mathbf{x}}}}\) and \(\mathbf{{y}^{\prime }}{\ge }\mathbf{y}_o \). We define \({{\hat{\mathbf{s}}}}^{-}:=\mathbf{x}_o -{{\hat{\mathbf{x}}}}\) and \({{\hat{\mathbf{s}}}}^{+}:={{\hat{\mathbf{y}}}}-\mathbf{y}_o \). Then, \(\left( {\mathbf{x}_o ,\mathbf{y}_o } \right) \in NE_I \) if and only if (i) \({{\hat{\mathbf{s}}}}^{-}\,{\geqq }\,\mathbf{0}_m \) and \({{\hat{\mathbf{s}}}}^{+}\,{\geqq }\,\mathbf{0}_s \), and (ii) \(\hat{{s}}_i^- >0\Leftrightarrow x_{io} >0\); or, if and only if \(\left( {\begin{array}{l} \mathbf{x}_o -{{\hat{\mathbf{s}}}}^{-} \\ \mathbf{y}_o +{{\hat{\mathbf{s}}}}^{+} \\ \end{array}} \right) \in \bar{{\Omega }}_o \) and \(n^{+}\left( {{{\hat{\mathbf{s}}}}^{-}} \right) =n^{+}\left( {\mathbf{x}_o } \right) \).
Part (viii) By Definition 2.3, \(\left( {\mathbf{x}_o ,\mathbf{y}_o } \right) \in NE_P\) if and only if there exists \(\left( {{{\hat{\mathbf{x}}}},{{\hat{\mathbf{y}}}}} \right) \in T^{DEA}\) that strongly dominates \(\left( {\mathbf{x}_o ,\mathbf{y}_o } \right) \). We define \({{\hat{\mathbf{s}}}}^{-}:=\mathbf{x}_o -{{\hat{\mathbf{x}}}}\) and \({{\hat{\mathbf{s}}}}^{+}:={{\hat{\mathbf{y}}}}-\mathbf{y}_o \). Then, \(\left( {\mathbf{x}_o ,\mathbf{y}_o } \right) \in NE_P \) if and only if \(\left( {\begin{array}{l} {{\hat{\mathbf{s}}}}^{-} \\ {{\hat{\mathbf{s}}}}^{+} \\ \end{array}} \right) >\mathbf{0}_{m+s} \); or, if and only if \(\left( {\begin{array}{l} \mathbf{x}_o -{{\hat{\mathbf{s}}}}^{-} \\ \mathbf{y}_o +{{\hat{\mathbf{s}}}}^{+} \\ \end{array}} \right) \in \bar{{\Omega }}_o \) and \(n^{+}\left( {{{\hat{\mathbf{s}}}}^{-}} \right) +n^{+}\left( {{{\hat{\mathbf{s}}}}^{+}} \right) =m+s\). \(\square \)
Proof of Lemma 3.2
First, assume that model (15) is infeasible. Then, system (14) is infeasible and, by Corollary 3.1, \(\left( {\mathbf{x}_o ,\mathbf{y}_o } \right) \) is an extreme-efficient DMU.
Now, let model (15) be feasible. Then, this model has an optimal solution since its objective function is upper bounded by \(m+s\). Hence, let \(\left( {{{\varvec{\updelta }} }^{*},\mathbf{t}^{-*},\mathbf{t}^{+*}} \right) \) be an optimal solution to model (15). We claim that the positive components of the vectors \(\mathbf{t}^{-*}\) and \(\mathbf{t}^{+*}\) are all equal to one. Since the proofs are similar for these vectors, we prove the assertion only for \(\mathbf{t}^{-*}\).
By way of contradiction, assume that \(0<t_g^{-*} <1\) for some \(g\in \left\{ {1,\ldots ,m} \right\} \). Dividing both sides of the constraints of (15) at optimality by \(t_g^{-*} \) yields
Then, according to (16), the vector \(\left( {{\varvec{\updelta }^{\prime }},\mathbf{t}^{{-}^{\prime }},\mathbf{t}^{{+}^{\prime }}} \right) \) defined by
is a feasible solution to model (15) whose objective function value is strictly greater than \(\mathbf{1}_m^T \mathbf{t}^{-*}+\mathbf{1}_s^T \mathbf{t}^{+*}\). This contradicts the optimality of \(\left( {{{\varvec{\updelta }} }^{*},\mathbf{t}^{-*},\mathbf{t}^{+*}} \right) \) and proves our claim. \(\square \)
Proof of Proposition 3.4
According Lemma 3.2, it is obvious that \(\mathbf{1}_m^T \mathbf{t}^{-*}=n^{+}\left( {\mathbf{t}^{-*}} \right) \) and \(\mathbf{1}_s^T \mathbf{t}^{+*}=n^{+}\left( {\mathbf{t}^{+*}} \right) \). Moreover, as in the paragraph below proposition 3.2, it is straightforward to show that \(\left( {\begin{array}{l} \mathbf{x}_o -\frac{1}{\sigma ^{*}}\mathbf{t}^{-*} \\ \mathbf{y}_o +\frac{1}{\sigma ^{*}}\mathbf{t}^{+*} \\ \end{array}} \right) \in \Omega _o \).
Let \(\left( {\begin{array}{l} \mathbf{x}_o -{{\hat{\mathbf{s}}}}^{-} \\ \mathbf{y}_o +{{\hat{\mathbf{s}}}}^{+} \\ \end{array}} \right) \in \Omega _o \) for which \(n^{+}\left( {{{\hat{\mathbf{s}}}}^{-}} \right) +n^{+}\left( {{{\hat{\mathbf{s}}}}^{+}} \right) \) is maximum. Then, we have \(\mathbf{1}_m^T \mathbf{t}^{-*}+\mathbf{1}_s^T \mathbf{t}^{+*}=n^{+}\left( {\mathbf{t}^{-*}} \right) +n^{+}\left( {\mathbf{t}^{+*}} \right) \le n^{+}\left( {{{\hat{\mathbf{s}}}}^{-}} \right) +n^{+}\left( {{{\hat{\mathbf{s}}}}^{+}} \right) \). From Proposition 3.2, we also know that there exists a feasible solution \(\left( {{{\hat{{\varvec{\updelta }} }}},{{\hat{\mathbf{t}}}}^{-},{{\hat{\mathbf{t}}}}^{+}} \right) \) for model (15) such that \(\mathbf{1}_m^T {{\hat{\mathbf{t}}}}^{-}+\mathbf{1}_s^T {{\hat{\mathbf{t}}}}^{+}=n^{+}\left( {{{\hat{\mathbf{s}}}}^{-}} \right) +n^{+}\left( {{{\hat{\mathbf{s}}}}^{+}} \right) \). Therefore, the proof is complete by the fact that model (15) is a maximization LP problem. \(\square \)
Rights and permissions
About this article
Cite this article
Mehdiloozad, M., Ahmadi, M.B. & Sahoo, B.K. On classifying decision making units in DEA: a unified dominance-based model. Ann Oper Res 250, 167–184 (2017). https://doi.org/10.1007/s10479-015-1989-z
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10479-015-1989-z