A minimax approach for selecting the overall and stage-level most efficient unit in two stage production processes

Abstract

Data envelopment analysis (DEA) is a very effective management tool in assessing the performance of a set of decision making units (DMUs). In the efficiency evaluation using classic single stage DEA models, the internal processes of the DMUs are often neglected. In most real-world problems, it may be more realistic to evaluate the efficiency evaluation in two-stage production systems. In some cases: however, the decision-maker must need to identify the most efficient single unit. Numerous methods have been introduced to find the most efficient unit in single stage systems whereas no methods have been proposed for this aim in two stage production systems. Therefore, a new model based on mixed-integer programming was proposed to determine the most efficient DMU in two-stage systems and sub-stages in this study. The most important innovation of the suggested approach is that the most efficient DMUs of both stages can be found separately using only one model. Numerical examples for real world problems and a simulation study were provided for the validity of the proposed model.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3

References

  1. Amin, G. R. (2009). Comments on finding the most efficient DMUs in DEA: An improved integrated model. Computers & Industrial Engineering, 56(4), 1701–1702.

    Article  Google Scholar 

  2. Amin, G. R., & Toloo, M. (2007). Finding the most efficient DMUs in DEA: An improved integrated model. Computers & Industrial Engineering, 52(1), 71–77.

    Article  Google Scholar 

  3. An, Q. X., Chen, H. X., Wu, J., & Liang, L. (2015). Measuring slacks-based efficiency for commercial banks in China by using a two-stage DEA model with undesirable output. Annals of Operations Research, 235, 13–35.

    Article  Google Scholar 

  4. Andersen, P., & Petersen, N. C. (1993). A procedure for ranking efficient units in data envelopment analysis. Management Science, 39(10), 1261–1264.

    Article  Google Scholar 

  5. Azizi, R., & Matin, R. K. (2016). Ranking two-stage production units in data envelopment analysis. Asia-Pacific Journal of Operational Research, 33(01), 1650002.

    Article  Google Scholar 

  6. Charnes, A., & Cooper, W. W. (1984a). Preface to topics in data envelopment analysis. Annals of Operations Research, 2(1), 59–94.

    Article  Google Scholar 

  7. Charnes, A., & Cooper, W. W. (1984b). The non-Archimedean CCR ratio for efficiency analysis: A rejoinder to Boyd and Färe. European Journal of Operational Research, 15(3), 333–334.

    Article  Google Scholar 

  8. Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring the efficiency of decision making units. European Journal of Operational Research, 2(6), 429–444.

    Article  Google Scholar 

  9. Chen, Y., & Zhu, J. (2004). Measuring information technology’s indirect impact on firm performance. Information Technology and Management, 5(1), 9–22.

    Article  Google Scholar 

  10. Chen, C., Zhu, J., Yu, J. Y., & Noori, H. (2012). A new methodology for evaluating sustainable product design performance with two-stage network data envelopment analysis. European Journal of Operational Research, 221(2), 348–359.

    Article  Google Scholar 

  11. Chilingerian, J. A., & Sherman, H. D. (2011). Health-care applications: From hospitals to physicians, from productive efficiency to quality frontiers. Handbook on data envelopment analysis (pp. 445–493). US: Springer.

    Google Scholar 

  12. Ertay, T., & Ruan, D. (2005). Data envelopment analysis based decision model for optimal operator allocation in CMS. European Journal of Operational Research, 164(3), 800–810.

    Article  Google Scholar 

  13. Foroughi, A. A. (2011). A new mixed integer linear model for selecting the best decision making units in data envelopment analysis. Computers & Industrial Engineering, 60(4), 550–554.

    Article  Google Scholar 

  14. Friedman, L., & Sinuany-Stern, Z. (1997). Scaling units via the canonical correlation analysis in the DEA context. European Journal of Operational Research, 100(3), 629–637.

    Article  Google Scholar 

  15. Friedman, L., & Sinuany-Stern, Z. (1998). Combining ranking scales and selecting variables in the DEA context: The case of industrial branches. Computers & Operations Research, 25(9), 781–791.

    Article  Google Scholar 

  16. Galagedera, D. U., Roshdi, I., Fukuyama, H., & Zhu, J. (2018). A new network DEA model for mutual fund performance appraisal: An application to US equity mutual funds. Omega, 77, 168–179.

    Article  Google Scholar 

  17. Golany, B. (1998). An interactive MOLP procedure for the extension of DEA to effectiveness analysis. Journal of the Operational Research Society, 39(8), 725–734.

    Article  Google Scholar 

  18. Gong, Y., Zhu, J., Chen, Y., & Cook, W. D. (2018). DEA as a tool for auditing: Application to Chinese manufacturing industry with parallel network structures. Annals of Operations Research, 263(1–2), 247–269.

    Article  Google Scholar 

  19. Hatami-Marbini, A., & Saati, S. (2020). Measuring performance with common weights: Network DEA. Neural Computing and Applications, 32, 3599–3617.

    Article  Google Scholar 

  20. Kao, C., & Hwang, S. N. (2008). Efficiency decomposition in two-stage data envelopment analysis: An application to non-life insurance companies in Taiwan. European Journal of Operational Research, 185(1), 418–429.

    Article  Google Scholar 

  21. Kao, C., & Hwang, S. N. (2014). Multi-period efficiency and Malmquist productivity index in two-stage production systems. European Journal of Operational Research, 232(3), 512–521.

    Article  Google Scholar 

  22. Kao, C., & Liu, S. T. (2019). Cross efficiency measurement and decomposition in two basic network systems. Omega, 83, 70–79.

    Article  Google Scholar 

  23. Karsak, E. E., & Ahiska, S. S. (2005). Practical common weight multi-criteria decision-making approach with an improved discriminating power for technology selection. International Journal of Production Research, 43(8), 1537–1554.

    Article  Google Scholar 

  24. Lam, K. F. (2015). In the determination of the most efficient decision making unit in data envelopment analysis. Computers & Industrial Engineering, 79, 76–84.

    Article  Google Scholar 

  25. Li, Y., Chen, Y., Liang, L., & Xie, J. (2012). DEA models for extended two-stage network structures. Omega, 40(5), 611–618.

    Article  Google Scholar 

  26. Li, Y., Lei, X., Dai, Q., & Liang, L. (2015). Performance evaluation of participating nations at the 2012 London Summer Olympics by a two-stage data envelopment analysis. European Journal of Operational Research, 243(3), 964–973.

    Article  Google Scholar 

  27. Li, X. B., & Reeves, G. R. (1999). A multiple criteria approach to data envelopment analysis. European Journal of Operational Research, 115(3), 507–517.

    Article  Google Scholar 

  28. Liu, J. S., & Lu, W. M. (2010). DEA and ranking with the network-based approach: A case of R&D performance. Omega, 38(6), 453–464.

    Article  Google Scholar 

  29. Liu, J. S., Lu, W. M., Yang, C., & Chuang, M. (2009). A network-based approach for increasing discrimination in data envelopment analysis. Journal of the Operational Research Society, 60(11), 1502–1510.

    Article  Google Scholar 

  30. Moreno, P., & Lozano, S. (2014). A network DEA assessment of team efficiency in the NBA. Annals of Operations Research, 214(1), 99–124.

    Article  Google Scholar 

  31. Örkcü, H. H., Özsoy, V. S., Örkcü, M., & Bal, H. (2019). A neutral cross efficiency approach for basic two stage production systems. Expert Systems with Applications, 125, 333–344.

    Article  Google Scholar 

  32. Premachandra, I. M., Zhu, J., Watson, J., & Galagedera, D. U. (2012). Best-performing US mutual fund families from 1993 to 2008: Evidence from a novel two-stage DEA model for efficiency decomposition. Journal of Banking & Finance, 36(12), 3302–3317.

    Article  Google Scholar 

  33. Seiford, L. M., & Zhu, J. (1999). Profitability and marketability of the top 55 US commercial banks. Management Science, 45(9), 1270–1288.

    Article  Google Scholar 

  34. Sexton, T. R., & Lewis, H. F. (2003). Two-stage DEA: An application to major league baseball. Journal of Productivity Analysis, 19(2), 227–249.

    Article  Google Scholar 

  35. Sexton, T. R., Silkman, R. H., & Hogan, A. J. (1986). Data envelopment analysis: Critique and extensions. New Directions for Evaluation, 32, 73–105.

    Article  Google Scholar 

  36. Sinuany-Stern, Z., Mehrez, A., & Barboy, A. (1994). Academic departments efficiency via DEA. Computers & Operations Research, 21(5), 543–556.

    Article  Google Scholar 

  37. Sueyoshi, T. (1999). DEA non-parametric ranking test and index measurement: Slack-adjusted DEA and an application to Japanese agriculture cooperatives. Omega, 27(3), 315–326.

    Article  Google Scholar 

  38. Toloo, M. (2012). On finding the most BCC-efficient DMU: A new integrated MIP–DEA model. Applied Mathematical Modelling, 36(11), 5515–5520.

    Article  Google Scholar 

  39. Toloo, M. (2014a). An epsilon-free approach for finding the most efficient unit in DEA. Applied Mathematical Modelling, 38(13), 3182–3192.

    Article  Google Scholar 

  40. Toloo, M. (2014b). The role of non-Archimedean epsilon in finding the most efficient unit: With an application of professional tennis players. Applied Mathematical Modelling, 38(21–22), 5334–5346.

    Article  Google Scholar 

  41. Toloo, M. (2015). Alternative minimax model for finding the most efficient unit in data envelopment analysis. Computers & Industrial Engineering, 81, 186–194.

    Article  Google Scholar 

  42. Toloo, M. (2016). A cost efficiency approach for strategic vendor selection problem under certain input prices assumption. Measurement, 85, 175–183.

    Article  Google Scholar 

  43. Toloo, M., & Ertay, T. (2014). The most cost efficient automotive vendor with price uncertainty: A new DEA approach. Measurement, 52, 135–144.

    Article  Google Scholar 

  44. Toloo, M., & Salahi, M. (2018). A powerful discriminative approach for selecting the most efficient unit in DEA. Computers & Industrial Engineering, 115, 269–277.

    Article  Google Scholar 

  45. Torgersen, A. M., Førsund, F. R., & Kittelsen, S. A. (1996). Slack-adjusted efficiency measures and ranking of efficient units. Journal of Productivity Analysis, 7(4), 379–398.

    Article  Google Scholar 

  46. Wang, Y. M., & Chin, K. S. (2010). Some alternative DEA models for two-stage process. Expert Systems with Applications, 37, 8799–8808.

    Article  Google Scholar 

  47. Wang, C. H., Gopal, R. D., & Zionts, S. (1997). Use of data envelopment analysis in assessing information technology impact on firm performance. Annals of Operations Research, 73, 191–213.

    Article  Google Scholar 

  48. Wang, Y. M., & Jiang, P. (2012). Alternative mixed integer linear programming models for identifying the most efficient decision making unit in data envelopment analysis. Computers & Industrial Engineering, 62(2), 546–553.

    Article  Google Scholar 

  49. Wanke, P., & Barros, C. (2014). Two-stage DEA: An application to major Brazilian banks. Expert Systems with Applications, 41(5), 2337–2344.

    Article  Google Scholar 

  50. Xie, J., Zhu, X., & Liang, L. (2020). A multiplicative method for estimating the potential gains from two-stage production system mergers. Annals of Operations Research, 288, 475–493.

    Article  Google Scholar 

  51. Zhu, J. (2000). Multi-factor performance measure model with an application to Fortune 500 companies. European Journal of Operational Research, 123(1), 105–124.

    Article  Google Scholar 

Download references

Author information

Affiliations

Authors

Corresponding author

Correspondence to Volkan Soner Özsoy.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendix

Appendix

Appendix A. Proofs of theorems

Theorem 1

The MILP model (4) is feasible.

Proof

To ease of notation, we let \( I_{r}^{u} = \frac{1}{{\left( {s + q + m} \right)\hbox{max} \left( {y_{rj} } \right)}};\; r = 1, \ldots ,s \), \( I_{p}^{w} = \frac{1}{{\left( {s + q + m} \right)\hbox{max} \left( {z_{pj} } \right)}}; \;p = 1, \ldots ,q \), and \( I_{i}^{v} = \frac{1}{{\left( {s + q + m} \right)\hbox{max} \left( {x_{ij} } \right)}}; \;i = 1, \ldots ,m \).

For \( I_{r}^{u} \) and \( I_{p}^{w} \), in the third set of constraints of model (4), there are two situations.

One is \( \sum\nolimits_{r = 1}^{s} {I_{r}^{u} y_{rj} } - \sum\nolimits_{p = 1}^{q} {I_{p}^{w} z_{pj} } \le I_{2j} \) for \( \forall j \in \left\{ {1, \ldots ,n} \right\} \) and the other is \( \sum\nolimits_{r = 1}^{s} {I_{r}^{u} y_{rj} } - \sum\nolimits_{p = 1}^{q} {I_{p}^{w} z_{pj} } \ge I_{2j} \) for \( \exists j \in \left\{ {1, \ldots ,n} \right\} \).

Let’s start with \( \sum\nolimits_{r = 1}^{s} {I_{r}^{u} y_{rj} } - \sum\nolimits_{p = 1}^{q} {I_{p}^{w} z_{pj} } \le I_{2j} \).

If \( \sum\nolimits_{p = 1}^{q} {I_{p}^{w} z_{pj} } - \sum\nolimits_{i = 1}^{m} {I_{i}^{v} x_{ij} } \le I_{1j} \), the model has a possible solution. Conversely, if \( \sum\nolimits_{p = 1}^{q} {I_{p}^{w} z_{pj} } - \sum\nolimits_{i = 1}^{m} {I_{i}^{v} x_{ij} } \ge I_{1j} \) \( \exists j \in \left\{ {1, \ldots ,n} \right\} \), we consider the set of deviation variable \( d_{1j} > 0 \left( {j = 1, \ldots ,n} \right) \) meet the conditions \( \sum\nolimits_{p = 1}^{q} {I_{p}^{w} z_{pj} } - \sum\nolimits_{i = 1}^{m} {I_{i}^{v} x_{ij} - d_{1j} } \le I_{1j} \). Let \( \delta_{1} = \mathop {\hbox{max} }\limits_{j} \left\{ {{{d_{1j} } \mathord{\left/ {\vphantom {{d_{1j} } {\sum\nolimits_{{{\text{i}} = 1}}^{m} {x_{ij} } }}} \right. \kern-0pt} {\sum\nolimits_{{{\text{i}} = 1}}^{m} {x_{ij} } }}} \right\} \) and \( v_{i}^{a} = I_{i}^{v} + \delta_{1} \) for \( \forall {\text{i}} \in \left\{ {1, \ldots ,m} \right\} \). Then we can write \( \delta_{1} > 0,v_{i}^{a} > I_{i}^{v} \) for \( \forall i \in \left\{ {1, \ldots ,m} \right\} \), and using the \( {{\delta_{1} \ge d_{1j} } \mathord{\left/ {\vphantom {{\delta_{1} \ge d_{1j} } {\sum\nolimits_{p = 1}^{q} {x_{ij} } }}} \right. \kern-0pt} {\sum\nolimits_{p = 1}^{q} {x_{ij} } }} \), we obtain

$$ \begin{aligned} & \mathop \sum \limits_{p = 1}^{q} I_{p}^{w} z_{pj} - \mathop \sum \limits_{i = 1}^{m} v_{i}^{a} x_{ij} \\ & = \mathop \sum \limits_{p = 1}^{q} I_{p}^{w} z_{pj} - \mathop \sum \limits_{i = 1}^{m} (I_{i}^{v} + \delta_{1} )x_{ij} \\ & = \mathop \sum \limits_{p = 1}^{q} I_{p}^{w} z_{pj} - \mathop \sum \limits_{i = 1}^{m} I_{i}^{v} x_{ij} - \mathop \sum \limits_{i = 1}^{m} \delta_{1} x_{ij} \\ & \le \mathop \sum \limits_{p = 1}^{q} I_{p}^{w} z_{pj} - \mathop \sum \limits_{i = 1}^{m} I_{i}^{v} x_{ij} - d_{1j} \\ & \le I_{1j} \\ \end{aligned} $$

for \( \forall j \in \left\{ {1, \ldots ,n} \right\} \). Thus, \( I_{p}^{w} \left( {p = 1, \ldots ,q} \right),v_{i}^{a} \left( {i = 1, \ldots ,m} \right) \), \( I_{r}^{u} \;\left( {r = 1, \ldots ,s} \right) \) are in feasible region.

Consider the \( \sum\nolimits_{r = 1}^{s} {I_{r}^{u} y_{rj} } - \sum\nolimits_{p = 1}^{q} {I_{p}^{w} z_{pj} } \ge I_{2j} \) \( \exists j \in \left\{ {1, \ldots ,n} \right\} \). Let a set of deviation variable \( d_{2j} > 0 \left( {j = 1, \ldots ,n} \right) \) meet the conditions \( \sum\nolimits_{r = 1}^{s} {I_{r}^{u} y_{rj} } - \sum\nolimits_{p = 1}^{q} {I_{p}^{w} z_{pj} } - d_{2j} \le I_{2j} \). Let \( \delta_{2} = \mathop {\hbox{max} }\limits_{j} \left\{ {{{d_{2j} } \mathord{\left/ {\vphantom {{d_{2j} } {\sum\nolimits_{p = 1}^{q} {z_{pj} } }}} \right. \kern-0pt} {\sum\nolimits_{p = 1}^{q} {z_{pj} } }}} \right\} \) and \( w_{p}^{a} = I_{p}^{w} + \delta_{2} \) for \( \forall p \in \left\{ {1, \ldots ,q} \right\} \) for this case. Then we can write \( \delta_{2} > 0, w_{p}^{a} > I_{p}^{w} \) for \( \forall p \in \left\{ {1, \ldots ,q} \right\} \), and using \( {{\delta_{2} \ge d_{2j} } \mathord{\left/ {\vphantom {{\delta_{2} \ge d_{2j} } {\sum\nolimits_{p = 1}^{q} {z_{pj} } }}} \right. \kern-0pt} {\sum\nolimits_{p = 1}^{q} {z_{pj} } }} \), i.e., \( \delta_{2} \sum\nolimits_{p = 1}^{q} {z_{pj} } \ge d_{2j} \) we have

$$ \begin{aligned} & \mathop \sum \limits_{r = 1}^{s} I_{r}^{u} y_{rj} - \mathop \sum \limits_{p = 1}^{q} w_{p}^{a} z_{pj} \\ & = \mathop \sum \limits_{r = 1}^{s} I_{r}^{u} y_{rj} - \mathop \sum \limits_{p = 1}^{q} \left( {I_{p + }^{w} \delta_{2} } \right)z_{pj} \\ & = \mathop \sum \limits_{r = 1}^{s} I_{r}^{u} y_{rj} - \mathop \sum \limits_{p = 1}^{q} I_{p}^{w} z_{pj} - \mathop \sum \limits_{p = 1}^{q} \delta_{2} z_{pj} \\ & \le \mathop \sum \limits_{r = 1}^{s} I_{r}^{u} y_{rj} - \mathop \sum \limits_{p = 1}^{q} I_{p}^{w} z_{pj} - d_{2j} \\ & \le I_{2j} \\ \end{aligned} $$

for \( \forall j \in \left\{ {1, \ldots ,n} \right\} \).

If \( \sum\nolimits_{p = 1}^{q} {w_{p}^{a} z_{pj} } - \sum\nolimits_{i = 1}^{m} {I_{i}^{v} x_{ij} } \le I_{1j} \) \( \forall j \in \left\{ {1, \ldots ,n} \right\} \), the model has a feasible region.

If \( \sum\nolimits_{p = 1}^{q} {w_{p}^{a} z_{pj} } - \sum\nolimits_{i = 1}^{m} {I_{i}^{v} x_{ij} } \ge I_{1j} \) \( \exists j \in \left\{ {1, \ldots ,n} \right\} \), we consider a set of deviation variable \( \tilde{d}_{1j} > 0 \left( {j = 1, \ldots ,n} \right) \) meet the conditions \( \sum\nolimits_{p = 1}^{q} {w_{p}^{a} z_{pj} } - \sum\nolimits_{i = 1}^{m} {I_{i}^{v} x_{ij} } - \tilde{d}_{1j} \le I_{1j} \). Let \( \tilde{\delta }_{1} = \mathop {\hbox{max} }\limits_{j} \left\{ {{{\tilde{d}_{1j} } \mathord{\left/ {\vphantom {{\tilde{d}_{1j} } {\sum\nolimits_{{{\text{i}} = 1}}^{m} {x_{ij} } }}} \right. \kern-0pt} {\sum\nolimits_{{{\text{i}} = 1}}^{m} {x_{ij} } }}} \right\} \) and \( v_{i}^{b} = I_{i}^{v} + \tilde{\delta }_{1} \) for \( \forall i \in \left\{ {1, \ldots ,m} \right\} \). Then we can write \( \tilde{\delta }_{1} > 0,v_{i}^{a} > I_{i}^{v} \) for \( \forall i \in \left\{ {1, \ldots ,m} \right\} \), and using the \( \tilde{\delta }_{1} \ge {{\tilde{d}_{1j} } \mathord{\left/ {\vphantom {{\tilde{d}_{1j} } {\sum\nolimits_{p = 1}^{q} {x_{ij} } }}} \right. \kern-0pt} {\sum\nolimits_{p = 1}^{q} {x_{ij} } }} \), we obtain

$$ \begin{aligned} & \mathop \sum \limits_{p = 1}^{q} w_{p}^{a} z_{pj} - \mathop \sum \limits_{i = 1}^{m} v_{i}^{b} x_{ij} \\ & = \mathop \sum \limits_{p = 1}^{q} w_{p}^{a} z_{pj} - \mathop \sum \limits_{i = 1}^{m} (I_{i}^{v} + \tilde{\delta }_{1} )x_{ij} \\ & = \mathop \sum \limits_{p = 1}^{q} w_{p}^{a} z_{pj} - \mathop \sum \limits_{i = 1}^{m} I_{i}^{v} x_{ij} - \mathop \sum \limits_{i = 1}^{m} \tilde{\delta }_{1} x_{ij} \\ & \le \mathop \sum \limits_{p = 1}^{q} w_{p}^{a} z_{pj} - \mathop \sum \limits_{i = 1}^{m} I_{i}^{v} x_{ij} - \tilde{d}_{1j} \\ & \le I_{1j} \\ \end{aligned} $$

for \( \forall j \in \left\{ {1, \ldots ,n} \right\} \). So, \( w_{p}^{a} \left( {p = 1, \ldots ,q} \right), \) \( v_{i}^{b} \) \( \left( {i = 1, \ldots ,m} \right) \), \( I_{r}^{u} \) \( \left( {r = 1, \ldots ,s} \right) \) are in feasible region. That is, the MILP model for two stages is always feasible. If lower bound solutions are infeasible then, these can be made feasible by increasing the values of input weight (\( v_{i} \)) or of the intermediate measures weights (\( w_{p} \)).

Summing the constraints \( \sum\nolimits_{p = 1}^{q} {w_{p} z_{pj} } - \sum\nolimits_{i = 1}^{m} {v_{i} x_{ij} } \le I_{1j} \) and \( \sum\nolimits_{r = 1}^{s} {u_{r} y_{rj} } - \sum\nolimits_{p = 1}^{q} {w_{p} z_{pj} } \le I_{2j} \) over \( j \) from \( j = 1 \) to \( n \) seperately, we have

$$ \mathop \sum \limits_{i = 1}^{m} v_{i} \left( {\mathop \sum \limits_{j = 1}^{n} x_{ij} } \right) - \mathop \sum \limits_{p = 1}^{q} w_{p} \left( {\mathop \sum \limits_{j = 1}^{n} z_{pj} } \right) \ge - 1 $$

and

$$ \mathop \sum \limits_{p = 1}^{q} w_{p} \left( {\mathop \sum \limits_{j = 1}^{n} z_{pj} } \right) - \mathop \sum \limits_{r = 1}^{s} u_{r} \left( {\mathop \sum \limits_{j = 1}^{n} y_{rj} } \right) \ge - 1. $$

Summing up the last two inequalities, we can write

$$ \mathop \sum \limits_{i = 1}^{m} v_{i} \left( {\mathop \sum \limits_{j = 1}^{n} x_{ij} } \right) - \mathop \sum \limits_{r = 1}^{s} u_{r} \left( {\mathop \sum \limits_{j = 1}^{n} y_{rj} } \right) \ge - 2. $$

Therefore, the objective function of the model (4) has a lower bound. As the MILP model is a minimization problem, model (4) is a feasible solution. This completes the proof.

Theorem 2

Model (4) and model (5) are equivalent.

Proof

From the \( \sum\nolimits_{p = 1}^{q} {w_{p} z_{pj} } - \sum\nolimits_{i = 1}^{m} {v_{i} x_{ij} + S_{1j} } = I_{1j} \) constraint, we can write

$$ S_{1j} = \mathop \sum \limits_{i = 1}^{m} v_{i} x_{ij} - \mathop \sum \limits_{p = 1}^{q} w_{p} z_{pj} + I_{1j} . $$

Summing up these constraints over \( j \) from \( j = 1 \) to \( j = n \) and considering \( \sum\nolimits_{j = 1}^{n} {I_{1j} } = 1, \) we can write

$$ \mathop \sum \limits_{j = 1}^{n} S_{1j} = \mathop \sum \limits_{i = 1}^{m} v_{i} \left( {\mathop \sum \limits_{j = 1}^{n} x_{ij} } \right) - \mathop \sum \limits_{p = 1}^{q} w_{p} \left( {\mathop \sum \limits_{j = 1}^{n} z_{pj} } \right) + 1. $$

Similarly, from the third type of constraints in the model (5), summing up these constraints over j from j = 1 to j = n, and considering \( \sum\nolimits_{j = 1}^{n} {I_{2j} } = 1, \) we can write

$$ \mathop \sum \limits_{j = 1}^{n} S_{2j} = \mathop \sum \limits_{p = 1}^{q} w_{p} \left( {\mathop \sum \limits_{j = 1}^{n} z_{pj} } \right) - \mathop \sum \limits_{r = 1}^{s} u_{r} \left( {\mathop \sum \limits_{j = 1}^{n} y_{rj} } \right) + 1. $$

By summing up the last two inequalities, we obtain

$$ \mathop \sum \limits_{j = 1}^{n} S_{1j} + \mathop \sum \limits_{j = 1}^{n} S_{2j} = \mathop \sum \limits_{i = 1}^{m} v_{i} \left( {\mathop \sum \limits_{j = 1}^{n} x_{ij} } \right) - \mathop \sum \limits_{p = 1}^{q} w_{p} \left( {\mathop \sum \limits_{j = 1}^{n} z_{pj} } \right) + 2. $$

If the constant 2 is ignored, the objective function of the model (5) is written as

$$ \mathop \sum \limits_{j = 1}^{n} S_{1j} + \mathop \sum \limits_{j = 1}^{n} S_{2j} = \mathop \sum \limits_{i = 1}^{m} v_{i} \left( {\mathop \sum \limits_{j = 1}^{n} x_{ij} } \right) - \mathop \sum \limits_{r = 1}^{s} u_{r} \left( {\mathop \sum \limits_{j = 1}^{n} y_{rj} } \right) $$

which is equal to the objective function of model (4). This completes the proof.

Theorem 3

The following model (6) and model (4) are equivalents.

Proof

In model (6), let \( d_{1j} = 1 - I_{1j} \) and \( \beta_{1j} = 1 - S_{1j} \) for \( j = 1, \ldots , n \). Summing up these equations over \( j \) from \( j = 1 \) to \( j = n \), this results in

$$ \mathop \sum \limits_{j = 1}^{n} d_{1j} = n - \mathop \sum \limits_{j = 1}^{n} I_{1j} = n - 1 $$

and

$$ \mathop \sum \limits_{j = 1}^{n} \beta_{1j} = - \left( {\mathop \sum \limits_{j = 1}^{n} S_{1j} - n} \right) = \left( {n - 1} \right) - \left( {\mathop \sum \limits_{i = 1}^{m} v_{i} \left( {\mathop \sum \limits_{j = 1}^{n} x_{ij} } \right) - \mathop \sum \limits_{p = 1}^{q} w_{p} \left( {\mathop \sum \limits_{j = 1}^{n} z_{pj} } \right)} \right). $$

Similarly, if \( d_{2j} = 1 - I_{2j} \) and \( \beta_{2j} = 1 - S_{2j} \) are taken for \( j = 1, \ldots , n \), it is obtained as

$$ \mathop \sum \limits_{j = 1}^{n} \beta_{2j} = \left( {n - 1} \right) - \left( {\mathop \sum \limits_{p = 1}^{q} w_{p} \left( {\mathop \sum \limits_{j = 1}^{n} z_{pj} } \right) - \mathop \sum \limits_{i = 1}^{m} v_{i} \left( {\mathop \sum \limits_{j = 1}^{n} x_{ij} } \right)} \right). $$

Hence, \( \sum\nolimits_{j = 1}^{n} {\left( {\beta_{1j} + \beta_{2j} } \right)} \) is obtained as \( \left( {2n - 2} \right) - \sum\nolimits_{i = 1}^{m} {v_{i} } \left( {\sum\nolimits_{j = 1}^{n} {x_{ij} } } \right) - \sum\nolimits_{r = 1}^{s} {u_{r} } \left( {\sum\nolimits_{j = 1}^{n} {y_{rj} } } \right). \)

Since \( 2n - 2 \) is a constant, the fact that \( \sum\nolimits_{j = 1}^{n} {\left( {\beta_{1j} + \beta_{2j} } \right)} \) is equal to minimum of

$$ \mathop \sum \limits_{i = 1}^{m} v_{i} \left( {\mathop \sum \limits_{j = 1}^{n} x_{ij} } \right) - \mathop \sum \limits_{r = 1}^{s} u_{r} \left( {\mathop \sum \limits_{j = 1}^{n} y_{rj} } \right) $$

completes the proof.

Theorem 4

The proposed minimax model (8) is always feasible and bounded.

Proof

Let \( \left( {u^{0} ,w^{0} ,v^{0} ,I_{1}^{0} ,I_{2}^{0} } \right) \) be a feasible solution to model (7) and we proved that such a solution exists in Theorem 1. Let \( d_{1j}^{0} = 1 - I_{1j}^{0} \), \( \beta_{1j}^{0} = 1 - I_{1j}^{0} + w^{0} z_{j} - v^{0} x_{j} \) for \( j = 1 \) to \( j = n \) and \( d_{max1}^{0} = max\left\{ {d_{1j}^{0} - \beta_{1j}^{0} :j = 1, \ldots , n } \right\} \) for the first stage. Summing up \( d_{1j}^{0} = 1 - I_{1j}^{0} \) over \( j = 1 \) to \( j = n \) results in \( \sum\nolimits_{j = 1}^{n} {d_{1j}^{0} } = n - 1 \). For \( \sum\nolimits_{p = 1}^{q} {w_{p} z_{pj} } - \sum\nolimits_{i = 1}^{m} {v_{i} x_{ij} + d_{1j} - \beta_{1j} } = 0 \) constraints, \( w^{0} z_{j} - v^{0} x_{j} + 1 - I_{1j}^{0} - 1 + I_{1j}^{0} - w^{0} z_{j} + v^{0} x_{j} = 0 \) could be proved. In the first stage, while for the most efficient unit, \( w^{0} z_{j} - v^{0} x_{j} \le 1 \) satisfies, for the other units, \( w^{0} z_{j} - v^{0} x_{j} \le 0 \) satisfies. Therefore, for \( d_{{max_{1} }} - d_{1j} + \beta_{1j} \ge 0 \) constraints, \( d_{max1}^{0} \ge v^{0} x_{j} - w^{0} z_{j} \) could be obtained. Hence, when considering the most efficient unit and other units for the first stage, \( d_{max1}^{0} \ge - 1 \) is obtained. Similarly, if \( d_{2j}^{0} = 1 - I_{2j}^{0} \) and \( \beta_{2j}^{0} = 1 - I_{2j}^{0} + u^{0} y_{j} - w^{0} z_{j} \) for \( j = 1 \) to \( j = n \) and \( d_{max2}^{0} = max\left\{ {d_{2j}^{0} - \beta_{2j}^{0} :j = 1, \ldots , n } \right\} \) are taken for the second stage, the same results are obtained. It is proved that \( \left( {u^{0} ,w^{0} ,v^{0} ,I_{1}^{0} ,I_{2}^{0} } \right) \) is a feasible solution to model (8).

To show that the proposed model has the bounded, \( \left( {u^{0} ,w^{0} ,v^{0} ,d_{1}^{0} ,\beta_{1}^{0} ,d_{2}^{0} ,\beta_{2}^{0} ,d_{max1}^{0} ,d_{max2}^{0} } \right) \) is taken any arbitrary feasible solution to model (8). From the constraints of this model,

$$ d_{max1}^{0} \ge d_{1j}^{0} - \beta_{1j}^{0} \ge - \beta_{1j}^{0} \ge - 1 $$

and similarly

$$ d_{max2}^{0} \ge d_{2j}^{0} - \beta_{2j}^{0} \ge - \beta_{2j}^{0} \ge - 1 $$

is obtained. Summing up the last two equations,

$$ d_{max1}^{0} + d_{max2}^{0} \ge d_{1j}^{0} - \beta_{1j}^{0} + d_{2j}^{0} - \beta_{2j}^{0} \ge - \beta_{1j}^{0} - \beta_{2j}^{0} \ge - 2 $$

means the objective function is bounded from below. Since the proposed model (8) is a minimization problem, the proof is completed.

Appendix B. Detailed graphs on cases in simulation

See Figs. 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20 and 21.

Fig. 4
figure4

Graph of correlation results in 100 repeats for Case 1

Fig. 5
figure5

Graph of correlation results in 100 repeats for Case 2

Fig. 6
figure6

Graph of correlation results in 100 repeats for Case 3

Fig. 7
figure7

Graph of correlation results in 100 repeats for Case 4

Fig. 8
figure8

Graph of correlation results in 100 repeats for Case 5

Fig. 9
figure9

Graph of correlation results in 100 repeats for Case 6

Fig. 10
figure10

Graph of correlation results in 100 repeats for Case 7

Fig. 11
figure11

Graph of correlation results in 100 repeats for Case 8

Fig. 12
figure12

Graph of correlation results in 100 repeats for Case 9

Fig. 13
figure13

Graph of correlation results in 100 repeats for Case 10

Fig. 14
figure14

Graph of correlation results in 100 repeats for Case 11

Fig. 15
figure15

Graph of correlation results in 100 repeats for Case 12

Fig. 16
figure16

Graph of correlation results in 100 repeats for Case 13

Fig. 17
figure17

Graph of correlation results in 100 repeats for Case 14

Fig. 18
figure18

Graph of correlation results in 100 repeats for Case 15

Fig. 19
figure19

Graph of correlation results in 100 repeats for Case 16

Fig. 20
figure20

Graph of correlation results in 100 repeats for Case 17

Fig. 21
figure21

Graph of correlation results in 100 repeats for Case 18

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Özsoy, V.S., Örkcü, M. & Örkcü, H.H. A minimax approach for selecting the overall and stage-level most efficient unit in two stage production processes. Ann Oper Res (2021). https://doi.org/10.1007/s10479-020-03827-x

Download citation

Keywords

  • Data envelopment analysis
  • Two-stage production systems
  • Most efficient DMU
  • Minimax model