ارزیابی های بهره وری روش های پوششی داده سنجش ویژه و وابسته به زمینه:یک برنامه بکارگرفته شده در پروژه پشتیبانی توسط بانک جهانی

We evaluate the efficiency of decision making units (DMUs) in a World Bank supported Social Risk Mitigation Project (SRMP) in Turkey through context-dependent and measure-specific data envelopment analysis (DEA) approaches. The results suggest that the efficiency evaluations with context-dependent and measure-specific DEA play various roles in an organization such as setting attainable targets to DMUs, setting long and short term targets to DMUs separately, grouping of DMUs, and improving internal competition between DMUs. Four main contributions of this study can be summarized as follows. Firstly, the study shows the applicability of context-dependent and measure-specific DEA methodologies in a World Bank supported large scale project to increase the effectiveness of the project. Secondly, it outlines some important managerial conclusions of context-dependent DEA clustering approach. Moreover, we propose an alternative approach for attractiveness scores computations in case of exogenous group formations. Finally, the study proposes and applies measure-specific version of context-dependent DEA approach.

Data envelopment analysis (DEA) is a non-parametric approach for identifying relative efficiency of “decision making units” (DMUs) when there are multiple inputs and outputs [1], [2] and [3]. DEA models have been widely applied for the efficiency evaluation throughout different industries, including public and private sectors. Recent applications can be found in Hua et al. [4], Gutiérrez-Nieto et al. [5], Kao and Hung [6], Yu and Lin [7], Erbetta and Rappuoli [8], Eilat et al. [9], Botti et al. [10], Wu et al. [11], Das et al. [12]. When the relative efficiency of the DMU set is evaluated using DEA approach, DMUs are differentiated into two groups: efficient DMUs and inefficient DMUs. Efficiency scores of inefficient DMUs can be distributed between 0 and 1 at all levels. This widespread distribution is particularly occuring in situations where the DMUs do not perform identical activities or in situations where the input–output compositions exhibit great variations due to exogenous factors. In a DEA based efficiency evaluation study performed by Zhu [13] for Fortune 500 companies, 75% of DMUs had efficiency scores below 0.5. When reference sets and efficiency targets for inefficient DMUs are determined using this approach, one often ends up with targets that are very difficult or are impossible to achieve. Therefore, it is important to determine achievable targets for inefficient DMUs. Otherwise, DEA approach will become useless for such problems. This is the main research subject of this paper. In such cases, there could be two alternatives to determine realistic input–output improvement for an inefficient DMU. One alternative could be to ask DMU gradually improve its input–output values and become efficient over time. However, this approach would lead to subjectivity in the determination of sub-targets that would gradually approach the efficient frontier over time. On the other hand, when the inefficient DMUs are large in number it would not be possible and convenient to determine separate sub-targets for each inefficient DMU, nor would it be realistic to assign the same sub-target for each inefficient DMU. As a result, such an approach would negate the strongest feature of DEA analysis, namely, objective performance evaluation in multiple input–output cases. Second alternative that is developed by Charnes [14] is based on dividing DMUs into homogenous sub-groups and then applying DEA analysis to this sub-groups separately. Major deficiency of this approach is difficulties during group formation process. Sometimes a DMU could belong to more than one group. In addition to this, reference set of an inefficient DMU would be only from its sub-group. To top all these, even though the sample can be separated into homogeneous sub-groups, it is still possible to end up with the DMUs that would have impossible to achieve sub-targets. In this study, we utilize a DEA based clustering methodology that enables assigning achievable goals to inefficient DMUs based on their own cluster reference set to solve the above mentioned problem. Although there are several studies in the literature that are dealing with ranking issues in data envelopment analysis [15], [16] and [17] the number of studies that concentrate on clustering through DEA is limited. The context-dependent DEA concept presented by Zhu [18] can be, in a manner, counted as one of these studies. The context-dependent DEA approach starts with clustering the DMUs and obtaining several performance levels. For this purpose, an algorithm is developed to remove the best-practice frontier to allow the remaining (inefficient) DMUs to form a new second-level best-practice frontier. If this new second frontier is removed, a third-level best-practice frontier is formed, and so on, until no DMU is left [18]. Each evaluation level represents an efficient frontier composed by DMUs in a specific performance level. Using this methodology, it is possible to cluster DMUs into groups with a non-parametric approach. The next step of context-dependent DEA approach after clustering the DMUs to several levels is the calculation of attractiveness and progress scores to produce a ranking at every performance level [19] and [20]. In this study, we examine the level-by-level improvement part of context-dependent DEA. We particularly concentrate on some significant managerial conclusions of this approach. Because context-dependent DEA clustering methodology enables assigning achievable goals to inefficient DMUs based on their own cluster reference. On the other hand, in some cases, it may be impossible for a DMU to improve all of the inputs or outputs proportionally at the same time. For these types of situations, measure specific data envelopment models can be used [13], [21] and [22]. Measure-specific models take sets of specific inputs or outputs of interest and give the target values for only those factors. The use of these models can be appropriate for the situations where only one or some of the inputs or outputs can be intervened. At this point, we also applied the measure-specific DEA model in order to obtain more achievable targets for DMUs. In addition to this, we applied measure-specific version of context-dependent DEA model. This study also examines context-dependent DEA based efficiency evaluation approach to measure the relative efficiency of DMUs in a World Bank supported Social Risk Mitigation Project (SRMP) in Turkey. Two main components of the SRMP, namely “Conditional Cash Transfers” and “Local Initiatives” components, were extensively carried out by Social Solidarity Foundations in all provinces of Turkey. We developed both the context-dependent DEA and the measure-specific context-dependent DEA based efficiency evaluation methodology of these Social Solidarity Foundations to determine their performance rankings. The context-dependent DEA methodology determines the efficient Social Solidarity Foundations, as well as targets need to be attained to become efficient and efficiency layers of DMUs. The results suggest that context-dependent DEA and the measure-specific context-dependent DEA approaches can effectively be used as a performance evaluation methodology in large scale projects. In addition to this superiority, the performance measure plays different roles in an organization such as setting attainable targets to DMUs, setting long and short term targets to DMUs separately, grouping of DMUs, and improving internal competition between DMUs. Eventually, four main contributions of this study can be summarized as follows. Firstly, the study shows the applicability of context-dependent and measure-specific DEA methodologies in a World Bank supported large scale project to increase the effectiveness of the project. Secondly, it outlines some important managerial conclusions of context-dependent DEA clustering approach. Moreover, we propose an alternative approach for attractiveness scores computations in case of exogenous group formations. Finally, the study proposes and applies measure-specific version of context-dependent DEA approach. The rest of the paper is arranged as follows. Second section of the study explains context-dependent DEA approach. Third section proposes mathematical foundation of measure- specific version of context-dependent DEA model. Fourth part summarizes the managerial conclusions of the context-dependent DEA approach. Social Risk Mitigation Project in Turkey is briefly explained in fifth part. We also present the data and the dimensions of the application in this section. An application of both the context-dependent and the measure specific context-dependent DEA methodologies in a World Bank supported large scale project is performed in the sixth part.

In this study, we have investigated the applicability of context-dependent and measure-specific DEA approaches in a World Bank supported Social Risk Mitigation Project (SRMP) to evaluate the efficiency of the project. For this purpose, various applications using different aspects of these two concepts have been conducted. Applications are performed using data consisting of 81 DMUs representing the cities of Turkey. Two inputs and 6 outputs are determined according to two main components of the SRMP which are “conditional cash transfers” and “local initiatives”. The analysis conducted in four dimensions. As a first dimension, a standard output-oriented constant returns to scale (CRS) and variable returns to scale (VRS) DEA methodologies are applied to data. As a result of this analysis, too many inefficient units and unrealistic targets for these inefficient DMUs obtained. These results were the motivation behind the application of different approaches of DEA such as context-dependent and measure-specific DEA methodologies. To cope with unrealistic targets, as a second dimension of the study, context-dependent DEA approach is applied to data in order to evaluate the DMUs in their efficiency levels and identify more realistic short-term targets. Many DMUs obtained as inefficient in the standard DEA analysis can now been considered as role models for the DMUs which take place in a lower level. Level by level targets for DMUs obtained so that an efficient DMU with an unrealistic target value in the standard DEA analysis can now set a more attainable target, for example, to carry itself to one or two levels higher. As well as the level-by-level targets, attractiveness and progress scores of DMUs in the highest and lowest levels are evaluated and ranking at these levels are obtained. Here, a new approach to attractiveness concept is also proposed such that the evaluation context for level 1 DMUs are taken as the DMUs in the same region but at different level. In other words, we compute the attractiveness scores in case of exogenous group formations. Unrealistic target improvement values obtained in the first dimension (standard DEA application) also drove us to apply measure-specific DEA methodology in order to obtain target values specified to each input or output. In a real world application such as ours, it is not always possible to intervene all the outputs or inputs. In some cases, it can be only possible to make progress in one output or input. In order to obtain targets by means of each output and input, the third dimension is designed. In this dimension, measure-specific DEA methodology is applied to data. Also, some concepts of measure-specific DEA such as benchmark share and industry efficiency (we used this concept as region efficiency) proposed by Zhu in [13] are also evaluated. As a fourth dimension, we proposed a measure-specific context-dependent DEA methodology so that to obtain targets both in a leveled manner and each input or output specific. In a manner we combined our efforts in the second and third dimensions into a methodology. In each dimension of the application we interpreted the results and reached various managerial conclusions for various DMUs. Eventually, four main contributions of this study can be summarized as follows. Firstly, the study showed the applicability of context-dependent and measure-specific DEA methodologies in a World Bank supported large scale project to increase the effectiveness of the project. Secondly, it outlined some important managerial conclusions of context-dependent DEA clustering approach. Moreover, we proposed an alternative approach for attractiveness scores computations in case of exogenous group formations. Finally, the study proposed and applied measure-specific version of context-dependent DEA approach. The results indicate that the efficiency evaluations with context-dependent and measure-specific DEA can play a role for DMUs in setting attainable targets, setting long and short term targets, grouping of DMUs, and improving internal competition between DMUs. Various managerial conclusions have been attained as the efficiencies of DMUs and target improvements are obtained in a leveled manner and also specified for inputs or outputs. We also evaluated the DMUs in their region context, as regions identified in Turkey for governing purposes have similar demographic and economic characteristics. Some managerial conclusions at each dimension of the application can be summarized as follows: •Average efficiency score obtained through standard DEA analysis is extremely low. Also, only 10 cities out of 81 cities is efficient. For inefficient cities, the targets to be efficient are considerably unattainable. This makes a deeper analysis inevitable to evaluate the performance of Social Risk Mitigation Project. At this point, the use of context-dependent and measure-specific DEA approaches is contributing. •Through application of context-dependent DEA, the cities that take place in SRMP were clustered to 10 efficiency levels. By using a leveled approach, inefficient cities became efficient cities at a lower level and acted as a role model for remaining cities. Also, for inefficient cities more achievable sub-targets were proposed to make progress in a leveled manner. For example, to a seventh level city, making progress to sixth level was proposed. This target was more realistic in the short-term than the target values to attain first level. •Through the calculation of attractiveness scores, a ranking of cities at each level were obtained. Identifying the best frontiers at each level was useful for benchmarking proposals. • Level 1 cities were basically located in three regions; Black Sea (2 cities), East Anatolia (3 cities) and Southeast Anatolia (5 cities). When attractiveness of these cities were evaluated by means of cities remaining in their regions, the best frontiers of the regions were determined. Rankings based on region attractiveness were useful in determining the role models for cities at the same region. For benchmarking purposes, evaluating the cities by taking regional properties into consideration as well were a more realistic way, as the cities at the same region have more similar economic and demographic background. •By use of measure-specific DEA models, target values by means of each input or output were obtained. For such a large scale project, it is not always possible to make progress in every input or output, so obtaining targets on a measure-specific basis proposed cities more attainable targets. •When region efficiencies were evaluated, the most efficient region was the Southeast Anatolia and Marmara seemed the most inefficient region. The reason of the Marmara's inefficiency mostly lied under having the most inefficient city, Istanbul. When the CRS and VRS models were compared, we concluded that the inefficiency of Istanbul was mainly caused by its scale. •For a region to be efficient, all cities in that region must have efficiency score of one. In our application, none of the regions had the efficiency score of 1. By examining the measure-specific DEA scores, we identified that “total distributed budget” was the most important factor generating the inefficiency of both cities and regions. •Measure-specific models and standard DEA model yielded the same best-practice frontiers and so the same 10 efficiency levels. Although the same efficiency levels were present, the measure-specific models yielded different values of the target improvements as only one of the inputs or output are of our interest. In such a real world case, obtaining the targets by means of each input and output and as well as in a leveled manner yields more realistic and attainable short-term objectives for units. In conclusion, the current research may be extended towards various directions. First of all, additional applications are recommended to be analyzed. Furthermore, the examination of operational efficiency over time would be interesting future study. Finally, similar analysis can be performed by various input/output combinations in order to test the sensitivity of the models.