Open Access

Sebastian Kohl

• This dissertation is concerned with the advancement of data envelopment analysis, especially for the implementation in hospital settings. It comprises three contributions: The first contribution is a literature review, which fills an unreviewed gap of over ten years and encompasses 262 publications. The second contribution develops a methodology to assess the accuracy of DEA models. The third contribution sets up an advanced teaching case to distribute knowledge on the DEA methodology among researchers and practitioners. After a motivation of the topic at the beginning of this theses, three research questions are derived that guide the dissertation. The findings of contribution 1 concerning research questions 1 and 2 lead the way towards the central topic of this thesis, which is stated in research question 3: What can be done to advance DEA in healthcare? The answers to this central question comprise findings of all contributions. Most importantly, the development of a method toThis dissertation is concerned with the advancement of data envelopment analysis, especially for the implementation in hospital settings. It comprises three contributions: The first contribution is a literature review, which fills an unreviewed gap of over ten years and encompasses 262 publications. The second contribution develops a methodology to assess the accuracy of DEA models. The third contribution sets up an advanced teaching case to distribute knowledge on the DEA methodology among researchers and practitioners. After a motivation of the topic at the beginning of this theses, three research questions are derived that guide the dissertation. The findings of contribution 1 concerning research questions 1 and 2 lead the way towards the central topic of this thesis, which is stated in research question 3: What can be done to advance DEA in healthcare? The answers to this central question comprise findings of all contributions. Most importantly, the development of a method to compare the accuracy of DEA models is an essential step for the advancement of DEA. For the first time, different DEA models can be compared and evaluated based on a neutral criterion. The benchmarking method allows the judgment of existing models, as well as the trial of new model developments. This procedure allows the formation of new gold standards in DEA, which can replace the basic CCR model. An experimental study in the second contribution shows that the SBM outperforms the CCR model and should be the new standard for the evaluation of constant returns to scale studies. This finding is generally valid for all DEA applications. Its particular relevance for healthcare DEA is supported by the answer to research question 2, which highlights an inadequate representation of sophisticated models in healthcare applications. One of the answers to research question 1 reveals another main field for advancing healthcare DEA studies: A trend to include quality measures into the analysis has started. The contributions 1 and 3 promote the topic, in order to raise the share of studies considering quality in their analysis. Besides a rising share of studies considering quality, meaningful implementation of quality indicators into the studies is of prime importance. Contribution 3 presents a two-stage approach using a Helmsman DEA in the first stage. Its benefits lie in a suitable inclusion of multiple indicators. Furthermore, it assures that the DMUs cannot evade the evaluation of quality. In doing so, the inclusion of quality parameters via the Helmsman DEA method is not automatically increasing the average efficiency in a data sample, as other conventional methods do. Finally, the method is parsimonious with inputs and outputs, which supports another core finding of this thesis: The adherence to meaningful settings for DEA is without any alternative. In this regard, the setting with the most substantial impact is the restrainment of inputs and outputs with regard to the available DMU number. Contribution 2 underpins the influence of this setting on the accuracy of results and shows that existing guidelines cannot guarantee a sufficient estimation accuracy. Therefore, a new guideline is developed, which ensures a higher quality of results for future studies. The contributions of this thesis open plenty of possibilities for future research. First of all, the presented benchmarking procedure for the accuracy of DEA models focusses on constant returns to scale settings. An extension of the method to variable returns to scale settings is a natural next step. Furthermore, the models under assessment in the experimental study of contribution 2 are a first selection. Further evaluations can reveal other existing models that outperform the SBM model. Examining a combination of the SBM and AR models, as suggested in Tone (2001) or testing for the best AR restrictions are as well suitable future research projects. Besides, advancing the development and application of the bootstrap for other models than the CCR is desirable for future research. As the method evolves to become state of the art in DEA, a shift of the standard DEA model to more sophisticated models should not be prevented by missing bootstrapping procedures for these models. In general, the contributions present encouraging results regarding the accuracy of DEA. Not every DEA model delivers satisfying results, and in some settings, DEA should not be applied at all. However, if sophisticated models are used in an appropriate environment, DEA is a valuable method for supporting decision makers. The contributions of this thesis are meant to advance DEA for healthcare settings and help the methodology to find acceptance and application in actual management support.…