A generalized composition approach in Network Data Envelopment Analysis for complex structures: An application of higher education institutions in Poland

We present in this paper the pitfalls of the most established approach in network data envelopment analysis for units with a parallel internal structure. We show that these pitfalls are the cause of deficiencies of prevalent models employed for general series structures. To overcome these issues, we build a general composition approach that can be applied to units with any type of structure. Our approach relies on multi-objective programming and, unlike existing methods in the literature, we identify the divisional efficiency scores in a min-max and max-min sense simultaneously. This allows us to identify unique and unbiased efficiency scores that are not affected by the different magnitude of scores that the divisions can attain. Comparisons with other approaches, under various structures and assumptions, highlight the advantages of the proposed approach. We further employ this new approach to evaluate the teaching and research efficiency of the top 19 public higher education institutions in Poland with data drawn from the period 2020–2021. The proposed assessment framework departs from the employment of standard metrics such as number of publications and journal rankings, commonly used to evaluate the quantity and quality of research outcomes, and relies on other proxies, such as field-weighted citation impact factor and volume of research grants, that may provide more reliable results.