An innovative approach to efficiency measurement: Combining convexified efficiency analysis trees and data envelopment analysis to benchmark U.S. museums

This study evaluates for the first time the efficiency of U.S. museums using a novel combination of Convexified Efficiency Analysis Trees (CEAT) and Data Envelopment Analysis (DEA). While DEA is widely used for efficiency measurement, it suffers from overfitting and limited discriminatory power in high-dimensional datasets. CEAT addresses these issues by integrating machine-learning techniques with production theory, providing interpretable and robust benchmarks. By categorizing museums into clusters based on input-output profiles, the analysis reveals distinct efficiency patterns across museums of varying sizes and operational foci. The results highlight CEAT's superior ability to differentiate efficiency levels, offering actionable insights into input reductions and output increases for underperforming museums. Specifically, the analysis classifies museums into four clusters according to their size. The first three comprise the largest, medium-large and medium sized museums, while the last cluster constitutes a subtree including smaller museums. It is further divided into several branches according to specific inputs like limited or extensive program expenses (budgets), and short and long opening hours, respectively correlating with non-commercial versus market-oriented business models. Peer benchmarks identified within clusters further guide managerial strategies for improvement. The largest museums can improve efficiency by reducing opening hours or program expenses, while medium-size museums would benefit from reducing staff size and gallery space. As for relatively smaller museums in terms of visitors, besides opening hours, reducing total compensation would also result in greater efficiency. This shows that one-size-fits-all recommendations for museums' efficiency improvement do not apply, adding diversity to the results obtained in previous literature.