What Is Data Envelopment Analysis?

Data envelopment analysis (DEA) is a nonparametric method for measuring the relative efficiency of a set of decision-making units (DMUs) that consume multiple inputs to produce multiple outputs. It was introduced by Charnes, Cooper, and Rhodes in a 1978 paper in the European Journal of Operational Research, building on earlier productivity work by Farrell. DEA belongs to the field of operations research and management science, and it is distinguished from regression-based approaches by requiring no prior assumption about the functional form of the production relationship between inputs and outputs. Instead, it constructs an efficiency frontier empirically from the observed data, then scores each DMU by how far it falls from that frontier.

The method has found wide application in economics, engineering, healthcare administration, and public policy, anywhere a decision-maker needs to compare the performance of similar operating units without imposing a parametric production model. A hospital network evaluating the cost efficiency of its facilities, a utility regulator benchmarking electricity distributors, and a university ranking departments by research output relative to faculty size are all natural DEA applications.

Efficiency Measurement and the Production Frontier

DEA constructs a production possibility set by finding, through linear programming, the smallest convex cone (under constant returns to scale) or convex hull (under variable returns to scale) that envelops all observed input-output combinations. This constructed boundary is the efficient frontier: the set of DMUs that cannot be improved by any linear combination of observed units. A DMU on the frontier receives an efficiency score of 1.0. A DMU below the frontier receives a score between 0 and 1, representing the proportion by which its inputs could be reduced (input orientation) or its outputs expanded (output orientation) while remaining feasible given the performance of other observed units. The FAO appendix on data envelopment analysis presents the foundational model formulation and interprets the efficiency scores in the context of agricultural productivity measurement.

DEA Model Types

The original Charnes-Cooper-Rhodes (CCR) model assumes constant returns to scale: a DMU that doubles its inputs should double its outputs. Banker, Charnes, and Cooper extended this in 1984 with the BCC model, which assumes variable returns to scale by adding a convexity constraint to the linear program. This extension separates overall technical efficiency into pure technical efficiency (measured against the variable-returns frontier) and scale efficiency (measuring whether the DMU is operating at its most productive scale size). A third major variant, the slack-based measure (SBM), addresses weaknesses in the radial efficiency scores of CCR and BCC models by directly incorporating input and output slacks into the objective function, producing scores that reflect all sources of inefficiency simultaneously. The Annals of Operations Research paper on DEA and its related linear programming models provides a systematic treatment of model variants and their mathematical relationships.

Extensions and Limitations

DEA has been extended in many directions: dynamic DEA tracks efficiency across time periods; network DEA models production as a chain of sub-processes rather than a single black box; stochastic DEA incorporates uncertainty in inputs or outputs. A persistent limitation of the basic model is its sensitivity to outliers: a single unusually efficient or mis-recorded DMU can shift the frontier and distort scores for all other units. DEA scores are also relative, not absolute; a score of 1.0 means only that no observed combination of peers performs better, not that the unit is globally optimal. The PMC article on DEA efficiency in the public sector using provider and patient data illustrates how these extensions and caveats apply in a real healthcare benchmarking study.

Applications

Data envelopment analysis has applications in a wide range of disciplines, including:

• Healthcare performance benchmarking, comparing hospital and clinic efficiency on cost, staffing, and patient outcome metrics
• Energy sector regulation, where utilities are benchmarked to set efficiency targets in price control reviews
• Higher education, evaluating research and teaching productivity across university departments
• Banking and financial services, measuring branch and institution efficiency relative to input costs
• Transportation and logistics, assessing fleet and route efficiency for airlines, ports, and freight carriers