Estimation of returns to scale using data envelopment analysis

Abstract Generalization of the measure of returns-to-scale from a single number to an interval permits extension of the concept to DEA data domains with multiple inputs and multiple outputs. The key new approach is a partition of the optimal frontier into three parts corresponding, respectively to increasing, constant, and decreasing returns to scale. These parts are characterized in terms of optimal primal solutions, and optimal dual solutions for both the original Charnes, Cooper, Rhodes model (1978) and the later Banker, Charnes, Cooper model (1984) and relying on concepts developed by R.D. Banker (1984) and R.M. Thrall (1988).