Data Envelopment Analysis with Output-Bounded Data
Abstract
In conventional data envelopment analysis (DEA), data are usually assumed to be non-negative with no specific bounds. However, many practical applications require some data, and thus their projections, to fall within certain limits. For example, percentage data such as the satisfactory rate cannot exceed 100% to make sense. This data characteristic is very likely to be violated under the assumption of constant returns to scale (CRS), due to its ray expansion property. In order to tackle this issue under CRS, a series of radial models are developed to constrain DEA projections within imposed bounds from the output side. Then efficient decision making units (DMUs) can be further discriminated simply by eliminating it from the reference set, avoiding the infeasibility problem existing in the VRS super-efficiency measures. The methodology is demonstrated with data consisting of 119 general acute care hospitals located in Pennsylvania, USA.
