The Use of Categorical Variables in Data Envelopment Analysis

Data Envelopment Analysis has now been extensively applied in a range of empirical settings to identify relative inefficiencies, and provide targets for improvements. It accomplishes this by developing peer groups for each unit being operated. The use of categorical variables is an important extension which can improve the peer group construction process and incorporate "on-off" characteristics, e.g., presence of drive-in window or not in a banking network. It relaxes the stringent need for factors to display piecewise constant marginal productivities. In so doing, it substantially strengthens the credibility of the insights obtained. The paper treats the cases when the categorical variable can be controllable or uncontrollable by the manager, for the cases of technical and scale inefficiency. The approach is illustrated using real data.