Categorical variables in DEA.

If a DEA model has a mix of categorical and continuous variables a standard LP formulation can still be used by entering all combinations of categorical and continuous variables as different types of inputs and/or outputs. Most units will then not have positive levels of all variables. The implications for selection of peers are investigated. Peers can have the same or fewer types of inputs than the unit under investigation, but either fewer or more types of outputs. There is a basic asymmetry between number of positive inputs and outputs of the peer units due to more of inputs reducing efficiency while more of outputs improving efficiency. The special cases of imposing a hierarchical structure on the categorical variables dealt with in the literature can easily be incorporated.