Stepwise DEA ; Choosing variables for measuring technical efficiency in Norwegian electricity distribution

A production technology specification in electricity distribution leads to many product aspects that together with detailed input data gives a high dimensionality. Non-parametric methods may give meaningful results when parametric methods lack degrees of freedom, but have problems with collinear or irrelevant variables. Aggregate efficiency estimates will be little affected, but rates of transformation will be corrupted and extreme observations will be measured as efficient by default. This paper uses tests suggested in the literature to measure the significance of the change in results when disaggregating or introducing an extra variable, and lets the data aid in deciding the specification. JEL Classification: D24, L94, C14