Abstract
Although the efficiency of the water sector has been studied extensively utilizing data envelopment analysis (DEA), most of the literature tends to use the conventional DEA model to compute efficiency scores. In addition, there are many studies that undertake a second stage regression of the efficiency scores against a set of covariates. However, such second-stage analyses are problematic due to potential serial correlation inherent in the estimated efficiency scores. To determine the bias-corrected efficiency scores of rural and urban water utilities in South Africa, this study uses a robust, non-parametric DEA bootstrapping model to generate them. Moreover, little is known about relative performance of the conventional DEA versus bias-corrected DEA. In addition, the truncated double-bootstrap regression results provide insights into the drivers of efficiency, including how water policies and the political and institutional contexts of water utilities can impact relative efficiencies. We found that there are significant differences between the rankings and efficiency scores generated by the conventional DEA model compared to the double-bootstrap DEA model, for both urban and rural samples. The regression model found the location and the ratio of metered to unmetered connections to be significant determinants of efficiency for both urban and rural water utilities.