Socio-economic predictors of electricity theft in developing countries: An Indian case study

More than a fifth of the total electricity production in India is lost due to theft. Previous research indicates that technical solutions, on their own, are not sufficient to curb electricity theft and that social and economic factors should additionally be taken into account. Using disaggregated district-level data from Uttar Pradesh, the largest state of India, over seven years (2006–2012), this study examines the socio-economic determinants of electricity theft behaviors. The study deploys an array of advanced machine-learning regression models to a) quantify the predictive power of socio-economic indicators in explaining the electricity theft behaviors at the district level, and, b) capture and illustrate non-linear relationships between relevant socio-economic indicators and electricity theft. In addition, the study explores the temporal-spatial correlation of electricity theft across districts and over the years. The results suggest that in using a random forest regression model in particular, 87% of the variability of loss rate could be explained by the underlying socio-economic attributes considered in this study. Specifically, crime rate, literacy rate, income, urbanization, and the average electricity consumption per capita are shown to be statistically significant. The results also suggest strong temporal-spatial correlations of electricity theft across some districts, when the average correlation was 0.39 to neighboring districts and only 0.14 to distant districts.