Understanding segmentation in rural electricity markets: Evidence from India

• Published in: Energy economics Vol. 87; p. 104697
• Main Authors: Robert Thomas, Daniel, Agrawal, Shalu, Harish, S.P., Mahajan, Aseem, Urpelainen, Johannes
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier B.V 01.03.2020; Elsevier Science Ltd
• Subjects: Clustering; Clusters; Cognitive style; Consumers; Customers; Datasets; Demand; Dependent variables; Electricity; Electricity consumption; Electricity meters; Energy; Energy economics; Households; India; Learning algorithms; Lighting; Machine learning; Markets; Membership; Rural areas; Rural development; Satisfaction; Segmentation; Segments; Villages; Willingness to pay; India; Demand; Segmentation; L94; C38; Q40; Clustering; Q41; India; O13; Rural development; Electricity; Energy; Q47
• ISSN: 0140-9883; 1873-6181
• DOI: 10.1016/j.eneco.2020.104697

How can demand for electricity be estimated without fine-grained usage data? Employing an original and large dataset, we develop a novel method for determining drivers of demand without electricity meter data. We first segment Indian consumers by their willingness to pay for electricity service, their level of usage, and their satisfaction with lighting, and then use cluster membership as a dependent variable in order to determine which household-level factors predict electricity usage. Our approach employs machine-learning and more traditional regression techniques to determine the optimal number of segments, generate the segments, and determine the predictors of segment membership. The dataset consists of more than 10,000 households in more than 200 villages in the states of Bihar, Odisha, Rajasthan, and Uttar Pradesh. We find that the rural Indian electricity market can be segmented into three clusters based on households' willingness to pay, satisfaction with lighting, and appliance wattage. The clusters consist of potential customers, low-demand customers, and high-use customers. We then determine the predictors of membership in these clusters. We show that different types of consumers can be identified along easily observable measures. Moreover, we show that there are clear groups of consumers that vary along their satisfaction, willingness to pay, and existing appliance usage. •Employing an original and large dataset of more than 10,000 households in more than 200 villages in India, we develop a method for determining drivers of demand without electricity meter data.•We segment Indian consumers by their willingness to pay for electricity service, their level of usage, and their satisfaction with lighting.•We use cluster membership as a dependent variable in order to determine which household-level factors predict electricity usage.•We show that different types of consumers can be identified along easily observable measures.