Adoption and Impacts of Renewable Energy: Evidence from a Randomized Controlled Trial in Rural Kenya

Abstract

Human-driven climate disruption and widespread energy poverty are among the major challenges of our time (Alstone, 2015; SEAll, 2017). An estimated 1.1 billion people remain without access to modern energy, most of whom rely on biomass and fossil fuels for lighting, cooking, and heating — energy sources that lead to indoor air pollution and global warming (SEAll, 2017; WHO, 2016). Policy makers, entrepreneurs, and researchers across the globe place high hopes on renewable energy, particularly off-grid solar, to provide cheap and clean energy to those without access to the electric grid. The hope is that off-grid solar can reduce harmful and warming emissions from kerosene combustion and, at the same time, improve access to modern energy for unelectrified households. Despite this excitement, there is still little empirical evidence that there is demand for this technology solution, that it brings about the hoped for environmental and health effects, and that it confers private returns. It seems particularly relevant to evaluate the impact of this technology in a real-world setting since, previously, both private and environmental gains from novel technologies, such as cookstoves, have been overestimated (Hanna, Duflo & Greenstone, 2016). This thesis investigates take-up, use, and impacts of off-grid solar lighting in rural Kenya. To this end we conducted a randomized control trial with over 1,400 households. We begin by providing an analysis of the demand for solar lights, their environmental and health effects as well as the private returns to households (Chapter 2, joint work with I. Günther). We find that access to a solar light leads to a reduction in kerosene consumption of over 1.4 liters per month, curbing emissions at a cost of less than USD 6 per ton of CO_{2} equivalent. This cost is low compared with the frequently cited Social Cost of Carbon (SCC) of USD 50 per ton of CO_{2}equivalent (Revesz et al., 2017; IWG, 2015). Children's symptoms related to dry eye disease and respiratory illnesses reduce by about a fourth standard deviation and a third standard deviation, respectively. In addition, households save around 2-3% of their monthly cash expenditure. However, we do not find any effect on children's test scores. Finally, we find that reducing transaction costs increases demand 19% to 44% at a market price of USD 9, however, large price subsidies (over 55%) are needed to increase adoption rates to 70%. Price also does not seem to affect use. We conclude this chapter by suggesting that environmental and health effects combined with the high price sensitivity of demand and the fact that subsidies do not decrease use might justify subsidies in some settings. The impact of solar lights on rural households and their environment depends heavily on whether households actually use them over time, which, as can be seen from the example of clean cookstoves, is not always the case (Hanna, Duflo & Greenstone, 2016). Measuring technology usage can be challenging, especially if respondents believe that it is socially desirable to use a device (social desirability bias). In this case, they might overreport use, which would lead to biased results. This bias has been found in several studies of technology adoption in developing countries (Wilson et al., 2016; Thomas et al., 2013). To address social desirability bias and other measurement issues, we used sensors to measure the use of solar lights. The sensors were developed for this study by an engineering team. Chapter 3 (joint work with I. Günther and Y. Borofsky) focuses on information provided by sensor data. Specifically, we deployed sensors to gather an objective measure of solar light use. We then compared this data with survey data in order to analyze the extent to which survey data is limited by systematic and/or random error and discuss what type of questions provided more accurate answers. We learn from sensor data that households used solar lights almost every day, for four hours per day on average, mostly in the evening and the morning hours. Furthermore, we find that, on average, self-reported estimates of solar light use are very similar to sensor measurements, however, the correlation of estimates at the individual household level are weak, suggesting that random errors are large. Our findings indicate that households that used the solar lights infrequently were more likely to overreport, whereas those who used them a lot were more likely to underreport use. We also find that asking about general usage provided more accurate information than asking about disaggregated use for each hour of the day. Finally, and as the Hawthorne effect would predict, frequent visits from surveyors to a random subsample increased solar light use initially, but it had no long-term effects. Due to the novelty of both affordable solar lighting and the sensors used, this study is the first to both use sensors to study solar light use and compare sensor data with survey data at a large scale. One of the key findings of Chapter 2 is that emissions reductions might justify subsidizing solar lights in some contexts, however, temporary subsidies can have complex and contradictory effects on take-up and use. Chapter 4 discusses the direct and indirect implications of subsidies on take-up and use in more detail. We begin by analyzing how subsidies affect use. They could lead to lower use, since paying a lower price might lead adopters to value the product less (sunk cost effect), or they might lead to poor targeting since households that do not actually need the subsidy might use it to purchase the product (selection effect). Social interaction effects could also affect adoption, as people might learn from early adopters or imitate them. We find that subsidies sharply increase demand for solar lights without compromising use, thus we do not find evidence for sunk cost or selection effects. Further, our results suggest that social interaction effects might increase the price sensitivity of demand, whereby demand decreases among households that received an offer to purchase at a high price, but tends to increase among households that received the low price. These findings have two implications. First, they suggest that social learning about the limited private returns are more likely than imitation, since in the latter case we would have expected to see increased adoption across the board. Second, they imply that social interaction effects are complements for subsidies, but that they do not increase adoption on their own. To summarize, we find that in our setting solar lights are used a lot and substantially reduce kerosene use. They reduce warming emissions at a low cost and provide cheaper and better quality light to rural households. Moreover, we observe that further price reductions are needed to increase adoption rates above 50% and that subsidies do not affect use. Taken together, these findings suggest that subsidizing off-grid solar can be justified in some settings.