Health Benefits of Replacing Kerosene Candles

Health Benefits of Replacing Kerosene Candles with Solar Lamps:

Evidence from Uganda

Chishio Furukawa†

First Draft: July 2012 This Draft: August 2017

Abstract A randomized controlled trial in rural Uganda shows that there can be improvement in air-quality-related health such as headaches, chest pain, fever, and eye irritation if non-electrified households switch from kerosene to solar lamps. This five-month study worked with a sample of 155 schoolchildren. Those who received solar lamps reported having better overall air-quality-related health (0.25 standard deviation of baseline distribution, 6% lower probability of any symptoms), although there was no statistically significant change in self-reported cough symptoms or in lung capacity, as measured by spirometer tests. The health improvements were concentrated in school exam periods, most likely because students who switched to studying under solar lamps were exposed to less indoor air pollution. While health benefits exist, a follow-up survey shows that poor maintenance and low adoption remain major challenges for scaling up this new technology. JEL Code: Q42, Q53, Q55 The key words for the paper include indoor air pollution, respiratory health, solar lighting, and technology adoption. †Contact: [email protected], 617-767-1209 Department of Economics, Massachusetts Institute of Technology, E52, 50 Memorial Drive, Cam- bridge, MA 02142 *I thank Andrew Foster for his valuable advice in economic analysis, Sriniketh Nagavarapu and David Weil for their initial encouragement, Esther Duflo, Kaivan Munshi, Abu Shonchoy, Daniel Prinz, Benjamin Marx, Alonso Bucarey, Roman Zarate, and anonymous referees for their feedback, and Bruce Kirenga at Mulago Hospital for medical advice. I am indebted to Ruth Nanteza, Abdulrazaq Nassir, John Ssebayigga, Miho Shinke, and Abdulwahid Ngobya, for their assistance in the fieldwork in Kyannamukaaka, and to the office of Barefoot Power Uganda Ltd., Dirk Kam, Benard Kalyango, Frank Yiga, Joyce Demucci, Annet Nalumansi, and Francis Ejuku, who provided guidance and supplied the solar lamps. Without their support and co-operation from students, parents, and teachers of the Kyannamukaaka community, this research could not have been completed. All errors are mine.

1 1 Introduction

Energy poverty, defined as a lack of access to clean and reliable sources of energy, affects billions of people who rely on fuels known to be inefficient and detrimental to air-quality- related health (AQRH). A report from the World Economic Forum states that, “One in five people lacks access to electricity. Energy poverty is not just about the lack of energy, it cuts across – and undermines – all aspects of development” [1]. As of 2013, there are 1.3 billion people who live without access to electricity in developing countries in Africa and Asia and, despite progress in adding communities to the grid, the total number living in energy poverty is not expected to decline by 2030 due to population growth [2].

Lighting sources are an area of household energy that have received relatively little atten- tion, despite their importance for living condition in developing countries. More than 80% of non-electrified households use kerosene-based lighting, usually simple candles in open metal containers that provide light by burning an uncovered wick dipped in kerosene. Although they are the cheapest lighting option in the short run, they provide only dim light (Appendix

Figure 1) and emit a level of indoor air pollution (IAP) that is significantly higher than the tolerable limit prescribed by the World Health Organization (Appendix Figure 3) [3].

Recent technological advancement in LED light bulbs and solar photovoltaic panels have made small, rechargeable lamps affordable even for rural households in developing countries.1

Unlike kerosene candles, solar lamps provide bright and clean light with almost zero marginal costs (Appendix Figure 2). If adopted successfully, they have the potential to improve productivity and health, and thereby contribute to alleviating the pervasive and persistent

1The cheapest, reliable solar lamp available in the market costs 10 USD. This is equivalent to six months of kerosene expenditures for households that use kerosene candles.

2 energy poverty in areas electrical grids have not yet reached.

This paper examines the new technology’s potential to improve air-quality-related health, which has been shown to be damaged by inhalation of kerosene smoke. In a five-month

field experiment in rural Uganda, I randomly distributed affordable and portable lamps to approximately half of the 155 upper primary school children. I find that students who received the lamps show improvements in self-reported overall air-quality-related health: 0.25 standard deviation of baseline distribution. This effect was especially concentrated during the students’ exam periods. This is likely because the children who did not receive solar lamps were more exposed to IAP as they had longer and more intensive exposure through increased study by kerosene candlelight at night. At the same time, the treatment effect on spirometer measurements, which gauge lung capacity, is very close to zero. In addition, improvement in self-reported cough is modest and not statistically significant except in the

first week after the intervention.

Additional variations demonstrate that this result is not driven by selective attrition, reporting bias, or income channel. First, the attrition rate is only 1% different between the two groups and thus, the trimming-based bound on treatment effect is tight. Second, the health benefits are concentrated in school’s exam periods, suggesting that reporting bias is unlikely to be important. Third, the health outcomes of diarrhea and malaria show no improvement, suggesting that both the income effect through saved kerosene expenditure and reporting bias are unlikely to be contributing factors.

Despite potential benefits in terms of health and quality of life the solar lamps deliver,

I find that their long-term adoption remains challenging. In this experiment, the research team conducted regular maintenance checks on the lamps. However, six months after they

3 left the study group, over 65% had returned to kerosene candles. The solar lamps were apt to malfunction due to the users’ inappropriate usage and maintenance. Furthermore, another paper reporting on academic outcomes [4] shows that the solar lamps may have hindered children’s study due to shortages in solar power, causing dim or flickering light. All in all, while solar lamps seem to have positive impact on some health outcomes, the technology in its present and most affordable form is not ready for wide and sustainable scale-up. These

findings of improvement in symptoms such as eye irritation, positive effect on study hours, negative effect on test scores, high willingness-to-pay for more lamps, and poor maintenance are consistent with a recent study in rural Bangladesh [5].

The paper proceeds as follows: Section 2 discusses the background and reviews previous literature; Section 3 explains the details of randomized evaluation; Section 4 presents the results; Section 5 discusses the challenges of technology adoption; and Section 6 concludes.

2 Kerosene Candles and Solar Lamps in Uganda

2.1 Lighting Sources in Rural Households

Kerosene is the most commonly used lighting fuel in non-electrified areas of Uganda. Gen- erally, lanterns are used by households that are economically better off whereas candles are used by poorer households. Although the percentage of households using kerosene has de- creased over the past decade, it still remains above 60% (Appendix Figure 6). As the map of lighting source distribution (Appendix Figure 7) shows, access to electricity is clustered around the capital, Kampala, and extends only to the major cities.

4 In contrast, solar lamps are a new lighting device that has recently been expanding

in Uganda. The World Bank and International Finance Cooperation (IFC) initiated the

Lighting Africa Program, which promotes modern lighting technology for rural electrification

in Africa. It reports that the market expansion has been dramatic with sales growth of over

90% per year and sold more than 4 million units by 2013 [6]. Solar lamps are part of a larger

attempt to bring modern electrical infrastructure to developing areas. For instance, solar

panels with cell phone charging capacity have also been a popular device for the villagers.

Given the surging concern about energy issues and innovation in renewable energy, the

United Nations organized an International Year of Sustainable Energy for All in 2012, which

also aimed to promote the application of solar energy in developing countries [7]. It also

organized the International Year of Light in 2015, with the goal of promoting sustainability and development in lighting technologies such as solar lanterns.

The expansion of solar lamps in Uganda has been relatively successful, thanks to its geography as well as policy support. Due to its location at the equator, Uganda experiences perpendicular sunrays as well as short rainy periods. Uganda’s tariff policy, which does not tax imported solar panels, also attracts solar businesses [8]. Although the Solar Lamp

Field Test implemented by the German Society for International Cooperation reports that there are some challenges stemming from rainy seasons and the dissemination of knowledge regarding how to fully recharge, use of solar lamps has been expanding steadily throughout the past decade (Appendix Figure 6) [9]. The spread of solar lamps has been documented in multiple literatures on social enterprise [10, 11].

5 2.2 Indoor Air Pollution and Air-Quality-Related Health

The indoor air pollution (IAP) emitted from kerosene candles poses a significant threat to

AQRH. Apple et al. report that simple wick kerosene candles emit PM2.52 in concentration per time, an order of magnitude higher than the WHO health guideline [3]. Fan et al. review literature and report that both kerosene candles and lanterns are significant sources of IAP

[12]. This level is considerably high for long-term indoor usage, especially for the children whose lungs are in a developmental phase [13]. In 2009, Pokhrel et al. provided the first evidence that use of kerosene candles, along with kerosene stoves, is correlated with the severity of pulmonary tuberculosis in Nepal [14].

In the past decade, a rapidly expanding discussion has focused on the impact of outdoor air pollution [15] and biofuel-based cooking on respiratory health outcomes. According to

Smith and Mehta, exposure to IAP from cooking and heating sources is responsible for 1.2 million premature deaths every year and account for 4-5% of the global burden of disease

[16]. Duflo et al. also report evidence of respiratory health harm from traditional cookstoves in India [17]. Pitt et al. reports on the significance of intra-household bargaining and endogeniety between health endowment and cooking habits, pointing to the difficulty in identifying the health impact of IAP exposure [18].

Compared to outdoor air pollution and cookstoves, however, the IAP from kerosene candles has received little attention. Grimm et al. uses a randomized experiment of solar lamp kits in rural Rwanda and reports the productivity, convenience, and budgetary benefits, but not the health benefits [19]. It is important to note that, when cooking, people tend not to stay inside the kitchen hut for an extended period of time. In contrast, all household

2PM2.5 is a fine particulate matter with diameter of 2.5 micrometers or less.

6 members are constantly exposed to the IAP from the lighting sources. Therefore, even if the particulate matter (PM) concentration may be significantly higher for the cooking sources, lighting sources may also have significant AQRH consequences.3

3 Data

This paper uses data from a randomized evaluation conducted by Barefoot Power Uganda

Ltd.4 between July and December 2011 in Kyannamukaaka Sub-county, 5 Masaka District,

Uganda. This randomized evaluation aimed to assess the extent to which replacing kerosene candles by solar lamps can alleviate the severity of the AQRH symptoms of students who study by kerosene candles.6 Since this was the first randomized evaluation on the IAP of kerosene candles, it aimed to achieve the maximum benefit in an ideal situation rather than the average benefit in a realistic situation. If high end of benefits were positive, then follow- up studies could identify ways to realize such positive benefits through encouraging take-ups and maintenance in a real world scenario.7 3Furthermore, the dose-response relation for PM exposure is reported to be exponential [20]. Although the PM concentration may be lower than cookstoves, additional PM from lighting sources exposure may seriously undermine the AQRH of those who have already been exposed to IAP from cooking sources. 4Barefoot Power Uganda Ltd. is a Ugandan social enterprise that distributes micro-solar lighting systems in collaboration with local businesses, non-governmental organizations, and micro-finance institutions. Its main products are Firefly series, pico lamps, and PowaPack series, larger solar lamp systems with rechargeable batteries and multiple lights. Both Firefly and Powapack received Lighting Africa 2010 Outstanding Product Awards. http://barefootpoweruganda.com/ and http://www.barefootpower.com/. 5The research team collaborated with seven primary schools in the region. 6As its second objective, it also aimed to evaluate the extent to which introducing solar lamps can improve students’ school performance. The school performance outcomes are reported in a separate paper “Do Solar Lamps Help Children Study? Contrary Evidence from Uganda” [4]. In this study, solar lamps did not improve the children’s test scores, but in fact, decreased by 5 points. This may be due to the common problem of flickering light due to lack of full battery recharge. 7This study received an approval from the Ministry of Education and Sports in Uganda and from the Kyannamukaaka Sub-county Office in Kyannamukaaka, Masaka District, Uganda. This trial was conducted using funding from the 2011-12 Brown International Scholars Program and the Barbara Anton Internship Grant from the Pembroke Center for Teaching and Research on Women at Brown University, and was conducted financially independently from the enterprise. It received the expedietd review from the Brown

7 3.1 Participants

The study population is comprised of 155 upper primary school students (grade 5 to grade

7) in rural Uganda who use kerosene candles and report some AQRH symptoms, such as (i) coughing, (ii) chest pain, or (iii) difficulty breathing. It targeted children between the age

10-15 because of (i) high vulnerability to IAP, (ii) short history of previous exposure, and

(iii) ease of administration from school. As this was the first study on the health benefit of solar lamps, we could not conduct a power calculation to approximate the necessary sample size; 155 was the largest number of students that the study could financially and logistically work with.

There were five eligibility criteria: first, the participants must use kerosene candles as their main source of lighting; second, the participants must report some AQRH symptoms in the baseline survey; third, the participants’ parents had to be present at the public lottery so that the basic household characteristics could be checked; fourth, the participant must not have older siblings who are also in the study group8; fifth, the participant must be day students who commute to schools, not boarding students. The last two criteria were used to avoid spillover effect. This reduced 563 students at the participating schools to 155 students, and all of them participated in the study at the baseline. An informed consent was signed both by the student and their guardians.910

University Institutional Review Board. 8This study included only one student from each household. 9There was a clerical mistake during the public lottery. Although 180 students were eligible to participate, the researchers mistakenly excluded the students whose parents did not come on the day of parents’ survey. 10Appendix Figure 5 shows the details of the recruitment, exclusion, intervention, and dropout throughout the study.

8 3.2 Interventions

Through a public lottery in July 201111, the research team randomly distributed solar desk lamps of the Firefly 512 type, produced by Barefoot Power, to approximately half of the students (73). Due to ethical restrictions and common belief that solar lamps are beneficial, this study employed delayed treatment methodology and gave the same solar lamps to the students who were in the control group upon the conclusion of the study in December 2011.

In order to avoid the reselling of solar lamps, the co-operating schools kept the panels for the

first two weeks and the students were required to bring back the solar lamp at the beginning of each school day for recharge. This ensured that the students and their family members were given a chance to appreciate the benefits of this new technology. After the first two weeks, the panels were given to the families. To ensure effective usage of solar lamps by the students, the research team continually exchanged the lamps that were faulty.13 There

were over 20 lamp replacements every week in the first month, and after introducing the 1W

panels, the number fell to less than 10 lamps in the subsequent months.

3.3 Assessments

The research team made follow-up observations of self-reported conditions of ten symptoms

and carried out spirometries14 every week in the first month (July-August 2011), and every

month for five months until December 2011. They also interviewed the parents at the begin-

11The research team conducted coin tosses where each student guessed heads or tails before the toss and s/he received a solar lamp only if s/he guessed correctly. This was done publicly in front of all participating students to ensure that the students felt the process was fair. 12Firefly 5 is the cheapest option at a price of $10 USD. 13Most of the issues involved recharging. The research team carried well-functioning lamps to each school during the observations, and replaced the lamp of any students reporting a problem. 14The research team used ndd EasyOne Frontline spirometer.

9 ning and after one month regarding living conditions and perceptions of kerosene candles.

3.4 Limitations: Threats to Internal Validity

There are five major threats to the internal validity in this research: validity of random-

ization, lack of full recharge, sensitization, differential dropouts, and income effects. These

altogether may have underestimated the benefits in the ideal situation of solar lamp provi-

sion.

1. Validity of Randomization: Table 1 presents a comparison of household and in-

dividual characteristics between treatment and control groups. It verifies the balance in the

observable characteristics between the kerosene candle group and solar lamp group, and con-

firms that the randomization was appropriately implemented (p-value=0.320). While many

wealth indicators show statistically insignificant differences, given the small sample size of

155 students, there are following two biases in the treatment assignment: first, the treatment

group is wealthier and its household head is more educated;15 second, the treatment group is less cautious about its health.16 These two biases seem contradictory since wealth and general preferences for health are expected to be positively correlated. Therefore, these may only be unsystematic and idiosyncratic variations.

2. Lack of Full Recharge: The solar lamps can last for four hours even if they were fully charged. Therefore, without fully recharging the lamp, it is impossible to continue its usage with bright and stable light throughout the night. It was a rainy season in Uganda between

15Probability of sleeping with kerosene candles in the past seven days, household head’s years of educa- tion, probability of wearing uniform, and number of children under 18 living together are weakly positively correlated with the treatment. 16Time students spend cooking, traditional three-stone cookstove usage (using the stones of same height is the most common and affordable cooking method), and bednet usage are weakly negatively correlated with the treatment.

10 September and November,17 thus the lack of full sunshine hampered the recharging process

for some. Furthermore, at the beginning of the study, it was common that the household

members used the panels without orienting them perpendicularly to the sun (Appendix

Figure 4). When the solar lamps could not be charged fully, the students occasionally had

to go back to using kerosene candles. Over half of the students reported that they used

kerosene candles occasionally even after receiving the solar lamps. This limitation leads

to an underestimation of the full benefits that solar lamps could have. It appears that the

households used the solar lamps as long as possible before resorting to using kerosene candles.

3. Sensitization among kerosene candle users: To receive cooperation from the

schools to collect students’ health data, the research team explained that the purpose of

the study was to measure the health benefits of solar lamps. In some schools, the teachers

reacted strongly by explaining to the students that the kerosene candles can harm health.

This led to some change of the control group’s attitudes and behaviors, including their usage

of other sources of lighting, such as electric torches, which are readily available at local shops.

24 students purchased torches after the study started. 32 students answered positively to

the question “Did you change the way you use kerosene candles after you started your

study?” This is a significant number of students that switched. This limitation leads to an

underestimation of the benefits from solar lamps when the control group remains uninformed

and contrinues to use kerosene lanterns in exactly the same way.

4. Attrition: Absenteeism led to missing spirometer measurements and self-reported health records. In rural Uganda, students miss schools due to sickness such as malaria or due

17Note that the solar panels are water resistant. Thus, it is still possible to recharge during the rainy time although sunlight is not as strong as at other times.

11 to their domestic work like digging or helping parents on the market days. This may have

biased the estimates in an ambiguous direction. However, the attrition rates are close to each

other: 19% missing from treatment group, and 20% missing from control group conditional

on the baseline period being observed. The attrition check is insignificant, implying that

this threat was likely not important.

There were some special cases of non-compliance that deserve some discussion. The first

is the theft of a lamp: two student’s solar lamps and panels were stolen when they left them

outside their home to recharge during the day. The second is neighbor student crossover:

three students in the control group had close friends assigned to the treatment group and

they studied together more than once a week. This analysis still included these students

because it is Intent-to-treat analysis, but included dummies for each of them.

5. Income effect: The households in the treatment group will save the money that they used to spend on kerosene.18 This may improve the students’ health level by putting

money towards better food or living conditions. If the income effect were significant, the

study would be overestimating the solar lamps’ benefits that occur through the channel

of removing the IAP. However, this is likely not the case for two reasons. First, students

commonly said that their parents would purchase salt, sugar, or shoes if they were given this

incremental income. None of these would directly impact AQRH. Second, the result that

other symptoms, such as diarrhoea, showed no improvement also suggests that income effect

was generally unimportant.

18Households spend 200USH, equivalent to 8 cents per day, on average, for kerosene.

12 4 Analysis

4.1 Regression Models

This analysis estimates the intent-to-treat effect, instead of the treatment-on-treated effect, to avoid inflating the estimate due to missing values correlated with the treatment. Here, the controls Xi include the class, gender, and dummies for students in treatment group whose solar lamps were stolen, and for those in control group who joined their friends to study with solar lamps. Table 2 summarizes the main outcomes, Table 3 reports the outcomes for each symptom using the equation 1, and Table 4 presents results with alternative outcome specifications. Figures 1-4 present corresponding event study graphs.19 The estimation uses the week fixed effect to account for seasonal variations and robust standard errors clustered at the individual level. The equation (2) shows the interaction with non-exam periods dummies

20 (Zit = 1(t∈ /Exam period) ) so that the coefficient β1 shows the treatment effect during the exam periods.21

Healthit = α0 + α1TRTi + α2Healthi0 + α3Xi + W eekt × Schooli + εit (1)

19A more complete series of analysis is available in the online repository. 20The exam periods were specified to be the Weeks 3, 15, and 18. The Week 3 was final exam for all but one school that had its final exam in Week 2. The Week 15 was the final exam for the Primary 7 members who had their national exams to graduate from their primary school. The Week 18 was the final exam for the Primary 5 and Primary 6 members. There were three schools that conducted the mid-term examinations. However, these did not coincide with any of the observation weeks and students placed less emphasis on the mid-terms. Since the final exam in the 3rd semester was critical for advancing to the next grade, students placed high importance and studied hard in the weeks prior to the exam. Therefore, the Week 15 was also included in the examination period. 21Although one limitation in this analysis is that the exam periods are also the periods with high numbers of missing values, it is highly likely that attrition bias is not the major driver of the estimated treatment effect. This is because a high number of missing values are caused not by the absence of students (in fact, during the exam periods, there were fewer absenses), but by the difficulty in obtaining cooperation from school administrations due to the scheduled exams.

13 Healthit = β0+β1TRTi+β2TRTi×Zit+β3Zit+β4Healthi0+β5Xi+W eekt×Schooli+εit (2)

The self-reported health survey recorded the severity of 10 symptoms: cough, difficulty breathing, chest pain, sore throat, eye irritation, fever, headache, diarrhoea, running nose, and malaria in the seven days prior to the interview. It classified the severity of symptoms into 4 categories: 1 for no symptoms, 2 for minor, 3 for moderate, and 4 for severe symp-

22 toms. The regressions in Table 2 take the outcome variable Healthit to be a binary variable that takes 1 if no symptom and 0 if any symptom regardless of severity. This approach has an advantage of not imposing a cardinal interval scale to the ordinal categorical measure and thereby allowing for intuitive interpretation of the magnitude. Table 4 shows the result for scores as well as binary outcomes for different levels of severity.

4.2 Outcomes

Table 2 shows the analysis of aggregate outcomes. AQRH includes cough, difficulty breath- ing, chest pain, sore throat, eye irritation, fever, headache, and non-AQRH includes diarrhoea and malaria.23 The coefficients in columns (3) and (4) show that the solar lamp decreased the fraction of students who have any air-quality-related symptoms by 6% among all stu- dents over all periods and by 9% during the exam periods. Since the standard deviation

22Note that, in case children could not accurately answer the question of severity, the interviewers asked how many days they suffered from respective symptoms in the past seven days. Approximately, 1 correspnods to 0 days, 2 corresponds to 1-2 days, 3 corresponds to 3-4 days, and 4 corresponds to 5-7 days. 23Since running nose is closely related to AQRH but not directly influenced by AQRH, this analysis left it out of either category.

14 of the AQRH outcome is 0.24, these results indicate moderate magnitude of improvement equivalent to 25% and 37% of standard deviation respectively.

Columns (7)-(8) show that the solar lamps had no impact on the spirometer measure- ments.24 As spirometers are used to diagnose severe lung illnesses such as asthma, they are not sensitive enough to detect change in health symptoms due to kerosene exposure in the short-period. Note that the respiratory health literature generally shows that spirometric outcomes are not responsive. For example, the RESPIRE study that found a significant de- crease in physician-diagnosed pneumonia also did not detect statistically and economically significant improvement in spirometric measurement.

Table 3 shows the results for each symptom. It shows that the improvement existed in most air-quality-related symptoms, and particularly large in fever as well as eye irritation, headache, and chest pain in the exam periods. There is small and statistically insigificant improvement for cough and difficulty breathing.

Table 4 shows that these results hold with alternative specifications of the self-reported health. The columns (4) and (5) show that the solar lamps reduce the probability of having more than moderate AQRH symptoms by 5%, and that the probability of having severe

AQRH symptoms by 2%. These results show that results in Table 1 hold across various outcome specifications.

Taken together, there are two important variations that shows that estimates benefits reflect true underlying health benefits. First, there is a more significant difference during

24Spirometer requires significant co-operation from the participants since the participants must inhale and exhale to the fullest of their capacity. The students found it difficult at the beginning of the study, however, they eventually got used to the spirometer after a few weeks. The study restricts the outcome to be the measurement with test quality higher than B in the scale of A-E, which is a standard in medical literature. However, the results are essentially the same even when the test quality thresholds are changed.

15 exam periods than in other periods. This variation implies that reporting bias and placebo effect are unlikely to be significant because, if so, there should be significant difference even during the non-exam periods. Second, there is a significant difference in AQRH symptoms, but not in non-AQRH symptoms. This variation implies that it is unlikely that income effect is significant because, if so, there should be significant difference for non-AQRH such as diarrhoea and malaria.

5 Difficulties of Long-term Technology Adoption

Although the results of this experiment demonstrate the potential health benefits of using solar lamps, the follow-up survey six months after the completion of the study – after the research team left to repair the lamps – presents alarming results regarding long-term adop- tion. Among 60% of those who were available for the resurvey (the attrition was large due to change of schools and dropouts), only 33% reported that they kept using the solar lamps and over 65% returned to using kerosene candles. The reasons for not using solar lamps include (i) panels spoiled, (ii) switch broken, and (iii) malfunctioning of bulbs. This rate of malfunctioning was much higher than the company had predicted from the laboratory experiments that showed the lamps can function for two years if properly maintained. Fur- thermore, the average lighting time of 1.5 hours was much shorter than the advertised length of 4 hours. These figures may be biased due to the high rate of dropout and cannot be eas- ily generalized because the quality of lamps may differ among various types and designs;25

25It is important to note that the solar and lighting technologies are constantly improving: the Firefly 5s are alraedy phased out and be replaced by the improved type. The latest version of Firefly has Surface Mount Diode (SMD), an improved version that has have 25% more lumens than general LEDs, and battery of Lithium Iron Phosphate (LiFePO4) that have a life span of 2 to 3 years, which is 3 times longer than the previous Nickel-cadimium (NiCd) batteries.

16 nevertheless, they are far from assuring that the current technology can be spread widely and sustainably.

These results have commonalities with the growing literature reporting weaknesses of health and environmental technologies such as cookstoves in the field. Duflo et al [21]. report that improved cookstoves did not reduce IAP and that health conditions did not improve without frequent maintenance by research team. Since this study design is similar to the

RESPIRE study26 [22], where researchers ensured the maintenance of technology, it is highly likely that the improvement in health would be much less without maintenance services. In a way similar to how the weak negotiation power of women in intra-household decision-making may inhibit spread of improved cookstoves [23], children who would benefit the most from improved lighting also have little influence on a households’ decision to purchase solar lamps, leading to low adoption rate.

It is noteworthy that, at the same time, 90% of the respondents also said that their parents would be willing to pay to repair the lamps. Furthermore, it frequently happened that the villagers approached the research team to ask where they could purchase the lamps.

In the villages, there was no local solar lamp distributor. In this way, lack of access to market and maintenance services was a significant obstacle to allowing the adoption and function of solar lamps. The improper recharge practice of hanging the panels on the wall was commonly seen at the initial stage of the study, as shown in Appendix Figure 4 , but all the respondents in the resurvey at the completion of the study said that they put panels either on the ground or on the roof. This shows that the villagers can learn how to recharge

26RESPIRE study was a large-scale randomized controlled trial of improved cookstoves conducted in Guatemala with significant maintenance efforts by the researchers.

17 properly after continuous reminders throughout the study.

6 Conclusion

The results of this paper suggest that substituting kerosene candles with solar panel-charged

LED lamps can lead to health improvements among children who suffer from air-quality- related symptoms. One must, however, consider two caveats for generalizing this result to other places. First, smooth and sustainable adoption of the technology will require im- provements in the robustness of the solar lamps themselves. Second, lighting source choice depends not only on lamps’ quality but also on access to market and suitability of such environmental conditions as strength of sunlight and length of rainy seasons. It is important to remember that this was the first, explorative field experiment rather than a confirmative large-scale one; the study design with solar lamp replacement likely played a significant role in ensuring health benefits. That said, given these results, continued research should focus on valuation, learning, and the maintenance of solar technology in the scalable setting in developing countries.

7 Conflict of Interest Statement

This study was funded by the Brown International Scholars Program 2011-12; the Barbara

Anton Internship Grant from the Pembroke Center for Teaching and Research on Women; and the Center for Environmental Studies at Brown University. I have no conflict of interest with the the Barefoot Power Uganda Ltd., which supplied the solar lamps. Even though I

18 worked with the company, I made payment to purchase their lamps.

19 References

[1] World Economic Forum. Ending energy poverty. http://www.weforum.org/sessions/summary/ending- energy-poverty.

[2] International Energy Agency. World energy outlook 2013: Energy access database. 2013.

[3] J. Apple et al. Characterization of particulate matter size distributions and indoor concentra- tions from kerosene and diesel lamps. Indoor Air, 20.5:399–411, 2010.

[4] Chishio Furukawa. Do Solar Lamps Help Children Study? Contrary Evidence from Uganda. Journal of Development Studies, 2014.

[5] Abu S. Shonchoy Yuya Kudo and Kazushi Takahashi. Impact of solar lanterns in geographically challenged locations: Experimental evidence from Bangladesh. IDE Discussion Paper No.502, 2015.

[6] The Lighting Africa Program. Solar lighting for the base of the pyramid -Overview of an emerging market-. 2010.

[7] United Nations. A vision statement by Ban Ki-moon Secretary-General of the United Nations: Sustainable Energy for All. 2011.

[8] Ministry of Energy and Mineral Development. The energy policy for Uganda. 2002.

[9] Anne Bruderle and the German Society for International Cooperation (GIZ). Solar lamps field test Uganda Final Report. 2011.

[10] Anil Chitrakar and Babu Raj Shrestha. The Tuki: Lighting up Nepal. Innovations, 2010.

[11] AKF Savings Groups Learning Initiative. Marketing solar lamps: emerging lessons from Uganda. Aga Khan Development Network, 2010.

[12] Cheng-Wei Fan and Junfeng Zhang. Characterization of emissions from portable household combustion devices: particle size distributions, emission rates and factors, and potential ex- posures. Atmospheric Environment, 35:1281–1290, 2001.

[13] Gauderman W.J. et al. The effect of air pollution on lung development from 10 to 18 years of age. The New England Journal of Medicine, 357 (11):1057–1067, 2004.

[14] Amod Pokhrel et al. Tuberculosis and indoor biomass and kerosene use in Nepal: a case-control study. Environmental Health Perspective, 118(4):558–564, 2009.

[15] Andrew Foster and Naresh Kumar. Health effects of air quality regulations in Delhi, India. Atmospheric Environment, 45 (9):1675–1683, 2011.

[16] Kirk Smithand Sumi Mehta. The burden of disease from indoorairpollution in developing countries: comparison of estimates. International Journal of Hygiene and Environmental Health, 2003.

[17] Esther Duflo Michael Greenstone and Rema Hanna. Cooking stoves, indoor air pollution and respiratory health in rural Orissa. Economic and Policy Weekly, 2008.

20 [18] Mark Rosenzweig Mark Pitt and Nazmul Hassan. Sharing the burden of eisease: gender, the household division of labor and the health effects of indoor air pollution in Bangladesh and India. 2006.

[19] Michael Grimm, Anicet Munyehirwe, Jrg Peters and Maximiliane Sievert. A first step up the energy ladder? Low cost solar kits and household welfare in rural rwanda. IZA Discussion Paper, 2014.

[20] Kenneth Chay and Michael Greenstone. The impact of air pollution on infant mortality: Evidence from geographic variation in pollution shocks induced by a recession. Quarterly Journal of Economics, 118:1121–1167, 2003.

[21] Rema Hanna Esther Duflo and Michael Greenstone. Up in smoke: The influence of household behavior on the long-run impact of improved cooking stoves. NBER Working Paper, 2012.

[22] Dan Pope Rolv T. Lie Anaite Diaz John P. McCracken Per Bakke Byron Arana Kirk R. Smith Smith-Sivertsen Tone, Esperanza Diaz and Nigel Bruce. Effect of reducing indoor air pollution on women’s respiratory symptoms and lung function: Respire Guatemala randomized trial. American Journal of Epidemiology, 170(2), 2009.

[23] Grant Miller and Mushfiq Mobarak. Intra-household externalities and low demand fora new technology: Experimental evidence on improved cookstoves. Working Paper, 2011.

21 Figures and Tables

Figure 1 Overall health treatment effect over time

Figure 2 Air-quality-related-health treatment effect over time Figure 3 Air-quality-related-health treatment effect over time 2

Figure 4 Other health treatment effect over time

Figures 1-4 plot the treatment effect estimates at each observation period with controls and school fixed effects. The bars represent 95% confidence interval when using the robust standard error specification. The shaded areas are the schools’ exam periods. Table 1: Balance Table Treatment Control Difference Mean Observations Mean Observations Mean (1) (2) (3) (4) (5) Household characteristics: Household head is father .5 74 .5072 69 -0.007 (.5034) (.5036) (0.084) Household head is farmer .8108 74 .8551 69 -0.044 (.3943) (.3546) (0.063) Household head’s years of education 6.8571 70 6.0484 62 0.809 (3.0988) (3.0911) (0.540) Main dwelling is permanent .5 74 .5072 69 -0.007 (.5034) (.5036) (0.084) Num. of kerosene candles at home 1.9444 72 1.9032 62 0.041 (.6897) (.6945) (0.120) Num. children below 18 living together 5.2394 71 4.3676 68 0.872** (2.728) (2.4671) (0.441) Num. older children living together 2.4058 69 2.2698 63 0.136 (2.6696) (2.1191) (0.418) Cookstove if 3-stone stove .527 74 .6377 69 -0.111 (.5027) (.4842) (0.083) Num. rooms in main dwelling 3.9054 74 4.0294 68 -0.124 (1.2405) (1.486) (0.231) Each child owns blanket at home .5811 74 .6029 68 -0.022 (.4967) (.4929) (0.083) Each member has at least two clothes .5405 74 .6029 68 -0.062 (.5018) (.4929) (0.084) Floor made of cement .2838 74 .2754 69 0.008 (.4539) (.45) (0.076) Wall made of bricks .6757 74 .6957 69 -0.020 (.4713) (.4635) (0.078) Water comes from unclean sources .5921 76 .5443 79 0.048 (.4947) (.5012) (0.080) Individual characteristics: Gender (male) 0.3947 76 .4937 79 -0.099 (0.492) (.5032) (0.080) Grade (either P5, P6, or P7) 5.7237 76 5.8101 79 -0.086 (.7411) (.7524) (0.120) Age 12.942 69 13.1452 62 -0.203 (1.5707) (1.7727) (0.294) Wear uniform .7027 74 .5588 68 0.144* (.4602) (.5002) (0.081) Wear shoes .2568 74 .2941 68 -0.037 (.4398) (.459) (0.076) Have registered lunch at school .6757 74 .6029 68 0.073 (.4713) (.4929) (0.081) Use bed net last night 0.257 74 0.25 68 0.007 (0.525) (0.436) (0.081) Sleeping with kerosene candles on .1831 71 .0984 61 0.085 (.3895) (.3003) (0.060) Did homework in the past 7 days .662 71 .6613 62 0.001 (.4764) (.4771) (0.083) Time students cooked 1.9571 70 2.803 66 -0.846** (1.7729) (2.4001) (0.364) Experienced illness or injury in 30 days .7397 73 .8182 66 -0.078 (.4418) (.3887) (0.070) Num. days suffered from illness or injury 3.5139 72 4.8462 65 -1.332 (3.3651) (5.8769) (0.830) * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors in parenthesis. Test for joint significance: χ2(26) = 28.81 and p-value= 0.320 Table 2: Treatment effect on overall health outcomes (RCT) (1) (2) (3) (4) (5) (6) (7) (8) Overall Overall AQRH AQRH NAQRH NAQRH FEV1% FEV1%

Treatment -0.057∗∗∗ -0.080∗∗∗ -0.065∗∗∗ -0.092∗∗∗ -0.007 -0.013 -0.004 -0.001 (0.019) (0.024) (0.020) (0.026) (0.021) (0.033) (0.013) (0.013) No exam 0.047 0.062 -0.194 -0.009 (0.051) (0.063) (0.130) (0.022) Trt*No exam 0.035 0.041 0.009 -0.004 (0.025) (0.027) (0.034) (0.010)

Constant 0.035 -0.017 0.102 0.035 -0.211∗∗ -0.018 0.638∗∗∗ 0.647∗∗∗ (0.075) (0.076) (0.088) (0.089) (0.085) (0.146) (0.083) (0.084) Obs 916 916 916 916 915 915 902 902 R2 0.320 0.322 0.333 0.335 0.337 0.337 0.127 0.127 * p<0.1, ** p<0.05, *** p<0.01. This table presents the treatment effects from the RCT and their interaction with non-exam periods. The columns (1)-(2) outcome “overall’ is binary and equal 1 if any symptom exists at any level (mild, moderate, serious), and 0 if no symptoms. The columns (3)-(4) outcome “AQRH” stands for air-quality-related-health, and includes cough, difficulty breathing, chest pain, sore throat, fever, and headache. The column (5)-(6) outcome “NAQRH” stands for non-air-quality-related-health, such as malaria and diarrhoea. The columns (7)-(8) outcome FEV1% takes the value 0-1, and measures the lung health using the spirometer with the higher values the better. The measurement is replaced by a missing value if the test quality was lower than B in the range of A to E. The variable “Treatment” and “Trt” represent treatment dummy; “No exam” represents the non-exam period dummy. Interactions between week and school dummies are included as fixed effects. Standard errors are clustered at individual level. Table 3: Treatment effect on each symptom (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Cough Diff breath Chest pain Sore throat Eye irritation Fever Headache Diarrhoea Malaria Running nose

Treatment -0.062 -0.065 -0.109∗∗ -0.064 -0.110∗∗ -0.135∗∗∗ -0.116∗∗ 0.000 -0.027 -0.013 (0.058) (0.047) (0.052) (0.054) (0.050) (0.049) (0.058) (0.032) (0.052) (0.063)

No exam 0.079 0.054 0.257∗∗ -0.265∗ -0.223 0.387∗∗∗ 0.636∗∗∗ -0.118 -0.269∗ 0.123 (0.202) (0.054) (0.127) (0.154) (0.179) (0.132) (0.129) (0.124) (0.161) (0.211)

Trt*No exam -0.008 0.010 0.079 0.058 0.115∗∗ 0.082 0.002 -0.018 0.036 -0.066 (0.060) (0.054) (0.061) (0.058) (0.051) (0.053) (0.061) (0.037) (0.052) (0.064) Obs 915 916 915 915 914 915 915 915 915 914 R2 0.181 0.281 0.164 0.121 0.111 0.279 0.335 0.252 0.260 0.121 * p<0.1, ** p<0.05, *** p<0.01. This table presents the treatment effects from the RCT and their interaction with non-exam periods. The outcome variables are binary and equal 1 if the symptom exists at any level (mild, moderate, serious), and 0 if no symptoms. “Diff breath” stands for difficulty breathing. The variable “Treatment” represents treatment dummy; “No exam” represents the non-exam period dummy; “Trt*No exam” their interactions. Interactions between week and school dummies are included as fixed effects. Standard errors are clustered at individual level. The Columns (1)-(7) are air-quality-related symptoms. The Columns (8)-(9) are non-air-quality-related symptoms. The Column (10) is left unclassified because it is only indirectly related to air-quality-related symptoms. Table 4 Treatment effect on overall health outcomes with alternative outcome measures (1) (2) (3) (4) (5) (6) (7) (8) (9) Overall Overall Overall AQRH AQRH AQRH NAQRH NAQRH NAQRH

Treatment -0.044∗∗∗ -0.015∗∗ -0.117∗∗∗ -0.051∗∗∗ -0.020∗∗ -0.117∗∗∗ -0.006 0.006 -0.007 (0.016) (0.007) (0.037) (0.018) (0.008) (0.040) (0.017) (0.006) (0.040) Obs 916 916 916 916 916 916 915 915 915 R2 0.295 0.227 0.323 0.320 0.236 0.360 0.263 0.115 0.309 Moderate X X X Severe X X X Total score X X X

* p<0.1, ** p<0.05, *** p<0.01. School-level fixed effects. Standard errors clustered at individual level. This table presents the treatment effect with alternative metric of outcomes. Columns (1), (4), (7) have the outcome of dummy whether the symptom is greater or equal to moderate. Columns (2), (5), and (8) have the outcome of dummy whether the symptom is severe. Columns (3), (6), and (9) present the pooled score. Appendix Figures

Appendix Figure 1 A student studying under a kerosene candle

Appendix Figure 2 A student studying under a solar lamp

Appendix Figure 1-3 are images of the student in the treatment group from the fieldwork, taken by the author with consent of the student. No adjustment of brightness is made for Appendix Figure 1 and 2. Appendix Figure 3 A student studying under a kerosene candle on the bed

Appendix Figure 3 shows how the student usually studies in the evening. The black soot on the wall confirms that she commonly studies with the candle on the edge of the bed.

Appendix Figure 4 Household’s inappropriate recharge practice

Appendix Figure 4 was taken during household visits in the village. To function properly, the solar panel needs to be put perpendicular to the sun; instead, it is hung on the wall whre it will not recieve maximum charge. Appendix Figure 5 Trend in lighting source over time Appendix Figure 5 shows the details of the recruitment, exclusion, intervention, and dropout throughout the study. Appendix Figure 6 Trend in lighting over time

Appendix Figure 6 uses the series of Uganda National Household Surveys to demonstrate the change in lighting sources over time. Appendix Figure 7 Map of lighting source