The World Bank Economic Review, 37(2), 2023, 283–304 https://doi.org10.1093/wber/lhac031 Article Time for Clean Energy? Cleaner Fuels and Women’s Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 Time in Home Production Farzana Afridi, Sisir Debnath, Taryn Dinkelman , and Komal Sareen Abstract In much of the developing world, cooking accounts for the largest share of women’s time in home production. Does relying on solid fuels drive this time burden? This study revisits a clean energy information experiment in rural India to assess the time savings’ potential of cleaner cooking technologies. Treatment villages were randomly assigned to receive information about negative health effects of cooking with solid fuels and about public subsidies for cleaner liquid petroleum gas (LPG). Time-use data indicate that primary cooks spend almost 24 hours cooking each week. Cleaner fuel use is correlated with about 140 minutes less cooking time each week. Yet households only reduce their weekly cooking time by about 35 minutes in response to the randomized clean energy information nudge. Factors limiting the impact of clean energy nudges on the choice of home production technologies and time use are discussed and an avenue for future research is suggested. JEL classification: O13, J22 Keywords: time use, home production, energy use, India 1. Introduction Cooking is the most time-intensive aspect of home production in developing countries today (Dinkelman and Ngai 2022). This essential activity still relies on solid fuels such as wood, charcoal, and animal dung. Much advocacy exists around freeing women in low-income countries from the drudgery of home pro- duction through the use of cleaner stoves and cleaner fuels.1 Yet there is limited direct evidence on how much time women (and households in general) spend using solid versus clean fuels, and how much time Farzana Afridi is a professor at the Indian Statistical Institute (Delhi) and the National University of Singapore, Lead Aca- demic at the IGC’s India Program, and a research fellow at IZA; her email address is fafridi@isid.ac.in. Taryn Dinkelman (corresponding author) is a professor at the University of Notre Dame, a research associate at NBER, and a research fellow at IZA and CEPR; her email address is tdinkelm@nd.edu. Sisir Debnath is a professor at the Indian Institute of Technology (Delhi); his email is sisirdebnath@iitd.ac.in. Komal Sareen is a graduate student at the Indian Institute of Technology (Delhi); her email is ksareen18@gmail.com. Afridi acknowledges financial support from the Bill and Melinda Gates Foundation’s Initiative for What Works for Women and Girls in the Economy (IWWAGE-IFMR). This study is registered with the Ameri- can Economic Association’s registry for randomized controlled trials AEARCTR-0003774. Excellent research assistance was provided by Akriti Dureja and Abhishek Arora. A supplementary online appendix for this article can be found at The World Bank Economic Review website. Data and code to reproduce the results in this paper are available in the ICPSR data repos- itory at https://www.openicpsr.org/openicpsr/project/183841/version/V1/view. 1 For example, see the Clean Cooking Alliance https://www.cleancookingalliance.org/home/index.html or Clean Cooking https://cleancooking.is/. © The Author(s) 2023. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 284 Afridi et al. might be saved from a switch to cleaner energy (Krishnapriya et al. 2021). Quantifying potential and actual time savings from a move towards clean energy cooking is important for considering the welfare gains to women of clean energy transitions in the home (Cecelski 2000). Such time savings would augment any health gains from lower exposure to indoor air pollution facilitated by clean energy cooking.2 This paper explores whether nudging households towards cleaner fuel use for cooking could generate significant time savings in home production in a developing country context. It revisits the information Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 experiment in Afridi, Debnath, and Somanathan (2021) in which rural Indian households were nudged away from solid fuel use and towards cleaner bottled liquid petroleum gas (henceforth LPG) for cooking. The information nudge was designed to draw the household’s attention to the severe negative health consequences of relying on solid fuels for cooking. Households were randomized (at the village level) to either receive information about the health benefits of using LPG and health costs of using firewood and dung in their homes (Health treatment); to receive this information along with information about available public subsidies for LPG (Health and Subsidy treatment); or to receive no information. Using household survey data collected from almost 3,000 households in 150 villages across Madhya Pradesh in India and a rich set of baseline and endline data on time use, the paper examines the impacts of the two information treatments on fuel use, fuel collection, and potential and actual time savings among primary cooks, almost all of whom are women. The analysis in this paper builds on Afridi, Debnath, and Somanathan (2021), who use the experiment to measure impacts on LPG consumption and other household adaptations (e.g., use of chimneys and a separate room as the kitchen) to mitigate indoor smoke. The first set of results in the paper provides new evidence that the time demands of home production in rural India are substantial. Primary cooks in the sample spend on average 60 hours per week in home production activities. Almost 24 of these hours are in cooking-related activities, the equivalent of a part- time job. Despite these significant time demands of home production, the baseline data suggest somewhat modest potential time savings from switching to cleaner fuels. Regardless of how observable character- istics of households are controlled for, total home production time is 30–40 minutes lower and cooking time is 11–36 minutes lower each day in households using only clean fuels, relative to those using only solid fuels. Moving towards cleaner cooking fuels could potentially save 5.8–7.7 percent of total home production time across the week. The main results on fuel use and actual time use exploit the randomized information intervention. Treated households are 6 percent less likely to collect solid fuel, and 4–5 percent less likely to exclusively use solid fuels for cooking. These estimates match results in Afridi, Debnath, and Somanathan (2021), who found that the intervention led households to adjust their fuel use patterns at the margin: treated households that were existing LPG users purchased more LPG cylinders each month, and treated house- holds also increased electric induction cooking to some extent. The information nudge shifted household fuel use in the direction of cleaner energy, although mixed fuel use is still common after the experiment. These changes in fuel use at the household level were not accompanied by large, or even moderate, shifts in total home production time. Experimental estimates indicate a 5 minutes per day time saving in home production, driven by a 5 minute reduction in cooking time for primary cooks exposed to the full information treatment (Health and Subsidy information). This represents about a 2.5 percent reduction in cooking time each week. Calculating the market value of actual time saved through the experiment helps to unpack the muted household response to the information intervention. Using the rural minimum wage as the opportunity cost of women’s time, the monthly time savings from making a marginal switch towards LPG use, as was 2 See Hanna, Duflo, and Greenstone (2016) for an example of a clean stove intervention that had no long-term impacts on health, and Verma and Imelda (2022) for an example of expansion of LPG cooking that did result in health improvements at population level. The World Bank Economic Review 285 induced by the experiment, is about 1.2 percent of monthly household income.3 The subsidized cost of a marginal increase in LPG use in the home is about 0.9 percent of household income. This suggests that households—at least those with existing LPG access—are not making vastly suboptimal decisions about LPG use on the intensive margin. The final section of the paper discusses several factors that limit more fundamental shifts in how cooking is organized in rural India. Specifically, the paper highlights the roles of credit constraints that limit LPG adoption, liquidity constraints that limit LPG use on the intensive Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 margin, and the low opportunity cost of female time reducing incentives for adoption and use. This study contributes to the large environmental literature that focuses on more efficient and cleaner cooking technologies, most notably the adoption and use of improved cookstoves (ICS) in developing countries. Research in this area typically measures the impact of ICS on indoor air pollution and the health of primary cooks, either in lab or in real-world settings (e.g., see Gould and Urpelainen (2018) and Beltramo et al. (2019) for a discussion of this literature, and Hanna, Duflo, and Greenstone (2016) and Bensch and Peters (2015) as specific examples of ICS studies). Improved cookstoves still use solid fuels for cooking, but less of it, and they direct smoke away from the primary cook. In contrast, the focus here is on a clean cooking fuel, one which does not require any biomass collection and does not produce smoke. To the best of our knowledge, this is the first study to examine the effect of randomized nudges towards clean fuel on time use in the household. Only a handful of studies in the improved cookstoves literature estimate the effects of these stoves on time use in cooking and fuel collection. Krishnapriya et al. (2021) provide a comprehensive review of this literature, noting that only 2 of 24 studies (Bensch and Peters 2015 and Hanna, Duflo, and Greenstone 2016) collect time-use data within a randomized evaluation (of improved cookstove methods) design (Berkouwer and Dean (2022) is a new addition to this list). Among the studies that do capture time use, both larger and smaller reductions in cooking time have been estimated in response to the new cookstove, compared with the time savings estimated from the experiment in this paper. Similar to the ICS literature, studies of the impact of LPG have also often focused on the health impacts of moving to cleaner fuels (e.g., see Verma and Imelda 2022). Williams et al. (2020) is an exception—that study measures time-use impacts of moving to LPG in a randomized study in Peru with 180 women who received free unlimited LPG for a year. The experimental results show that free unlimited access to LPG saves on the order of 3.3 hours of cooking time per week—a substantially larger effect than found here, but still a small share of total cooking hours per week (23 hours in their sample). The main takeaway is that, against a backdrop of 24 hours per week spent in cooking, technologies that lead to small time shifts each week cannot have transformational impacts on how time is organized in the home. Financial constraints to adoption and use of new technologies (in general) and clean cookstoves (in particular) have been well studied in the development literature. For example, Bensch, Peters, and Grimm (2015) find that liquidity constraints may dampen adoption of ICS in Burkina Faso. Berkouwer and Dean (2022) randomly subsidize adoption of energy efficient stoves in urban Kenya and find low adoption rates in response to the subsidy, concluding that credit constraints prevent profitable adoption of the improved cookstove. In Uganda, Levine et al. (2018) show that timely credit payments that address liquidity con- straints, along with reducing information asymmetries, increases efficient cookstove adoption by 40–50 percentage points. And in Hanna, Duflo, and Greenstone (2016), while adoption of (highly subsidized) improved cookstoves was large, consistent use of the new technology tapered off over time, partly due to the recurring costs of stove maintenance. This paper shows that, despite the fact that LPG access and use is highly subsidized in low-income Indian households, the impact of switching to cleaner fuels on time use is limited. Credit and liquidity constraints likely still inhibit household responses to randomized information nudges. Finally, the result from this information experiment can be tied to the broader literature on female labor-force participation over the development process, as it has been observed in currently developed 3 These back-of-the-envelope calculations do not account for any health benefits from moving towards cleaner fuels. 286 Afridi et al. countries. The time use of American women changed dramatically over the 20th century, with large reduc- tions in home production time and importantly in food production time. These shifts were accompanied by new cooking technologies in the home. As a point of comparison, 100 years ago, American farmwives spent about 23 hours per week in food preparation and 52 hours each week in home production (Ramey 2009). This is very similar to the time use of the sampled primary cooks in our paper. And, as is the case in the current paper’s setting, cooking in the United States 100 years ago was still heavily reliant on solid fuel Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 (Bose, Jain, and Walker 2022). Yet over decades, households in the United States adopted time-saving new cooking technologies that depended on cleaner fuels, as the cost of these technologies fell and as growing employment opportunities for women made it easier to adopt these new technologies (e.g., Greenwood, Seshadri, and Yorukoglu 2005; Ngai and Pissarides 2008; Vidart 2022). By 1965, every household in America had an electric or gas stove, and by the 2010s, women in the United States spent fewer than 8 hours per week cooking. As Bose, Jain, and Walker (2022) notes, an important part of the dramatic energy transition in the United States was the availability of market work for women. In rural India (and in the specific setting of this paper), female labor-force participation has fallen over time, even as levels of education have risen (Fletcher, Pande, and Moore 2017; Afridi, Dinkelman, and Mahajan 2018. While financial constraints may contribute to the low responsiveness of households in our experiment, the lack of employment opportunities for women in rural India could be another key barrier to increased clean energy use. Future research might usefully test whether a combination of relaxing financial constraints and providing new access to female employment could support a more robust transition to clean cooking fuels, with larger impacts on women’s time use. The paper begins by describing the context of the clean cooking fuel (LPG) subsidy in India. The second section describes time-use patterns in the baseline survey data and links these time-use patterns to patterns of solid or clean fuel use, examining how much time might be saved if households switch to cleaner energy. The third section of the paper outlines the information treatments provided in the randomized experiment and presents the impacts of the experiment on fuel choices and time allocation in the home. The final section places our estimates in context and discusses key remaining constraints to cleaner fuel adoption. 2. Context 2.1. Subsidies to Expand Access and Use of LPG in India Access to LPG for household use in India has increased substantially in the last decade. Since 2011, access to bottled LPG (or LPG cylinders) has risen from 28.5 percent (Census 2011) to 79 percent (PPAC Report 2018). This expansion in access was facilitated by the Pradhan Mantri Ujjwala Yojana (PMUY) program which began in April 2016. The largest program in the world to facilitate access to clean fuel, the PMUY reached 72 million low-income Indian families between April 2016 and June 2019. Under the PMUY, a woman in a rural, socioeconomically disadvantaged household can obtain an LPG connection (henceforth, account) at no upfront cost.4 The total cost of registration for an LPG account is equivalent to about two weeks of monthly household income in rural areas.5 Of this total cost of INR 4 An LPG “connection” is the official term that refers to registration for obtaining the pressure regulator and consumer booklet along with the first LPG cylinder. To register for an LPG account, a consumer must provide proof of identity and address and submit a security deposit equivalent to INR 1,875, unless they qualify for the PMUY connection subsidy. The security deposit is for the empty 14.2 kg capacity cylinder and the pressure regulator. The consumer then has to pay the market price separately for the gas in the cylinder (INR 750) and a stove (INR 750). While the stove can be purchased on the open market, the regulator and refill cylinders are supplied only by the Oil Marketing Companies (OMCs) through their LPG dealers. 5 Average rural household income was approximately INR 7,215 per month in 2011, the latest year for which reliable estimates are available (IHDS-II 2011). The USD–INR exchange rate is approximately USD 1 = INR 75. The World Bank Economic Review 287 3,375, the government covers the INR 1,875 security deposit and administrative charges under the PMUY. PMUY-eligible households, in addition, are mandated to receive an interest-free loan of INR 1,500 from the Oil Marketing Companies (OMCs) to purchase a stove and the first cylinder. Between 2013 and 2020, all households (i.e., both PMUY and non-PMUY households) were subsidized for up to 12 standard (14.2 kg) LPG cylinders annually. The subsidy was implemented as an electronic cash-back scheme—a consumer bought a refill cylinder at the market price and the subsidy was credited to Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 their bank account as cash back within the next 2–3 days. Assuming that a household with 4–5 members requires one standard LPG cylinder per month to facilitate exclusive cooking with LPG, the LPG cost post-subsidy was no more than INR 20 per day or about INR 500–600 per month.6 This translates into a not-insubstantial LPG fuel cost of about 7 percent of monthly household income, post-subsidy (i.e., 500/7,215 = 6.93 percent).7 Not surprisingly, despite rapid increases in LPG access in rural areas, and subsidized LPG refill con- sumption, regular LPG use in India remains low. The mean LPG refill consumption in rural areas is about four cylinders per year, compared to eight cylinders per year in urban areas.8 This study takes place in an environment of widespread access, but relatively low intensive-margin use. 2.2. The Technology of Home Production in Rural India 2.2.1. Home Production Is Extremely Time Intensive The focus of this study is rural Madhya Pradesh (MP), one of the largest northern states in the country, where about 67 percent of households had an LPG account in 2018. At the baseline of a cluster ran- domized controlled trial (RCT) that is further described in the section on experimental design, a time-use survey was administered to the primary cook (the member of the household with the primary responsibil- ity of cooking food for the family, identified by household members during the survey) of 3,000 randomly sampled households (with and without LPG accounts) in 150 villages that were also randomly sampled in the district of Indore in MP during November–December 2018. With few exceptions (i.e., 0.07 percent of the total baseline sample), all primary cooks were women. (Further details of the experimental design and implementation are provided in supplementary online appendix S1.) Primary cooks reported their time use over the 24 hours prior to the survey day (or the last “normal” day).9 To get a sense of the time intensity of different types of activities, weekly hours spent on each activity are calculated by multiplying the unconditional average time spent per day in each activity by 7 (variation in time use across activities by week and weekend days was minimal). Several features of the primary cook’s (henceforth PC’s or women’s) time use stand out. First, in fig. 1(a) almost 60 hours per week are classified as rest, and just over 60 hours per week are spent in home produc- tion (domestic work). Time in home production is greater than time spent in a full-time job. In addition to these home production hours, some women are also engaged in market work. The average amount 6 The market price of LPG refills varies each month in tandem with the international market price of crude petroleum. The refill subsidy varied with the market price to keep the post-subsidy refill price in the range of INR 500–600 per cylinder during the period of this study. 7 The expenditure estimate is in line with a recent report (Council on Energy, Environment and Water), which indicates that if LPG were used exclusively for cooking, the proportion of household expenditure on cooking fuel would vary from 9.2 percent for the bottom wealth decile to 3.9 percent for the top wealth decile in India (at post-subsidy LPG refill price of INR 580 in March 2020). 8 Authors’ estimates from data shared by OMCs for the study area and media reports (Hindu Businessline). LPG refill data are not available publicly. See also Pillarisetti et al. (2019). 9 The surveyor first asked the respondent what time they woke up on the last regular day. The respondent was then asked to recall their activities throughout the day in 15-minute intervals until they went to sleep. This is one standard way to collect time diary information. These activities were then coded using the 24-hour time-use recall. See supplementary online appendix S1 for details of the design of the survey and time-use classification. 288 Afridi et al. Figure 1. Time Use of Primary Cooks (at Baseline) Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 Source: Authors’ analysis based on data in Afridi et al. (2021) and the related experiment. Note: The figure reports average weekly hours spent in different activities, calculated using self-reported time-use survey of primary cooks at baseline in 2018. Time use reported for a regular day prior to the survey is aggregated to weekly totals. Average hours in panel (a) sum to 168 weekly hours. The category “Others” in panel (b) represents the remaining time on domestic work after excluding time spent on fuel and water collection, cooking, cleaning, childcare, shopping, and tending animals. The sample is restricted to 2,942 households where the total recorded time spent by the primary cook (PC) equals 24 hours or 1,440 minutes. The World Bank Economic Review 289 of time in market work among these women is about one-third of the time spent in home production (19.3 hours). Figure 1(b) further subcategorises the time spent on home production. Cooking and cleaning are the most time-intensive components of home production. Cooking occupies almost 24 hours each week, or 3.4 hours per day. Cleaning is also very time intensive, at almost 20 hours per week. In contrast, the time spent on fuel collection is only a small share of total time in domestic work, at under 2 hours per week. Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 This is because collecting fuel (primarily firewood) is a household activity—of the 2.2 trips (on average 117 minutes per trip) made by the household to collect firewood in the previous month in our sample, the primary cook made 1.3 trips (67 minutes per trip), either alone or with other household members. Fuel collection is an occasional, rather than daily, activity. A remarkable characteristic of the pattern of women’s time use across home production categories (especially cooking time) in fig. 1(b) is that it closely matches time use among rural Indian women recorded in national (Afridi, Dinkelman, and Mahajan 2018) and other surveys (Anderman et al. 2015; Maji, Mehrabi, and Kandlikar 2021), time-use patterns among women recorded by studies in other developing countries (e.g., Peru (Williams et al. 2020) and Malawi (Cundale et al. 2017)), and time-use patterns of rural housewives in the historical United States (Ramey 2009; Dinkelman and Ngai 2022).10 Average weekly time in cooking in the United States in the 1920s was 23.5 hours, with cleaning and laundry reaching almost 21 hours—almost identical to the time Indian women in our sample spent cooking and cleaning. 2.2.2. Fuel Stacking Is Prevalent Although primary cooks in the sample survey rely on traditional wood/charcoal stoves (chulha) for cook- ing, mixed fuel use is common, as it is in other developing countries. The baseline survey data indicate that households collect, purchase, and use a mixture of clean and dirty fuels. Complete reliance on clean fuels is unusual. Table 1 shows households’ access (Panel A), use (Panel B), collection (Panel C), and purchase (Panel D) of cooking fuels in the previous month. Consistent with the discussion of widespread access to LPG in the previous section, over two-thirds of households have an LPG account. Yet, while 67.5 percent of the sample reported using LPG for cooking in the previous month, a large share of households report also using dirty fuels (74.8 percent firewood, 87.8 percent dung cakes, and 11.3 percent crop residue) for cooking. Over 70 percent of households reported collecting solid fuels in the previous month (Panel C), while a lower proportion (14.3–29.1 percent) purchased them (Panel D). Households tend to use solid fuels frequently and regularly for cooking regardless of LPG account status, as shown in table 2. Primary cooks in the households were asked to list all the fuels used in cooking over the last month (column 1) and in preparing the most recent meal (column 2). Only 7.3 percent of households used clean fuels exclusively in the last month, with LPG making up the bulk of clean fuel use. In the last meal, only 29 percent of households report using LPG exclusively. Fuel stacking is prevalent: over half of all households used a mix of clean and dirty fuels in the month before the survey. 2.2.3. Fuel Mix Is Correlated with Home Production Time Table 3 uses the entire sample to show the association between time spent in cooking and in home pro- duction with the household’s use of mixed fuels and clean LPG. The outcomes—minutes spent cooking the last meal (column 1), cooking over the last day (column 2), and in total home production activities the day before the survey (column 3)—are regressed on controls for the type of fuel used in the reference 10 Figure S2.1 in the supplementary online appendix shows the distribution of time use by rural women in India in the nationally representative Time Use Survey of 2019, for the sample restricted to the same age group and season as in our study. 290 Afridi et al. Table 1. Fuel access, use, Collection and Purchase (at Baseline) Mean Std. error Obs. (1) (2) (3) Panel A: Fuel access Has LPG account 0.672 0.009 2,784 Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 Panel B: Fuel use last month Firewood 0.748 0.008 2,784 Dung cake 0.878 0.006 2,784 Crop residue 0.113 0.006 2,784 LPG 0.675 0.009 2,784 Electric stove 0.062 0.005 2,784 Panel C: Fuel collection last month Firewood 0.704 0.010 2,083 Dung cake 0.697 0.009 2,444 Crop residue 0.726 0.025 314 Panel D: Fuel purchase last month Firewood 0.143 0.008 2,083 Dung cake 0.291 0.009 2,441 LPG refills 0.322 0.011 1,880 Source: Authors’ analysis based on data in Afridi et al. (2021) and the related experiment. Note: The sample is restricted to households where the total reported time spent by the primary cook (PC) equals 24 hours or 1,440 minutes, and excludes eight villages that did not comply with assigned treatment status. The mean and standard errors in Panels A and B are computed using this sample. The sample is further restricted to households that use the respective fuels in Panels C and D. For Panel B, the households were asked “Did you cook with the fuel in the last month?,” where fuel refers to firewood, crop residue (including twigs and leaves), dung cakes, LPG and electric stove, respectively. In Panel C, the respondents were asked to list all household members who collected firewood or crop residue (including twigs and leaves) in a typical week in the last month. The question for dung cakes was “Did you or anyone in the household either collect or make dung cakes in a typical week in the last month?” For Panel D, the households were asked “Did you buy firewood or dung cake in the last month?” The purchase of LPG in Panel D is calculated for 30 days prior to the survey. The LPG refill data are missing for 74 households at baseline. Fuels excluded from each panel: LPG and electricity cannot be collected (Panel C); crop residue cannot be purchased; no data available on electricity purchase (Panel D). Table 2. Fuel Use for Cooking (at Baseline) Last month Last meal Share using: (1) (2) Only clean fuels 0.073 0.306 (0.005) (0.009) Only LPG 0.064 0.290 (0.005) (0.009) Only electricity 0.001 0.008 (0.001) (0.002) Only solid fuels 0.302 0.542 (0.009) (0.009) Mixed fuels 0.543 0.138 (0.009) (0.007) Observations 2,784 2,784 Source: Authors’ analysis based on data in Afridi et al. (2021) and the related experiment. Note: The table reports mean (standard errors) fuel use in cooking last month and last meal in columns (1) and (2), respectively. The sample is restricted to households where the total recorded time spent by the primary cook (PC) equals 24 hours or 1,440 minutes, and excludes 8 villages that did not comply with assigned treatment status. The variable “Only clean fuels” takes the value 1 if the household uses LPG or/and electricity, and 0 otherwise. The variable “Only LPG” takes value 1 if the household reports using LPG exclusively for cooking and 0 otherwise. The variable “Only electricity” takes the value 1 if the household uses only electric/induction stove and 0 otherwise. The variable “Only solid fuels” equals 1 if the household solely uses solid fuels such as firewood, dung, crop residue, and sigdi, chulha, kande, and 0 otherwise. The variable “Mixed fuels” takes the value 1 if the household uses both LPG and solid fuels (as detailed above), and 0 otherwise. Each column does not add up to 100 because other fuels like gobar gas, bio-gas, consumed by approximately 0.3 percent of households, are excluded. The World Bank Economic Review 291 Table 3. Fuel Use and Time in Home Production (at Baseline) Cooking Cooking Home production last meal yesterday yesterday (1) (2) (3) Only clean fuels −18.540*** −17.806*** −31.014*** Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 (1.19) (4.99) (9.94) Mixed fuels −6.400*** −7.543*** −12.138** (1.67) (2.64) (5.90) # trips by PC to collect firewood — −0.169 4.689*** (0.72) (1.75) # times dung made and collected — 0.653** 2.615*** (0.32) (0.72) Subdistrict FE Yes Yes Yes R2 0.181 0.105 0.074 N 2,784 2,773 2,773 Source: Authors’ analysis based on data in Afridi et al. (2021) and the related experiment. Note: The dependent variable is the daily time spent in cooking last meal, cooking yesterday and home production yesterday in columns (1), (2) and (3), respectively (in minutes). The sample is restricted to households where the total recorded time spent by the primary cook (PC) equals 24 hours or 1440 minutes, and excludes 8 villages that did not comply with assigned treatment status. The independent variable i.e. fuel use in column (1) corresponds to use in last meal, whereas fuel use during last month is considered for columns (2) and (3). ‘Only Clean Fuels’ and ‘Mixed Fuels’ are defined in Table 2. ‘Only Solid Fuels’ is the reference category and has been excluded from the regression. We measure the trips for firewood collection made solely by the PC, however, dung collection and dung making (in a typical week in last month) is recorded for all the household members, including the PC. Further, there are 11 missing observations for total trips to collect firewood at baseline. Controls include household size and assets, education and primary occupation of the household head, education and age of the primary cook, indicators for household religion and caste and indicators for the presence of private primary schools, health sub-center, distance to block headquarters, all-weather road access and proportion of irrigated land, and sub-district fixed effects. The standard errors, clustered at the village level, are reported in parentheses. *p < 0.10, ** p < 0.05 and *** p < 0.01. period (last meal or last day), and a set of household-level observables that may affect fuel and time use. The indicator variable for only clean fuels captures households that use either LPG or electricity (or both) in their cooking. Mixed fuels captures households that use a mix of some clean and some solid fuels. The omitted category is exclusive use of solid fuels in cooking. The first column shows that if a primary cook used only clean fuels in preparing the last meal, they reported spending a statistically significant 18.5 fewer minutes in meal preparation, relative to primary cooks using only solid fuels. This time saving is about the same (17.8 minutes) for meal preparation in the prior day. Aggregating these estimates over a week implies that exclusive use of clean fuels entails about 2 hours (18 minutes × 7 days = 126 minutes) less cooking time each week, or about 3.6 hours less home production time each week. These correlations between fuel use and cooking time use captured in the baseline survey correspond with studies that test for efficiency in cooking with LPG and non-LPG fuels.11 Conditional on household-level observables (as in table 3), households with different patterns of fuel use may still differ in unobservable ways that affect time-use patterns. Implementing a propensity-score matching estimator more flexibly controls for observable differences across households that do and do not use LPG in their most recent meal preparation. Using the rich baseline data, adjusted time-use gaps are compared across matched households that did and did not use LPG in cooking the last meal.12 11 For example, Budya and Arofat (2011) note that LPG is 2.5 times more efficient than kerosene: 1 liter of kerosene can be replaced by 0.4 liters of LPG. 12 Matching the households that used LPG and those that did not use LPG to cook their last meal was done using ob- served characteristics at baseline: village, household, and primary-cook characteristics. Household characteristics in- clude household size, dummy for household head’s education above primary level, head’s occupation, primary cook’s age, dummy for primary cook’s education above primary level, dummy for non-Hindu households, household caste, in- dicators for ownership of agricultural land, house, vehicles, animals, and durable consumer goods. Village-level controls include dummies for a private primary school, access to health subcenters, all-weather road, the proportion of irrigated land, and distance to block headquarters. The specification also controls for seven deprivation indices required for 292 Afridi et al. Table 4. Fuel Use and Time in Fuel Collection (at Baseline): Propensity Score Matched Estimates Firewood Dung making Dung collection (in mins, last trip) (in mins, last month) (in mins, last month) LPG use (last meal) −23.28*** −66.70*** −45.21*** (4.73) (15.70) (15.99) Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 Control group mean 86.25 301.17 156.94 Observations 1,973 1,973 1,973 Source: Authors’ analysis based on data in Afridi et al. (2021) and the related experiment. Note: The dependent variable in column (1) is the total time spent by the primary cook in firewood collection (in minutes) during the last trip for fuel collection. There are 11 missing observations for total trips to collect firewood at baseline. The dependent variable in columns (2) and (3) records the total time taken (in minutes) to make and collect dung in the last month by household, including PC. The propensity score matched estimates for LPG use at last meal are calculated using nearest neighbor matching on observations with identical propensity scores between households “using” LPG at last meal and households “not using” LPG at last meal. Control variables include household size, education, and primary occupation of the household head, education and age of the primary cook, indicators for household religion and caste, and indicators for the presence of private primary schools, health subcenter, distance to block headquarters, all-weather road access, and proportion of irrigated land. We also control for seven deprivation indices for PMUY eligibility, i.e., dummies for households with only one room, kucha walls and kucha roof, female-headed households with no adult male member, and scheduled caste/scheduled tribe (SC/ST) households. Controls also include dummies for ownership and lease of agricultural land, ownership of house, animal-drawn cart, scooter/motorcycle, bicycle, watch, refrigerator, sewing machine, TV, cooler, pressure cooker, tractor, thresher, water pump, car, and dummy for ownership of animals. *p < 0.10, **p < 0.05, and ***p < 0.01. Table 4 shows that primary cooks in LPG-using households spend less time in fuel collection and production than primary cooks in non-LPG-using households. Column (1) indicates that primary cooks spend about 23 fewer minutes each trip collecting firewood, while columns (2) and (3) show 45–66 fewer minutes on dung-collection and dung-making activities in the last month.13 These are fairly minimal weekly time savings. Primary cooks in LPG-using households also tend to spend less time in home production on average, and more time on leisure activities, as shown by the propensity-score matched estimates in table 5. The outcomes in this table are captured from the time-use reports of the primary cook measured for the day prior to the baseline survey (in contrast to time spent in the last month and on the last trip reported in table 4). Panel A, column (1) of table 5 indicates that primary cooks in LPG-using households spend 38 fewer minutes per day on home production, and the majority of this time gap shows up as additional minutes spent in leisure (Panel A, column 4). About 7 minutes of the potential time savings come from less time in fuel collection on the prior day (Panel B, column 1), while 11 minutes of time each day was saved in cooking and 9 minutes of time was saved in the “Others” category (the majority of which was in making dung cakes; Panel B, column 5). None of this potential time saved appears to be allocated towards the market (Panel A, column 2). As one might expect, using LPG does not directly impact time spent on home production activities like childcare (Panel B column 4). Adding up columns (1), (2), and (5) in Panel B, the matching estimates suggest that LPG use in the last meal could save around 28 minutes per day for the primary cook, or 3.3 hours per week in home production; or 4.4 hours if the estimate from column (1) of Panel A is used. This equates to a 5.5–7.3 percent reduction of weekly time in home production. Taken together, the baseline time-use survey demonstrates that home production is generally very time intensive. Specifically, it is cooking and cleaning, rather than fuel collection or fuel (dung cake) production PMUY eligibility: indicator for households having one room, kucha walls and kucha roof, indicators for female-headed households without adult male member, and scheduled caste/scheduled tribe (SC/ST) households. Figure S2.2 shows that this matching exercise selects a set of non-LPG-using households that resemble the LPG-using households more closely on a large group of observables. Figure S2.3 shows the reduction in standardized bias in covariates post-matching. All of our results are robust to alternative matching methods, e.g., kernel-based matching. 13 Note that information on fuel collection time used in this table is not from the time diary, but from the main house- hold survey. Time spent in firewood collection was asked for all household members (including the PC) who undertook this activity in the previous month, and time spent by the PC is reported only in column (1) of the table. Dung col- lection/making was asked for the household as a whole (although this activity is almost always undertaken by the PC alone). The World Bank Economic Review 293 Table 5. Fuel Use and Primary Cook’s Total Time Use (at Baseline): Propensity Score Matched Estimates Time use of primary cook (mins/day) Home production Market work Personal care Leisure (1) (2) (3) (4) Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 Panel A: Overall time use LPG use (last meal) −38.01*** −5.19 7.06** 33.07*** (7.92) (9.57) (2.94) (7.34) Control group mean 527.93 185.56 152.77 572.68 Observations 1,973 1,973 1,973 1,973 Fuel collection Cooking Cleaning Childcare Others (1) (2) (3) (4) (5) Panel B: Time spent in home production LPG use (last meal) −7.65*** −11.02*** −8.15 −1.52 −9.67** (2.04) (3.45) (5.12) (3.79) (4.26) Control group mean 19.03 204.38 166.65 56.96 80.89 Observations 1,973 1,973 1,973 1,973 1,973 Source: Authors’ analysis based on data in Afridi et al. (2021) and the related experiment. Note: The dependent variable is the daily time spent by the primary cook in different activities (in minutes per day). The category “Home Production” in Panel A includes childcare and the category “Leisure” includes sleep. The category “Others” in Panel B is the residual for domestic work that includes time spent on water collection, shopping, tending animals, and remaining activities. For both Panels A and B, the matched propensity score estimates of LPG use at last meal are calculated using nearest neighbor matching on observations with identical propensity scores between households using and not using LPG at last meal. Controls include household size, education and primary occupation of the household head, education and age of the primary cook, indicators for household religion and caste, and indicators for the presence of private primary schools, health subcenter, distance to block headquarters, all-weather road access, and proportion of irrigated land. We also control for seven deprivation indices for PMUY eligibility, i.e., dummies for households with only one room, kucha walls and kucha roof, female-headed households with no adult male member, and SC/ST households. Controls also include dummies for ownership and lease of agricultural land, ownership of house, animal-drawn cart, scooter/motorcycle, bicycle, watch, refrigerator, sewing machine, TV, cooler, pressure cooker, tractor, thresher, water pump, car, and dummy for ownership of animals. *p < 0.10, **p < 0.05, and ***p < 0.01. that uses the most time among PCs’ daily activity; and cooking with dirty fuels takes somewhat more time per meal than using cleaner fuels. The patterns here suggest some scope for limited time savings as households increase their use of LPG and reduce their reliance on solid fuels for cooking. Potential time savings suggested by the observational data are still small, relative to the 24 hours of total time spent on cooking each week. The next set of results examines the causal effect of a shift towards cleaner fuel use, by estimating how time use changes across households that are randomly nudged to adopt and use the clean fuel. 3. Information Nudges towards Clean Fuel: The RCT 3.1. Experimental Design and Sampling Afridi, Debnath, and Somanathan (2021) designed two information campaigns to nudge households to use cleaner fuels at home. The cluster-randomized RCT was conducted in the same 150 villages selected randomly to be part of the baseline study discussed above, with 50 villages (and the 20 randomly sampled households within each village) randomly assigned to one of two treatment groups, or a control group. See supplementary online appendix S1 and Afridi, Debnath, and Somanathan (2021) for further details on sampling and design of the experiment. In November and December 2018, the baseline survey was administered to the 3,000 households se- lected for inclusion in the study. The information intervention was then conducted in the first nine months of 2019, and a follow-up survey was conducted in the last three months of 2019. The information cam- paign was designed to increase the adoption and regular use of LPG. It consisted of an awareness campaign on the health and financial benefits of switching to regular use of LPG for cooking. The campaign centered around improving households’ understanding of the adverse health impacts of solid fuels and measures to 294 Afridi et al. mitigate inhalation of indoor smoke (Health) and the government subsidy to LPG consumers (Subsidy). The information intervention had two treatment arms: the health awareness arm (Health treatment) and the health awareness and LPG subsidy (Health and Subsidy treatment) arm. Government health workers delivered information under either arm to sampled households in the relevant randomized villages. Infor- mation was delivered in a series of video vignettes by a health worker following a written script.14 The goal was to get households to understand the health cost of continued biomass use for home cooking, Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 the benefits of alternatives like using chimneys or LPG, and how households can reduce the costs of us- ing LPG through the government-run, cash-back subsidy program. No information was provided in the control group of villages. Both baseline and follow-up surveys collected detailed time-use data from the primary cook in the household. Time-use data were self-reported in a typical time diary fashion: with women reporting all activities conducted 24 hours prior to the date of the survey. A description of how the time-use modules were administered is in the supplementary online appendix and table S1.1. See tables S2.1–S2.4 for balance on observable characteristics across treatment arms at baseline. There are no differences in average time spent in personal care and domestic work across treatment and control arms; there are some differences in leisure and work time at baseline. However, the joint significance test of all of the time-use variables across treatment group status has a p-value of 0.38 and 0.26 on Health treatment and Health and Subsidy treatment arms, respectively. 3.2. Empirical Strategy Two related empirical specifications are used to causally estimate the reduced-form effects of the infor- mation awareness campaign on fuel use and the primary cook’s time use. First, exposure to the Health treatment or Health and Subsidy treatment campaign is combined into a single indicator of treatment status that takes a value of 1 if a household was exposed to either treatment, and 0 otherwise (control group), as follows: Yi1 0 v = βc + βT Tv + β0Yiv + βX Xiv + βZ Zv + iv , (1) where Yi1 v is fuel use or minutes per day spent by the primary cook in the ith household in village v at endline; Yi0v is the baseline fuel/time use by the same primary cook in the previous year, in approximately the same season (October–December); Tv is a dummy variable indicating whether village v is assigned to either treatment; and Xiv are a set of baseline characteristics for household i in village v. These controls include household size and assets, education and primary occupation of the household head, education and age of the primary cook, indicators for household religion and caste.15 Village characteristics are also included: Zv , the proportion of irrigated land, and indicators for the presence of private primary schools, health subcenter, distance to block headquarters, and all-weather road access.16 To account for variation in the administration of the local health department, which may have impacted the delivery of the intervention across subdistricts, the specification also includes subdistrict fixed effects. 14 Implementation of the experiment leveraged the existing public health system by engaging Accredited Social Health Activists (ASHAs) to deliver the information—female residents of the village, who had at least secondary schooling, were between 25 and 45 years of age, and were employed by the state government to provide public health services. ASHAs were paid INR 50 per visit per household, comparable to their regular remuneration. 15 Since the ownership of different household assets is likely to be highly collinear, a first principal components analysis over several indicators is used to measure the economic status of a household. These indicators include ownership of land and farm animals, pucca house, and a list of consumer durables. Education of the head of the household and the primary cook is measured by an indicator that takes value 1 for more than primary education and 0 otherwise. 16 Census data on “distance of village to block headquarters” are missing for 260 households (13 villages). Instead, traveling distance is imputed using Google Map’s Distance Matrix Application Programming Interface (API). The correlation between this imputed traveling distance and census data is 0.84 (mean census (Google API) distance is 18.07 (20.01) km, as against the mean straight-line distance of 13.70 km) for the 137 villages with census distance data. The results do not vary if a dummy for missing distance data for the 13 villages is added into the regression analysis. The World Bank Economic Review 295 Table 6. Effect of Information Treatment on Fuel Collection: Experimental Results Firewood Dung (1) (2) (3) (4) Overall treatment −0.039 — −0.060* — Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 (0.025) (0.033) Treatment H — −0.019 — −0.061 (0.027) (0.041) Treatment H+S — −0.061** — −0.059* (0.029) (0.035) Baseline collection 0.210*** 0.209*** 0.073*** 0.073*** (0.020) (0.020) (0.025) (0.025) Subdistrict FE Yes Yes Yes Yes H = H+S [p-value] [0.115] [0.950] Control group mean 0.540 0.540 0.255 0.255 R2 0.100 0.101 0.077 0.077 N 2,545 2,545 2,784 2,784 Source: Authors’ analysis based on data in Afridi et al. (2021) and the related experiment. Note: The dependent variable takes value 1 if the household collected firewood (columns (1) and (2)) and collected dung cakes (columns (3) and (4)) in the last month, respectively, and 0 otherwise. The sample is restricted to households where the total time spent by the primary cook equals 24 hours, and excludes eight villages that did not comply with assigned treatment status. H denotes only health information and H+S refers to both health and subsidy information treatment arms. There are 239 missing observations for firewood collection at the endline. Controls include household size and assets, education and primary occupation of the household head, education and age of the primary cook, indicators for household religion and caste, and indicators for the presence of private primary schools, health subcenter, distance to block headquarters, all-weather road access and proportion of irrigated land. Standard errors, clustered at the village level, are reported in parentheses. *p < 0.10, **p < 0.05, and ***p < 0.01. The main parameter of interest is β T , representing the impact of the awareness campaign (either Health or Health and Subsidy) on the outcomes of interest. Since the treatment status was randomly assigned to the sampled villages, households’ exposure to treatment was exogenous. The OLS estimate of β T from equation (1) represents the intention to treat effect of the awareness program. The second specification distinguishes between the two types of treatments and estimates and compares the impact of the two arms on each outcome: Yi1 h h hs hs 0 v = βc + βT Tv + βT Tv + β0Yiv + βX Xiv + βZ Zv + νiv , (2) where Tvh is a dummy for assignment of village v to the Health treatment and a dummy for assignment Tvhs to the Health and Subsidy treatment. The other variables are as explained above. Standard errors in both equations (1) and (2) are clustered at the village level. 3.2.1. Fuel Use Impacts The first set of results in table 6 presents the impact of the information treatments on solid fuel collection for home production: specifically, firewood and dung. Over half of the control households spend any time collecting firewood and around one-quarter of control households spend time collecting dung for making dung cakes for home energy use. For each outcome, results are first presented for the combined treatment first, and then for the two separate treatments. All columns contain the full set of additional controls. Across the board, exposure to the Health and Subsidy treatment reduced the incidence of solid fuel collection for household use. Firewood collection declines by 6.1 percentage points and dung collection declines by 5.9 percentage points, respectively.17 Although the point estimates on the Health treatment are 17 These results are similar to the results from tables 11 and 12 presented in Afridi, Debnath, and Somanathan (2021). Sample size varies slightly between Afridi, Debnath, and Somanathan (2021) and this paper due to seven missing obser- vations on time use. 296 Afridi et al. Table 7. Effect of Information Treatment on Fuel Used in Cooking Last Meal: Experimental Results Only clean fuels Mixed fuels Only solid fuels (1) (2) (3) (4) (5) (6) Overall treatment 0.048* — −0.025 — −0.040* — Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 (0.026) (0.019) (0.023) Treatment H — 0.049* — −0.022 — −0.046* (0.029) (0.020) (0.026) Treatment H+S — 0.047 — −0.029 — −0.033 (0.030) (0.023) (0.028) Baseline usage 0.333*** 0.333*** 0.059** 0.058** 0.306*** 0.306*** (0.022) (0.022) (0.023) (0.023) (0.019) (0.019) Subdistrict FE Yes Yes Yes Yes Yes Yes H = H+S [p-value] [0.970] [0.721] [0.652] Control group mean 0.314 0.314 0.140 0.140 0.533 0.533 R2 0.169 0.169 0.019 0.019 0.170 0.171 N 2,784 2,784 2,784 2,784 2,784 2,784 Source: Authors’ analysis based on data in Afridi et al. (2021) and the related experiment. Notes: The dependent variable is the type of fuel used in cooking the last meal. The sample is restricted to households where the total recorded time spent by the primary cook (PC) equals 24 hours or 1,440 minutes, and excludes eight villages that did not comply with assigned treatment status. H denotes only health information and H+S refers to both health and subsidy information treatment arms, respectively. “Only clean fuels,” “Mixed fuels,” and “Only solid fuels” are defined in table 2. Controls include household size and assets, education and primary occupation of the household head, education and age of the primary cook, indicators for household religion and caste, and indicators for the presence of private primary schools, health subcenter, distance to block headquarters, all-weather road access, and proportion of irrigated land. Standard errors, clustered at the village level, are reported in parentheses. *p < 0.10, **p < 0.05, and ***p < 0.01. not statistically significant, they also suggest that exposure to information moved households away from solid fuel use. The point estimates suggest that one cannot reject that the two information treatments had the same impact on reducing household activity towards collection of solid fuels for energy use. Next, the paper looks at how primary cooks use fuels in the most recent or last meal cooked. Table 7 shows the impacts of randomized treatment nudges on how primary cooks use fuels in the most recent last meal cooked. The three outcomes are only clean fuels, which bundles LPG with electricity (columns 1–2); a mix of clean and solid fuels (columns 3–4); and only solid or dirty fuels (columns 5–6).18 Exclusive use of solid fuels falls by around 4 percentage points (or 7.5 percent of the mean of the control group), while exclusive use of clean fuels rises by about the same magnitude. The point estimates on Health treatment and Health and Subsidy treatment are statistically the same. Note that the point estimates on mixed fuel use in the last meal are negative, but not statistically significantly different from zero. One reason why it may be difficult to detect a change in mixed fuel use is that composition of households changes in each group: some households move from exclusive solid fuel use into the mixed fuels category, while other households move from mixed fuels into the clean fuels only category. Tables 6 and 7 show that the experiment marginally shifted household choice of cooking fuels. This move away from solid fuel use towards cleaner energy sources is in line with results in Afridi, Deb- nath, and Somanathan (2021), who find that the intervention significantly increased consumption of LPG refills. Using administrative LPG refill consumption data, that paper shows that monthly LPG consump- tion (unconditional on household’s LPG access) increased by 12.5 percent in the combined treatment 18 Women typically report two meal cooking times–morning (before 1 p.m.) and evening (after 1 p.m.). There are no systematic differences in the types of fuel used between morning and evening meals, either at baseline or endline as shown in fig. S2.4, although there is a significant increase in “only clean fuels” at endline. The World Bank Economic Review 297 arm (Health and Subsidy treatment) as shown in table S2.5 (reproduced from Afridi, Debnath, and So- manathan 2021).19 3.2.2. Time-Use Impacts Table 8 shows the intent to treat impacts of the information intervention on average time used by primary cooks in the day before the endline survey. Panels A–D present outcomes for different categories of time Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 use: total home production (including childcare), market work, personal care, and leisure. For each out- come, results are shown for the combined treatment effect in column (1), and then effects for the Health treatment and the Health and Subsidy treatment separately in column (2). Exposure to treatment does not have a large impact on total time allocated to any category. In almost all columns, the point estimates are small. In Panel A, the coefficient on the Health and Subsidy treatment is larger, indicating a reduction of about 14 minutes of home production time per day among this treated group relative to controls. However, none of the point estimates are significantly different from zero. Table 9 investigates whether the composition of total home production time changed after exposure to the information nudge. Results for minutes per day spent on fuel collection, cooking, cleaning, childcare, and others are reported. In Panel B, column (2), the Health and Subsidy treatment reduces time spent in cooking by a marginally significant 5 minutes per day, with no significant change in any other category. It is worth noting that the first stage (in table 7) for whether the treatment moved households towards more clean fuels is not sufficiently strong to pursue an IV strategy. While exposure to the Health and Health and Subsidy information nudged households towards using cleaner fuels in cooking, it did not have significant impacts on time use of primary cooks in exposed households. 4. Discussion 4.1. Valuing Time Savings from Cleaner Cooking at Home The analysis using baseline survey data demonstrated that home production as a category, and cooking in particular, is time-consuming work in rural Indian households. The patterns of fuel use in these data suggested modest potential time savings from changing the technology of cooking, i.e., by using clean fuels rather than dirty fuels. The experimental results indicated that the information nudge towards clean fuels resulted in marginal shifts towards cleaner energy in the home and only minimal time savings in the home. What is the market value of these time savings? How does this value compare with estimates from other clean cooking initiatives in the literature? And how does the market value of time saved compare with the costs of switching to cleaner fuel within the context of the experiment, and in this Indian setting? What was the market value of the actual time saved? Assuming that the government-mandated un- skilled wage rate of INR 280 per day (INR 35 per hour) captures the opportunity cost of women’s time in the labor market, then the 5 minutes per day of actual cooking time savings for households estimated through the experiment in table 8 is worth about INR 86 per month. This represents a saving of about 1.2 percent of rural monthly household income.20 This estimate may be a lower bound on the market value of time saved in the entire household, since non-primary cooks (for whom daily time-use data were not collected) may also have saved some time in fuel collection (see table 6).21 19 Besides LPG, the RCT identified a significant shift towards another clean fuel—electricity or induction cooking. Self- reported electric stove cooking increased by over 50 percent in the Health and Subsidy treatment group, but off a very low base. 20 The market value of this time saved is calculated as (35/60) × 5 minutes × 7 days × 4.2 weeks per month. Rural monthly household income of INR 7,215 is taken from the 2011 IHDS, widely regarded as the most careful and most recent survey of household income in India (IHDS-II 2011) 21 Of households surveyed, 70 percent report spending 44 hours in the previous month, on average, collecting firewood. 298 Afridi et al. Table 8. Effect of Information Treatment on Time Use of Primary Cook: Experimental Results Panel A: Home production (1) (2) Overall treatment −5.072 — (7.284) Treatment H — 3.815 Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 (9.105) Treatment H+S — −14.433 (8.843) Baseline time use 0.127*** 0.126*** (0.021) (0.021) Subdistrict FE Yes Yes H = H+S [p-value] — [0.089] Control group mean 513.145 513.145 R2 0.097 0.099 N 2,784 2,784 Panel B: Market work (1) (2) Overall treatment 5.201 — (9.286) Treatment H — 2.143 (11.518) Treatment H+S — 8.419 (11.214) Baseline time use 0.153*** 0.152*** (0.022) (0.022) Subdistrict FE Yes Yes H = H+S [p-value] — [0.634] Control group mean 165.922 165.922 R2 0.096 0.096 N 2,784 2,784 Panel C: Personal care (1) (2) Overall treatment −4.103 — (2.757) Treatment H — −4.959 (3.212) Treatment H+S — −3.200 (3.285) Baseline time use −0.001 −0.001 (0.018) (0.018) Subdistrict FE Yes Yes H = H+S [p-value] — [0.611] Control group mean 152.162 152.162 R2 0.050 0.050 N 2,784 2,784 Panel D: Leisure (1) (2) Overall treatment 4.030 — (6.034) Treatment H — −0.724 (7.168) Treatment H+S — 9.068 (6.679) Baseline time use 0.143*** 0.145*** (0.017) (0.017) The World Bank Economic Review 299 Table 8. Continued Subdistrict FE Yes Yes H = H+S [p-value] — [0.158] Control group mean 606.527 606.527 R2 0.107 0.108 N 2,784 2,784 Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 Source: Authors’ analysis based on data in Afridi et al. (2021) and the related experiment. Note: The dependent variable is the time spent daily in different categories (in minutes) of domestic work. The category “Others” is the residual for domestic work that includes time spent on water collection, shopping, tending animals, and residual activities. The sample is restricted to households where the total recorded time spent by the primary cook (PC) equals 24 hours or 1,440 minutes, and excludes eight villages that did not comply with assigned treatment status. H denotes only health information and H+S refers to both health and subsidy information treatment arms, respectively. The controls remain the same as in table 7. The standard errors, clustered at the village level, are reported in parentheses. *p < 0.10, **p < 0.05, and ***p < 0.01. It is useful to compare these estimates of the market value of time savings estimates from studies that measure how improved cookstoves affect time use. While cookstoves do not use cleaner fuel, consistent use of improved cookstoves (ICS) may yield time savings from more efficient (quicker) cooking over charcoal and/or from lower demand for fuel collection. Even in the large improved cookstove literature, though, Krishnapriya et al. (2021) note that few studies capture time use from surveys embedded in a randomized design. Time savings in our study are lower than estimated time savings from the ICS literature, and the market value relative to household income is also somewhat lower. Bensch and Peters (2015) estimate that house- holds in a sample of rural Senegalese villages randomized into receiving a free ICS saved approximately 75 minutes per day in cooking time, the market value of which represents about 5 percent of rural median household income.22 In a more urban setting, Berkouwer and Dean (2022) estimate a significant 54 min- utes of cooking time saved per day in response to randomly assigned subsidies for ICS in Nairobi, Kenya. The market value of the cooking time saved is calculated to be about 4.8 percent of monthly household income. In both studies, the larger impact on cooking time makes sense, since the free/subsidized ICS facilitated a much larger change in cooking technology than information nudges towards clean fuel. Are households in the sample making optimal choices at the margin, given the price of using cleaner fuels? Recall that access to an LPG account in India is subsidized for PMUY households while LPG refills are subsidized for all households, and the out-of-pocket price of a cylinder is approximately INR 500 (at the time of our study). Afridi, Debnath, and Somanathan (2021) estimates that among households with an existing LPG account, monthly LPG consumption increases by 12.5 percent in nudged households in the combined Health and Subsidy treatment (see table S2.5). The cost of this increased consumption is about 0.125 × 500, or INR 62 per month, which equates to about 0.9 percent of monthly household income (INR 62/INR 7,215). At the margin, the market value of time saving only slightly outweighs the cost of using a little more LPG. A complete transition to LPG cooking would be considerably more expensive. Assuming that one LPG cylinder would allow a small family (4–5 members) to cook all meals over gas for one month, the cost of one additional LPG cylinder entails expenditures of about 7 percent of monthly household income (at the subsidized price). To make it worthwhile to exclusively cook over LPG, a household would need to save 14.5 hours of cooking time per week, or over 2 hours of time per day, and put all of it into the market to earn the minimum wage.23 22 The national minimum wage for unskilled workers in Senegal was CFA 182.95 per hour in 2011, so 75 minutes of cooking time saved each day is valued at around CFA 227 (or CFA 6,956 each month). Median rural household income in Senegal is estimated at CFA 139,000 in Houweling et al. (2012). 23 These calculations ignore the cost and time savings for households to stop purchasing or collecting wood. 300 Afridi et al. Table 9. Effect of Information Treatment on Home Production Time of Primary Cook: Experimental Results Panel A: Fuel collection (1) (2) Overall treatment −1.539 — (2.043) Treatment H — −0.689 Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 (2.193) Treatment H+S — −2.435 (2.290) Baseline time use 0.011 0.012 (0.013) (0.013) Subdistrict FE Yes Yes H = H+S [p-value] — [0.344] Control group mean 16.153 16.153 R2 0.020 0.020 N 2,784 2,784 Panel B: Cooking (1) (2) Overall treatment −2.721 — (2.412) Treatment H — −0.233 (2.971) Treatment H+S — −5.344* (2.866) Baseline time use 0.058*** 0.057*** (0.017) (0.017) Subdistrict FE Yes Yes H = H+S [p-value] — [0.124] Control group mean 197.756 197.756 R2 0.073 0.074 N 2,784 2,784 Panel C: Cleaning (1) (2) Overall treatment −0.079 — (4.334) Treatment H — 1.879 (5.181) Treatment H+S — −2.136 (4.806) Baseline time use 0.094*** 0.093*** (0.018) (0.018) Subdistrict FE Yes Yes H = H+S [p-value] — [0.428] Control group mean — 165.676 165.676 R2 0.046 0.046 N 2,784 2,784 Panel D: Childcare (1) (2) Overall treatment 2.784 — (3.537) Treatment H — 5.156 (4.100) Treatment H+S — 0.281 (4.238) Baseline time use 0.240*** 0.239*** (0.027) (0.027) The World Bank Economic Review 301 Table 9. Continued Subdistrict FE Yes Yes H = H+S [p-value] — [0.272] Control group mean 60.984 60.984 R2 0.184 0.185 N 2,784 2,784 Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 Panel E: Others (1) (2) Overall treatment −3.400 — (3.713) Treatment H — −2.285 (4.609) Treatment H+S — −4.578 (4.319) Baseline time use 0.112*** 0.112*** (0.018) (0.018) Subdistrict FE Yes Yes H = H+S [p-value] [0.646] Control group mean 72.577 72.577 R2 0.058 0.058 N 2,784 2,784 Source: Authors’ analysis based on data in Afridi et al. (2021) and the related experiment. Note: The dependent variable is the time spent daily in different categories (in minutes) of domestic work. The category “Others” is the residual for domestic work that includes time spent on water collection, shopping, tending animals, and residual activities. The sample is restricted to households where the total recorded time spent by the primary cook (PC) equals 24 hours or 1,440 minutes, and excludes eight villages that did not comply with assigned treatment status. H denotes only health information and H+S refers to both health and subsidy information treatment arms, respectively. The controls remain the same as in table 7. The standard errors, clustered at the village level, are reported in parentheses. *p < 0.10, **p < 0.05 and ***p < 0.01. 4.2. Constraints on Clean Energy Use and Time Saving At the heart of these results is the fact that the marginal shift in fuel mix generated by exposure to the clean energy information nudge is insufficient to generate transformative changes in cooking practices, and hence time use of primary cooks. Researchers have documented that the transition to cleaner energy in developing countries is often slow, due to mixed fuel use. Households rarely switch from exclusive solid fuels to exclusive clean fuels. As households grow richer, they tend to climb a step at a time towards cleaner and more efficient fuels (Van der Kroon, Brouwer, and Van Beukering 2013): a concept commonly known as the energy ladder. In our setting, the awareness campaign nudges allowed some households to take one tiny step on this ladder—but not sufficiently high to generate substantial time savings. Several factors likely constrain how far up the energy ladder the households could plausibly have moved in response to the nudge, thereby limiting time savings. Financial constraints affect adoption and use of LPG on both intensive and extensive margins (e.g., see Puzzolo et al. 2016; Alem and Ruhinduka 2020; Sharma et al. 2020). Because of the way the Indian LPG subsidy works—households must pay full price up-front, and wait 2–4 days for a refund of the subsidy portion—a lack of liquidity limits intensive margin use of LPG. Ongoing work by Afridi, Barnwal, and Sarkar (2022) estimates that when the out-of- pocket price for an LPG cylinder rises by 1 percent (even when the post-subsidy price remains unchanged), LPG use in low-income or PMUY households falls by 0.2 percent. That these households are sensitive to variation in the pre-subsidy price suggests they are liquidity constrained. In addition, credit constraints in the poorest households likely affect LPG uptake on the extensive margin (e.g., as in Berkouwer and Dean (2022) who find that credit constraints retard adoption of cleaner cookstoves in Nairobi). A second factor limiting the impact of information nudges relates to the lack of a market for women’s time. Recent studies of the historical United States suggest that the availability of women’s formal employ- ment in the labor market played an important role in facilitating household adoption of new technologies 302 Afridi et al. for cooking and cleaning (Bose, Jain, and Walker 2022; Vidart 2022). Greater opportunities for women to work outside the home changed the opportunity costs of female time at home, and raised family income. Yet in our study villages, the average female employment rate is very low, at 15.3 percent. The puzzle of low female labor-force participation in India has been noted elsewhere in the literature (see Fletcher, Pande, and Moore 2017; Afridi, Dinkelman, and Mahajan 2018). Even if credit and liquidity constraints can be overcome, limited labor-market opportunities for women may not make it worthwhile for house- Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 holds to try to move 35 minutes of primary cook time into the market each week. Moreover, clean fuel adoption and use decisions may be subject to intra-household bargaining constraints. Since women bear the main burden of cooking (Parikh 2011) and have lower returns to market work relative to men (Gronau 1973), men may not internalize the benefits of cooking using cleaner fuels (Bloomfield 2015; Beltramo et al. 2015), because these benefits do not accrue to them directly. The ability of women to negotiate the intra-household allocation of resources can be another challenging constraint on households’ transition to clean fuels (Duflo, Greenstone, and Hanna 2008; Doss 2013; Miller and Mobarak 2013; Gould and Urpelainen 2018) and one which may be alleviated by greater female labor-market participation. 5. Conclusion This paper asks whether reliance on biomass for cooking fuel contributes to the large amount of time women spend cooking across the developing world. While exclusive use of solid fuels is associated with greater time in fuel collection and cooking, the evidence from this clean energy information experiment in rural India suggests that more than a nudge is needed for households to shift their mix of cooking fuel use towards cleaner cooking fuels. The main result is that there are very small impacts on time spent in home production in response to the information nudge. The value of the time saved does not seem to vastly outweigh the costs of using LPG at the margin, even at highly subsidized prices. However, these results should not be interpreted as evidence that clean energy is unable to reduce the drudgery of home production among women in low- income settings. Rather, expecting transformative changes in time allocation from informational nudges towards cleaner fuels alone seems unrealistic. Future research might usefully test whether a more robust transition to clean cooking fuels could be supported by combining a relaxation of financial constraints with new access to female employment and income-earning opportunities. References Afridi, F., P. Barnwal, and S. Sarkar. 2022. “On the Design of Subsidy Programs: Access to Clean Energy and Liquidity Constraints.” Working Paper, https://sites.google.com/site/prabhatbarnwal/. Afridi, F., S. Debnath, and E. Somanathan. 2021. “A Breath of Fresh Air: Raising Awareness for Clean Fuel Adoption.” Journal of Development Economics 151: 102674. Afridi, F., T. Dinkelman, and K. Mahajan. 2018. “Why Are Fewer Married Women Joining the Work Force in Rural India? A Decomposition Analysis over Two Decades.” Journal of Population Economics 31: 783–818. Alem, Y., and R. D. Ruhinduka. 2020. “Saving Africa’s Tropical Forests through Energy Transition: A Randomized Controlled Trial in Tanzania.” Ruhr Economic Papers 885. Anderman, T., R. DeFries, S. Wood, R. Remans, R. Ahuja, and S. Ulla. 2015. “Biogas Cook Stoves for Healthy and Sustainable Diets? A Case Study in Southern India.” Frontiers in Nutrition 2. Beltramo, T., G. Blalock, S. Harrell, D. I. Levine, and A. M. Simons. 2019. “The Effects of Fuel-Efficient Cookstoves on Fuel Use, Particulate Matter, and Cooking Practices: Results from a Randomized Trial in Rural Uganda.” Working paper series No. WPS-85. Center for Effective Global Action. UC, Berkeley. Beltramo, T., G. Blalock, D. I. Levine, and A. M. Simons. 2015. “The Effect of Marketing Messages and Payment over Time on Willingness to Pay for Fuel-Efficient Cookstoves.” Journal of Economic Behavior and Organization 118. The World Bank Economic Review 303 Bensch, G., and J. Peters. 2015. “The Intensive Margin of Technology Adoption–Experimental Evidence on Improved Cooking Stoves in Rural Senegal.” Journal of Health Economics 42: 44–63. Bensch, G., J. Peters, and M. Grimm. 2015. “Why Do Households Forego High Returns from Technology Adoption? Evidence from Improved Cook Stoves in Burkina Faso.” Journal of Economic Behavior and Organization 116 (August): 187–205. Berkouwer, S. B., and J. T. Dean. 2022. “Credit, Attention, and Externalities in the Adoption of Energy Efficient Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 Technologies by Low-Income Households.” American Economic Review 112(10): 3291–3330. Bloomfield, E. 2015. “Gender and Livelihood Impacts of Clean Cook Stoves in South Asia. Executive Summary. Global Alliance for Clean Cook Stoves.” https://cleancooking.org/binary-data/RESOURCE/file/000/000/363-1.pdf. Bose, G., T. Jain, and S. Walker. 2022. “Women’s Labor Force Participation and Household Technology Adoption.” European Economic Review 147: August. Budya, H., and M. Y. Arofat. 2011. “Providing Cleaner Energy Access in Indonesia through the Megaproject of Kerosene Conversion to LPG.” Energy Policy 39(12): 7575–86. Cecelski, E. 2000. “The Role of Women in Sustainable Energy Development.” Technical report, National Renewable Energy Laboratory: Energy, Environment, and Development. June. https://www.nrel.gov/docs/fy00osti/26889.pdf. Census. 2011. “Registrar General of India, Ministry of Home Affairs, Government of India.” Cundale, K., R. Thomas, J. K. Malava, D. Havens, K. Mortimer, and L. Conteh. 2017. “A Health Intervention or a Kitchen Appliance? Household Costs and Benefits of a Cleaner Burning Biomass-Fuelled Cookstove in Malawi.” Social Science & Medicine 183 (June): 1–10. Dinkelman, T., and L. R. Ngai. 2022. “Time Use and Gender in Africa in Times of Structural Transformation.” Journal of Economic Perspectives 36(1): 57–80. Doss, C. 2013. “Intrahousehold Bargaining and Resource Allocation in Developing Countries.” World Bank Research Observer 28(1): 52–78. Duflo, E., M. Greenstone, and R. Hanna. 2008. “Cooking Stoves, Indoor Air Pollution and Respiratory Health in Rural Orissa.” Economic and Political Weekly: 71–6. Fletcher, E. K., R. Pande, and C. T. Moore. 2017. “Women and Work in India: Descriptive Evidence and a Review of Potential Policies.” Harvard Kennedy School Faculty Research Working Paper Series. Working Paper No. RWP18- 004. Gould, C. F., and J. Urpelainen. 2018. “LPG as a Clean Cooking Fuel: Adoption, Use, and Impact in Rural India.” Energy Policy 122 (November): 395–408. Greenwood, J., A. Seshadri, and M. Yorukoglu. 2005. “Engines of Liberation.” Review of Economic Studies 72(1): 109–33. Gronau, R. 1973. “The Intrafamily Allocation of Time: The Value of the Housewives’ Time.” American Economic Review 63(4): 634–51. Hanna, R., E. Duflo, and M. Greenstone. 2016. “Up in Smoke: The Influence of Household Behavior on the Long-Run Impact of Improved Cooking Stoves.” American Economic Journal: Economic Policy 8(1): 80–114. Houweling, E., R. Hall, A. Diop, J. Davis, and M. Seiss. 2012. “The Role of Productive Water Use in Women’s Liveli- hoods: Evidence from Rural Senegal.” Water Alternatives 5(3): 658–77. IHDS-II. 2011. “Desai, Sonalde and Vanneman, Reeve. National Council of Applied Economic Research, New Delhi. India Human Development Survey, 2011-12. Inter-university Consortium for Political and Social Research [distrib- utor].” Krishnapriya, P. P., M. Chandrasekaran, M. Jeuland, and S. K. Pattanayak. 2021. “Do Improved Cookstoves Save Time and Improve Gender Outcomes? Evidence from Six Developing Countries.” Energy Economics 102(October): 105456. Levine, D. I., T. Beltramo, G. Blalock, C. Cotterman, and A. M. Simons. 2018. “What Impedes Efficient Adoption of Products? Evidence from Randomized Sales Offers for Fuel-Efficient Cookstoves in Uganda.” Journal of the European Economic Association 16(6): 1850–80. Maji, P., Z. Mehrabi, and M. Kandlikar. 2021. “Incomplete Transitions to Clean Household Energy Reinforce Gender Inequality by Lowering Women’s Respiratory Health and Household Labour Productivity.” World Development 139. Miller, G., and A. M. Mobarak. 2013. “Gender Differences in Preferences, Intra-Household Externalities, and Low Demand for Improved Cookstoves.” National Bureau of Economic Research (NBER) Working Paper No. 18964. April. 304 Afridi et al. Ngai, L. R., and C. A. Pissarides. 2008. “Trends in Hours and Economic Growth.” Review of Economic Dynamics 11(2) 239–56. Parikh, J. 2011. “Hardships and Health Impacts on Women Due to Traditional Cooking Fuels: A Case Study of Himachal Pradesh, India.” Energy Policy 39(12): 7587–94. Pillarisetti, A., M. Ghorpade, S. Madhav, A. Dhongade, S. Roy, K. Balakrishnan, and S. Sankar et al. 2019. “Promoting LPG Usage during Pregnancy: A Pilot Study in Rural Maharashtra, India.” Environment International 127 (June): Downloaded from https://academic.oup.com/wber/article/37/2/283/6984984 by Joint Bank-Fund library user on 04 September 2023 540–49. PPAC Report. 2018. “LPG Profile: Data on LPG Marketing as on 01.01.2018, Petroleum Planning and Analysis Cell (PPAC),” Available at http://www.ppac.org.in/WriteReadData/Reports/201802221141024777342LPG01Jan2018. pdf Puzzolo, E., D. Pope, D. Stanistreet, E. A. Rehfuess, and N. G. Bruce. 2016. “Clean Fuels for Resource-Poor Settings: A Systematic Review of Barriers and Enablers to Adoption and Sustained Use.” Environmental Research 146 (April): 218–34. Ramey, V. A. 2009. “Time Spent in Home Production in the Twentieth-Century United States: New Estimates from Old Data.” Journal of Economic History 69(1): 1–47. Rosenbaum, P. R., and D. B. Rubin. 1985. “Constructing a Control Group Using Multivariate Matched Sampling Methods That Incorporate the Propensity Score.” American Statistician, 39(1): 33–38. Sharma, D., K. Ravindra, M. Kaur, S. Prinja, and S. Mor. 2020. “Cost Evaluation of Different Household Fuels and Identification of the Barriers for the Choice of Clean Cooking Fuels in India.” Sustainable Cities and Society 52: 101825. Van der Kroon, B., R. Brouwer, and P. J. Van Beukering. 2013. “The Energy Ladder: Theoretical Myth or Empirical Truth? Results from a Meta-Analysis.” Renewable and Sustainable Energy Reviews 20 (April): 504–13. Verma, A., and Imelda. 2022. “Clean Energy Access: Gender Disparity, Health, and Labor Supply.” Forthcoming in Economic Journal. Vidart, D. 2022. “Human Capital, Female Employment and Electricity: Evidence from the Early 20th-Century United States.” Working Paper. Williams, K. N., J. L. Kephart, M. Fandiño-Del-Rio, S. M. Simkovich, K. Koehler, S. A. Harvey, and W. Checkley. 2020. “Exploring the Impact of a Liquefied Petroleum Gas Intervention on Time Use in Rural Peru: A Mixed Methods Study on Perceptions, Use, and Implications of Time Savings.” Environment International 145 (December): 105932.