The World Bank Economic Review, 39(2), 2025, 410–438 https://doi.org10.1093/wber/lhae024 Article (Joint) Bank Savings, Female Empowerment, and Child Labor in Rural Ethiopia Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Jose Galdo Abstract This study examines whether the random allocation of single and joint savings accounts to cash crop farmers in rural Ethiopia is associated with increased savings and changes in decision-making authority and control over resources that could ultimately affect child labor and schooling resource allocations. Consistent with posited channels of intrahousehold bargaining models, women from households assigned to the joint saving treatment group show significant gains in autonomy and control of savings resources, broader financial empowerment, and increased labor participation. Positive effects on school participation and attendance are reported for girls, although point estimates are measured imprecisely. In a setting where schooling and child labor are not mutually exclusive, children work more when joint deposit accounts are available. In the absence of impacts on household income, this increase in child labor is explained by complementarities between adult farm labor and child labor in the household production function, which is reinforced by lumpy investments in labor-intensive agricultural inputs that likely increased the opportunity costs of children’s time. JEL classification: C93, D14, G21, J43, I21, O12, R20 Keywords: bank savings, child labor, schooling, women’s empowerment, RCT 1. Introduction Savings by the poor have become a priority in the development agenda, and savings mobilization strategies for the poor are widely seen as key initiatives for agricultural development, food security, and economic growth (World Bank 2008). However, access to formal deposit accounts is still far from a reality for 1.7 billion adults worldwide, most of them living in developing countries. This market failure is particularly prevalent in agriculture-based economies due to structural barriers associated with the supply and demand of financial services, which, according to the Global Findex Database, are mainly driven by banking costs, Jose Galdo is a professor of Economics and Public Policy at Carleton University, Ottawa, Canada; his email is jose.galdo@carleton.ca. The author gratefully acknowledges financial support from the IZA/DFID Growth and Labour Mar- kets in Low Income Countries (GLM-LIC), grant agreement GA-C3-RA5-323. Research activities were approved by the Research Ethics Board Review at Carleton University, Ottawa, Canada. AEA RCT Registry AEARCTR-0006295. The au- thor thanks the Ethiopian Economic Policy Research Institute for their institutional support and Gustavo Bobonis, Alberto Chong, Ana Dammert, Degnet Abebaw, Jason Garret, and Jessica Goldberg for their comments and suggestions. The author also thanks the editor and three anonymous referees. Carlos Perez provided excellent research assistance. The author has no conflict or financial interest in the issues studied in this article. All errors are the author’s own. A supplementary online appendix is available with this article at The World Bank Economic Review website. C The Author(s) 2024. 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 The World Bank Economic Review 411 physical distance, and a lack of national identification systems. In Ethiopia, the setting of this study, bank savings account penetration reaches only 12 percent for adults with primary education, while the financial gender gap has widened in recent years even as total financing available to the world’s poor has increased steadily (World Bank 2017). Following a growing body of literature regarding the impacts of ownership (e.g., Ashraf et al. 2010; Dupas and Robinson 2013a; Prina 2015; Dupas, Keats, and Robinson 2017) and control (e.g., Ashraf 2009; Schaner 2018; Field et al. 2021) of bank savings accounts on women’s financial inclusion and Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 economic empowerment, this study implements a financial intervention that provides small subsidies to cover the pecuniary costs of opening formal savings accounts to 1200 agricultural households in rural Ethiopia. The intervention randomly allocates these subsidies into three groups: a single deposit account in the name of the household head (90 percent of them are males), a joint deposit account in the name of both husband and wife, and a pure control group. Thus, this paper attempts to determine whether access to single or joint accounts leads to differential patterns of saving behavior and female financial empowerment that, in turn, would affect schooling and child labor decisions. Assessing downstream outcomes, such as household income, adult labor supply, and labor-intensive agricultural investments, makes it possible to evaluate complementary channels to explain the children’s outcomes. By randomly assigning single savings accounts to heads of households and joint savings accounts to both spouses, the study exogenously shifts the household income-sharing rule between these two treatment groups by providing female spouses an economic opportunity to manage and control household savings through joint savings accounts. In rural Ethiopia, differences in intrahousehold welfare can be traced to differences in bargaining power (Dercon and Krishnan 2000). There is a commonly held norm that Ethiopian women who have more control of household assets have more voice in household decisions, and thus the welfare of women after marriage depends on the control they have over assets during the marriage (Fafchamps et al. 2009). This kind of behavior is consistent with collective models of household decision making in which asymmetric preferences for goods and services within the household depend on bargaining power, which in turn depends on rules regarding the management of household assets during the marriage (e.g., Lundberg and Pollak 1993; Browning and Chiappori 1998). This study hypothesizes that relative to single accounts solely owned by the heads of households, joint savings accounts would provide female spouses higher autonomy and control of financial resources, lead- ing to increased decision-making power within the household. In turn, shifts in bargaining power would have positive consequences for children’s school enrollment and attendance. These differential effects are inconsistent with unitary models. Instead, these changes are consistent with collective models of intra- household allocation wherein increasing women’s control of household resources could lead to higher schooling investment allocations for children (e.g., Quisumbing and Maluccio 2000; Duflo 2003). In a context where schooling and child labor are not mutually exclusive, how changes in school participation would affect child labor is not clear from a theoretical standpoint (Wydick 1999; Edmonds 2007, de Hoop et al., 2014). On the one hand, one would expect a reduction in child labor demand following the “lux- ury axiom” in Basu and Van (1998) if a positive income effect emerges from this saving encouragement intervention. For instance, access to formal savings accounts could spawn increases in crop output and, thus, agricultural income (Karlan et al. 2014; Brune et al. 2016; Callen et al. 2019). On the other hand, the value of a child’s time could increase if the relaxation of capital constraints through increased sav- ings leads to changes in the marginal product of family labor through complementarities between labor- intensive investments and child labor, raising the opportunity cost of a child’s time (e.g., Wydick 1999; Edmonds and Theoharides 2020). Likewise, gains in decision-making authority by female beneficiaries would lead to expansions of female labor supply (Kabeer 2005; Field et al. 2021), which could increase the demand for child labor due to job complementarities in farm production. Hence, whether child labor is affected by saving encouragement interventions is an empirical question for which there is only thin evidence. 412 Galdo Several findings emerge from this study. Administrative bank data show the overall take-up rate for this intervention is 57 percent, with similar rates between single (54 percent) and joint (60 percent) deposit accounts treatment groups. In terms of usage of the accounts, the share of active users, defined as having at least 5 deposits over 24 months, reaches 65 percent of those who take up the treatment. This usage leads to a sizable increase in total savings for single and joint accounts treatment groups over the next 27 months after the intervention. Consistent with posited channels of intrahousehold bargaining models, one observes that female beneficiaries in the joint account treatment group show statistically significantly Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 higher autonomy and control of saving resources and higher gains in financial empowerment relative to female beneficiaries in the single account treatment group (0.91 SD versus 0.41 SD). Moreover, the study observes positive differential impacts on school enrollment and school expenditures for girls aged 6–16 in households assigned to the joint-account treatment group, although these impacts are measured imprecisely. Regarding child labor effects, and relative to a pure control group, the study finds positive and significant effects among children from households assigned to the joint account group at the extensive (15 percent) and intensive (19 percent) margins, with no statistically significant differential impacts between boys and girls. No meaningful labor changes are observed among household members assigned to the single account group 27 months after the treatment setup. The expansion of child labor in the joint account treatment group is accompanied by significant increases in women’s farm labor supply, which suggests complementarities between adult and child work in this agricultural setting. The study also observes labor- intensive agricultural investments (e.g., organic fertilizer, coffee trees, seedlings) in households assigned to the joint account group (0.25 SD), which can also increase the opportunity cost of child work within the household. On the other hand, the study finds negligible and imprecisely measured mean impacts on household income, a variable that is shown to be one important driver associated with reductions in child labor (e.g., Edmonds 2007). This study contributes to the literature on financial inclusion across four important dimensions. First, it expands the growing literature on saving interventions (e.g., Ashraf, Karlan, and Yin. 2006a; Dupas and Robinson 2013a, 2013b; Prina 2015; Karlan and Leigh 2014) by focusing on an often-neglected key sec- tor: small-holder agricultural households. This study is one of a few existing RCTs, (e.g., Brune et al. 2016; Aker et al. 2020) that target formal saving opportunities for cash crop farmers in poor, rural settings. Since the agricultural sector in Ethiopia contributes close to half of the country’s GDP, this intervention provides useful information for policies that aim to foster agricultural development in impoverished settings. Sec- ond, the article assess the interaction between savings product design and intrahousehold labor decision making among cash crop farming households. While the literature on microcredit interventions offers am- ple but mixed evidence on child labor effects (e.g., Islam and Choe 2013; Angelucci, Karlan, and Zinman 2015; Tarozzi, Desai, and Johnson 2015), the article expands on this important work by randomly allo- cating different saving products and assessing farm labor responses by adults and children alike. A few studies have evaluated the link between formal savings and child labor supply responses in developing settings. For instance, Berry, Karlan, and Pradhan (2018) assess the impact of financial education on sav- ings and labor for youth in Ghana. Third, this study contributes to a large body of research investigating the role of bank product design in promoting access and usage of deposit accounts in undeveloped set- tings. These savings products include basic formal bank accounts (e.g.’ Dupas and Robinson 2013a; Prina 2015), commitment savings accounts (e.g., Ashraf et al. 2006a; Ashraf et al. 2010; Dupas and Robinson 2013b; Brune et al. 2016), deposit collection services (Ashraf, Karlan, and Yin 2006b; Callen et al. 2019), digital payments (e.g., Aker et al. 2020; Jack and Suri 2016; Schaner 2017), among others. This study contributes to this literature by randomly allocating standard single and joint accounts and directly com- paring their effects on intrahousehold issues. In this regard, this study evaluates a straightforward policy: that is, whether commercial banks should promote individual or joint accounts as they expand their op- erations among agricultural households, which can be easily implemented and replicated across different settings. The World Bank Economic Review 413 Finally, this study speaks to the literature on women’s economic empowerment as one of a recent hand- ful of studies examining the effects of bank account control on intrahousehold decision-making power. Ashraf (2009) uses an experimental approach to identify the role of information and communication in intrahousehold financial decisions among married couples in the Philippines. Couples are given money to save in the bank or take consumption. Consistent with a monitoring mechanism, results show that men deposit money into their own accounts when choices are private. When required to communicate, men deposit money in their wives’ accounts. Both women and men whose spouses control savings decisions Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 are more likely to put money away when choices are private. Schaner (2015) provides large temporal sub- sidies to married couples in rural Kenya to incentivize using single and joint bank accounts at randomly assigned interest rates. Results show that couples well matched on discount factors are less likely to use costly single and joint accounts, while couples who received high interest rates in joint accounts reported higher levels of agreement about the account usage. Schaner (2018) uses the same intervention in Kenya to directly compare the long-run impacts of single and joint deposit accounts on downstream outcomes. She finds increased entrepreneurship and household income rates for single accounts, while no significant impacts emerge for joint accounts. Field et al.’s (2021) study is of particular interest as it assesses the impacts of control over women’s earnings by randomizing deposits of women’s wages from India’s public workfare program into their bank accounts and accounts owned by the male head of household. Having the money deposited into women’s accounts increased their labor market participation, particularly in households with lower levels of, and stronger norms against, female work. Our article offers evidence of control of single versus joint savings accounts for women’s financial empowerment in rural Ethiopia. In restricting legal control to one individual, usually the head of household, single accounts create a formal barrier to wives that the account holder can use in bargaining, while joint accounts offer a path to shift the income-sharing rule, and thus, women’s decision-making power. The remainder of this paper is organized as follows: section 2 describes the setting of this study; sec- tion 3 provides institutional details on the saving intervention, sampling framework, data sources, time- line, and a discussion of the gender dimension of the bank savings product design. Section 4 assesses the take-up rates and usage of accounts, while section 5 describes the empirical framework and presents the main findings. Section 6 provides concluding remarks. 2. The Setting The setting of this study covers two remote agricultural areas of Ethiopia within the preeminent coffee- producing regions of the country. Our financial intervention covers 12 rural districts (Kebeles) across Jimma and Sidama provinces in the country’s west- and south-central parts. These two areas entail significant cultural variation. Orthodox Christian households (55 percent of the sample) mainly pop- ulate Sidama, while Jimma is a predominantly Muslim area (45 percent). Ethiopia is one of the poor- est countries in the world, the second-most populous country in Africa, and its agricultural sector ac- counts for 46 percent of GDP, 85 percent of total employment, and 90 percent of export revenues (FAO 2015). The farming system is typical of rain-fed agriculture in Sub-Saharan Africa, with agricultural income subject to uncertainty due to weather shocks. Each household in the sample constitutes a small-holder farming unit with an average of one hectare of land distributed in seven plots mainly dedicated to culti- vating coffee crops, followed by enset, maize, banana, and avocado. These coffee farmers live in dispersed areas due to low population densities and difficult terrain, with poor communication systems and phys- ical infrastructure, but share membership in Fairtrade coffee cooperatives. In the sample, for instance, the proportion of houses with mud floors (70 percent), no electricity (78 percent), and no ventilated pit latrines (78 percent) reflect the living conditions of most farmers in Sub-Saharan Africa. As shown in the supplementary online appendix table S1.1, all farmers in the sample are poor, with an average annual 414 Galdo income per capita of around US$100 and an average of four years of formal schooling among heads of households. Rural Ethiopia is where gender segmentation of work and social life is rooted in the cultural and historical influence of East Africa’s triple heritage, that is, African, Islamic, and colonial (Bass 2004). Administrative data show that around 90 percent of heads of households are males in the baseline. There is a commonly held norm that a farmer is a man who is the primary breadwinner, while the female spouse typically has more household chores responsibilities (Badstue et al. 2020; Galdo, Dammert. and Abebaw Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 2021). In this regard, baseline data shows that heads of households spend, on average, twice as much time as their spouses farming household plots, while the latter devote more than three times as much time to household chores than the former. The agricultural sector in Ethiopia is the epitome of financial exclusion. The provision of formal fi- nancial services to small-holder farmers are institutionally considered “high risk” because of the inherent variability of agricultural production, seasonality, and lack of formal insurance mechanisms.). As a result, access to formal credit markets is extremely limited in these remote areas due to high transaction costs and inadequate contractual enforcement mechanisms that lead households to resort to informal financial arrangements. No household in the baseline data, for instance, reports lending from commercial banks at the onset of this study. 3. The Intervention The Gender Dimension of Bank Savings Product Design Following intrahousehold bargaining models, allocating single and joint savings accounts would shift the “distribution factors” that influence the relative power of individuals within the household differently (e.g., Chiappori 1992). The power of joint accounts to shift women’s intrahousehold bargaining power would depend on individual preferences and constraints, economic roles and responsibilities of household members, the social and cultural norms that shape gender roles and norms, and practices about resource allocation that affect access and control over the bank account. One would expect changes in women’s decision-making power within the household if female benefi- ciaries gain autonomy and control over joint savings accounts, that is, they would not be solely the de jure owners of the account, but they would also have effective control over deposits and withdrawals since information is public (savings are observable), and monitoring is less costly (Ashraf 2009). This is a key distinction vis-à-vis households wherein a male head of household owns a single savings account that en- tails private information (savings are not observable) and limited saving monitoring and control between spouses. When both spouses participate in financial decisions and savings goals, household long-term fi- nancial planning and risk management may be enhanced through collaborative financial decision making between spouses, particularly in households where spouses have better alignment in time preferences over financial decisions (Schaner 2015). Likewise, access to financial resources through joint savings accounts can lead women to enhance their ability to monitor and control household savings, strengthening women’s bargaining power because they may be more able to negotiate for their preferences regarding resource allocation (Ashraf, Karlan, and Yin 2010). Because joint savings accounts allow female beneficiaries to monitor and control income, gains in women’s decision-making power would bias household choices toward their preferences (e.g., Thomas 1990; Quisumbing and Maluccio 2000). Moreover, joint savings might give women greater economic independence, enabling them to participate more actively in economic activities because of shifts in norms around women’s work (Field et al. 2021). Furthermore, joint savings accounts can empower women by giving them access to and control over financial resources, making them more likely to contribute to resource allocation decisions in spheres such as education and healthcare for children (Duflo 2003). Enhancing these investments in human capital development may be more likely to be achieved when women are more informed and educated. Finally, access to bank accounts often goes The World Bank Economic Review 415 hand in hand with improved financial literacy. Women can learn about budgeting tools or new finan- cial concepts, empowering them with the knowledge and practice to make financial decisions. Because empowerment indicators have a more significant impact on financial decisions in low-income households than in higher-income households, the positive relationship between women’s decision-making power and access to joint banking accounts may be particularly salient in households where resources are relatively scarce. On the other hand, it is not clear a priori that co-ownership of joint savings accounts would necessar- Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 ily disrupt a “dictatorial” household decision maker if female spouses are only de jure owners of joint accounts under the complete control of the husbands. This would lead to female spouses not making de- posits that the husbands could access and control, a particular issue for individuals with low bargaining power. Likewise, joint savings accounts would have negligible effects in households where a husband and a wife disagree about how to spend their income, and there is interpersonal pressure to share savings by a high-bargaining-power spouse. Schaner (2017), for instance, provides a striking illustration of the unintended effects of a new bank product (ATM) that was meant to reduce the costs of accessing bank accounts but instead reduced the overall account use for low-bargaining-power individuals. Therefore, it is likely that joint savings accounts might be more effective for households that are not at the extremes of the decision-making power distribution. Unlike Western developed countries, where household finances and assets are often held in common, control over productive assets and finances is more intricate in Ethiopia due to a mix of legal and customary rules that result in a lack of dissociation of management from ownership (Fafchamps 2001). It is well-known that African women who bring more assets into the household have more voice in financial and farming decisions as control and management of assets within the marriage affect women’s own labor decisions and individual income (e.g., Jones 1986; von Braun and Webb 1989). Thus, addressing the gender dimension of this study requires knowing who controls the deposit ac- counts. Specifically, in what percentage of cases do male heads of households control the single account? Administrative data from the bank partner shows that among households that take up a single bank ac- count, close to 90 percent belong to the male heads of households. Since in Ethiopia, individual ownership of financial assets within marriage is closely related to controlling and managing those assets (Fafchamps and Quisumbing 2002), male beneficiaries are expected to exert control over the newly single savings accounts. Since Ethiopian farms function as unified units under the control of heads of households due to traditional ox-plow agriculture that requires significant physical strength (McCann 1995) and returns to scale in management and experience (Boserup 1965), farming decisions are conducted mainly by male heads of households (Fafchamps, Kebede and Quisumbing 2009). Thus, it is expected that the exclu- sive control of single deposit accounts by male heads of households would lead to strong preferences for agricultural investments rather than over child human capital investments due to extended patrimonial customs on disposition rules that provide a larger share of the productive asset to those with greater control (Fafchamps and Quisumbing 2002). It is, therefore, important to determine whether female beneficiaries typically made bank deposits and withdrawals on their own. Although the study does not have specific information for each transaction, survey information collected 27 months after the intervention asks female beneficiaries whether they visited the bank on their own in the past 12 months before the survey. For this variable, results show that women from households assigned to the single and joint account treatment groups have a statistically significantly higher likelihood of visiting a bank than women from the control group. Importantly, women in the joint account treatment group have a statistically higher likelihood of visiting a bank branch in the past 12 months relative to women in the single account treatment group, as attested by the p-value = 0.056 that rejects the null of no differential impacts between women in the single and joint account treatment groups. Whether control over the new bank joint accounts translates into changes in decision-making 416 Galdo power that alter the intrahousehold allocation of resources is a fundamentally empirical question that is assessed in section 5. Sampling Framework The sampling framework follows from a population of 5100 agricultural households belonging to 4 Fairtrade coffee cooperatives in rural Ethiopia in Jimma (2) and Sidama (2). Access to administrative data from the cooperatives was granted following an agreement that considers the head of the house- Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 hold, whether male or female, as the person who represents the household in any business affairs with the Fairtrade cooperatives, and as such, it is expected that he or she has more and better information about production, payments, and operations with the cooperative. This sampling frame comes from co- operatives’ 2014 administrative records. These records show that membership at these Fairtrade coffee cooperatives ranges from 800 to 1500 farming households with a yearly red-cherry coffee production that ranges from 600 to 1122 kilograms per household. Indeed, this population framework entails variation in geographic, cultural, and agricultural output. Based on this population, the study selected a represen- tative sample based on a 2 × 2 stratified design applied independently to each Fairtrade cooperative. Stratification is based on two variables of interest: the level of coffee production and the gender of the heads of households. In doing so, the analysis split the population of Fairtrade members into high- and low-production groups according to whether households’ production is above or below the 2014 per- household median of coffee production. The resulting stratified sample is composed of 1,200 households. The allocation of household units to treatment and control groups is then implemented through random- ization at the household level using simple random methods. As a result, 450 households were assigned to the single-account treatment group, 450 to the joint-account treatment group, and 300 to a pure control group. Outreach and Marketing Efforts Fieldwork was implemented through a research partnership with the Commercial Bank of Ethiopia (CBE) and the Ethiopian Economic Policy Research Institute (EEPRI); CBE is the leading private bank in the country with over 13 million nationwide clients, distributed across 1160 local branches, including four in the study’s areas of intervention. CBE offers standard deposit accounts, which are interest-bearing deposits that pay account users an annual nominal interest rate of 5 percent in the analysis period. Anyone signing up for a savings account must attend a local bank branch, present a valid identification card (ID) and two passport pictures, and have a minimum balance of birr 25 (US$1.20). No other deposits or withdrawal fees apply to holders of deposit accounts at CBE. In collaboration with Fairtrade Cooperative leaders and CBE marketing officials, and at the onset of the intervention in December 2015, the study offered a marketing and financial education workshop following the assessment of focus groups in which it became apparent that knowledge of banking operations and lack of trust in commercial banks were important demand barriers for saving products in this rural setting. Both heads and households and spouses from treatment and control groups were invited to public events at the cooperative premises in which they received a half-day financial training on basic bank operations and bank regulations on single and joint deposit accounts, as well as messages reinforcing social capital (trust) on formal banking institutions. These public events were also used as a platform to disseminate public information on the role of savings in coping with emergencies, financing agricultural investments, and serving as a buffer between seasonal income and consumption. At the end of the workshop, farmers were publicly informed that some households would be randomly selected to receive monetary subsidies to open one account, either a single or joint account, that exclusively targets heads of households and spouses, the beneficiaries of this intervention. Likewise, they were publicly informed that the chosen treatment households would receive house visits in the coming days following the public workshop. The World Bank Economic Review 417 Subsidies Between December 2015 and January 2016, the public workshop was followed by individual visits to households, in which field workers provided both heads of households and spouses together, detailed information on the size of subsidies and the program’s rules and distributed the corresponding vouchers as one-time subsidies to cover the pecuniary costs involved in the opening of formal savings accounts. The timeline for these visits to households overlapped with the coffee harvesting season because it is in this season that farmers receive windfalls of cash from coffee crops. The voucher was tailored according Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 to the randomly assigned treatment group. The study set a one-time, small subsidy of 70 birr (US$ 3.20) and 100 birr (US$4.50) for the single and joint account treatment groups. These marginally differential incentives are intended to cover the higher costs of opening joint accounts relative to single accounts, that is, transportation costs and picture IDs for two people rather than one. No maintenance or withdrawal fees exist across these two standard saving products. It was up to the farmers to choose whether to visit the local bank branch that same day or later, as it was also up to the farmers to select the holder of the single savings account, either the household head or the spouse. The pure control group did not receive these subsidies, but they were free to open formal savings accounts, single or joint, at the same commercial bank if they chose to do so. Some important treatment features deserve attention. First, farmers received subsidies through time-limited, household- specific vouchers. The value of the voucher was deposited directly into the new savings account when heads of households and/or spouses visited the bank office to set up the account. Second, individuals who take the offer are encouraged to make an out-of-pocket cash deposit of any size to replicate realistic conditions when opening a savings account. Third, although the fieldwork is designed to comply with the original saving product design allocation, the partner bank cannot deny services to treatment households that approached its local branches, complied with all bank procedures, and wanted to open a different type of account from the one that was allocated under treatment. Thus, a household with a voucher assigned to the joint account group could have a single savings account and a subsidy of 70 birr instead of 100 birr. Similarly, a household assigned to the single account group could switch to a joint savings account and receive a subsidy of 100 birr instead of 70 birr. As seen in section 4, the overall noncompliance rate reached 25 percent. Data Sources and Timeline This study makes use of four different data sources. The identification of the final, stratified sample of 1200 households comes from institutional data from four Fairtrade coffee cooperatives. Through a partnership with the Commercial Bank of Ethiopia (CBE), the study also gained access to administrative records for each new savings account opened as part of the intervention. This institutional data include the date of account opening, type of account, name of account holders, deposits, withdrawal transactions, and saving balances from December 2015 to January 2018. Household survey data were collected in three rounds. The study gathered an initial baseline dataset in July/August 2015 that collects household- and individual-levels information about socio-demographic and labor market variables, land and assets ownership, agricultural production, environmental and per- sonal adverse shocks, food security, and women’s empowerment, among other dimensions. A limited first follow-up household survey was collected in December 2016/January 2017, 12 months after the start of the financial intervention. Unlike the baseline household survey, it contained a smaller set of variables, focusing mainly on household-level financial matters such as formal and informal savings, loans, and household income. No schooling or labor-market variables for each household member were gathered in this data effort. Lastly, the study conducted a final follow-up household survey in March/April 2018, 27 months following the financial intervention. This survey collected individual- and household-level infor- mation, focusing on schooling and labor-market information from each household member and standard 418 Galdo Figure 1. Timeline of Financial Intervention. Source: Author’s elaboration. Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 data on agricultural input and output, income, expenditures, and food consumption. The analysis focuses on this second wave of follow-up data. The fieldwork was conducted by personnel from the Ethiopian Economic Policy Research Institute (EEPRI), the research arm of the Ethiopian Economics Association. Figure 1 summarizes the timeline of this intervention along with information on the seasonality of the agricultural production calendar. Importantly, attrition rates are low, involving around 2 percent of the original sample over the full-time length of this study. While 1197 Fairtrade coffee households were interviewed in the baseline survey out of the original target of 1200 households, 1185 and 1166 households participated in the first and second follow-up household surveys 12 and 27 months after the intervention. There is some missing reported survey information for some variables, particularly monetary ones; thus, the reported number of observations across columns in some results tables may show some variation. In any case, this problem affects around 2 percent of observations; thus, the analysis does not make any imputations of missing data. Table S1.1 in the supplementary online appendix reports comprehensive balance tests using the full treatment split and a rich set of variables, including most baseline outcomes of interest. Columns 3, 4, and 5 show the p-values of the equality for means test for all treatment groups against each other, that is, single account vs. control, joint account vs. control, and single account vs. joint account. These statistics add confidence to the random allocation design since the study does not find imbalances in most baseline variables and cannot reject the equality of means across these randomly assigned groups. Likewise, joint orthogonality tests, shown at the bottom of the table, cannot reject balance on household variables across each pair-wise sample comparison. 4. Take-Up Rates and Savings Account Usage Panel A in table 1 shows administrative bank data on take-up rates. Out of 900 households that were offered the subsidies, 512 (57 percent) opened deposit accounts at the partner bank, including 54 per- cent of households assigned to the single account group and 60 percent to the joint account group. All bank transactions are managed in person at the bank branches since mobile money/ATM technology is unavailable. No direct costs are associated with withdrawals except travel costs and travel time. On the other hand, only 5 percent of control group households opened accounts at the partner bank following the intervention. Table S1.2 in the supplementary online appendix displays results for the determinants of take-up for the single and joint accounts treatment groups. One can observe that women’s financial empowerment in the baseline is not statistically relevant for take-up decisions in both treatment groups. Single and joint account treatment groups show different take-up drivers. Larger households, Orthodox Christian households, households with male heads, and households without bank accounts are more likely to open a joint deposit account. In contrast, households with prime-age heads, more isolated households as measured by access to the electrical grid, households that loan money to relatives and neighbors, and households that do not own a nonfarm business are more likely to open single accounts. Among farmers who take up the treatment, three out of four opened accounts in their initially assigned treatment group The World Bank Economic Review 419 Table 1. Take-Up Rates Single account Joint account Control treatment group treatment group group (1) (2) (3) Panel A: Take-Up Number of observations in each treatment group 450 450 300 Opened account at partner bank 54% 60% 6% Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Open account in the initially assigned treatment group 43% 41% —- Panel B: Usage (unconditional) Ever used account (at least one deposit) 51% 56% 5% Made at least 3 deposits within the first 24 months 40% 40% 1% Active user: made at least 5 deposits within first 24 37% 38% 0.30% months Mean saving balance within first 6 months [birr] 177 (651) 228 (620) 2 (9) Mean saving balance within first 12 months [birr] 618 (4266) 526 (4666) 4 (32) Mean saving balance within first 24 months [birr] 425 (2297) 387 (2090) 32 (376) Panel C: Usage (conditional on opening account) Mean saving balance within first 6 months [birr] 308 (888) 413 (702) 35 (21) Mean saving balance within first 12 months [birr] 1026 (5410) 980 (6424) 79 (118) Mean saving balance within first 24 months [birr] 809 (3159) 597 (2486) 573 (1521) Source: Author’s analysis based on administrative data from the bank partner from December 2015 to December 2017. Note: Standard deviation of monetary saving balances in parenthesis. US$/birr exchange rate was around 21 in 2015, 22 in 2016 and 26 in 2017. (80 percent of those takers assigned to the single account group and 70 percent of takers assigned to the joint account group). These numbers show a somewhat higher demand for single rather than joint sav- ings accounts. Table S1.3 in the supplementary online appendix presents the results of the determinants of compliance to the initial treatment assignment separated by single and joint accounts treatment groups. Among households initially assigned to the single account group, compliance/switching decisions seem random as only one variable—access to the electrical grid—is statistically associated with decisions to switch and open a joint account. In contrast, there are some patterns among households initially assigned to the joint account: Muslim households, households where the head of household is a female, households that are located closer to the bank branches, and households that do not own a nonfarm family business are statistically more likely to switch and open a single account. Panel B in table 1 shows bank administrative information on account usage for over 24 months. Al- most all treatment households who opened a savings account used the account at least once. Of the 900 households offered the subsidy, 37.5 percent became “active” users, as defined by making at least five deposits in the first two years after setting up the account. Conditional on take-up status, panel C reveals nonlinear positive accumulation of saving balances over time, which on average amounts to 356,1005, and 713 birr for the treatment group after 6, 12, and 24 months following the intervention. Moreover, some differences in the savings accumulation pattern emerge between those households who opened single (809 birr) and joint (597 birr) savings accounts 24 months after the intervention. Regard- less of take-up status, the accumulation of monetary savings among treatment households shows higher balances in the first year relative to the second one, suggesting some resource depletion over time. Un- like the analysis of determinants of take-up rates, there is no clear pattern for the drivers of active us- age for the joint account treatment group except for a negative association with distance to the bank branch. For the single account group, on the other hand, more affluent households, as measured by the absence of exposure to food shortage and access to ventilated pit latrine, and households that re- ceived remittances show higher usage of the account as seen in table S1.2 in the supplementary online appendix. 420 Galdo 5. Empirical Framework and Main Findings For each outcome of interest, the estimation framework focuses on the intent-to-treat (ITT) parame- ter.1 The analysis estimates the mean effects of being assigned either to single- or joint-account treatment groups using the following specification: yihs = α + β1 SRA RA hs + β2 Jhs + μs + εihs (1) where yihs is the outcome of interest (e.g., farm child labor) for individual i from household h in strata s; Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 SRA hs RA and Jhs denote indicators for the household’s single- and joint-account treatment status. μs represents strata dummies from the sample stratifying variables, while ε ihs is the idiosyncratic mean-zero error term.2 Given the random assignment to treatment and the lack of differential attrition across treatment groups, the OLS framework provides unbiased estimates of β 1 and β 2 . Due to the imperfect compliance with the treatment assignment and in order to consider any potentially nonrandom compliance decisions, the study also estimates the effective mean treatment effect on the treated (TOT) through two-stage least squares: ˜ 2 Jhs + μs + εihs ˜ 1 Shs + β yihs = α + β (2) ˜ 2 represent the average effect of actual ownership of single and joint ˜ 1 and β In this specification, β savings accounts on the outcomes of interest. According to unitary models of household allocation, β˜1 − β˜ 2 should be equal to 0 since changes in the sharing rules of household resources should not af- fect schooling or labor decisions. Yet, since accounts take-up are endogenous variables in equation (2), the study instruments the actual take-up variables of single and joint accounts with the corresponding randomly allocated treatment indicators, Shs = π01 + π11 SRA RA hs + π21 Jhs + μs + ε1, ihs (3) Jhs = π02 + π12 SRA RA hs + π22 Jhs + μs + ε2,ihs (4) Under IV conditions, one needs robust partial correlations between the instrumental variables and en- dogenous regressors in equations (3) and (4). The first-stage results for the IV estimation are presented in table S1.4 in the supplementary online appendix. The analysis rejects the null of weak instruments at conventional levels by looking at the F-values of the multivariate regression and the p-values of the parameters of interest. Identification conditions also required the lack of correlations between the instru- mental variables and the error term in equation (3) to estimate consistent parameters β ˜ 2 . This ˜ 1 and β condition is guaranteed by the random allocation of households to single and joint saving groups. All qualitative results and conclusions emerging from this study are similar regardless of the estimation of ITT and TOT parameters, and thus, the analysis focuses in the text on reduced-form/intent-to-treat esti- mates in line with standard practices, while reporting the corresponding TOT parameters in table S1.5 in the supplementary online appendix. The analysis discusses the TOT results only in those few cases where there are differences in the statistical significance of the parameters with respect to the ITT ones.. Importantly, while information on socio-demographic and outcome variables is reported by heads of households 27 months after the start of the treatment, indicators for treatment assignment, take-up, and 1 ITT parameters can underestimate the efficacy of an intervention if some treated individuals do not adhere to the initial random assignment. In the presence of heterogeneous impacts, the reported ITT mean effect is not representative of group mean impacts for participants who adhere to or switch from the assigned treatment. 2 Following Bruhn and McKenzie (2009), the study used 16 strata dummies from the sample stratifying variables (4 Fair- trade Cooperatives x high/low baseline coffee production levels x gender of head of household). Similar quantitative and qualitative findings emerge from an alternative specification with no stratifying or control variables. The World Bank Economic Review 421 stratification variables come only from administrative sources. All tables’ first and second rows report the single and joint accounts’ treatment impacts following equation (1). Because the unit of randomization is the household, clustered standard errors at the household level are presented when more than one observation per household is included in the estimation sample (schooling and child labor outcomes). Otherwise, robust standard errors are presented. The p-values for both the test of equal treatment effects for single- and joint-account groups and the pooled F-test for the joint significance of the two terms are presented at the bottom of the tables, as well as the mean of the corresponding outcome variable Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 in the control group and the estimation sample size. In any study of this type, where there are several outcomes, there is a potential problem of overinterpreting any single significant result. To avoid that, the analysis reports multiple outcomes across several dimensions of interest, ensuring no selection of outcomes based on significance levels. Moreover, following standard practices for each “family” of outcomes, the study reports a normalized index of all the outcomes in the family taken together.3 Finally, the analysis reports the sharpened False Discovery Rate (FDR) q-values for each index outcome to adjust for multiple hypotheses testing following the Anderson procedure (Anderson 2008). Table S1.6 in the supplementary online appendix describes all outcomes used in this study. Finally, all findings, insights, and conclusions reported in section 5 are robust to changes in the specification model that include as control covariates a small set of unbalanced baseline variables and changes to the sample estimation that include only male- headed households and co-resident couples at the time of the follow-up survey. These results reported for the index of each family of outcomes are shown in tables S1.12, S1.13, and S1.14 in the supplementary online appendix. Impacts on Bank Monetary Savings This section makes use of household survey data on ownership of bank savings accounts to complement the administrative data-driven analysis of take-up and usage given in section 4. Columns 1–3 in table 2 depict intent-to-treat effects on three outcomes of interest: ownership of bank savings accounts, size of bank deposits, and size of bank balances, 27 months following the treatment. Relative to administrative bank data, this survey-based assessment includes information from only one specific period and is prone to measurement error. Yet, at the same time, it provides bank information from accounts owned by heads of households and spouses in any bank, whether held at the partner bank or not. Estimation is performed at the household level after aggregating information from bank savings accounts owned by heads of households, spouses, or both. While the first outcome of interest is not sensitive to extreme values, the size of savings deposits and balances shows a skewed distribution due to the presence of outlier values.4 For this reason, the analysis presents point estimates after taking the inverse hyperbolic sine transformation for all monetary outcomes in table 2, while table S1.7 in the supplementary online appendix shows results in levels (with 98 percentile truncation). The findings and conclusions are generally aligned regardless of how the outlier values are deal with. Column 1 in table 2 shows intent-to-treat effects on the ownership of formal deposit accounts in any bank of 29 and 32 percentage points for single and joint account treatment groups, both statistically significant at the 1 percent level. These results are driven by an increase in the likelihood of account ownership in the partner bank, showing sizable and significant effects of 48 and 54 percentage points in 3 For each “family” of outcomes, the study reports a standardized index of dependent variables by computing a simple average of all (normalized) outcome variables in the corresponding family. The analysis normalized variables by sub- tracting the mean in the control group and dividing by the standard deviation in the control group. The resulting index is normalized relative to the control group’s mean for ease of interpretation. After computing these indexes, the study reports the estimated regression coefficients of treatment dummies on the standardized index of dependent variables. 4 Figure S1.1 in the supplementary online appendix shows the CDF of mean deposits and mean balances, two outcomes that follow highly skewed non-normal distributions. In both cases, the CDF for the treated group dominates the CDF for the control group. Around 2 percent of the sample within each group has mean balances above 30,000 birr. 422 Table 2. ITT Impacts on Savings Total Total Index of saving saving dependent Bank savings Other saving deposits deposits balances variables Has Saving Saving deposit deposits in balances in Under the account in any bank any bank mattress ROSCAS Other Bank + all Bank + all any bank (Birr) (Birr) (Birr) (Birr) (Birr) other (birr) other (birr) (1) (2) (3) (4) (5) (6) (7) (8) (9) inverse hyperbolic sine transformation Single account 0.298∗∗∗ 1.202∗∗∗ 2.022∗∗∗ −0.807∗∗∗ −0.122 0.087 0.087 0.933∗∗ 0.501∗∗ group (0.035) (0.294) (0.290) (0.283) (0.224) (0.200) (0.200) (0.490) (0.229) [0.083] Joint account 0.318∗∗∗ 1.008∗∗∗ 1.951∗∗∗ −0.277 0.167 −0.053 −0.053 0.867∗∗ 0.533∗∗ group (0.035) (0.291) (0.286) (0.283) (0.220) (0.198) (0.198) (0.444) (0.223) [0.070] p-value: 0.519 0.485 0.774 0.033 0.137 0.439 0.439 0.912 0.857 Tsingle = Tjoint p-value: Joint 0.000 0.000 0.000 0.011 0.328 0.738 0.738 0.041 0.043 significance Mean of control 0.360 2.195 2.793 3.240 4.393 1.384 6.640 7.471 group Std. dev. 0.481 3.795 3.964 4.005 3.031 2.849 3.269 3.341 N 1166 1166 1166 1166 1166 1166 1166 1166 Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Robust standard errors in parenthesis. Regression specification includes strata dummies from the sample stratifying variables. The inverse hyperbolic sine transformation is applied to all continuous (birr) variables (columns 2–8). The standardized index of variables is computed as a simple average of the z-scores of the dependent variables in columns 1–7 after standardizing each variable by subtracting the mean and dividing by the standard deviation in the control group. The exchange rate US$/birr was around 27 in 2018. Anderson’s sharpened False Discovery Rate (FDR) q-values for multiple hypotheses in brackets. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Galdo Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 The World Bank Economic Review 423 the single and joint account treatment groups. Moreover, columns 2 and 3 show sizable and statistically meaningful mean effects on bank monetary deposits in the past 12 months before the survey date and bank monetary balances for the single and joint account treatment groups. The magnitude of the treatment effects is similar for both treatment groups and statistically significant at the 1 percent level. None of these results could arise from the subsidies themselves, as they represent around 3 percent of the average bank saving balances 27 months after the intervention. When measuring the bank savings impacts in levels in table S1.7 in the supplementary online appendix, columns 2 and 3 show sizable and statistically Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 meaningful mean effects on monetary deposits (61 percent) in the past 12 months before the survey date and monetary balances (69 percent) for the single account treatment group. The joint treatment group’s mean gains are 31 percent and 32 percent, although measured without statistical significance due to the large outcome variance. The p-value does not reject the null of equal effects for single and joint account treatment groups. One important question is whether these large gains in bank savings crowded out informal (“under the mattress”) savings or any other form of family and community–saving networks (that is, ROSCAs) that could lead to the breakdown of social insurance. Columns 4, 5, and 6 in table 2 report the ITT mean effects for savings deposits associated with ownership of other types of savings’ vehicles in the past 12 months before the survey. While the study finds negligible and statistically insignificant impacts on monetary savings at other microfinance institutions and cooperatives, it does see negative and statistically significant reductions at home (“under the mattress”). The point estimates are statistically significant at the 1 percent level only for the single account treatment group. Consistently, the p-values reject the null of equal coefficients between single and joint account treatment groups. Looking at the corresponding results in levels in table S1.7 in the supplementary online appendix, one observes a higher mean reduction in savings at home among households assigned to the single account treatment group (31 percent reduc- tion) than the joint account treatment group (22 percent reduction). The study also observes small but meaningful reductions in savings at ROSCAS, particularly for households assigned to the single account treatment group. Column 8 in table 2 shows the point estimates for the overall saving balances. One observes simi- lar and statistically meaningful impacts for both treatment groups when applying the inverse hyperbolic sine transformation to total monetary balances. The corresponding column 8 in table S1.7 in the supple- mentary online appendix shows positive effects for the single (988 birr) and joint (378 birr) treatment groups. While the point estimates in levels are statistically significant only for the single account treat- ment group, the null of no differential impacts by type of savings account is not rejected. Furthermore, column 9 in table 2 shows the estimated effects associated with the family index of eight normalized variables. There is strong evidence of saving accumulation patterns equivalent to 0.50 and 0.53 standard deviations for the single and joint treatment groups, a result statistically significant at the 5 percent level. The unadjusted and Anderson’s adjusted q-values reject the null of zero mean effects for this family of outcomes. Impacts on Female Beneficiaries’ Decision-Making Power This subsection assesses the effects of bank savings product design on women’s decision-making power across different dimensions. In columns 1–3 in table 3, the analysis presents ITT mean effects for three outcomes related to females’ autonomy and effective control over bank savings accounts: ownership of any single or joint savings account in any bank, an indicator that equals 1 if a woman has equal or more decision-making power role on household bank deposits relative to her male spouse, 0 otherwise, and a similar indicator for bank savings withdrawals. These three variables refer to a recall period of 12 months before the survey takes place and collected 27 months following the intervention. Looking at the ITT estimated parameters, one observes sizable and statistically significant mean gains in women’s management and control of savings for single and joint saving design products. For each one of these 424 Table 3. ITT Impacts on Decision-Making Power for Women Decision-making Ownership of Decision-making on bank Time Index of bank acc in any on bank deposits in withdrawals in any Financial allocation Production dependent bank any bank bank index index index variables (1) (2) (3) (4) (5) (6) (7) Single account 0.178∗∗ 0.141∗∗∗ 0.174∗∗∗ 0.118∗ 0.048 0.027 0.407∗∗∗ (0.026) (0.031) (0.032) (0.069) (0.070) (0.070) (0.080) [0.001] Joint account 0.378∗∗∗ 0.285∗∗∗ 0.333∗∗∗ 0.164∗∗ 0.084 0.077 0.908∗∗∗ (0.028) (0.033) (0.033) (0.066) (0.068) (0.068) (0.081) [0.001] p-value: 0.000 0.000 0.000 0.405 0.537 0.411 0.000 Tsingle = Tjoint p-value: Joint 0.000 0.000 0.000 0.046 0.468 0.487 0.000 Significance Dep. var. mean in 0.072 0.160 0.156 0.000 0.000 0.000 control group Std. dev. 0.260 0.367 0.364 1.000 1.000 1.000 N 1094 1093 1092 1095 1092 1094 Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Robust standard error in parenthesis. Regression specification includes strata dummies from the sample stratifying variables. Empowerment indexes are computed as a simple average of the z-scores of relevant variables after standardizing each variable by subtracting the mean and dividing by the standard deviation in the control group. For the financial index, the analysis considers two variables: decision-making power on spending farm and nonfarm household income. For the time allocation index, the study used five variables: decision-making power on sending children to school, assigning children to work, assigning children to household chores, own farm labor participation, own farm labor participation outside the house. For the production empowerment index, the analysis includes five variables: decision-making power on buying/renting farm tools and equipment, selecting crops, negotiating the price of coffee, using agriculture inputs, and attending FT Coops meetings. The index of dependent variables is calculated as a simple average of the z-scores of the dependent variables included in columns 1–6 after standardizing each variable by subtracting the mean and dividing by the standard deviation in the control group. Anderson’s sharpened False Discovery Rate (FDR) q-values for multiple hypotheses in brackets. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Galdo Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 The World Bank Economic Review 425 outcomes, the magnitude of these impacts is around two times higher in households assigned to the joint account treatment group relative to the single account treatment group: 0.38 vs. 0.18 for ownership of bank account, 0.28 vs. 0.14 for decision-making power for bank deposits, and 0.33 vs. 0.17 for decision- making power for bank withdrawals. As a result, and across these three intermediate outcomes, the p- value for evaluating the equality of single and joint accounts parameters decisively rejects the null. The meaningful results for the single account treatment group are explained by the presence of female heads of households who opened a single account in the partner bank and female spouses who own individual Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 or shared deposit bank accounts in other banks. Next, the study assesses whether this effective control of bank savings by women in the treated house- holds leads to increased decision-making power within the household in three important domains: con- trol over financial matters, agricultural production, and time allocation. Importantly, these outcomes are measured without having a direct link with bank account ownership or management of the savings, and thus they are not mechanically related to an increase in savings account usage by female beneficia- ries. The financial empowerment index is based on two intermediate outcomes that refer to women’s decision-making power over allocating agricultural and nonagricultural revenues. For the time-allocation empowerment index, the analysis used five intermediate variables related to decision-making power over sending children to school, assigning children’s time to farm work, assigning children’s time to household chores, and one’s labor participation in the household farm and outside the house. For the agricultural production empowerment index, the analysis used female beneficiaries’ decision-making power over five intermediate variables: buying/renting farm tools/equipment, selecting crops, negotiating the price of cof- fee crops, using agricultural inputs such as fertilizer and pesticides, and attending Fairtrade cooperative meetings. Each intermediate variable is categorical and receives the value of 2 if she is the only decision maker or the most important decision maker, 1 if she equally shares the decision with a male spouse, and 0 otherwise. A full description of these variables is given in table S1.6 in the supplementary online appendix. The analysis computes a standardized index for each empowerment domain by computing a simple average of the z-scores. The index is normalized relative to the control group’s mean for ease of interpretation. Some original variables have missing survey values, and thus the sample size is re- duced slightly. It is important to emphasize that women self-report information on these intermediate outcomes. By looking at column 4 in table 3, one observes positive and statistically significant impacts on women’s decision-making power for both single- and joint-account treatment groups: while mean effects for the former is 0.12 standard deviations and statistically significant at the 10 percent level, the latter show mean effects of 0.16 standard deviations, a result statistically significant at the 5 percent level. The p-value for the test of equality between single and joint treatment group parameters does not reject the null of equal impacts. Columns 1–3 in table S1.8 in the supplementary online appendix show the impacts of each one of the two subcomponents of the financial index. Positive impacts are led by increased decision-making power on the allocation of nonagricultural household income (0.11 SD) followed by decision-making power on agricultural income (0.08 SD), both statistically significant at the 1 percent and 10 percent levels. A positive though less pronounced effect emerges when considering decision-making power in the time allocation and agricultural production domains for which the analysis cannot reject the null of no effects. In column 5 in table 3, one observes the ITT mean impact for time-allocation is 0.08 SD above the mean of the control group for women in the joint account treatment group, while corresponding impacts for the single account treatment group are half of that and lack statistical significance. The p-value for testing equal parameters across saving products does not reject the null. By looking at the impacts for each one of the five subcomponents in table S1.8 (columns 4–8) in the supplementary online appendix, one can observe positive impacts for all intermediate outcomes. However, the only component that shows a statistically meaningful effect is women’s decision-making power on their labor supply outside the household farm 426 Galdo (0.12 SD). The analysis cannot distinguish between the children’s time allocation variables between boys and girls, which potentially dilutes meaningful gains in women’s decision-making power over girls’ time allocation. Furthermore, the agricultural production index results reported in column 6 in table 3 show a similar picture. Mean impacts are 0.08 SD for women beneficiaries in the joint account treatment group, while corresponding impacts for the single account treatment group reach only 0.03. Neither is statistically informative, nor does the p-value for testing equal parameters across saving products reject the null. Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Looking at the impacts for each of the five subcomponents in table S1.8 in the supplementary online appendix (columns 9–13), one can observe positive impacts for all intermediate outcomes with the larger impacts associated with the joint account group relative to the single account group. However, these differences are not statistically informative, as reported by the p-values at the bottom of the table. This is a setting in which gender segmentation of work is rooted in the cultural and historical influence of East Africa’s triple heritage, that is, African, Islamic, and colonial (Bass 2004), where a man is considered “the” primary decision maker in agriculture production. Agriculture tasks in coffee production require some specific skills and specialization. Achieving meaningful effects on female decision-making power in agricultural production within the household would have required an important shift in family production practices and a speedy change in the household productive specialization. Column 7 in table 3 groups all variables from columns 1–6 in a composite index of standardized outcomes. One observes a mean impact of 0.91 SD for female beneficiaries in the joint account treatment group, a value more than twice the size of that observed for female beneficiaries in the single account group (0.41). This leads to a rejection of equal mean effects across these treatment groups. All results are statistically significant at the 1 percent level, even after using Anderson’s adjusted q-values for multiple hypotheses. An important question is whether these sizable impacts on the composite index of women’s empowerment is driven by women with lower (below the median) or higher (above the median) levels of financial decision-making power in the baseline. Table S1.9 in the supplementary online appendix reports that women with lower financial empowerment in the baseline show larger gains in the composite index of decision-making power 27 months after the intervention. These differences, however, are not statistically meaningful. This result suggests that the power of joint savings accounts to advance women’s empowerment is not driven by households with particularly high or low levels of baseline decision-making power, while initial low levels of decision-making power do not impede strong gains in women’s decision- making power status. Impacts on Child Schooling One would expect that women’s (co)ownership of formal savings accounts might benefit child invest- ments in education if households are credit-constrained. Table 4 presents the ITT mean effects for three schooling-related outcomes of interest: whether a child is currently enrolled in school, daily hours of school attendance in the past seven days before the survey date, and the overall school attainment (“years of formal school”). These variables are collected 27 months following the outset of the treatment and reported by the heads of households on behalf of each child in school aged 6 to 16. Differential but uncertain ITT impacts emerge across treatment groups, while noticeable qualitative differences between the ITT and TOT parameters are also observed. Positive school effects emerge for girls in households assigned to the joint account treatment group as the corresponding ITT gains reach 0.036 (4.3 percent) for currently attending school and 0.152 (4.5 percent) for daily hours of school at- tendance. On the other hand, negative mean effects are observed for girls in the single account treatment group. As a result, the p-values for the test of equality of parameters between single and joint account treatment groups show statistically differential impacts for girls for the first outcome. For boys, on the other hand, the study observes negative and not meaningful effects for both single and joint account treatment groups. This leads the p-values for the test of equality of parameters between boys and girls Table 4. ITT Impacts on Schooling Outcomes by Gender of the Child Annual school Index of dependent Currently attending school Daily hours of school Years of formal schooling expenditures (invlog birr) variables Boys Girls Boys Girls Boys Girls Boys Girls Boys Girls (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Single account −0.035 −0.020 −0.141 −0.037 0.050 0.054 −0.104 −0.105 −0.077 −0.024 The World Bank Economic Review (0.034) (0.032) (0.145) (0.133) (0.207) (0.203) (0.213) (0.212) (0.092) (0.087) [0.450] [0.644] Joint account −0.019 0.036 −0.132 0.152 0.176 0.225 −0.026 0.229 −0.010 0.124 (0.031) (0.031) (0.131) (0.128) (0.194) (0.205) (0.196) (0.199) (0.082) (0.082) [0.644] [0.210] p-value: 0.616 0.043 0.947 0.107 0.472 0.347 0.689 0.063 0.433 0.050 Tsingle = Tjoint p-value: Joint 0.602 0.120 0.533 0.233 0.607 0.483 0.875 0.160 0.657 0.107 Significance Test p-value: 0.752 0.573 0.988 0.966 0.416 Tsingle (boys)=Tsingle (girls) p-value: 0.195 0.101 0.860 0.333 0.091 Tjoint (boys)=Tjoint (girls) Dep. var. mean in 0.845 0.836 3.444 3.374 2.926 2.872 4.791 4.790 control group Std.dev. 0.361 0.370 1.515 1.535 2.400 2.561 2.312 2.367 N 1013 1013 1011 1011 1015 1021 1015 1021 Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Clustered standard errors at the household level in parenthesis. Inverse hyperbolic sine transformation applied to school expenditures (birr). Regression specification includes strata dummies from the sample stratifying variables. The index of dependent variables is calculated over the boys’ and girls’ subsamples separately as a simple average of the z-scores of the four dependent schooling variables after standardizing each variable by subtracting the mean and dividing by the standard deviation in the control group. Anderson’s sharpened False Discovery Rate (FDR) q-values for multiple hypotheses in brackets. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. 427 Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 428 Galdo within the joint account group to reject the null for the second outcome of interest. The corresponding TOT parameters reported in table S1.5 in the supplementary online appendix show positive (11 per- cent) and meaningful impacts for girls in the joint account treatment group for the school attendance outcome. Furthermore, when looking at the ITT school attainment impacts in columns 5 and 6 in table 4, one observes threefold and fourfold larger impacts for boys (6.0 percent) and girls (7.8 percent) who belong to the joint account treatment group relative to those in the single account treatment group. However, Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 these point estimates lack statistical precision. If the increased decision-making power by female beneficiaries in the joint account treatment group is the mechanism that explains these positive albeit weak schooling impacts for girls of the joint account treatment group, then one would expect to observe higher school-related expenditures in households as- signed to the joint account treatment group. The study then assesses whether assignment to the joint account treatment group ultimately leads to a higher allocation of resources towards schooling expenses than the single account treatment group by child gender. For that purpose, the study has information on yearly school expenditures for each child in the sample, as reported by the heads of households 27 months following the intervention. Columns 7 and 8 in table 4 present the ITT impacts after taking the inverse hyperbolic sine transformation for this monetary outcome. Girls from households assigned to the joint account treatment group are the only group with positive impacts of 0.23 (26 percent). The direction of the impacts is reversed for school expenses associated with the single account treat- ment group. The p-value (0.063) for the test of equality of parameters confirms the distinctive responses to school expenditures between the single and joint account treatment groups for girls. For boys, on the other hand, the size of the estimated parameters for school expenditures is negative and measured imprecisely. Finally, columns 9 and 10 in table 4 show the composite index of standardized schooling outcomes. One can observe negative impacts for boys for either arm of the saving intervention. For girls, on the other hand, positive albeit uncertain impacts are driven by households assigned to the joint-account treatment group that shows mean gains of 0.12 SD. The p-values for the test of equality of parameters reported at the bottom of the table confirm the distinctive schooling impacts according to the specific saving product for girls (p-value = 0.050). Likewise, the p-values for the test of equality of parameters between boys and girls within the joint account group also reject the null (p-value = 0.091). Finally, the corresponding TOT impacts reported in table S1.5 in the supplementary online appendix show statistically significant impacts of 0.28 standard deviations for girls in the joint account treatment group. All in all, these positive albeit weak results are in line with independent evidence for Ethiopia that points out that the allocation of schooling resources within the household is targeted towards girls when female spouses experience positive changes in autonomy and control of household resources (e.g., Quisumbing and Maluccio 2000; Gebremedhin and Mohanty 2016). The study acknowledges, however, that other mechanisms can be in play given the weak schooling results. Impacts on Child Labor This section presents farm child labor effects for children aged 6 to 16 at both the extensive and intensive margins of work. Labor information, reported by heads of households on behalf of each family member, is based on a relatively long rather than a short questionnaire design. The survey contains 12 questions to elicit information about the specific farm and nonfarm labor activities. The recall time length of these labor variables refers to the past 30 days before the survey date, which overlaps with the coffee harvesting season, the busiest season of the agricultural calendar. In columns 1 and 2 in table 5, one observes that joint ownership of bank savings accounts triggers child labor responses among agricultural households. Statistically meaningful positive impacts on farm labor participation (15 percent) and hours worked (19 percent) emerge for children in households allocated to Table 5. ITT Impacts on Farm Child Labor, Last 30 Days Index for Index for All children Boys Girls boys girls Hours Hours Hours Participation worked Participation worked Participation worked (1) (2) (3) (4) (5) (6) (7) (8) The World Bank Economic Review Single account −0.019 −0.135 −0.053 2.228 0.012 −1.954 −0.013 −0.024 (0.037) (1.974) (0.046) (2.618) (0.047) (2.349) (0.097) (0.090) [0.685] [0.644] Joint account 0.077∗∗ 3.607∗ 0.063 4.486∗ 0.083∗ 2.793 0.166∗ 0.146 (0.036) (2.006) (0.044) (2.494) (0.046) (2.481) (0.091) (0.092) [0.130] [0.196] p-value: 0.003 0.038 0.003 0.362 0.097 0.021 0.041 0.035 Tsingle = Tjoint p-value: Joint 0.009 0.075 0.012 0.197 0.126 0.070 0.070 0.086 significance p-value 0.233 0.153 0.926 Tsingle (boys)=Tsingle (girls) p-value 0.714 0.567 0.910 Tjoint (boys)=Tjoint (girls) Dep. var. mean in 0.525 19.047 0.603 20.828 0.446 17.223 control group St. dev. 0.499 26.843 0.490 26.054 0.498 27.561 N 2048 2048 1024 1024 1024 1024 Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Clustered standard errors at the household level in parenthesis. Farm participation and farm hours worked do not include household chores activities. The recall time length for these outcome variables refers to the 30 days before the survey date. Regression specification includes strata dummies from the sample stratifying variables. The index of dependent variables is calculated separately over the boys’ and girls’ subsamples as a simple average of the z-scores of the dependent variables, farm participation and farm hours worked. Anderson’s sharpened False Discovery Rate (FDR) q-values for multiple hypotheses in brackets. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. 429 Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 430 Galdo the joint account treatment group. This is equivalent to an increase of almost four hours of child labor per month. In a context in which the daily hours of school attendance by children are less than four hours, schooling and child labor are not mutually exclusive, and thus, changes in the time devoted to child labor do not necessarily offset school time. For children from households assigned to the single account treatment group, on the other hand, child labor impacts are statistically not informative. As a result, the p-values for the equality of the point estimates by type of savings account reject the null. Columns 3 to 6 display point estimates according to the child’s gender. One can observe positive and Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 mostly comparable effects for boys and girls who belong to households assigned to the joint account treatment group, although the statistical significance of the individual point estimates is weakened due to smaller sample sizes. The assessment of the p-values for the equality of the point estimates between boys and girls, thus does not reject the null for both single and joint treatment groups, as observed at the bottom of the table. Columns 7 and 8 in table 5 report the mean impacts for the index of dependent variables estimated separately by the child’s gender. The ITT point estimates associated with the joint account treatment group show 0.17 and 0.15 standard deviations for boys and girls, a result statistically significant at the 10 percent level only for the boys. On the other hand, the labor index impacts associated with the single account treatment group are negligible and statistically different from the corresponding estimates for the joint account treatment group for boys (p-value = 0.041) and girls (p-value = 0.035). By looking at the corresponding TOT impacts depicted in table S1.5 in the supplementary online appendix, one observes stronger and more meaningful positive impacts for boys and girls that belong to the joint account treatment group. The mean impacts for the index of dependent variables associated with the joint account treatment group are 0.33 and 0.31 standard deviations for boys and girls, a result statistically significant in both cases. On the other hand, the point estimates for the single account treatment group are mostly negative and statistically uninformative. Table S1.10 in the supplementary online appendix presents alternative indicators of child labor mea- sured over the full agricultural calendar to assess whether the increased child labor in the last 30 days before the survey was also observed in the full agricultural season. The analysis uses three additional vari- ables of child labor: the average number of months a child worked over the past 12 months, the average number of days per week a child worked over the past 12 months, and the average number of hours per day a child worked over the past 12 months. Consistently, results show positive and statistically signifi- cant effects on child labor for children in households assigned to the joint account treatment group across these three variables. The p-values that test the equality of mean effects between single and joint treat- ment groups decisively reject the null for these variables. These results show evidence that the expansion of farm child labor among households assigned to the joint account treatment happened over the entire agricultural cycle. On the other hand, unreported results for nonfarm child labor (that is, wage labor, nonfarm household business labor) show negligible results since only a small fraction of children (around 2 percent) work outside the household farm. On the other hand, most children in the sample spent time on household chores, with an average of around 15 and 19 hours per week for boys and girls. Following the same specification, the analysis computes the corresponding treatment effects for the time children spend on household chores. Table S1.11 in the supplementary online appendix depicts these additional results. One observes small and statistically uninformative treatment effects for single and joint account treatment groups. Understanding the Child Labor Effects The impacts of savings encouragement interventions on child labor are theoretically ambiguous and de- pend on the relative contributions of alternative mechanisms that might offset each other. This section assesses three competing mechanisms: complementarities with adult labor, lumpy investments in labor- The World Bank Economic Review 431 intensive agricultural resources, and changes in household income. Figure S1.2 in the supplementary online appendix presents a summary table of mechanisms. Adult Labor Supply Responses Child labor can respond to expansions of adult work within households as work complementarities in the farm production function can emerge (Edmonds 2007). Theoretically, the link between formal sav- ings account ownership and expansions of labor supply is consistent with collective household bargaining models (Field et al. 2021). In this framework, women’s financial inclusion leads to gains in their decision- Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 making power. Greater bargaining power raises a woman’s labor supply because of changes in gender norms within the households. Women became more accepting of female work, while their husbands per- ceived fewer social costs to having a partner who works (Field et al. 2021). This prediction is inconsistent with neoclassical labor supply models with unitary decision making wherein greater bargaining power would reduce female labor supply through the positive income effect. In the present context, it is expected that control over household assets would impact female beneficiaries’ decisions on how to allocate one’s labor effort as customary rules provide incentives for women to bring more assets into the household and thus to have more say in the household decision making (Jones 1986; Fafchamps and Quisumbing 2002). Since in Ethiopia child labor is seen as an extension of, and subordinate to, women’s work (e.g., Bass 2004; Galdo et al. 2021), one can expect labor supply expansions by female beneficiaries may, in turn, lead to changes in the value of children’s time due to changes in the household production func- tion (Morduch 1999). As women work more, they might need to do something with their out-of-school children. This might also explain the expansion of child labor, provided women are working with their children. Table 6 shows farm labor responses by male and female beneficiaries at both the extensive and intensive margins of work. Like the child labor information, the analysis follows the same labor instrument and the same recall time length that refers to 30 days before the survey date. Results show positive and significant effects on farm labor participation (10 percent) and hours worked (15 percent) for female beneficiaries assigned to the joint account treatment group. On the other hand, the analysis observes negligible and un- informative mean effects for the single account treatment group, which leads to a rejection of the equality of coefficients by the treatment group, as reported by the p-values of the test. Moreover, male beneficia- ries assigned to the joint account treatment group also show positive farm labor supply responses, albeit smaller and statistically uncertain. As a result, the p-values reported at the bottom of table 6 do not reject the null of equal effects by gender of the beneficiaries. Finally, the last two columns in table 6 report the mean effects for an index of dependent variables estimated separately for male and female beneficiaries. One observes statistically significant impacts for female beneficiaries assigned to the joint account treat- ment group, equal to 0.14 standard deviations. This result is statistically informative, even considering Anderson’s adjusted p-values for multiple hypotheses. Overall, these results are consistent with evidence that shows the connection between female empowerment and female labor supply expansions (e.g., Field et al. 2021). More generally, the study’s results align with studies that show positive adult labor supply responses when formal savings accounts are available (e.g., Callen et al. 2019). Importantly, the results suggest work complementarities between adult and child labor in the farm production function among small-holder farming households. Labor-Intensive Agricultural Investments The relaxation of financial constraints might lead to labor-intensive investments that can increase the demand for child labor (e.g., Wydick 1999; Edmonds and Theoharides 2020). This section presents the mean effects of this saving initiative on agricultural resource investments. The head of household reports the information for the last agricultural season, measured 27 months following the intervention. The one-year recall period likely reduces concerns about seasonality, although it may have exacerbated recall errors due to the absence of record-keeping, a common feature in this type of setting. The econometric 432 Table 6. ITT Impacts on Adult Farm Labor Index of Dep. Index of Dep. variables for variables for Male beneficiaries Female beneficiaries males females Hours Hours Participation worked Participation worked (1) (2) (3) (4) (5) (6) Single account 0.001 −2.258 −0.002 −0.389 −0.021 −0.008 (0.026) (4.047) (0.033) (2.570) (0.077) (0.068) [0.644] [0.685] Joint account 0.034 4.148 0.060∗ 4.953∗∗ 0.102 0.140∗∗ (0.025) (4.116) (0.032) (2.651) (0.075) (0.068) [0.238] [0.094] p-value: Tsingle = Tjoint 0.145 0.086 0.034 0.027 0.067 0.016 p-value: Joint 0.247 0.224 0.063 0.058 0.153 0.032 significance p- 0.929 0.720 0.830 value:Tsingle (male)=Tsingle (female) p- 0.391 0.783 0.557 value:Tjoint (male)=Tjoint (female) Dep. var. mean in 0.870 76.000 0.613 32.312 control group St. dev. 0.336 52.899 0.487 36.330 N 1007 1007 1099 1099 Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Robust standard error in parenthesis. Farm participation and farm hours do not include household chores activities. The recall time length for these outcome variables refers to the 30 days before the survey date. Regression specification includes strata dummies from the sample stratifying variables. The index of dependent variables is calculated separately over the men and women subsamples as a simple average of the z-scores of the dependent variables, farm participation and farm hours worked. Anderson’s sharpened False Discovery Rate (FDR) q-values for multiple hypotheses in brackets. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Galdo Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 The World Bank Economic Review 433 Table 7. ITT Impacts on Agricultural Investments New agri. Organic fertilizer: Chemical New coffee tools and Index of manure, compost fertilizer plants Seeds equipment dependent (kg/hectare) (kg/hectare) (units/hectare) (kg/hectare) (birr) variables (1) (2) (3) (4) (5) (6) Single account 120.060∗ −0.407 2.294∗∗ 0.333 0.072 0.233∗∗ Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 (63.638) (1.617) (1.096) (0.308) (7.569) (0.099) [0.070] Joint account 73.397 1.428 2.483∗∗ 0.581∗ 1.084 0.254∗∗∗ (62.028) (1.689) (1.035) (0.317) (7.411) (0.093) [0.044] p-value: 0.471 0.215 0.891 0.393 0.874 0.846 Tsingle = Tjoint p-value: Joint 0.159 0.446 0.012 0.188 0.984 0.009 Significance Dep. var. mean in 676.355 10.170 0.634 1.551 46.996 control group St. dev. 959.199 21.878 6.726 3.95 99.845 N (2% trimming) 1140 1140 1142 1142 1141 Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Robust standard error in parenthesis. Regression specification includes strata dummies from the sample stratifying variables. The exchange rate US$/birr was around 27 in 2018. The index of dependent variables is computed as a simple average of the z-scores of the dependent variables included in columns 1–7. Anderson’s False Discovery Rate (FDR) q-values for multiple hypotheses in brackets; 2 percent trimming applied to all dependent variables. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. estimation is done after trimming 2 percent of the top-value observations to reduce the influence of outlier values. Columns 1 and 2 in table 7 report mean impacts for two key agricultural inputs: organic fertilizer (that is, manure and compost) and chemical fertilizer, both measured in kilograms per hectare. Sizable and statistically significant effects emerge only for the more labor-intensive organic fertilizer for the sin- gle (18 percent) and joint (11 percent) account treatment groups. The p-value for the test of equality of coefficients does not reject the null. Moreover, columns 3 and 4 show the mean effects for the number of new coffee trees (in units per hectare) and coffee seedlings (in kilograms per hectare) purchased by farmers in the past 12 months. The ITT impacts for coffee plants are positive for both treatment groups and statistically significant. For coffee seedlings, on the other hand, one can observe positive (37 percent) and statistically significant effects for the joint account treatment group, although the equal effects across saving product design are not rejected. Column 5 reports investments in agricultural tools and equipment such as plows, slasher sickles, axes, and yokes, which are small-scale, labor-intensive agricultural produc- tion means. Results show positive but imprecisely measured effects for both treatment groups. Overall, these results suggest that agricultural investments are driven toward labor-intensive resources that involve the participation of all household members. Column 6 in table 7 aggregates these agricultural investment outcomes in an index of standardized variables. One can observe positive impacts of 0.23 and 0.25 SD for the single and joint treatment groups, measured with statistical precision by looking at the point es- timates. The null of no impacts is rejected even when using the more conservative Anderson’s adjusted q-values for multiple hypotheses. Impacts on Income The complementary role of farm adult labor and labor-intensive lumpy investments in child labor can be counterbalanced by impacts on income or wealth that could relax financial constraints, allowing house- 434 Galdo Table 8. ITT Impacts on Annual Income Nonfarm hh business Index of dependent Agriculture income (birr) (birr) Remunerated labor (birr) variables (1) (2) (3) (4) Inverse hyperbolic sine transformation Single account −0.033 0.173 −0.195 −0.031 (0.066) (0.324) (0.301) (0.078) Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 [0.644] Joint account 0.026 −0.122 −0.421 −0.070 (0.068) (0.320) (0.301) (0.081) [0.450] p-value:Tsingle = Tjoint 0.400 0.309 0.392 0.608 p-value: Joint 0.698 0.595 0.361 0.681 Significance Dep. var. mean in 9.942 2.473 2.365 control group Std. dev. 0.883 4.314 4.147 N 1166 1166 1166 Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Robust standard errors in parenthesis. Regression specification includes strata dummies from the sample stratifying variables. The inverse hyperbolic sine transformation is applied to all income variables. The index of dependent variables is computed as a simple average of the z-scores of the dependent variables included in columns 1–3. Anderson’s sharpened False Discovery Rate (FDR) q-values for multiple hypotheses in brackets. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. holds to reduce the demand for child labor (e.g., Basu and Van 1998; Edmonds 2007). Some studies have reported positive income effects associated with saving interventions in developing settings (e.g., Brune et al. 2016; Callen et al. 2019). Table 8 shows ITT effects on yearly income separately across three important categories: agriculture, nonfarm household business, and labor income. The head of the household reports all this information. Given the large variance associated with income variables, the study presents point estimates after taking the inverse hyperbolic sine transformation for all monetary outcomes. Column 1 in table 8 shows that access to bank savings accounts does not translate into higher agri- cultural income. Estimated effects are small and lack statistical precision for both single- and joint- account treatment groups. Similar null mean effects are also observed for nonfarm household busi- ness income (column 2), a marginal supplementary activity for a quarter of the sample. Moreover, la- bor income, which contributes around 9 percent to the total household income in the sample, shows negative and imprecise effects in column 3. Finally, column 4 in table 8 shows negligible point esti- mates for both treatment groups. Since the reliability of recall in agricultural data is important (Beegle, Carletto, and Himelein 2012), it is worth noting that these results are independent of the time length of the recall period used in the survey. Unreported results for a shorter length of the recalled period (e.g., 90 days) show similar findings, as is the case for alternative transformations of the income vari- ables (e.g., trimming), and aggregation of household income (e.g., total income, nonagricultural total income). There are some possible explanations for the lack of meaningful impacts on income that might help understand this result. First, agricultural income presents a remarkably high variance. While the aver- age total household income is 27,636 birr 27 months after the intervention, the standard deviation is 44,329 birr. This level of variability makes it harder to reject the true underlying relationship between income and treatment variables in this agrarian setting. Second, the sample is composed of small-holder Fairtrade farmers. The Fairtrade cooperative decides the number of coffee kilos purchased and the price The World Bank Economic Review 435 per kilo paid to the individual farmers, which cooperative boards regulate. This implies that most farm income variation among households depends on decisions made by the Fairtrade cooperative regarding the quantity of coffee purchased and the price per kilo paid to farmers. In this regard, there is no sig- nificant correlation between coffee production per household and the share of coffee purchased by the Fairtrade coffee cooperative in the sample. Put differently, Fairtrade cooperatives do not mechanically absorb or reward potential increases in coffee production and quality among treated households. Thus, modest increases in either coffee production or quality due to the reallocation of household labor among Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 these treated households do not mechanically translate into higher farm income paid by cooperatives to farmer associates. Third, the extent of labor increases in farm activities by females and children in the joint account treatment group is modest, that is, four hours per month for children and five hours per month for women. This labor supply expansion might not be enough to cause meaningful changes in farm output that mostly depends on male heads primarily in charge of agricultural tasks, including planting and seedlings. In sum, the analyses of mechanisms to explain the expansion of child labor among households as- signed to the joint account treatment suggest complementarities between adult farm labor and child labor in the household production function. These complementarities are reinforced by investments in labor- intensive agricultural inputs that likely increased the opportunity costs of child time. The lack of impacts on income that otherwise would have had the opposite effects strengthens these positive effects on child labor. 6. Conclusions This study provided small-holder farmers one-time, person-specific small subsidies to cover the monetary costs of opening single or joint formal deposit accounts in remote areas of Ethiopia. This research showed strong demand in terms of take-up and usage of formal bank deposit accounts, regardless of the saving product design, which led to important overall savings gains among farmers 27 months after the setup of the intervention. Importantly, this study highlighted the role of savings product design on intra- and inter- household behavioral responses to ownership of deposit accounts. In settings in which gender stratification of work and social lives is deeply rooted in social norms that work against women, small subsidies that effectively provided control of bank savings to women helped to achieve positive impacts on women’s financial decision-making power within the household. Thus, whether women’s financial inclusion is key to tackling Africa’s gender inequality, this intervention shows that the architecture of savings product design matters for agricultural households. Consistent with posited channels of intrahousehold bargaining models, women’s higher control of saving resources led to positive albeit weak schooling impacts within the households. These results are not observed in households with single deposit accounts. While there is a legitimate interest in the child labor policy community on the link between financial inclusion and child labor, this study showed that child labor for boys and girls systematically increased by four hours per month in house- holds assigned to the joint account treatment group. These numbers suggest that the expansion of child labor was not detrimental to schooling attainment but leisure time. Gains in financial control and autonomy by female beneficiaries assigned to the joint account treatment group are accompanied by important expansions of their farm labor supply. Because women in this rural setting direct chil- dren’s time allocation, job complementarities likely emerged, increasing the overall demand for child labor. Furthermore, in the presence of multiple-factor market failures, these job complementarity ef- fects on child labor are reinforced by the lack of effects on income and the emergence of positive im- pacts on labor-intensive agricultural investments. A relevant question from these findings is whether the increased demand for child labor affected school performance (test scores). Unfortunately, the study does not have data to answer this question, which could have added more insights into the overall 436 Galdo welfare impacts of formal savings interventions on children of agricultural households in Sub-Saharan Africa. Data Availability The data that support the findings of this study are available on the author’s website (https://sites.google .com/view/josegaldo). Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 References Aker, J., M. Sawyer, M. Goldstein, M. O’Sullivan, and M. McConnell. 2020. “Just a Bit of Cushion: The Role of a Simple Savings Device in Meeting Planned and Unplanned Expenses in Rural Niger.” World Development 128(1044772): 1–15. Anderson, M. 2008. “Multiple Inference and Gender Differences in the Effects of Early Intervention: A Reevaluation of the Abecedarian, Perry Preschool, and Early Training Project.” Journal of the American Statistical Association 103(484): 1481–1495. Angelucci, M., D. Karlan, and J. Zinman. 2015. “Microcredit Impacts: Evidence from a Randomized Microcredit Program Placement Experiment by Compartamos Banco.” American Economic Journal: Applied Economics 7(1): 151–82. Ashraf, N. 2009. “Spousal Control and Intra-household Decision Making: An Experimental Study in the Philippines.” American Economic Review 99(4): 1245–77. Ashraf, N., D. Karlan, and W. Yin. 2006a. “Tying Odysseus to the Mast: Evidence from a Commitment Savings Product in the Philippines.” Quarterly Journal of Economics 121(2): 635–72. ———. 2006b. “Deposit Collectors.” Advances in Economic Analysis and Policy 6(2): Article 5. ———. 2010. “Female Empowerment: Further Evidence from a Commitment Savings Product in the Philippines.” World Development 38: 333–44. Badstue, L. , Petesch, P., C. Farnworth, L. Roeven, and M. Hailemariam. 2020. “Women Farmers and Agricultural Innovation: Marital Status and Normative Expectations in Rural Ethiopia.” Sustainability 12(23): 9847. Bass, L. 2004. Child Labor in Sub-Saharan Africa. London: Lynne Rienner Publishers. Basu, K., and P.H. Van. 1998. “The Economics of Child Labor.” American Economic review 88(3): 412–27. Beegle, K., G. Carletto, and K. Himelein. 2012. “Reliability of Recall in Agricultural Data.” Journal of Development Economics 98(1): 34–41. Berry, J., D. Karlan, and M. Pradhan. 2018. “The Impact of Financial Education for Youth in Ghana.” World Devel- opment 102: 71–89. Boserup, E. 1965. The Conditions of Agricultural Growth. Chicago: Aldine Publishing. Browning, M., and P.-A. Chiappori. 1998. “Efficient Intra-Household Allocation: A General Characterization and Empirical Tests.” Econometrica 66(6): 1241–78. Bruhn, M., and D. McKenzie. 2009. “In Pursuit of Balance: Randomization in Practice in Development Field Experi- ments.” American Economic Journal: Applied Economics 1(4): 200–32. Brune, L., X. Gine, J. Goldberg, and D. Yang. 2016. “Facilitating Savings for Agriculture: Field Experimental Evidence from Malawi.” Economic Development and Cultural Change 64(2): 187–220. Callen, M., S. De Mel, C. McIntosh, and C. Woodruff. 2019. “What Are the Headwaters of Formal Savings? Experi- mental Evidence from Sri Lanka.” Review of Economic Studies 86(6): 2491–529. Chiappori, P.-A. 1992. “Collective Labor Supply and Welfare.” Journal of Political Economy 100(3): 437–67. de Hoop, J., and F. Rosati. 2014. “Does Promoting School Attendance Reduce Child Labor? Evidence from Burkina Faso’s BRIGHT Project.” Economics of Education Review 39: 78–96. Dercon, S., and P. Krishnan. 2000. “In Sickness and in Health: Risk-Sharing within Households in Ethiopia.” Journal of Political Economy 108(4): 688–727. Duflo, E. 2003. “Old Age Pension and Intra-Household Allocation in South Africa.” World Bank Economic Review 17(1): 1–25. Dupas, P., A. Keats, and J. Robinson. 2017. “The Effect of Savings Accounts on Interpersonal Financial Relationships: Evidence from a Field Experiment in Rural Kenya.” Economic Journal 129(617): 273–310. The World Bank Economic Review 437 Dupas, P., and J. Robinson. 2013a. “Savings Constraints and Microenterprise Development: Evidence from a Field Experiment in Kenya.” American Economic Journal: Applied Economics 5(1): 163–92. ———. 2013b. “Why Don’t the Poor Save More? Evidence from Health Savings Experiments.” American Economic Review 103(4): 1138–71. Edmonds, E. 2007. “Child Labor.” Handbook of Development Economics. Vol. 4: 3607–3709, edited byPaul , and- John . North-Holland, Amsterdam. . Edmonds, E., andTheoharides. 2020. “The Short Term Impact of a Productive Asset Transfer in Families with Child Labor: Experimental Evidence from the Philippines.” Journal of Development Economics 146: 102486. Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Fafchamps, M. 2001. “Intrahousehold Access to Land and Sources of Inefficiency: Theory and Concepts.” In Land Reform Revisited: Access to Land, Rural Poverty, and Public Action, edited by A. de Janvry, E. Sadoulet, and J.-P. Platteau, 000–000. Oxford: Oxford University Press/WIDER. Fafchamps, M., B. Kebede, and A. Quisumbing. 2009. “Intrahousehold Allocation in Rural Ethiopia.” Oxford Bulletin of Economics and Statistics 71(4): 567–599 Fafchamps, M., and A. Quisumbing. 2002. “Control and Ownership of Assets within Rural Ethiopian Households.” Journal of Development Studies 38(6): 47–82. FAO. 2015. “State of Food Security. Ethiopia Country Report.” Food and Agriculture Organization of the United Nations. Rome, Italy. Field, E., R. Pande, N. Rigol, S. Schaner, and C. Troyer-Moore. 2021. “On Her Own Account: How Strengthening Women’s Financial Control Affects Labor Supply and Gender Norms." American Economic Review 111(7): 2342– 75. Galdo, J., A. Dammert, and D. Abebaw. 2021. “Gender Bias in Agricultural Child Labor: Evidence from Survey Design Experiments.” World Bank Economic Review 35(4): 872–891. Gebremedhin, T.A., and I. Mohanty. 2016. “Child Schooling in Ethiopia: The Role of Maternal Autonomy.”PLoS ONE 11(12):1–20. Islam, A., and C. Choe. 2013. “Child Labor and Schooling Responses to Access to Microcredit in Rural Bangladesh.” Economic Inquiry 51(1): 46–61. Jack, W., and T. Suri. 2016. “The Long-Run Poverty and Gender Impacts of Mobile Money.” Science 354(6317): 1288–1292. Jones, C.W. 1986. “Intra-Household Bargaining in Response to the Introduction of New Crops: A Case Study from North Cameroon.” In Understanding Africa’s Rural Households and Farming Systems, edited by J.L. Moock, 000– 000. Boulder, CO and London: Westview Press, Kabeer, N. 2005. “Gender Equality and Women’s Empowerment: A Critical Analysis of the Third Millennium Devel- opment Goal 1.” Gender Development 13: 13–24. Karlan, D., and L. Leigh. 2014. “Loose Knots: Strong versus Weak Commitments to Save for Education in Uganda.” Center Discussion Papers 162693. Economic Growth Center. Yale University. New Haven, CN, USA. Karlan, D., R. Osei, I. Osei-Akoto, and C. Udry. 2014. “Agricultural Decisions after Relaxing Credit and Risk Con- straints.” Quarterly Journal of Economics 129(2): 597–652. Lundberg, S., and R. Pollak. 1993. “Separate Spheres Bargaining and the Marriage Market.” Journal of Political Economy 101(6): 988–1010. McCann, J. C. 1995. People of the Plow: An Agricultural History of Ethiopia, 1800–1990. Madison: University of Wisconsin Press. Morduch, J. 1999. “The Microfinance Promise.” Journal of Economic Literature 37(4): 1569–1614. Prina, S. 2015. “Banking the Poor via Savings Accounts: Evidence from a Field Experiment.” Journal of Development Economics 115: 16–31. Quisumbing, A., and J. Maluccio. 2000. “Intrahousehold Allocation and Gender Relations: New Empirical Evidence from Four Developing Countries.” FCND Discussion Paper 84, International Food Policy Research Institute. Wash- ington D.C, USA. Schaner, S. 2015. “Do Opposites Detract? Intrahousehold Preference Heterogeneity and Inefficient Strategic Savings.” American Economic Journal: Applied Economics 7(2): 135–74. ———. 2017. “The Cost of Convenience? Transaction Costs, Bargaining Power, and Savings Account Use in Kenya.” Journal of Human Resources 52(4): 919–45. ———. 2018. “The Persistent Power of Behavioral Change: Long-Run Impacts of Temporary Savings Subsidies for the Poor.” American Economic Journal: Applied Economics 10(3): 67–100. 438 Galdo Tarozzi, A., J. Desai, and K. Johnson. 2015. “The Impacts of Microcredit: Evidence from Ethiopia.” American Eco- nomic Journal: Applied Economics 7(1): 54–89. Thomas, D. 1990. “Intra-household Resource Allocation: An Inferential Approach.” Journal of Human Resources 25(4): 635–64. von Braun, J., and P. Webb. 1989. “The Impact of New Crop Technology on the Agricultural Division of Labor in a West African Setting.” Economic Development and Cultural Change 37(3): 513–34. World Bank. 2008. “World Development Report: Agriculture for Development.” The World Bank Group, Washington, DC, USA. Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 ———. 2017. Global Findex Database 2017. The World Bank Group, Washington, DC, USA. Wydick, B. 1999. “The Effect of Microenterprise Lending on Child Schooling in Guatemala.” Economic Development and Cultural Change 47(4): 853–69. Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Supplementary Online Appendix (Joint) Bank Savings, Female Empowerment, and Child Labor in Rural Ethiopia Jose Galdo S1: Additional Figures and Tables Figure S1.1. CDF of Mean Deposits and Balances, 27 Months Later. Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Source: Author’s elaboration based on follow-up survey data collected in March/April 2018. Figure S1.2. Joint Savings Accounts Mechanisms. Source: Author’s elaboration. Table S1.1. Balancing Test Control p-value p-value p-value Treatment mean Tsingle =Tcontrol Tjoint =Tcontrol Tsingle =Tjoin mean (std. dev.) (std. dev.) (1) (2) (3) (4) (5) Panel A: Household socio-demographics Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Household size 5.59 5.64 0.71 0.74 0.96 (2.06) (1.98) Christian (%) 0.55 0.57 0.82 0.46 0.57 (0.49) (0.49) Muslim (%) 0.44 0.42 0.68 0.39 0.61 (0.49) (0.49) Ownership of dwelling (%) 0.97 0.96 0.55 0.33 0.68 (0.16) (0.18) Access to water from protected 0.78 0.76 0.45 0.53 0.89 well/spring (%) (0.41) (0.43) Access to water from unprotected 0.16 0.20 0.19 0.25 0.87 well/spring (%) (0.36) (0.39) Main source of lighting is 0.23 0.21 0.78 0.45 0.59 electricity/generator (%) (0.42) (0.41) Mud floor (%) 0.72 0.69 0.45 0.30 0.75 (0.44) (0.46) Corrugated iron roof (%) 0.77 0.79 0.28 0.76 0.38 (0.42) (0.40) Pit latrine ventilated (%) 0.20 0.17 0.06 0.91 0.05 (0.40) (0.38) Pit latrine not ventilated (%) 0.78 0.81 0.04 0.67 0.01 (0.41) (0.39) Owns a mobile phone (%) 0.66 0.64 0.52 0.59 0.91 (0.47) (0.48) Household wealth index −0.00 0.03 0.86 0.74 0.87 (1.79) (1.71) Panel B: Household Finances / Savings Yearly average total monthly 1228.22 1361.85 0.49 0.10 0.25 income (birr) (1315.04) (1877.12) Yearly average agriculture monthly 820.78 994.96 0.39 0.00 0.01 income (birr) (927.89) (1635.45) Monthly income allocated to food 596.50 631.28 0.75 0.42 0.47 (birr) (590.49) (1120.97) Table S1.1. Continued Control p-value p-value p-value Treatment mean Tsingle =Tcontrol Tjoint =Tcontrol Tsingle =Tjoin mean (std. dev.) (std. dev.) (1) (2) (3) (4) (5) Monthly income allocated to agric. 56.32 55.46 0.93 0.81 0.72 Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 inputs & equip. (birr) (149.08) (138.44) Monthly income allocated to 73.85 60.49 0.06 0.52 0.10 school/education (birr) (145.77) (107.98) Monthly income allocated to hired 92.94 84.26 0.31 0.99 0.25 labor (birr) (223.16) (163.13) Monthly income allocated to 78.03 103.69 0.36 0.35 0.93 monetary savings (birr) (330.24) (424.21) Monthly income allocated to loan 16.67 14.35 0.35 0.43 0.07 repayment (birr) (110.44) (96.32) Household received remittances in 0.14 0.15 0.61 0.66 0.93 the past year (0.34) (0.36) Have a bank savings account (%) 0.16 0.15 0.91 0.98 0.88 (0.36) (0.36) Distance from dwelling to the 5.44 5.48 0.84 0.84 1.00 closest bank branch (km) (2.71) (2.92) Have savings in microcredits, 0.15 0.13 0.55 0.68 0.83 Coop. Bank, NGOs. (0.35) (0.34) Save in dried coffee beans. 0.56 0.55 0.38 0.87 0.24 (0.49) (0.49) Amount of savings in dried coffee 83.19 83.56 0.86 0.81 0.62 beans (kg) (155.58) (175.45) Household borrowed credit in the 0.27 0.29 0.64 0.49 0.81 past 12 months (0.44) (0.45) Panel C: Household agricultural output/inputs Land size (hectares) 1.08 1.07 0.54 0.48 0.23 (1.27) (0.78) Coffee cultivated % total area 0.58 0.58 0.67 0.86 0.78 (0.23) (0.23) Production of cherry coffee beans 488.80 484.25 0.64 0.70 0.37 (kg) (612.65) (496.45) Production of dried coffee beans 235.21 225.28 0.63 0.92 0.49 (kg) (559.47) (470.86) Share of red cherry sold to FT 94.50 94.60 0.90 0.80 0.67 Coop (%) (18.89) (20.02) Selling price of cherry coffee beans 9.53 9.46 0.80 0.59 0.43 to FT Coop (birr) (5.61) (3.04) Use of chemical fertilizer (kg) 9.97 10.95 0.71 0.68 0.94 (36.10) (31.26) Table S1.1. Continued Control p-value p-value p-value Treatment mean Tsingle =Tcontrol Tjoint =Tcontrol Tsingle =Tjoin mean (std. dev.) (std. dev.) (1) (2) (3) (4) (5) Use of organic fertilizer (kg) 836.61 896.26 0.83 0.33 0.47 Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 (1426.02) (1428.87) Number of coffee plants 1503.20 1460.97 0.74 0.94 0.73 (2740.88) (1938.02) Units of hired labor in last season 3.78 4.03 0.91 0.32 0.36 (6.20) (6.03) Panel D: Head of household Gender (1=Male) 0.87 0.88 0.61 0.52 0.20 (0.33) (0.32) Age 49.82 49.95 0.78 0.95 0.70 (14.91) (14.97) Schooling 3.69 3.55 0.60 0.53 0.92 (3.63) (3.42) Ever attended formal school (%) 0.65 0.66 0.82 0.97 0.83 (0.47) (0.47) Marital status (1=married) 0.84 0.86 0.96 0.28 0.21 (0.36) (0.35) Hours worked in household farm 79.68 84.44 0.11 0.44 0.36 last month (53.59) (55.07) Hours worked in nonfarm 8.18 6.40 0.31 0.35 0.93 household business last month (25.61) (21.75) Panel E: Children’s schooling and labor Farm work for children aged 6–16 0.50 0.53 0.30 0.41 0.80 last month (%) (0.49) (0.49) Farm monthly hours worked for 19.11 20.86 0.21 0.35 0.72 children aged 6–16 (29.98) (30.48) Children aged 6–16 attending 0.80 0.81 0.33 0.83 0.39 school (%) (0.40) (0.39) Average years of schooling for 3.11 3.03 0.80 0.33 0.43 children aged 6–16 (2.57) (2.51) Household variables: F-test of joint 0.71 0.80 0.60 significance (p-value) Source: Author’s analysis based on baseline survey data collected in July/August 2015. Note: Sample mean and standard deviation (in parenthesis). There are 449, 450, 299 households in the single account treatment group, joint account treatment group, and control group, respectively. Table S1.2. Determinants of Take-Up and Frequent Usage of Accounts Take-up Frequent user if take-up=1 Single account Joint account Single account Joint account Baseline covariates Women’s financial decision-making − 0.030(0.034) − 0.000(0.024) − 0.057(0.037) − 0.060(0.060) power Household size − 0.009(0.015) 0.045∗∗∗ (0.012) − 0.002(0.016) − 0.017(0.020) Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Christian household 0.075(0.064) 0.232∗∗∗ (0.056) − 0.047(0.074) 0.004(0.101) Gender hh 0.098(0.087) 0.358∗∗∗ (0.070) − 0.126(0.112) − 0.101(0.259) 30–45 years hh 0.225∗∗ (0.094) − 0.148(0.099) 0.063(0.141) 0.180(0.127) 46–65 years hh 0.179∗ (0.096) − 0.019(0.100) 0.125(0.139) 0.198(0.125) +65 years hh 0.039(0.107) − 0.085(0.112) 0.026(0.147) 0.059(0.144) Literacy of hh 0.062(0.062) 0.047(0.055) 0.010(0.080) − 0.010(0.088) Land size (hectare) 0.001(0.033) − 0.051(0.036) − 0.050(0.039) − 0.034(0.055) Share of land dedicated to coffee − 0.021(0.126) − 0.114(0.109) − 0.247∗ (0.144) − 0.184(0.187) crop Yields of coffee crop (x1000 kg) 0.008 (0.013) − 0.000(0.014) 0.022(0.015) 0.000(0.020) Have a nonfarm family business − 0.148∗∗ (0.075) 0.010(0.067) − 0.065(0.101) 0.091(0.092) Monthly household income (x1000 − 0.022(0.019) 0.028(0.020) − 0.023(0.036) 0.003(0.022) birr) Remittances (x1000 birr) 0.002(0.006) 0.006∗ (0.004) 0.011∗∗ (0.004) 0.007(0.004) Has a bank account − 0.106(0.079) − 0.163∗∗∗ (0.057) 0.033(0.100) 0.119(0.122) Lend money to neighbors or family 0.178∗∗∗ (0.068) 0.049(0.068) 0.104(0.088) − 0.134(0.093) in the last 12 months Experienced food shortage in the last 0.061(0.059) − 0.053(0.054) − 0.240∗∗∗ (0.075) − 0.017(0.087) 12 months Has protected water well − 0.056(0.058) − 0.005(0.051) 0.073(0.073) − 0.086(0.084) Has access to electricity − 0.156∗∗ (0.062) 0.019(0.049) 0.050(0.086) − 0.076(0.073) Has mud floor − 0.028(0.059) − 0.003(0.057) − 0.005(0.080) − 0.067(0.074) Has a ventilated pit latrine − 0.007(0.059) − 0.011(0.057) 0.199∗∗ (0.079) − 0.046(0.081) Distance from dwelling to the bank − 0.010(0.009) 0.014∗ (0.007) − 0.010(0.012) − 0.037∗∗∗ (0.012) branch (km) N 414 422 257 229 R2 0.14 0.29 0.13 0.11 Source: Author’s analysis based on baseline survey data collected in July/August 2015. Note: Linear probability model. Robust standard errors are in parentheses. “Take up” estimation is based on the sample of treatment group units. ‘Frequent users’ analysis is conditional on the sample of treated group units that opened a bank account. Frequent user is defined as 1 for those making at least five deposits in the first two years after setting up the account, 0 otherwise. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Table S1.3. Compliers and Switchers Single account Joint account take-up take-up 1=complier, 1=complier, 0=switcher 0=switcher Baseline covariates Women’s financial decision-making power − 0.032(0.036) 0.043(0.031) − 0.002(0.015) Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Household size 0.005 (0.014) Christian household − 0.041(0.067) 0.192∗∗∗ (0.063) Gender hh (1=male) − 0.169(0.107) 0.799∗∗∗ (0.093) 30–45 years hh − 0.091(0.138) 0.101(0.097) 46–65 years hh − 0.061(0.141) 0.064(0.092) +65 years hh − 0.057(0.147) − 0.024(0.111) Literacy of hh (1=literate) − 0.033(0.069) − 0.014(0.055) Land size (hectares) 0.024(0.038) − 0.064(0.042) Share of coffee crop land − 0.077(0.134) 0.035(0.129) Yields of coffee crop (×1000 kg) − 0.002(0.014) − 0.002(0.017) Have a nonfarm family business − 0.088(0.099) 0.113∗∗ (0.051) Monthly household income (×1000 birr) − 0.022(0.032) 0.016(0.018) Remittances (×1000 birr) 0.000(0.007) 0.001(0.004) Has a bank account − 0.151(0.107) 0.018(0.093) Lend money to neighbors or family in the last 12 0.076(0.065) 0.001(0.072) months Experienced food shortage in the last 12 months − 0.021(0.067) − 0.060(0.065) Has protected water well − 0.039(0.062) − 0.073(0.056) Has access to electricity − 0.167∗∗ (0.084) − 0.013(0.049) Has mud floor 0.022(0.073) − 0.062(0.053) Has a ventilated pit latrine − 0.035(0.072) 0.004(0.050) distance from dwelling to bank branch (km) − 0.003(0.009) 0.022∗∗ (0.008) N 230 256 R2 0.09 0.43 Source: Author’s analysis based on baseline data collected in July/August 2015. Note: Linear probability model. Robust standard errors are in parentheses. The estimation sample is conditional on households that opened a bank account in the partner bank. For the single account treatment group, a complier indicator takes the value 1 for those households that opened a single deposit account and 0 for those households that opened a joint deposit account (switchers). Similarly, the complier indicator for the joint account treatment group takes the value 1 for those households that opened a joint deposit account and 0 for those households that opened a single deposit account (switchers). ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Table S1.4. First-Stage Regressions Take single account Take joint account Single account 0.437∗∗∗ 0.107∗∗∗ (0.024) (0.016) Joint account 0.169∗∗∗ 0.432∗∗∗ (0.017) (0.022) F-test 26 27 N 1166 1166 Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Robust standard in parenthesis. See table 2 for full specification details. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Table S1.5. Treatment on the Treated (TOT) Impacts p-value: p-value: joint Dep. var. mean Single account Joint account Tsingle =Tjoint significance control group Panel A: savings Has bank account 0.555∗∗∗ (0.074) 0.518∗∗∗ (0.069) 0.704 0.000 0.360 Size of bank deposits (invlog birr) 2.409∗∗∗ (0.660) 1.386∗∗ (0.629) 0.279 0.000 2.195 Size of bank balances (invlog birr) 3.892∗∗∗ (0.629) 2.983∗∗∗ (0.573) 0.262 0.000 2.793 Under the "mattress" (invlog birr) − 0.807∗∗∗ (0.283) − 0.277(0.283) 0.033 0.011 3.240 ROSCAs (invlog birr) − 0.416(0.493) 0.550(0.461) 0.143 0.329 4.393 Other savings vehicles (invlog birr) 0.254(0.441) − 0.223(0.421) 0.432 0.734 1.384 Total savings deposits (invlog birr) 0.572(0.504) 0.255(0.479) 0.638 0.470 6.640 Total savings balance (invlog birr) 0.932∗ (0.490) 0.867∗∗ (0.444) 0.912 0.041 7.471 Index of dependent variables 0.527∗∗∗ (0.151) 0.517∗∗∗ (0.140) 0.962 0.000 0.000 Panel B: Women’s empowerment Ownership of bank account 0.208∗∗∗ (0.056) 0.760∗∗∗ (0.058) 0.000 0.000 0.072 Decision-making on bank deposits 0.173∗∗ (0.069) 0.571∗∗∗ (0.067) 0.000 0.000 0.160 Decision-making on bank withdrawals 0.225∗∗∗ (0.069) 0.660∗∗∗ (0.066) 0.000 0.000 0.156 Financial empowerment index 0.191(0.148) 0.295∗∗ (0.128) 0.556 0.045 0.000 Productive empowerment index 0.020(0.153) 0.163(0.135) 0.465 0.484 0.000 Time allocation empowerment index 0.067(0.150) 0.161(0.133) 0.615 0.464 0.000 Index of dependent variables 0.524∗∗∗ (0.178) 1.803∗∗∗ (0.170) 0.000 0.000 0.000 Panel C: Schooling Boys 6-16 Currently attending school − 0.073(0.075) − 0.019(0.057) 0.553 0.598 0.845 Daily hours of school − 0.266(0.310) − 0.195(0.238) 0.848 0.527 3.444 Years of formal schooling 0.033(0.430) 0.347(0.348) 0.527 0.604 2.926 Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Table S1.5. Continued p-value: p-value: joint Dep. var. mean Single account Joint account Tsingle =Tjoint significance control group Annual expenses (invlog birr) − 0.227(0.441) 0.007(0.362) 0.665 0.873 4.791 Index of dependent variables − 0.173(0.193) 0.023(0.151) 0.405 0.653 0.000 Girls 6-16 Currently attending school − 0.068(0.065) 0.092∗(0.056) 0.047 0.119 0.836 Daily hours of school − 0.172(0.270) 0.353(0.232) 0.120 0.229 3.374 Years of formal schooling − 0.005(0.437) 0.456(0.377) 0.379 0.477 2.872 Annual expenses (birr) − 0.377(0.429) 0.566(0.358) 0.070 0.156 4.790 Index of dependent variables − 0.126(0.178) 0.284∗ (0.149 0.060 0.104 0.000 Panel D: Child labor All children Extensive margin − 0.086(0.078) 0.180∗∗ (0.069) 0.007 0.012 0.525 Intensive margin − 2.216(4.067) 7.916∗∗ (3.763) 0.051 0.079 19.047 Boys 6-16 Extensive margin − 0.158(0.099) 0.171∗∗ (0.081) 0.004 0.015 0.603 Intensive margin 3.018(5.598) 8.261∗ (4.688) 0.459 0.192 20.828 Index of child labor for boys − 0.103(0.188) 0.333∗∗ (0.154) 0.057 0.071 0.000 Girls 6-16 Extensive margin − 0.019(0.096) 0.172∗∗ (0.087) 0.233 0.126 0.134 Intensive margin − 6.196(4.703) 7.387(4.576) 0.024 0.078 17.223 Index of child labor for girls − 0.131(0.168) 0.308∗ (0.158) 0.046 0.095 0.000 Panel E: Adult Labor Males Extensive margin − 0.017(0.055) 0.075(0.048) 0.162 0.243 0.870 Intensive margin − 7.840(8.449) 10.583(7.852) 0.089 0.223 76.000 Index of dep. var. for males − 0.113(0.160) 0.240∗ (0.142) 0.074 0.151 0.000 Females Extensive margin − 0.043(0.072) 0.147∗∗ (0.065) 0.045 0.062 0.613 Intensive margin − 4.087(5.571) 12.317∗∗ (5.440) 0.033 0.056 32.312 Index of dep. var. for females − 0.108(0.147) 0.346∗∗ (0.138) 0.023 0.031 0.000 Panel F: Labor-Intensive investment Organic fertilizer (kg per hectare) 255.295∗ (146.001) 71.917(139.803) 0.396 0.160 676.355 Chemical fertilizer (kg per hectare) − 1.965(3.554) 4.074(3.570) 0.224 0.440 10.1706 Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Table S1.5. Continued p-value: p-value: joint Dep. var. mean Single account Joint account Tsingle =Tjoint significance control group New coffee plants (units per hectare) 4.248 (2.721) 4.081(2.681) 0.971 0.011 0.634 Seeds (kg per hectare) 0.333(0.308) 0.581∗ (0.317) 0.393 0.188 1.551 New tools and equipment (birr) 0.720(7.569) 1.084(7.411) 0.954 0.989 46.996 Index of dependent variables 0.432∗ (0.233) 0.412∗ (0.214) 0.953 0.008 0.000 Panel G: Household income Agricultural income (invlogbirr) − 0.102(0.152) 0.100(0.156) 0.392 0.693 9.942 Nonfarm hh business income (invlog 0.516(0.719) − 0.485(0.677) 0.308 0.594 2.473 Bir) Labor income (invlog birr) − 0.228(0.660) − 0.886(0.629) 0.459 0.356 2.365 Index of dependent variables − 0.036(0.175) − 0.149(0.176) 0.654 0.677 0.000 Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Robust standard in parenthesis. Specification includes strata dummies from the sample stratifying variables. The first-stage results are reported in table S1.4, while table S1.6 provides a full description of the variables. The inverse hyperbolic sine transformation is applied to monetary and income variables (birr). See table 2 for further details about specifications. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Table S1.6. Description of Outcome Variables Bank Savings Description “Own bank account” 1=if head of household and/or spouse owned any commercial bank deposit account in the past 12 months before the survey, 0=otherwise “Size of saving deposits ($)” Monetary value (in birr $) of all saving deposits made in the past 12 months before the survey by the head of household and/or spouse “Size of saving balances ($)” Monetary value (in birr $) of saving balances at the time of survey held in all Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 savings accounts owned by the head of household and/or spouse Other monetary saving channels Saving deposits under the mattress Monetary value (in birr $) of all saving deposits made in the past 12 months before the survey by the head of household and/or spouse Saving deposits in ROSCAs (Iqqub, Iddir) Monetary value (in birr $) of all saving deposits made in the past 12 months before the survey by the head of household and/or spouse Saving deposits in other institutions Monetary value (in birr $) of all saving deposits made in the past 12 months before the survey by the head of household and/or spouse Total saving deposits Monetary value (in birr $) of the bank plus nonbank saving deposits made in the past 12 months before the survey by the head of household and/or spouse Total saving balances Monetary value (in birr $) of the bank plus nonbank saving balances at the time of survey by the head of household and/or spouse. Adult labor Participation (extensive) 1=whether worked in the household farm in the past 30 days before the survey, 0=otherwise. Hours worked (intensive) Number of worked hours in the household farm in the past 30 days before the survey Child Labor Participation (extensive) 1=whether children aged 6–16 worked in the household farm in the past 30 days before the survey, 0=otherwise. Hours worked (intensive) Number of worked hours by children aged 6–16 in the household farm in the past 30 days before the survey Months per year Number of months worked in the household farm in the past 12 months by children aged 6-16 Days per week The average number of days per week worked in the household farm in the past 12 months by children aged 6–16 Hours per day The average number of hours per day worked in the household farm in the past 12 months by children aged 6–16 Schooling Currently attending school 1=whether child aged 6–16 is currently attending formal school institution at the time of the survey, 0=otherwise Daily hours of school Number of daily hours of formal school by a child aged 6–16 at the time of the survey Years of formal schooling Number of completed years of formal schooling for children aged 6–16 at the time of the survey School expenditures School expenditures (birr $) in the past 12 months before the survey for children aged 6–16 Women empowerment Ownership of bank deposit account 1=if woman (head of household or spouse) owned a single or joint bank deposit account in the last 12 months before the survey, 0=otherwise. Decision-making on bank account deposits 1=if woman (head of household or spouse) is typically the main decider or equally decides with her male husband regarding bank saving deposits, 0=otherwise Decision-making on bank account 1=if woman (head of household or spouse) is typically the main decider or withdrawals equally decides with her male husband regarding bank saving withdrawals, 0=otherwise Table S1.6. Continued Bank Savings Description Financial empowerment index PCA index is computed based on two intermediate outcomes: (1) women’s decision-making power on the allocation of agricultural revenues in the past 12 months and (2) women’s decision-making power on the allocation of nonagricultural revenues in the past 12 months. Each variable is coded as 2=if the woman head of household or spouse is the only decision-maker or the most Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 important decision maker, 1=if the woman head of household or spouse equally shares the decision along with her male head of household or spouse, and 0=otherwise. Time allocation empowerment index PCA index is computed based on five intermediate outcomes: women’s decision-making power in sending children to school in the past 12 months, women’s decision-making power in assigning children to household chores in the past 12 months, women’s decision-making power in assigning children to farm work, women’s decision-making power to work on the household farm in the past 12 months, women’s decision-making power to work in farm or nonfarm activities outside the house in the past 12 months. Each of these five variables is coded as 2=if woman head of household or spouse is the only decision maker or the most important decision maker, 1=if woman head of household or spouse equally shares the decision with her male head of household or spouse, and 0=otherwise. Agricultural Production empowerment index PCA index is computed based on five intermediate outcomes: women’s decision-making power on buying/renting farm tools/equipment in the past 12 months, women’s decision-making power on selecting crops in the past 12 months, women’s decision-making power on using agricultural inputs such as fertilizer and pesticides in the past 12 months, women’s decision-making power on negotiating the price of coffee crops in the past 12 months, and women’s decision-making power on attending Fairtrade cooperative meetings in the past 12 months Each one of these five variables is coded as 2=if woman head of household or spouse is the only decision maker or the most important decision maker, 1=if woman head of household or spouse equally shares the decision along with her male head of household or spouse, and 0=otherwise. Agriculture Investments Organic fertilizer Kilograms per hectare of manure and compost applied to household plots in the last agricultural season (March 2017–February 2018) Chemical fertilizer Kilograms per hectare of chemical fertilizer applied to household plots in the last agricultural season (March 2017–February 2018) New coffee plants Number of new coffee trees per hectare purchased in the last agricultural season (March 2017–February 2018) Seeds Kilograms per hectare of seeds applied to household plots in the last agricultural season (March 2017–February 2018) New agricultural tools and equipment Value (in birr $) for household purchases of any new tool or equipment used in agricultural production in the last agricultural season (March 2017-February 2018). Income Agriculture income Household annual income (birr $) from agricultural activities in the last 12 months before the survey Nonfarm household business income Household annual income (birr $) from nonfarm household business sales in the last 12 months before the survey Remunerated labor income Household annual income (birr $) from remunerated labor in the last 12 months before the survey Source: Author’s own elaboration. Table S1.7. ITT Impacts on Savings (2% Trimming for Monetary Variables) Index of Total saving Total saving dependent Bank savings Other saving deposits deposits balance variables Saving Saving Has deposit deposits in balances in Under the Bank+ all Bank+ all account in any bank any bank mattress ROSCAS other other any bank (birr) (birr) (birr) (birr) Other (Birr) (Birr) (Birr) (1) (2) (3) (4) (5) (6) (7) (8) (9) Single account 0.298∗∗∗ 665.812∗∗ −47.783∗∗∗ 28.383 534.554 988.931∗∗ 0.267∗∗∗ 782.5146∗∗∗ −298.746∗∗ (0.035) (278.1174) (263.333) (127.274) (18.471) (23.301) (337. 283) (453.268) (0.078) Joint account 0.318∗∗∗ 336.5617 369.162 −212.255∗ −21.181 9.297 272.195 378.943 0.192∗∗ (0.035) (281.9545) (263.280) (127.883) (19.173) (22.889) (335.625) (442.486) (0.077) p-val.: Tsingle =Tjoint 0.519 0.251 0.126 0.406 0.074 0.393 0.450 0.148 0.301 p-val.: Joint Significance 0.000 0.057 0.012 0.063 0.024 0.457 0.284 0.084 0.002 Dep. var. mean in control 0.360 1087.885 1133.507 951.764 242.212 99.067 2597.861 4568.993 group St. dev. 0.481 3379.823 3127.615 1839.236 283.670 285.182 4058.500 5756.658 N (2% trimming for 1166 1143 1143 1143 1143 1143 1143 1143 monetary var.) Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Robust standard errors in parenthesis. Regression specification includes strata dummies from the sample stratifying variables; 2% trimming is applied to monetary (birr) variables. The index of dependent variables is computed as a simple average of the z-scores of the dependent variables in columns 1–7 after standardizing each variable by subtracting the mean and dividing by the standard deviation in the control group. The exchange rate US$/birr was around 27 in 2018. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Table S1.8. ITT Impacts on Female Empowerment, Subcomponents Analysis Financial empowerment Time allocation Agric. production empowerment index index index Spending Spending hh Own Own Buying and Negotiating Using attending hh farm nonfarm Children’s Children’ children’s farm work renting agr. Selecting coffee agr. Coop income income school work hh chores work outside tools crops price inputs meetings (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Single account 0.058 0.077∗ −0.010 0.030 0.067 0.056 0.056 0.076 0.007 0.010 0.005 0.037 (0.043) (0.042) (0.075) (0.073) (0.076) (0.070) (0.073) (0.072) (0.073) (0.072) (0.072) (0.069) Joint account 0.075∗ 0.113∗∗∗ 0.045 0.085 0.007 0.064 0.122∗ 0.098 0.0197 0.082 0.077 0.060 (0.041) (0.040) (0.073) (0.071) (0.073) (0.068) (0.070) (0.070) (0.071) (0.070) (0.069) (0.067) p-val.: 0.627 0.279 0.373 0.354 0.360 0.895 0.304 0.720 0.668 0.243 0.245 0.939 Tsingle =Tjoint p-val.: Joint 0.196 0.018 0.644 0.425 0.576 0.625 0.212 0.371 0.907 0.379 0.400 0.571 Significance Dep.var. mean 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 in control group St. dev 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 N 1095 1095 1093 1093 1094 1095 1095 1095 1095 1095 1095 1095 Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Robust standard errors in parenthesis. Each subcomponent of women empowerment indices was standardized relative to the control group’s mean for ease of interpretation. Table S1.6 provides a full description of the variables. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Table S1.9. The role of Baseline Financial Decision-Making Power on Women’s Empowerment Above median baseline index Below median baseline index Dep. var: women’s empowerment index Single account 0.388∗∗∗ (0.100) 0.538∗∗∗ (0.159) Joint account 0.805∗∗∗ (0.104) 1.023∗∗∗ (0.163) Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 p-val.: Tsingle =Tjoint 0.000 0.000 p-val.: joint significance 0.000 0.000 p-val.: Tsingle_above =Tsingle_below 0.377 p-val.: Tjoint_above =Tjoint_below 0.198 Dep. var. mean in control group 0.00 0.00 St. dev. 1.00 1.00 N 709 376 Source: Author’s analysis based on baseline (July/August 2015) and follow-up (March/April 2018) surveys. Note: Robust standard errors in parenthesis. The median value for the baseline index of financial decision-making power is 0.067 and is based on four intermediate outcomes that refer to women’s decision-making power over the allocation of agricultural revenues, nonagricultural revenues, informal and formal savings management, and household loans. Each intermediate variable is categorical and receives the value of 2 if she is the only decision-maker or the most important decision-maker, 1 if she equally shares the decision with a male spouse, and 0 otherwise. The analysis computes a standardized baseline index by computing a simple average of the z-scores. The dependent variable is the empowerment family index used in column (7) in table 3. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Table S1.10. ITT Impacts on Child Work in the Last 12 Months All children aged 6-16 Boys Girls Months Days per Hours per Months Days per Hours per Months Days per Hours per per year week day per year week day per year week day (1) (2) (3) (4) (5) (6) (7) (8) (9) Single account −0.202 −0.111 −0.195 0.026 0.003 −0.195 −0.360 −0.159 −0.126 (0.366) (0.205) (0.162) (0.461) (0.259) (0.213) (0.456) (0.248) (0.183) Joint account 0.745∗∗ 0.385∗ 0.189 0.767∗ 0.208 −0.007 0.727 0.569∗∗ 0.409∗∗ (0.360) (0.203) (0.157) (0.432) (0.244) (0.201) (0.459) (0.254) (0.186) p-val.: Ho: Tsingle =Tjoint 0.004 0.006 0.007 0.060 0.350 0.294 0.012 0.001 0.002 p-val.: Joint Significance 0.011 0.019 0.027 0.089 0.562 0.516 0.033 0.004 0.006 Dep. var. mean in control group 4.320 2.511 2.000 4.568 2.733 2.290 4.062 2.280 1.698 St. dev. 4.519 2.456 2.038 4.413 2.464 2.113 4.622 2.432 1.916 N 2035 2035 2035 1015 1015 1015 1020 1020 1020 Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Clustered standard errors at the household level in parenthesis. Farm participation and farm hours worked do not include household chores activities. The recall time length for these outcome variables refers to the 12 months before the survey date. Regression specification includes strata dummies from the sample stratifying variables. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Table S1.11. Weakly Hours of Household Chores for Children Boys Girls Single account −1.044(1.121) −1.497(1.221) Joint account −0.433(1.055) 0.262(1.237) p-val.: Tsingle =Tjoint 0.504 0.083 p-val.: joint significance 0.627 0.189 Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 Dep. var. mean in control group 14.992 19.373 St. dev 11.060 13.759 N 999 1007 Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Clustered standard errors at the household level in parenthesis. See table 2 for specification details. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Table S1.12. ITT Impacts for Only Male-Headed Households p-value: p-value: Joint Dep. var.: mean Single account Joint account Tsingle = Tjoint significance control group Family of outcomes indices Savings 0.287∗∗∗ (0.075) 0.330∗∗∗ 0.493 0.000 0.000 (0.074) Women’s empowerment 0.429∗∗∗ 0.916∗∗∗ 0.000 0.000 0.000 (0.079) (0.081) Boys’ schooling −0.097 (0.096) −0.004 0.293 0.504 0.000 (0.087) Girls’ schooling −0.027 0.127 0.047 0.105 0.000 (0.091) (0.086) Boys’ child labor −0.073 0.186∗ 0.044 0.011 0.000 (0.102) (0.095) Girls’ child labor −0.031 0.176∗ 0.015 0.038 0.000 (0.093) (0.095) Male adult labor −0.019 0.105 0.069 0.153 0.000 (0.077) (0.076) Female adult labor −0.065 0.122∗ 0.004 0.016 0.000 (0.073) (0.073) Agricultural Investments 0.257∗∗ (0.112) 0.266∗∗ 0.935 0.015 0.000 (0.106) Household income 0.074 0.039 0.624 0.637 0.000 (0.079) (0.079) Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Robust standard errors in parenthesis. See table 2 for specification details. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Table S1.13. ITT Impacts for Coresident Couples p-value: p-value: joint Dep. var.: Mean Single account Joint account Tsingle = Tjoint significance control group Family of outcomes Indices Savings 0.300∗∗∗ (0.091) 0.189∗∗ 0.183 0.004 0.000 (0.089) Women’s empowerment 0.407∗∗∗ 0.908∗∗∗ (0.081) 0.000 0.000 0.000 Downloaded from https://academic.oup.com/wber/article/39/2/410/7684482 by The World Bank user on 02 May 2025 (0.080) Boys’ schooling −0.047 (0.098) 0.000 0.509 0.770 0.000 (0.092) Girls’ schooling −0.035 0.120 0.053 0.125 0.000 (0.093) (0.086) Boys’ child labor −0.047 0.206∗∗ 0.006 0.013 0.000 (0.104) (0.097) Girls’ child labor −0.008 0.192∗∗ 0.020 0.041 0.000 (0.095) (0.097) Male adult labor −0.044 0.125 0.014 0.039 0.000 (0.080) (0.077) Female adult labor −0.059 0.123∗ 0.005 0.018 0.000 (0.073) (0.073) Agricultural Investments 0.268∗∗ (0.112) 0.272∗∗ 0.980 0.014 0.000 (0.110) Household Income 0.086 0.076 0.885 0.535 0.000 (0.081) (0.082) Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Robust standard errors in parenthesis . See table 2 for specification details. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. Table S1.14. ITT Impacts with Control Covariates. p-value: p-value: joint Dep. var. mean control Single account Joint account Tsingle = Tjoint significance group Family of outcomes indices Savings 0.284∗∗∗ (0.068) 0.322∗∗ 0.526 0.000 0.000 (0.068) Women’s empowerment 0.432∗∗∗ 0.892∗∗∗ 0.000 0.000 0.000 (0.083) (0.084) Boys’ schooling −0.089 (0.095) −0.013 0.385 0.594 0.000 (0.083) Girls’ schooling −0.035 0.134 0.026 0.060 0.000 (0.088) (0.082) Boys’ child labor −0.050 0.133 0.043 0.103 0.000 (0.100) (0.093) Girls’ child labor −0.021 0.158∗ 0.028 0.061 0.000 (0.091) (0.090) Male adult labor −0.030 0.100 0.056 0.134 0.000 (0.077) (0.074) Female adult labor −0.001 0.134∗∗ 0.032 0.050 0.000 (0.069) (0.067) Agricultural investments 0.248∗∗ (0.102) 0.274∗∗∗ 0.817 0.005 0.000 (0.094) Household income −0.014 −0.071 0.464 0.644 0.000 (0.078) (0.081) Source: Author’s analysis based on follow-up survey data collected in March/April 2018. Note: Robust standard errors in parenthesis. See table 2 for specification details. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.10. C The Author(s) 2024. 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