Policy Research Working Paper 11111 Mitigating the Impact of Household Expropriation on Female Entrepreneurship Experimental Evidence from Ghana Francisco Campos Adriana Conconi Elwyn Davies Marine Gassier Markus Goldstein Africa Gender Innovation Lab & Finance, Competitiveness and Investment Global Department May 2025 Policy Research Working Paper 11111 Abstract How do intrahousehold dynamics affect the investment of there is no evidence that the allocation of resources within female entrepreneurs? This paper presents findings from households is efficient, the joint decision-making interven- a randomized controlled trial in Ghana that assesses the tion leads to increased household support for the women’s impacts of four alternative support mechanisms on wom- businesses but does not impact business performance. The en-owned businesses: (a) an unconditional grant provided savings support mechanism leads to a 15 percent increase through a mobile money account equivalent to two months in sales and a 10 percent increase in profits. These effects of median profits, (b) an unconditional grant disbursed are largest among female entrepreneurs who faced high to the female entrepreneurs’ spouses in similar conditions; expropriation pressure at baseline. This subgroup obtains (c) a grant conditional on participating with their spouses a 29 percent increase in sales and a 23 percent increase in in a training on joint decision-making; and (d) a grant profits. The paper tests for alternative mechanisms, includ- conditional on reaching a savings goal under a dedicated ing self-control issues, liquidity constraints, and access to bank account. In line with Fafchamps et al. (2014), the savings, but these do not explain the results. The findings study finds no impacts of the unconditional grants on the substantiate that intrahousehold dynamics matter for wom- business performance of female entrepreneurs. The dis- en’s investment decisions, and highlight the importance of bursement to the spouse also has no impact on the sales, promoting autonomy in the face of expropriation pressures, profits, or investment of female entrepreneurs. Although for increased growth and investment. This paper is a product of the Gender Innovation Lab, Africa Region and the Finance, Competitiveness and Investment Global Department. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at fcampos@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Mitigating the Impact of Household Expropriation on Female Entrepreneurship: Experimental Evidence from Ghana Francisco Campos, Adriana Conconi, Elwyn Davies, Marine Gassier, Markus Goldstein1 Keywords: Gender, Entrepreneurship, Informality, Small Enterprises, Grants, Microfinance JEL codes: G31, J16, L25, L26, O12, O16 1 Francisco Campos, Adriana Conconi, Elwyn Davies, and Marine Gassier are from the World Bank. Markus Goldstein is from Center for Global Development. The authors thank Fidelity Bank and FMMS Ghana for their partnership in this study, as well as Innovations for Poverty Action (IPA) for project management. Lorraine d’Anglejan and Lorenzo Uribe provided excellent field coordination in managing the project implementation and data collection. This paper is a product of the World Bank’s Africa Gender Innovation Lab and the Finance, Competitiveness, and Investment (FCI) Department. We gratefully acknowledge funding from the World Bank Group’s Umbrella Facility for Gender Equality and the World Bank FCI Department. 1. Introduction Around the world, female entrepreneurship is growing. However, gender gaps in business performance continue to be large, and few interventions have achieved transformational impacts on the performance of female-owned enterprises (World Bank, 2019; Hardy and Kagy, 2020). In particular, the evidence on the effects of cash grants shows that such interventions have limited to no impacts on the performance of existing women-owned businesses (De Mel et al. 2008, 2009, 2012 in Sri Lanka; Fafchamps et al. 2014 in Ghana; Berge et al. 2015 in Tanzania; Fiala 2017 in Uganda). 2 Funding to female entrepreneurs is often not invested in their businesses, whether by their own choice or not (Jayachandran, 2021). Among the different constraints faced by women-owned enterprises, intra-household dynamics have received increased attention (World Bank, 2019; Buvinić and Furst-Nichols, 2016; Baland and Ziparo, 2018). A first explanation for the pervasive gender differences in the decisions that entrepreneurs make regarding investments in their own businesses is that such decisions are guided by efficiency considerations. Bernhardt et al. (2019) hypothesize that grant recipients consider all household businesses when making an investment decision, and channel their investments towards the business with the highest return to capital, even if it is not their own. They argue that, since average returns to capital are higher in male-owned businesses, this calculation will often lead women to invest the grants they receive in their spouses’ businesses as a means to improve overall household income. Another set of explanations emphasize that women face greater pressure than men to share their income and the capital they have access to with their spouse or with other household members. Several studies discuss patterns associated with household expropriation. Schaner (2015) shows that in Kenya, individuals are willing to forego interest earnings to control a greater share of household differences when their temporal preferences differ from those of their spouse. Jakiela and Ozier (2016) find that in Kenya, women prefer investment strategies that allow them to hide their income from family members even if these strategies reduce their expected earnings. Observational studies also suggest that married individuals commonly hide income and 2 The gender differences from grants programs discussed in this paper is related to relatively small grants to existing businesses in urban settings. Large grants to a selected set of women/men entrepreneurs in the context of business plan competitions (e.g. McKenzie 2017 in Nigeria) have led to positive effects on business performance and job creation. Conditional and unconditional cash transfers to households including those led by women have also had large impacts, notably through new entrepreneurship. 2 consumption from their spouse, and are willing to invest resources to be able to do so (Baland et al. 2016). In Uganda, Fiala (2017) observes that the impacts of interventions expanding women’s access to capital and training were significantly lower for those that do not hide money from their spouse. Based on the results from an experiment in Kenya, Brudevold-Newman et al. (2017) find that in addition to credit constraints, women face savings constraints that shape their financial behaviors. Carranza et al. (2022) show that in Côte d’Ivoire workers increase their labor supply when they are given the means to hide their income from others. Riley (2024) demonstrates that disbursing a loan using mobile money rather than cash, which has the potential of reducing intra- household sharing pressure, has a significant impact on investments and profits. Finally, Friedson- Ridenour and Pierotti (2019) collected qualitative data from a subset of the women included in our study sample, and show that they often find it necessary to hide money and limit investment to ensure continued support from their husband. In this study, we present the results of a randomized controlled trial designed to help uncover how these two types of intra-household dynamics –efficiency considerations and expropriation pressures– shape the investment decisions of women entrepreneurs. As part of this study, we sampled women entrepreneurs in Accra, Ghana living with a domestic spouse (from now on “married” couples), and randomly offered unconditional or conditional grants. We assessed four alternatives of supporting women-owned small businesses: (a) an unconditional grant of two months of median profits paid to a mobile money account; (b) a similar unconditional grant disbursed to their spouses in the same way; (c) a similar grant to women paid to a mobile money account conditional on participating with their spouses in a training on joint decision- making; and (d) a similar grant to women paid to a mobile money account conditional on reaching a savings goal using a dedicated bank account. The unconditional grants had no effect on average business performance for women entrepreneurs irrespective of being provided to the female entrepreneur or their spouse. We use the data on the profitability of the businesses of the recipients’ spouses to have a first approximation on whether efficiency considerations at the household level explain the lack of impact of the unconditional grant to female entrepreneurs. If women were channeling the grant towards the more profitable business in the household, we would see differences in impact between recipients who run a business that is more profitable than their spouses. This is not the case. 3 Furthermore, we investigate whether the training on joint decision-making helped foster greater intra-household cooperation, thereby supporting the allocation of resources at the household level. The training emphasized the importance of efficiency in the allocation of resources. The disbursement of the grant to the mobile money account was conditioned on both spouses attending the training. The intervention largely had the intended effects on attitudes and behaviors. It also had positive effects on business investment in female-owned enterprises. But even under this condition, the grant had no impact on average business performance, and there was no difference in the impact of the grant based on the profitability of the recipients’ businesses relative to that of their spouses. We then turn to the expropriation pressure women entrepreneurs face inside the household. We use three strategies to examine how these pressures shape their investment decisions. First, we show that as women’s opportunities to improve their income expand, these pressures tend to become more severe. In particular, women face greater pressures as a result of receiving an unconditional grant, which helps explain the lack of impact of such transfers. Second, we test how the impact of the cash grant varies when it is disbursed under conditions that may limit expropriation pressures. A savings account was opened for a group of grant recipients. These businesswomen had to make regular deposits for a period of 6 months and reach a specific savings goal before receiving the grant through the mobile account. These conditions have the potential to reduce expropriation pressures or improve women’s ability to resist these pressures. They can create incentives for women to actively evade expropriation, since giving in to these pressures could mean foregoing the grant. These also provide intended recipients with a compelling rationale they can offer when resisting demands to share their resources during the savings period, which may lower the social costs of doing so. Finally, the savings accounts may make it easier for the women entrepreneurs who start using them to shield or hide part of their income from other household members, as the balance, but not necessarily the existence, is private. We find that when the disbursement of the cash grant is conditioned to the savings intervention, the grants have a positive impact on the investment, sales, and profits of the women- owned businesses. Moreover, we find that the impact is greater and economically large for women who were more vulnerable to expropriation at baseline. 4 Thirdly, we explore alternative mechanisms for these results including self-control issues, liquidity constraints, and access to savings in banking, and do not find evidence supporting those. Targeting and mechanisms of overcoming expropriation are hence of critical importance to policy. Our results indicate that intra-household dynamics and specifically expropriation pressures matter for female entrepreneurs’ investment decisions. They also suggest that well-designed interventions –such as commitment savings mechanisms tailored to the constraints that women entrepreneurs face– can enhance their autonomy in the face of these pressures and help them grow their business. The rest of the paper is structured as follows: Section 2 describes the data collection process, the sample, and the interventions. Section 3 introduces the estimation strategy and presents the main results of the experiment. Section 4 reviews the evidence on expropriation pressures specifically. Section 5 concludes by discussing policy implications. 2. Data and impact evaluation design 2.1 Sample selection and randomization To constitute a sample frame for this the study sample, we first conducted a listing collecting basic information on 10,483 women entrepreneurs in Accra. To do so, we selected a sample of 528 Enumeration Areas (EAs), plus an additional back-up sample of 133 EAs, among the 2,132 EAs within the boundaries of the Accra Metropolitan Area. 3 We then used a random walk process to select women entrepreneurs in each EA included in the study sample. This random walk process was conducted separately for (i) businesses that were visible from the street, (ii) household businesses that were not visible from the street, and (iii) businesses located in markets. The second step of the sampling process was to identify women entrepreneurs who were eligible, willing to participate in the study, and whose spouses were also willing to participate in the study. 4 Given our focus on intra-household dynamics, we only included in the sample frame women business owners living with a domestic partner (“married” 5) aged 20 to 60 years old. This 3 The selection of the EAs was random, with a probability proportional to size, defined as the population of each area, after stratifying by the number of women entrepreneurs and the average room occupancy (a proxy of wealth) in each EA. This sampling method is common in environments where the majority of firms are informal. 4 The requirement to have joint participation of women and men in the study (participate in a baseline survey) may mean that couples with low levels of interest in join participation/low cooperation are missing from our sample, but it is hard to predict its implication as other reasons such as lack of time or business commitments could also explain their lack of participation. 5 In the definition of “married”, we include living with someone even if not officially married. Of the women in our sample, 4.2% were in polygamy marriages. 5 yielded a sample of 7,006 women. We contacted them in a random order until identifying 3,096 couples who met the criteria for participating in the study. 6 When structuring our experimental sample, we wanted to anticipate general equilibrium effects, and more generally ensure that the results would not be influenced by the positive or negative effects, that the interventions could have on control firms. To do so, we completed the randomized allocation at the EAs level, rather than at the individual level. This means that our standard errors are clustered at the EA level. Still, given the low level of correlation of the main outcome of interest (profits) within these clusters in the listing and baseline surveys, the clustering does not materially affect our statistical power. For the randomization process, we stratified the sampled EAs by (i) the number of sampled firms in the EA; (ii) whether most of the EA was covered by a market; (iii) average capital stock of in-sample businesses in the EA; and (iv) average profits of in-sample businesses in the EA. Using these four stratification dimensions, we conducted a block randomization. More specifically, we created a first level of strata using the first three dimensions, and sorted the EAs within each of these primary strata by average profits. We then used within-strata profits ranks to create blocks of EAs that were randomly assigned to one of the five groups as follows: • control group • unconditional grant paid to mobile money account (T1) • unconditional grant disbursed to the woman entrepreneur’s spouse (T2) • grant conditional on participating in a household decision-making training (T3) • grant conditional on reaching a savings goal (T4) Given the expectation pre-interventions of high take-up of the unconditional cash grants and of lower take-up for treatment groups 3 and 4, as well as the importance of studying heterogeneity impacts for groups 3 and 4, we have purposedly oversampled treatment 3, treatment 4 and the control group as follows: 354 women entrepreneurs in treatment 1; 381 in treatment 2; 767 in treatment 3; 823 in treatment 4; and 771 in the control group. 6 The initial sampling objective was 3,000 women entrepreneurs due to budgetary concerns, but the field work implementation led to slightly overachieving this target. 6 To check for balance between the different experimental groups, we estimate for each outcome of interest and heterogeneity dimension variable the following equation: 0 = + 1 1 + 2 2 + 3 3 + 4 4 + (1) where Yi0 is the baseline value of the outcome/heterogeneity variable considered and Group1/Group2/Group3/Group4 capture the treatment assignment. Errors are clustered by EA. Table 1 provides balance checks on a set of baseline variables, showing balance on observables across the five groups. It also provides information about the characteristics of the entrepreneurs, spouses, and firms in our experiment. The women entrepreneurs were on average 39 years old (with spouses averaging 45 years of age) and have completed, on average, 7 years of formal education with 18% of them having completed secondary education or more. Also, women had at baseline around 10 years of experience in their businesses. Eighty percent of women had at least one dependent child, and the average number of dependent children was two. Half of the women had an entrepreneur husband, 37% had a wage worker husband, and 7% report that their husbands were unemployed. Fifty-four percent of the women worked selling goods, 22% in food transformation or restaurants, 12% in the textile sector, and 11% in beauty care services. The average number of employees excluding herself in the woman’s business was less than one, and the maximum number of employees is 13 (7 employees if we consider only paid employees). The women’s average business sales for the past month was USD 350 (winsorized the top at the 1% level). The business profits of women-owned firms in the past month were, on average, USD 83 (winsorized the bottom and top at the 1% level). 2.2 Description of the interventions and take-up The interventions were implemented with the support of mobile phone operators including MTN and Airteltigo for the transfer of the cash grant, FMMS Ghana who conducted the training as part of treatment 3, and Fidelity Bank for the savings product as part of treatment 4. The work was coordinated by Innovations for Poverty Action (IPA). In the first treatment, we provided an unconditional cash grant of 500 Ghana cedis (equivalent to then USD 120) privately to female entrepreneurs through mobile bank accounts. 7 The grant amount, roughly equal to two months of median profits for women-owned businesses, is comparable to the amount disbursed in an earlier experiment conducted by Fafchamps et al (2014). One goal of this treatment arm was to further test the hypotheses outlined in Bernhardt et al. (2019). Specifically, we consider how such grants affect the performance of the businesses of the recipients’ spouses, and whether these effects depended on whether the recipient’s business is more profitable than that of her spouse. Moreover, we compare the effects of these unconditional grants to the effects of transfer mechanisms in other treatment arms. These unconditional grants also serve as a benchmark to test the effects of the interventions in the other treatment arms. We tried to maximize the impact of the unconditional grants by completing the payment done through mobile money accounts, which have become increasingly common in Ghana. 7 The mobile payment could, in comparison to cash grants, reduce risks of not being delivered and help protect the usage as they are done privately. While it may reduce the pressure to redistribute to household members, 8 it is important to note that other household members might still have obtained knowledge of the payment. Specifically, within recruitment, women entrepreneurs and their spouses were told that their application made them eligible to be randomly selected to receive a grant. Still, after selection, spouses (except those in T2) were not told about which condition the women entrepreneurs were selected into. Secondly, it may be possible that recipients told their spouse and other household members about the funds they received. Finally, the other household members could potentially have observed any changes in the expenditures of the beneficiaries. For the mobile payment, women entrepreneurs were contacted over the phone by the service providers between August and September 2017 to confirm on which number to send the mobile money grant. Out of 354 women in T1, the program was able to speak with 348 women (98%) and was able to disburse 337 grants of 500 Ghana cedis (95%) by October 2017. The take- up of less than 100% is due to the implementing team not being able to get confirmation from some beneficiaries of their mobile phone numbers. In the second treatment arm, the grant was transferred to the female entrepreneur’s domestic partner/husband. The transfer to the spouse was not announced to her, as the 7 Account ownership at a financial institution or with a mobile-money service provider in Ghana was at 57.72% in 2017 among those with 15+ years, an increase from 40.51% in 2014. It went up to 68.23% in 2021 according to the World Development Indicators: https://data.worldbank.org/indicator/FX.OWN.TOTL.ZS?locations=GH. 8 Riley (2024) finds that loans paid through mobile money in Uganda lead to higher investment and business performance compared to providing loans as cash. 8 interventions were independent from data collection. This treatment could lead to a range of potential effects on female entrepreneurs. First, spouses who receive a grant could cover a greater share of the household’s expenditures, which would free part of their wives’ income. Second, spouses could transfer part of the grants directly to their female entrepreneur, so that it can be invested in her business as an opportunity for the household. Third, the spouse could spend the cash grant directly on private consumption or invest it in his own business activities. Spouses were contacted over the phone between August and September 2017 (at the same time as the other grant female beneficiaries). Out of the 381 spouses in this group, we could reach 376 men (99%) and eventually provided 370 grants to spouses (97%) in October 2017. In the third treatment arm, the grant was only disbursed if the female entrepreneur and her spouse attended a household support training which included a lab-in-the-field experiment containing a variety of public good games followed by a debriefing, emphasizing the value of cooperation within the household. The inclusion of the lab-in-the-field experiments was to encourage experiential learning. An example of a successful use of experiments to learn is the study by Abel et al. (2021) in South Africa, where participants that are unsuccessful in winning a probabilistic game were subsequently also less likely to participate in a lottery. In our case, the game highlights that low cooperation between domestic partners can lead to lower outcomes, a message that is further reinforced by the debriefing. Between July and August 2017, 767 women entrepreneurs and their spouses in this treatment were invited to participate in a training program that included a lab-in-the-field experiment. In total, 98 rounds were held, each with a maximum of 11 couples. The training included two sessions. The first session (3 hours) included a sequence of lab-in-the-field experiments on the household allocation of resources, as well as a debriefing discussion session on intra-household cooperation. The second session (1 hour, two weeks after the first session) served as a refresher training to cement the lessons from the first session. The lab-in-the field experiment was designed with the goal of evaluating two related hypotheses: (i) limited cooperation within the household is associated with a loss of income and (ii) this limited cooperation is associated with gender biases in resource allocation. During the training experiments, we controlled for differential returns to investment by experimentally varying whether the man or the woman owns a business with a profitable investment opportunity. In a sequence of different decision-making games with women entrepreneurs and their spouse, participants faced a tradeoff between keeping the money for themselves, investing this in a 9 business opportunity owned by one of them, or investing this in a shared opportunity. This way we were able to identify gender-driven differences in investment behavior of couples and use the average results to inform the debriefing sessions. Women who completed the full training program with their spouse were eligible to receive the 500 Ghana cedis grant through the mobile payment. At the end of the program, out of the 767 couples invited to the training, 586 women successfully completed the training with their spouse and were eligible for the grant (76%), and 584 women eventually received their grant through mobile money in October 2017. Finally, in the last treatment arm, the 500 Ghana cedis grant was disbursed only if the woman entrepreneur met a savings goal of 160 Ghana cedis (approximately a third of the grant amount), deposited in at least eight different installments in a free bank account with Fidelity Bank, within six months. Between February and August 2017, 823 women were invited to join a 6-month savings plan program with Fidelity Bank, a commercial bank in Ghana which we partnered with for this study. The savings program consisted in organizing the visit of a field agent (“smart friend”) every week to women entrepreneurs to help them save regularly. During their first visit, the smart friends received the consent of the participants and opened a “smart account” for them at Fidelity Bank. This bank account was opened for free (Fidelity Bank sponsored the 20 Ghana cedis - approximately $4.5 - minimum deposit), with no deposit nor withdrawal charges, and an annual interest rate of 3%. If a beneficiary agreed to join the program, the smart friends visited her once a week until the end of the 6 months to collect weekly deposits (it was not mandatory to deposit). At the end of the program, out of the 823 women entrepreneurs in this group, 744 women joined the program (90%), 665 women were successfully eligible for the grant, and 663 women (81%) eventually received the grant through mobile money in October 2017. The savings plus grant intervention could help overcome constraints in investing in the business through the following mechanisms: (i) utilizing the bank account as a means of protecting from expropriation pressures; (ii) using the bank account and engagement as a self-commitment device to achieve a goal towards investment; (iii) use the savings as a source of liquidity - increasing the lump-sum amount available to invest (the minimum saving at the end would represent 32% extra funding on top of the grant); and (iv) finally, the marginal access to banking services. We test these mechanisms to help explain our results. 10 2.3 Timeline of the data collection and interventions The analysis presented in this paper was conducted using the baseline survey, completed in late 2016, and two follow-up surveys performed respectively in late 2017 (partial survey) and early 2019 (full sample survey). The baseline survey recorded detailed information on 3,096 women entrepreneurs and with interviews with their spouses. This information includes data such as individual and business characteristics, employment, investment, usage of financial services and finances, business performance, and indicators of harassment. The different interventions were implemented between January and October 2017, and all participants received the grants at the same time, in October 2017. In November-December 2017, a first partial follow-up survey was administered to a subset of the women in the sample (1,350 women entrepreneurs), randomly selected from the original baseline sample. The objective of this quick survey was to get detailed information on immediate grant usage. Approximately 93% of the targeted sample in the control group responded to this first follow-up survey and there was no differential attrition across treatment status (Table A1 in the Appendix). In January to May 2019, a second and full follow-up survey was conducted targeting all respondents interviewed at baseline, and gathering information on both the women entrepreneurs and their spouses. The 1.5 years post-intervention follow-up survey elicited information on business performance, business practices, access to banking, finance, and grants, and household dynamics. The participation rate of women entrepreneurs from the baseline was of 92% in the control group and no significant difference at 5% level across treatment arms (Table A1 in the Appendix). 9 The analysis presented here focuses mostly on the full follow-up survey. An additional round of the endline survey was scheduled for 2020 but could not be carried out because of the COVID-19 pandemic. 9For the full follow-up survey, participation of women is slightly lower for the treatment where the husband receives the cash grant (T2), but this is not significant at the 5% level (p = 0.087). The attrition was larger for interviews with husbands (17.6% for the control group), and there is less attrition in groups receiving only cash grants. 11 Figure 1: Timeline of the surveys and interventions 3. Measuring impact 3.1 Estimating impacts We estimate the impacts on both individual and household level outcomes using the following ANCOVA specification: 1 = + 0 0 + 1 1 + 3 2 + 5 3 + 7 4 + ∑ =1 + (2) Where Yi is the outcome measure considered for individual i, Yi0 is the baseline measure of this outcome variable, and Group1/Group2/Group3/Group4 capture the treatment assignment. We also use dummy variables (blockn) to control for the blocks created as part of the randomization process. Errors are clustered by EAs. 3.2 Impacts of the interventions on beneficiary behavior, business performance, and household efficiency We lead with three sets of results. The first result is that the two conditional treatments, the training treatment (T3) and the savings treatment (T4), resulted in changes in household allocation of resources and savings behavior respectively, in line with their design. The second result is that they also led to increased investment and in some measures of profits, notably the savings treatment (T4). This was not the case for the unconditional treatments, regardless of whether the cash grant was given to the woman entrepreneur (T1) or her spouse (T2). The third result is that although women in dual entrepreneur couples seem to invest less of the grant received and make 12 lower profits than the households where the spouse is not an entrepreneur, there is no evidence that this is the result of efficiency concerns. In fact, the allocation of resources between dual entrepreneur couples appears to be inefficient, with grants not allocated to the most profitable businesses within the households. Impacts on household-business behavior and savings The training treatment (T3) led to changes in the household allocation of resources. As shown in Table 2, female entrepreneurs that were assigned to the training treatment (T3) received higher levels of “chop money” (the name given in Ghana to the regular monetary transfers that many men give to their spouses for household maintenance), felt that their husband is supportive of their business, and felt that they would get their husband’s support if their business had a difficult month. Female entrepreneurs in the treatment where the husband received the transfer (T2) were also more likely to feel higher degrees of support, but there is no significant impact on chop money. Treatments T3 and T2 were designed to influence the decision-making within the household by providing an interactive training (in T3) or providing liquidity to men (in T2). Treatment T4 did not target household decision-making and had no significant effects at that margin. Turning to savings, Table 3 shows that approximately 1.5 years after the grants, there is an increase in the share of women who save across all treatments, compared to the control group. The amount saved is also higher in all treatments except for the unconditional cash grant given to the woman (T1). Both the savings (T4) and training treatment (T3) led to a higher long-term share of women saving using a bank account. The share was the highest in the savings treatment (T4), highlighting that this treatment increased the adoption of bank accounts for savings purposes, as it was designed to do. Compared to the training treatment (T3), the savings treatment (T4) encouraged the use of bank accounts at the extensive margin (whether a bank account is used) and less at the intensive margin (the amount used). Impacts on business performance Table 4 shows how these different treatments affected investment and business performance. In line with earlier studies, the unconditional cash grant did not lead to an increase 13 in investment or business performance of women entrepreneurs. There are no significant impacts regardless of whether it is given to the woman entrepreneur (T1) or her spouse (T2). Table 4 shows that, for these two groups, there is no significant impact on investment, sales or profits and the point estimates are in fact in some instances negative. The treatments with a grant conditional on participating in the training (T3) or reaching a savings goal (T4) had positive and significant impacts on female entrepreneurs’ business investment, 0.07 (T4) to 0.11 (T3) standard deviations above the control group mean. 10 In addition, conditioning the cash grant on reaching a savings goal (T4) led to an impact of 15% on monthly sales and 10% on monthly profits for the women entrepreneurs. These differences are significant at 5 and 10 percent levels respectively, but do not remain significant for all measures of sales and profits (impacts on z-scores are positive but not significant). Compared to the unconditional cash grant recipients (T1), the difference in monthly profits is significant at the 1% level (p = 0.006). The effects of the training treatment (T3) on sales and profits although positive are not significant for women, suggesting that their intensity was insufficient to change performance on women- owned businesses at this margin. For husbands, table A6 shows that impacts on sales, profits and total income (incorporating the multiple sources of earnings irrespectively of having or not a business) are not significant for T1 and T2. 11 For all the groups, the grant had no impact on the husband’s business performance, even when restricting these results to the men with a business at baseline. 12 Still, the conditional grants (T3 and T4) had impacts respectively of 19% and 13% on husband’s total income. Furthermore, table A7 shows that husbands within households where women faced high pressure to share resources at baseline 13 benefited in terms of business performance (sales and profits) and total income from the grant conditional on participating in the joint-decision making training (T3) and to a lesser extent from the savings intervention (T4). We cannot reject that these are different from T1 and T2 among the households where women were vulnerable to high- 10 The first partial follow-up also indicates that the women report using the funds for investing in the business, especially in T3 and T4, and significantly less in T2. 11 The partial first follow-up suggests that the grants to men were on average relatively more spread between his business, some investment in her business, and other non-business purposes. 12 The results for the men with a business at baseline are available upon request. 13 The questions used for measuring the intra-household sharing score include: 1- Whenever I have money on hand, either my spouse or other family members always end up requesting some of it; 2- People who do well in their business here are likely to receive additional requests from family and friends for money to help out with some expense or another; 3- Machines and equipment held in my business are a good way of saving money so that others don’t take it; 4- Without the income earned in my business, my household would have a hard time having enough money to buy food or pay school related expenses. 14 pressure at baseline, but the findings against the control group suggest that the husbands in this group have benefited personally from the grant provided to women, which may have mitigated the overall impact for women of these interventions. Analysis in dual-entrepreneur households The next question we analyze is whether the lack of impacts on women’s investment and performance in the unconditional cash grant treatments is due to efficiency considerations. We focus on dual entrepreneur couples, where efficiency considerations can be measured as the performance of both partners’ businesses is known. Bernhardt et al. (2019) find that cash grants typically have no positive impacts for female business owners because the beneficiaries invest the grant in their husband’s activities when he is also an entrepreneur. They further suggest that this is because women entrepreneurs seek to optimize income at the household-level, investing in the activities in which returns to capital are the highest, even if it is not their own. We analyze these results for those receiving the unconditional grant and the control group. Similarly to Bernhard et al. (2019), women from dual-entrepreneur households in our study who received the cash transfer invested less in their businesses than women who were the only entrepreneurs in their households, although that result is not significant (Table 5). Moreover, we do not find evidence that the lower investment in dual-entrepreneur households is the result of efficiency considerations. From a household income maximization perspective, investing the cash grant in the business with the highest return on investment would be more efficient. Using profits 14 at baseline of both female and male businesses as a measure for profitability and restricting to the sample of dual-entrepreneur households (51 percent of the sample), we do not find evidence for this hypothesis. Table 6 breaks down the impacts on investment, sales, and profits by interacting the treatment effects with a variable for whether her business is more profitable at baseline. 15 The unconditional cash transfer does not lead to higher 14 A more adequate measure would be the marginal returns to capital, which is not observed, as it is not the same as past returns on investment. Profits are not necessarily correlated with marginal returns to capital, as firms in higher profit levels may have reached a peak and not have higher returns going forward. However, the household may use the business performance as a proxy of future returns when considering efficient allocation. 15 While we did not have a full measure of capital invested for men at baseline, we have also computed table 6 with a “imperfect” measure of return of investment and still found no positive impact of cash grants for women businesses with higher return on investment than her husband at baseline. The coefficients in that case are positive for some outcomes but not significant. Results available upon request. The imperfect measure we are using for capital investment is the answer to ‘How much do you purchase at a time when you purchase inventories or raw materials’ instead of a stock measure of the value of assets + inventory. 15 investment, profits, or sales when the women’s business is more profitable than their spouse at baseline. The point estimates of the interaction effect and combined effect are negative, although not significantly different from zero. 16 4. Mechanism: The role of expropriation pressure Next, we provide evidence for expropriation pressures as an explanation for the lack of impact of grants on the performance of women-owned businesses. We first show that receiving an unconditional grant (T1) increases expropriation pressures for women. As noted, the payments were made privately but done in a real setting, so not guaranteed to be fully hidden. We also show that this increased pressure is not unique to receiving a grant, but can also be seen among women in the control group who report an increase in profits. Second, we show how the savings treatment (T4) is effective in leading to higher sales and profits for women with high expropriation pressures at baseline. Among other effects, the savings treatment allows them to increase spending on raw materials, a type of spending that involves a more regular expenditure. To measure expropriation, we use an intra-household sharing score, which is an index based on women’s assessment of the pressure they face within the household to redistribute any income or assets they have. This score was collected for women at baseline and follow-up. 17 4.1 Evidence of household pressures on earnings Women who receive an unconditional cash transfer (treatment T1) are more likely to report that sharing pressures within the household increase, even for those less vulnerable at baseline. Table 7 highlights the effects of the interventions on various measures of within and outside household pressures. Women who faced less pressure at baseline and who received the unconditional cash grant (T1) report a higher sharing score 1.5 years after receiving this grant. This increase is not statistically significant for other treatments. Furthermore, for women in T1 who already faced high pressures at baseline, their pressures remained unchanged (the combined effect of the treatment 16 In the case of the husbands, as with women, the unconditional cash transfer treatment (T1) does not lead to higher investments, profits, or sales when the husband’s business is more profitable. Results available upon request. 17 The questions used for measuring the intra-household sharing score include: 1- Whenever I have money on hand, either my spouse or other family members always end up requesting some of it; 2- People who do well in their business here are likely to receive additional requests from family and friends for money to help out with some expense or another; 3- Machines and equipment held in my business are a good way of saving money so that others don’t take it; 4- Without the income earned in my business, my household would have a hard time having enough money to buy food or pay school related expenses. 16 and the interaction of the treatment with a high-sharing score is statistically not different from zero). These results suggest that the unconditional grant (T1) can lead to an increased and persisting vulnerability to expropriation. Correlational evidence based on the control group suggests a similar relationship between earnings and pressure, with receiving a grant and pressure. 18 Table 8 shows a regression of the intra-household sharing score on increases in profits and investments for the control group. Women in the control group who increase profits between the baseline and endline survey, also face an increase in the intra-household sharing score at endline. This positive conditional correlation is only present for outcomes (profits) but not for inputs (investments). Nevertheless, since this is correlational, reverse causality cannot be ruled out in this relationship (for example, women who face increases in household pressures, might seek to expand earnings to be able to meet these pressures). The presence of expropriation pressure provides a rationale for women to give the impression to their spouses that their income is lower in order to reduce these pressures. Reports on own income and the spouse’s income suggest that men indeed underestimate their wife’s income. Figure 2 shows the difference in reporting of profits for both the wife and the husband. Each dot represents a couple. A positive difference means that the spouse is underestimating profits, while a negative difference means that the spouse overestimates profits. Many of the men in the sample report that their spouses have lower profits than they actually have. This underestimation gets larger when the profit of the woman increases. Conversely, for men’s earnings the opposite pattern occurs: many women overestimate their husband’s profits, and this also increases when his profits increase. Similar results can be found for earnings. These results are consistent with the presence of income-hiding by women. 18 A grant can be seen as a windfall and therefore be subject to a different form of expropriation than business income (although this is less likely after 1.5 years following receipt of the grant). 17 Figure 2. Differences in reporting of profits Women Husbands 2000 2000 1000 1000 diff. in profits in USD 0 0 -1000 -1000 -2000 -2000 -2 0 2 4 -2 0 2 4 z-scores of profits z-scores of profits Note. The difference in profits for women is her self-reported profits minus her profits as reported by her husband. The difference in profits for men is his self-reported profit minus his profits as reported by his wife. A positive difference therefore means that the domestic partner underestimates the spouse’s income, while a negative difference means that the domestic partner overestimates the spouse’s income. All values are from the endline. 4.2 The role of the savings treatment in resisting expropriation pressure As highlighted earlier, the savings treatment is the only treatment that sees positive impacts on profits and sales. The savings treatment (T4) offers women a technology to resist expropriation pressure. Heterogeneity analyses suggest that the savings treatment is especially effective with women who face high household pressure at baseline. Table 9 shows the impact of the various treatments interacted with whether the women had high intra-household pressure at baseline. Among the women with high intra-household pressures at baseline, only in the savings treatment (T4) are there positive and significant coefficients on sales and profits. These effects are large – over 0.2 standard deviations above the control group mean. This indicates that the savings 18 mechanism was effective in increasing sales and profits for this group. 19 For the other treatments, there is no significant impact. There is also no impact on the group that reported low intra- household pressures at baseline. 20 The savings treatment (T4) did not change reported pressures, as the earlier results from Table 7 highlight. There is no significant change in the intra-household sharing score for women with high and low sharing scores. This suggests that the treatment did not change pressure, but allowed women to have a better way of coping with it. An implication for investment is that many women invest more in raw materials. Table 10 highlights impacts on investment 1.5 years after the receipt of the cash grant. Women that saw high pressures at baseline increased their investments especially by increasing their purchases of raw materials. Compared to bulky investments that are often one-off, raw materials are typically recurring expenses and require regular cash at hand, which can be expropriated more easily. Women who can make larger or more regular purchases of raw materials may be able to do so at a lower unit cost. Interventions that increase women entrepreneurs’ savings or assets without improving their ability to cope with expropriation pressure, may not affect this important determinant of profitability. 4.3 Alternative mechanism explanations We test for alternative mechanisms to expropriation pressure that could help explain the positive impacts on the performance of women-owned businesses of conditioning the grant on the regular savings intervention (T4). Specifically, we examine (i) self-control issues; (ii) liquidity constraints; and (iii) access to savings in banking. Firstly, the bank account could help overcome individual self-control issues related to time- inconsistent preferences. Women with lack of self-control may use the regular savings mechanism with a grant for conducting an investment they normally would not do, which could have payoff 19 We have also conducted additional heterogeneity analysis of the impacts on business performance of each intervention and found that the savings treatment (T4) has large positive effects on business performance for women that at baseline transferred money out in the past 12 months. 20 A potential reason for no impact of the savings intervention on women facing low pressure is that the treatment led to less of behavior changes for women with already low pressure at baseline. These women were less likely to see a change in the long-term savings behavior compared to women reporting high pressures at baseline. The increase of long-term adoption of a savings bank account was about twice higher for women with high baseline pressures. There is no significant long-term impact on the amount saved in a bank account for the low-pressure group either. Nevertheless, the low-pressure group in the savings treatment (T4) did manage to avoid an increase in pressures, unlike the low-pressure group in the unconditional cash grant treatment (T1). 19 in the future. If that is the case, we would expect to see the conditional grant with savings (T4) to have a positive impact on investment and performance for the women with time-inconsistent preferences at baseline. Fafchamps et al. (2014) provide suggestive evidence that (relative to a cash grant) an in-kind grant intervention helps entrepreneurs with self-control problems. On the other hand, Riley (2024) uses a self-control index composed of whether a woman had hyperbolic time preferences, whether she was impatient at baseline, and whether she did not save for her business, and do not find heterogeneous effects on providing a loan in-cash or in a mobile payment. In our study, Table A2 in the Appendix shows that women with high discount rates at baseline that received the savings support intervention (T4) did not perform better than those with low discount rates in the same group, and there is no clear indication in any of the groups that overcoming self- control issues is driving the main effects. Secondly, the regular savings in preparation for the grant could increase the accumulated size of the investment available to the entrepreneur by a minimum of 160 Ghana cedis over the 500 Ghana cedis of the grant (at least by 32%) and hence support those with liquidity constraints. We test the importance of investing a lump-sum amount that could be higher than in the other groups by assessing the heterogenous impacts of the savings treatment (T4) by the size of the firm at baseline. If the mechanism of change in the savings treatment was through increased liquidity associated to the intervention, one would expect that the women with smaller levels of sales would benefit more from the intervention. Table A3 presents the heterogeneity of the impacts on business performance by the sales of the women-owned firms at baseline. We do not find that firms with lower sales have higher impacts than those with higher sales from the savings treatment (T4). In fact, the results show the opposite, again relatively strong impacts on business performance (0.15 to 0.19 standard deviations above the control mean) for women with larger sales at baseline, which is consistent with women with larger businesses facing higher expropriation pressure and hence benefitting more from the intervention. Thirdly, the intervention helped women entrepreneurs open a bank account. There is evidence from other studies (e.g. in Kenya by Dupas and Robinson, 2013) that providing female entrepreneurs with access to savings accounts enabled investments in the business and increased consumption. If the positive impacts of the savings treatment are specifically due to increased access to banking with the associated products and services that come with it, rather than the reduced expropriation mechanism, we should expect to see larger effects for those without savings 20 in banks at baseline. About 30% of women saved in a bank account at baseline. Table A4 shows that women with no savings in a bank account at baseline do not have larger effects from T4 than those that were already saving in formal instruments, suggesting that this is likely not the main driver of the results. In sum, relative to expropriation, these three alternative mechanisms do not seem to explain the impacts of conditioning the grant payment to the savings treatment. 5. Conclusion In this paper we confirm that women who receive unconditional (relatively small) grants for micro- entrepreneurship have limited changes in business performance. Women who receive unconditional grants do not efficiently invest in their business due to intra-household dynamics. The grant itself may be expropriated, or the anticipation of expropriation of any returns on the investment deters them from investing (appropriately) in their business. This study developed and tested a promising relatively cheap mechanism of involving couples in capacity-building efforts to promote joint decision-making within the household on the allocation of resources. This conditional intervention led to increased support to women in their business operations, higher savings albeit not in formal accounts, and as presented in Appendix Table A5, reduction in women's contribution to household expenses, increased level of happiness, and expanded hours worked in the business. The female beneficiaries of this intervention invested further in their business, but that investment did not generate significant returns on business performance by the time of our follow-up survey. Finally, the paper shows strong positive effects on business performance for women who received conditional grants to a savings support and started with high vulnerability to expropriation. The grant allows them to sustainably reduce their vulnerability to expropriation, through a savings account. Because of the intervention, they can protect their working capital, which leads to improvement in investment and business success. This evidence expands from the existing literature that unconditional grant payments are not effective, and demonstrates the importance of family pressure in the allocation of resources. The paper is a contribution to highlight the importance of intrahousehold dynamics in female entrepreneurship in developing economies. Women often lack authority over the allocation 21 of household assets and may face pressure to share resources, and the study identifies a scalable mechanism of addressing them. It also sets the pathway for expanding the depth of the mechanisms involving couples to help ease constraints related to the distribution of household resources. Interventions to promote firm growth should consider the heterogeneity within households and enterprises. When they do, policies can be scalable, targeted, and impactful. 22 References Abel, M., S. Cole, and B. Zia. 2021. Changing Gambling Behavior through Experiential Learning, World Bank Economic Review, Volume 35, Issue 3, October 2021 Baland, J-M., I. Bonjean, C. Guirkinger, and R. Ziparo. 2016. The economic consequences of mutual help in extended families. Journal of Development Economics. vol. 123 (C). pages 38-56. Baland, J.M. and Ziparo, R., 2018. Intra-household bargaining in poor countries. Towards gender equity in development, 69(1). UNU-Wider Studies in Development Economics. Bernhardt, A., E. Field, R. Pande, and N. Rigol. 2019. Household matters: Revisiting the returns to capital among female microentrepreneurs. American Economic Review: Insights, 1(2), pp.141- 60. Berge, L., K. Bjorvatn, and B. Tungodden. 2015. Human and financial capital for micro enterprise development: Short-term and long-term evidence from a field experiment in Tanzania. Management Science, Vol. 61, No. 4 (April 2015), pp. 707-722 (16 pages). Brudevold-Newman, A. P., M. Honorati, P. Jakiela, and O. Ozier. 2017. A firm of one's own: experimental evidence on credit constraints and occupational choice. World Bank Policy Research Working Paper, (7977). Buvinić, M., and R. Furst-Nichols. 2016. Promoting Women's Economic Empowerment: What Works? The World Bank Research Observer, Volume 31, Issue 1, February, pp.59-101. Carranza, E., A. Donald, F. Grosset, and S. Kaur. 2022. The social tax: Redistributive pressure and labor supply (No. w30438). National Bureau of Economic Research. De Mel, S., D. McKenzie, and C. Woodruff. 2008. Returns to capital in microenterprises: evidence from a field experiment. The Quarterly Journal of Economics, 123(4), 1329-1372. De Mel, S., D. McKenzie, and C. Woodruff. 2009. Are women more credit constrained? Experimental evidence on gender and microenterprise returns. American Economic Journal: Applied Economics, 1(3), 1-32. De Mel, S., D. McKenzie, C. Woodruff. 2012. One-time transfers of cash or capital have long- lasting effects on microenterprises in Sri Lanka. Science, 335(6071), 962-966. Dupas, P., and J. Robinson. 2013. Savings Constraints and Microenterprise Development: Evidence from a Field Experiment in Kenya. American Economic Journal: Applied Economics. Fafchamps, M., D. McKenzie, S. Quinn, and C. Woodruff. 2014. Microenterprise growth and the flypaper effect: Evidence from a randomized experiment in Ghana. Journal of Development Economics 106 (2014) 211–226. 23 Fiala, N. 2017. Business is tough, but family is worse: Household bargaining and investment in microenterprises in Uganda. Working Paper. Friedson-Ridenour, S., and R. Pierotti. 2019. Competing priorities: Women’s microenterprises and household relationships. World Development, 121, 53-62. Hardy, M., and G. Kagy. 2020. It’S Getting Crowded in Here: Experimental Evidence of Demand Constraints in the Gender Profit Gap, The Economic Journal, Volume 130, Issue 631, October 2020, Pages 2272–2290. Liebman, J. B. and L. Katz. 2007. Experimental analysis of neighborhood effects. Econometrica, 75(1), 83-119. Jakiela, P. and O. Ozier. 2016. Does Africa Need a Rotten Kin Theorem? Experimental Evidence from Village Economies. Review of Economic Studies, 83 (1), pages 231-268. Jayachandran, S. 2021. Microentrepreneurship in developing countries. Handbook of labor, human resources and population economics, 1-31. McKenzie, D. 2017. Identifying and spurring high-growth entrepreneurship: Experimental evidence from a business plan competition. American Economic Review, 107(8), 2278-2307. Riley, Emma. 2024. Resisting Social Pressure in the Household Using Mobile Money: Experimental Evidence on Microenterprise Investment in Uganda. American Economic Review, 114 (5): 1415–47 Schaner, S. 2015. Do Opposites Detract? Intrahousehold Preference Heterogeneity and Inefficient Strategic Savings. American Economic Journal: Applied Economics, 7 (2), pages 135-174. World Bank. 2019. Profiting from Parity: Unlocking the Potential of Women's Business in Africa. © World Bank, Washington, DC. 24 Table 1: Balance table and sample characteristics Treatment assignment Differences T1: Cash T2: Cash T3: HH T4: Savings + Control F test grant (W) grant (M) Training + CG CG N 354 381 767 823 771 Firm Characteristics Years running a business (W) 10.102 9.747 9.930 10.199 10.225 0.216 Number of employees in business (W) 0.531 0.556 0.524 0.513 0.547 0.188 Number of paid employees in business (W) 0.195 0.241 0.214 0.222 0.231 0.305 Woman sells goods 0.537 0.522 0.563 0.520 0.546 0.675 Business practices index (0 to 5) (W) 1.285 1.323 1.330 1.332 1.313 0.122 Individual Characteristics Woman's age 39.692 39.115 39.047 39.441 39.239 0.375 Husband's age 45.275 44.892 45.150 45.218 45.171 0.079 Years of formal education completed (W) 6.893 6.869 7.312 7.194 7.276 0.778 Secondary education or more (W) 0.167 0.163 0.191 0.200 0.182 0.713 Lives in the same household as her husband 0.907 0.903 0.917 0.911 0.911 0.193 Number of dependent children (W) 1.980 1.885 1.845 1.913 2.003 1.091 Number of dependent children (M) 1.946 1.829 1.820 1.787 1.968 1.810 Husband is an entrepreneur 0.525 0.499 0.490 0.492 0.501 0.396 Husband is a wage worker 0.333 0.404 0.368 0.378 0.374 1.011 The woman entrepreneur's husband is 0.076 0.058 0.078 0.083 0.060 1.344 unemployed Economic activities Sales (US$) past month (W) 399.618 364.583 355.595 348.163 402.759 0.637 Profits (US$) past month (W) 81.899 87.101 79.119 80.982 86.857 0.392 Husband's sales (US$) past month 307.322 310.985 252.519 163.488 215.486 3.278** Husband's profits (US$) past month 87.487 93.323 92.281 68.047 75.185 1.624 Woman saves 0.935 0.913 0.927 0.919 0.914 0.439 Woman's total savings amount (US$) 155.314 128.504 184.158 129.872 149.429 1.249 Woman saves in bank account 0.274 0.281 0.301 0.292 0.335 1.270 Woman saves in her own bank account 0.269 0.279 0.292 0.287 0.326 1.114 Woman saves in MFI/ROSCA 0.452 0.478 0.460 0.458 0.468 0.137 Household support Woman receives chop money 0.819 0.848 0.821 0.790 0.778 2.628** Chop money received past month (US$) 69.386 70.470 77.183 71.482 72.440 0.930 Husband supportive of her running a 0.893 0.885 0.890 0.855 0.865 1.455 business Gets husband support if her business had a 0.717 0.744 0.723 0.662 0.709 1.912 difficult month Woman's sharing score normalized 0.464 0.434 0.458 0.467 0.468 1.343 Joint Orthogonality Test (p-value): 0.566 0.148 0.660 0.484 Notes: F test is calculated with regressions that include only treatment groups dummies (where the dummy excluded is the control group variable). In each case the dependent variable is the row variable. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. Husband’s sales and profits impute those who do not have business with 0. Joint Orthogonality Test includes all vars. in the table except the total savings amount, the amount chop money received, husband support if her business had a difficult month due to missing values. The p-value of the F-test is presented. 25 Table 2: Treatment effects on intra-household support Amount chop Husband is Gets husband Husband money supportive of support if her support not Receives chop received in Z-scorea woman business had a reduced if her money the past running a difficult business had a month (US$) business month good month winsorized T1: Cash grant (W) 0.009 -0.027 -2.409 0.043 -0.032 0.053 (0.030) (0.031) (4.093) (0.030) (0.031) (0.041) T2: Cash grant (M) 0.077** 0.030 -2.838 0.075** 0.048** 0.022 (0.035) (0.031) (4.392) (0.031) (0.024) (0.038) T3: HH Training + Cash grant 0.103*** 0.065*** 8.529** 0.095*** 0.075*** 0.067* (0.024) (0.022) (3.739) (0.022) (0.022) (0.035) T4: Savings + Cash grant 0.016 -0.010 1.257 0.041* 0.001 0.014 (0.026) (0.023) (3.854) (0.022) (0.023) (0.035) Control group mean -0.055 0.735 75.660 0.771 0.775 0.361 Sample size 2855 2855 2134 2668 2536 2536 p-value: Treatment 1 = Treatment 2 0.075 0.130 0.921 0.385 0.011 0.462 p-value: Treatment 1 = Treatment 3 0.001 0.003 0.004 0.066 0.001 0.727 p-value: Treatment 1 = Treatment 4 0.821 0.604 0.334 0.937 0.282 0.328 p-value: Treatment 2 = Treatment 3 0.425 0.259 0.004 0.514 0.215 0.220 p-value: Treatment 2 = Treatment 4 0.073 0.209 0.298 0.251 0.041 0.828 p-value: Treatment 3 = Treatment 4 0.000 0.001 0.027 0.005 0.001 0.106 p-value test of equality 0.000 0.003 0.012 0.000 0.000 0.285 Notes: Specifications use the main follow-up survey. Specifications include strata dummies, a variable representing the initial outcome at baseline, and a variable indicating missing data at baseline. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. a Z-score includes different measures of household level dynamics, winsorized and inverse hyperbolic sine. Variables are: 1- woman receives chop money from her husband, 2- the amount of chop money that the woman entrepreneur reports having received from her husband in the past month winsorized (top 1%) in USD, 3- inverse hyperbolic sine of the amount of chop money that the woman entrepreneur reports having received from her husband in the past month in LCU, 4- woman entrepreneur feels that her husband is supportive of her running a business, 5- woman entrepreneur feels that her husband would provide support if her business had a difficult month, 6- woman entrepreneur feels that her husband would not reduce his support if her business had a good month, 7- the sharing score normalized (from 0 to 1), 8- woman has not transferred money out in the past 12 months. 26 Table 3: Impact on savings practices Amount in Total savings Saves in Saves in bank bank account Z-scorea Saves amount (US$) MFI/ROSCA account (US$) winsorized or SACCO winsorized T1: Cash grant (W) 0.027 0.061*** 4.812 -0.003 -1.193 -0.002 (0.037) (0.018) (15.035) (0.025) (8.194) (0.034) T2: Cash grant (M) 0.119*** 0.043** 46.114*** 0.024 5.575 0.091*** (0.039) (0.021) (14.714) (0.028) (7.413) (0.033) T3: HH Training + Cash grant 0.193*** 0.073*** 58.109*** 0.074*** 27.699*** 0.005 (0.034) (0.017) (13.834) (0.021) (6.948) (0.028) T4: Savings + Cash grant 0.216*** 0.047*** 53.805*** 0.155*** 16.495*** 0.032 (0.034) (0.018) (12.049) (0.025) (6.201) (0.029) Control group mean 0.001 0.856 139.6 0.224 36.69 0.417 Sample size 2855 2855 2855 2855 2827 2855 p-value: Treatment 1 = Treatment 2 0.035 0.376 0.017 0.379 0.469 0.009 p-value: Treatment 1 = Treatment 3 0.000 0.446 0.002 0.002 0.001 0.808 p-value: Treatment 1 = Treatment 4 0.000 0.423 0.001 0.000 0.035 0.274 p-value: Treatment 2 = Treatment 3 0.074 0.096 0.466 0.080 0.006 0.006 p-value: Treatment 2 = Treatment 4 0.016 0.834 0.604 0.000 0.143 0.060 p-value: Treatment 3 = Treatment 4 0.512 0.083 0.761 0.001 0.106 0.302 p-value test of equality 0.000 0.000 0.000 0.000 0.000 0.032 Notes: Specifications use the main follow-up survey and include strata dummies, a variable representing the initial outcome at baseline, and a variable indicating missing data at baseline. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. a Z-score includes different measures of financial services, winsorized and inverse hyperbolic sine. Variables include 1- woman has any savings, 2- total amount of savings of the woman entrepreneur winsorized (top 1%) in USD, 3- inverse hyperbolic sine of total amount of savings of the woman entrepreneur in LCU, 4- woman entrepreneur saves in a bank account, 5- total amount of savings in bank account winsorized (top 1%) in USD, 6- inverse hyperbolic sine of total amount of savings in bank account in LCU, 7- woman entrepreneur saves in a MFI or ROSCA. 27 Table 4: Impact on the performance of women-owned businesses Monthly Monthly sales Z-score Z-score profits (US$) Z-score Salesb (US$) profitsa Investmentc winsorized winsorized T1: Cash grant (W) -0.066 -7.874 -0.012 2.927 0.013 (0.060) (5.561) (0.054) (31.463) (0.040) T2: Cash grant (M) -0.002 2.939 -0.044 -20.406 0.063 (0.063) (5.226) (0.051) (24.447) (0.043) T3: HH Training + Cash grant 0.001 2.089 0.008 18.186 0.109*** (0.050) (4.187) (0.047) (21.255) (0.033) T4: Savings + Cash grant 0.067 7.450* 0.069 50.560** 0.071** (0.048) (4.246) (0.043) (21.863) (0.033) Control group mean 0.000 72.83 0.000 325.60 0.003 Sample size 2,868 2,863 2,858 2,849 2856 p-value: Treatment 1 = Treatment 2 0.346 0.079 0.575 0.486 0.335 p-value: Treatment 1 = Treatment 3 0.257 0.068 0.702 0.622 0.026 p-value: Treatment 1 = Treatment 4 0.021 0.006 0.106 0.129 0.176 p-value: Treatment 2 = Treatment 3 0.966 0.864 0.304 0.101 0.308 p-value: Treatment 2 = Treatment 4 0.247 0.364 0.019 0.004 0.858 p-value: Treatment 3 = Treatment 4 0.154 0.178 0.160 0.130 0.266 p-value test of equality 0.202 0.082 0.154 0.038 0.014 Notes: Specifications use the main follow-up survey and include strata dummies, a variable representing the initial outcome at baseline, and a variable indicating missing data at baseline. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively.a Z-score of profits includes 4 variables. Those are: 1- profits generated by the woman entrepreneur’s main business in the past 7 days winsorized (top and bottom 1%) in USD, 2- inverse hyperbolic sine of the profits generated by the woman entrepreneur’s main business in the past 7 days in Ghana Cedis, 3- profits generated by the woman entrepreneur’s main business in the past month winsorized (top and bottom 1%) in USD, 4- inverse hyperbolic sine of the profits generated by the woman entrepreneur’s main business in the past month in Ghana Cedis. b Z-score of sales includes 4 variables. Those are: 1- revenues generated by the woman entrepreneur’s main business in the past 7 days winsorized (top 1%) in USD, 2- inverse hyperbolic sine of the revenues generated by the woman entrepreneur’s main business in the past 7 days in Ghana cedis, 3- revenues generated by the woman entrepreneur’s main business in the past month winsorized (top 1%) in USD, 4- inverse hyperbolic sine of the revenues generated by the woman entrepreneur’s main business in the past month in LCU. cZ-score of investment includes different measures of investment including productive assets, inventory, working capital and inverse hyperbolic sine. The variables are: 1- the total value of productive assets owned by the woman entrepreneur in her main business winsorized (top 1%) in USD, 2- inverse hyperbolic sine of the total value of productive assets owned by the woman entrepreneur in her main business in LCU, 3- the total value of inventory held in the woman entrepreneur’s business winsorized (top 1%) in USD, 4- inverse hyperbolic sine of the total value of inventory held in the woman entrepreneur’s business in LCU, 5- the total value of raw materials or inputs expenses in woman's business in the past month winsorized (top 1%) in USD, 6- inverse hyperbolic sine of the total value of raw materials or inputs expenses in woman's business in the past month in LCU. 28 Table 5: Heterogeneity in treatment effects of unconditional grant (T1) on the investments and profits of women entrepreneurs by employment status of their spouses (Sub-sample: Treatment 1 and control group) Z-Score Investment Z-Score Profits Heterogeneity at baseline Frequencies Treatment 1: Cash grant (W) Treatment 1: Cash grant (W) Husband is an entrepreneur (No) 49% 0.083 0.004 (0.058) (0.084) Husband is an entrepreneur (Yes) 51% -0.054 -0.070 (0.062) (0.079) p-value of difference [0.153] [0.537] Notes: Specifications use the main follow-up survey and include strata dummies, a variable representing the initial outcome at baseline, a variable indicating missing data at baseline, and all treatment arms as well as their interaction with spouse being an entrepreneur. Sub-sample: just Treatment 1 and control group. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. Variables are detailed in Table 4. Table 6: Heterogeneity in treatment effects of unconditional grant (T1) among women entrepreneurs whose spouse is also an entrepreneur by relative performance of the spouses’ businesses (Sub-sample: Treatment 1 and control group) Women's outcomes Z-score sales Z-score Z-score Z-score and profitsa Sales profits investment Treatment 1: Cash grant (W) 0.033 0.071 0.009 0.004 (0.079) (0.086) (0.095) (0.073) Treatment 1 * Her business is more profitable at BL (according to profits -0.117 -0.066 -0.163 -0.067 reported by each one) (0.135) (0.155) (0.152) (0.116) Her business is more profitable at BL (according to profits reported by 0.050 0.114 0.012 0.088 each one) (0.090) (0.094) (0.102) (0.083) Combined effect: Treatment 1 + interaction with her business is more -0.084 0.005 -0.155 -0.062 profitable at BL (according to profits reported by each one) b (0.122) (0.141) (0.133) (0.094) Control group mean 0.027 0.020 0.029 0.063 Sample size 482 479 482 478 Notes: Specifications use the main follow-up survey and include strata dummies, a variable representing the initial outcome at baseline, a variable indicating missing data at baseline, and all treatment arms as well as their interaction with women’s business being more profitable at baseline. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. a Z-score of sales and profits includes same measures of sales and profits including weekly and monthly outcomes winsorized and inverse hyperbolic sine non- winzorized. Variables are detailed in Table 4. b Combined effect is the coefficient related to the sum of the Treatment 1 and the interaction variable (the first and second variables of this regression). 29 Table 7: Effects on women entrepreneurs’ vulnerability to expropriation across treatment groups Sharing score inputs (each of them normalized Does from 0 to 1)b not Sharing transfer Bought score money People assets normalized Without out in Whenever who do is a (from 0 to the income the past I have well in good 1)a earned in 12 money on their way of months my hand … business saving business … ... money … T1: Cash grant (W) 0.043** -0.031 0.044 0.058** 0.022 0.041 (0.018) (0.035) (0.030) (0.026) (0.033) (0.028) T2: Cash grant (M) 0.033 -0.022 0.052* 0.043 0.036 0.029 (0.021) (0.040) (0.029) (0.030) (0.035) (0.033) T3: HH Training + Cash grant 0.014 -0.048 0.019 0.031 0.030 -0.029 (0.015) (0.031) (0.021) (0.022) (0.025) (0.026) T4: Savings + Cash grant -0.002 0.014 0.001 0.002 0.003 -0.023 (0.016) (0.033) (0.023) (0.022) (0.026) (0.028) T1: Cash grant (W) * High intra-hh pressure BL (W) (Above median) -0.050** -0.039 -0.073** -0.044 -0.024 -0.044 (0.025) (0.065) (0.037) (0.032) (0.041) (0.035) T2: Cash grant (M) * High intra-hh pressure BL (W) (Above median) -0.005 -0.034 -0.010 -0.045 -0.005 -0.015 (0.024) (0.068) (0.041) (0.037) (0.039) (0.037) T3: HH Training + CG * High intra-hh pressure BL (W) (Above -0.041** -0.001 -0.031 -0.057** -0.049 -0.047 median) (0.019) (0.053) (0.034) (0.029) (0.034) (0.030) T4: Savings + CG* High intra-hh pressure BL (W) (Above median) -0.007 -0.020 -0.012 -0.011 -0.002 -0.002 (0.019) (0.052) (0.032) (0.027) (0.035) (0.030) High intra-hh pressure at BL (W) (Above median) 0.006 0.066* -0.005 0.029 -0.015 0.029 (0.013) (0.039) (0.021) (0.018) (0.026) (0.022) Combined effect: T1 + high intra-hh pressure at BL c -0.007 -0.070 -0.029 0.015 -0.001 -0.003 (0.021) (0.057) (0.031) (0.027) (0.041) (0.034) Combined effect: T2 + high intra-hh pressure at BL 0.028 -0.056 0.042 -0.002 0.031 0.014 (0.019) (0.062) (0.034) (0.028) (0.035) (0.029) Combined effect: T3 + high intra-hh pressure at BL -0.027* -0.049 -0.013 -0.026 -0.019 -0.076** (0.016) (0.048) (0.029) (0.024) (0.032) (0.026) Combined effect: T4 + high intra-hh pressure at BL -0.009 -0.006 -0.011 -0.009 0.001 -0.026 (0.015) (0.047) (0.025) (0.023) (0.031) (0.025) 30 Control group mean 0.635 0.308 0.437 0.655 0.633 0.787 Sample size 2,855 2,855 2,855 2,855 2,668 2,855 p-value: Treatment 1 = Treatment 2 0.652 0.823 0.818 0.644 0.735 0.729 p-value: Treatment 1 = Treatment 3 0.091 0.609 0.395 0.293 0.805 0.011 p-value: Treatment 1 = Treatment 4 0.012 0.190 0.160 0.030 0.551 0.030 p-value: Treatment 2 = Treatment 3 0.361 0.497 0.248 0.688 0.868 0.071 p-value: Treatment 2 = Treatment 4 0.099 0.365 0.087 0.167 0.334 0.122 p-value: Treatment 3 = Treatment 4 0.262 0.045 0.415 0.178 0.265 0.831 p-value: T1 + interaction = T2 + interaction 0.157 0.834 0.062 0.622 0.457 0.644 p-value: T1 + interaction = T3 + interaction 0.373 0.709 0.624 0.175 0.671 0.036 p-value: T1 + interaction = T4 + interaction 0.918 0.247 0.565 0.416 0.966 0.506 p-value: T2 + interaction = T3 + interaction 0.008 0.913 0.132 0.425 0.162 0.003 p-value: T2 + interaction = T4 + interaction 0.059 0.411 0.111 0.802 0.381 0.170 p-value: T3 + interaction= T4 + interaction 0.294 0.355 0.966 0.521 0.551 0.054 p-value test of equality: T1 = T2 = T3 = T4 0.066 0.293 0.264 0.099 0.646 0.065 p-value test of equality T1 + interaction = T2 + interaction = T3 + 0.118 0.613 0.413 0.713 0.739 0.016 interaction = T4 + interaction Notes: Specifications use the main follow-up survey. Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. a The sharing score is normalized with a mix-max normalization by seeing how much greater the score is than the minimum value and scaling this difference by the range. That is, sharing score normalized = (sharing score - min score)/(max score - min score). b The four variables that compose the sharing score are: 1- Whenever I have money on hand, either my spouse or other family members always end up requesting some of it; 2- People who do well in their business here are likely to receive additional requests from family and friends for money to help out with some expense or another; 3- Machines and equipment held in my business are a good way of saving money so that others don’t take it; 4- Without the income earned in my business, my household would have a hard time having enough money to buy food or pay school related expenses. c Combined effects are the coefficient related to the sum of the treatment and the corresponding interaction variable of the treatment with the high intra-household pressure variable. Table 8: Change in the intra-household pressure in the control group following increases in profits and investments Change in intra-hh Change in intra-hh Change in intra-hh pressure score (W) pressure score (W) pressure score (W) (1) (2) (3) Change of profits a 0.002** 0.002** (0.001) (0.001) Change of investments a 0.000 0.000 (0.000) (0.000) Sample size 682 699 679 R-squared 0.110 0.097 0.110 Control group mean -0.277 -0.253 -0.270 Notes: Specifications include strata dummies. Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. The change is computed as (var Follow-Up - var baseline). aThe change of profits and investments does not include those observations with missing at baseline. 31 Table 9: Heterogeneity in treatment effects on women entrepreneurs’ performance by baseline level of vulnerability to expropriation Monthly Monthly Z-score Z-score sales profits Z-scorea Salesb profitsc (US$) (US$) winsorized winsorized T1: Cash grant (W) -0.078 -0.055 -0.094 -15.821 -10.888** (0.062) (0.064) (0.067) (32.334) (5.488) T2: Cash grant (M) -0.018 -0.049 0.018 -13.233 6.037 (0.064) (0.064) (0.071) (32.310) (5.960) T3: HH Training + Cash grant -0.037 -0.031 -0.037 -0.009 -1.783 (0.051) (0.053) (0.056) (24.600) (4.622) T4: Savings + Cash grant -0.030 -0.033 -0.013 11.793 1.186 (0.052) (0.052) (0.058) (25.858) (5.060) T1: Cash grant (W) * High intra-hh pressure BL (W) (above median) 0.084 0.108 0.068 47.314 7.664 (0.111) (0.120) (0.112) (52.887) (8.952) T2: Cash grant (M) * High intra-hh pressure BL (W) (above median) -0.038 -0.002 -0.071 -26.367 -10.042 (0.104) (0.104) (0.117) (47.686) (10.085) T3: HH Training + CG * High intra-hh pressure BL (W) (above median) 0.110 0.105 0.106 51.055 11.059 (0.091) (0.091) (0.098) (44.071) (7.920) T4: Savings + CG * High intra-hh pressure BL (W) (above median) 0.246*** 0.265*** 0.207** 101.581** 16.221** (0.084) (0.088) (0.091) (44.142) (7.326) High intra-hh pressure at BL (W) (above median) -0.052 -0.071 -0.031 -12.308 -0.982 (0.059) (0.060) (0.063) (28.922) (4.925) Combined effect: T1 + high intra-hh pressure at BL d 0.006 0.053 -0.026 31.49 -3.224 (0.094) (0.099) (0.098) (51.00) (8.926) Combined effect: T2 + high intra-hh pressure at BL -0.056 -0.0513 -0.053 -39.6 -4.004 (0.089) (0.083) (0.104) (34.91) (8.863) Combined effect: T3 + high intra-hh pressure at BL 0.073 0.073 0.069 51.05 9.276 (0.081) (0.080) (0.087) (37.86) (7.105) Combined effect: T4 + high intra-hh pressure at BL 0.216*** 0.232*** 0.194*** 113.4*** 17.410*** (0.068) (0.072) (0.074) (36.82) (6.126) Control group mean 0.001 0.000 0.000 325.60 72.83 Sample size 2,872 2,858 2,868 2,849 2,863 p-value: Treatment 1 = Treatment 2 0.404 0.928 0.157 0.945 0.009 p-value: Treatment 1 = Treatment 3 0.504 0.705 0.393 0.611 0.085 p-value: Treatment 1 = Treatment 4 0.440 0.718 0.237 0.396 0.036 p-value: Treatment 2 = Treatment 3 0.770 0.777 0.416 0.664 0.163 p-value: Treatment 2 = Treatment 4 0.857 0.792 0.658 0.428 0.413 32 p-value: Treatment 3 = Treatment 4 0.891 0.979 0.659 0.631 0.522 p-value: T1 + interaction = T2 + interaction 0.554 0.314 0.813 0.155 0.942 p-value: T1 + interaction = T3 + interaction 0.500 0.844 0.374 0.715 0.192 p-value: T1 + interaction = T4 + interaction 0.018 0.061 0.020 0.116 0.020 p-value: T2 + interaction = T3 + interaction 0.172 0.156 0.261 0.018 0.153 p-value: T2 + interaction = T4 + interaction 0.001 0.000 0.013 0.000 0.013 p-value: T3 + interaction= T4 + interaction 0.061 0.042 0.131 0.120 0.245 p-value test of equality: T1 = T2 = T3 = T4 0.791 0.913 0.594 0.904 0.098 p-value test of equality T1 + interaction = T2 + interaction = T3 + 0.002 0.003 0.022 0.001 0.014 interaction = T4 + interaction Notes: Specifications use the main follow-up survey and include strata dummies, a variable representing the initial outcome at baseline, and a variable indicating missing data at baseline. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. aZ-score includes different measures of sales and profits including weekly and monthly outcomes. b Z-score of sales includes 4 variables. Those are: 1- revenues generated by the woman entrepreneur’s main business in the past 7 days winsorized (top 1%) in USD, 2- inverse hyperbolic sine of the revenues generated by the woman entrepreneur’s main business in the past 7 days in LCU, 3- revenues generated by the woman entrepreneur’s main business in the past month winsorized (top 1%) in USD, 4- inverse hyperbolic sine of the revenues generated by the woman entrepreneur’s main business in the past month in LCU. c Z-score of profits includes 4 variables. Those are: 1- profits generated by the woman entrepreneur’s main business in the past 7 days winsorized (top and bottom 1%) in USD, 2- inverse hyperbolic sine of the profits generated by the woman entrepreneur’s main business in the past 7 days in LCU, 3- profits generated by the woman entrepreneur’s main business in the past month winsorized (top and bottom 1%) in USD, 4- inverse hyperbolic sine of the profits generated by the woman entrepreneur’s main business in the past month in LCU. d Combined effects are the coefficient related to the sum of the treatment and the corresponding interaction variable of the treatment with the high intra-household pressure variable. 33 Table 10: Heterogeneity in treatment effects on women entrepreneurs’ investments by baseline level of vulnerability to expropriation Money Total Raw Total spent on Total Materials Investment Total inputs/raw Productive bought in (LCU) Inventory materials Z-scorea Assets the past IHSb (LCU) IHS each time (LCU) IHS month [stock] [stock] of purchase [stock] (LCU) IHS (3)+(4) (LCU) IHS [flow] [flow] (1) (2) (3) (4) (5) (6) T1: Cash grant (W) 0.010 0.069 0.131 0.001 0.166 0.264 (0.055) (0.190) (0.179) (0.268) (0.242) (0.183) T2: Cash grant (M) 0.082 0.134 0.214 0.205 -0.221 -0.130 (0.053) (0.178) (0.178) (0.203) (0.234) (0.185) T3: HH Training + Cash grant 0.110** 0.164 0.265* 0.240 0.142 0.269* (0.043) (0.137) (0.141) (0.170) (0.186) (0.138) T4: Savings + Cash grant 0.046 0.043 0.070 -0.057 -0.029 0.031 (0.042) (0.145) (0.146) (0.174) (0.188) (0.147) T1: Cash grant (W) * High intra-hh pressure BL (W) 0.007 -0.127 -0.114 -0.222 0.217 -0.132 (above median) (0.086) (0.319) (0.295) (0.389) (0.413) (0.327) T2: Cash grant (M) * High intra-hh pressure BL (W) -0.060 -0.080 0.004 -0.141 0.214 0.488* (above median) (0.085) (0.315) (0.288) (0.359) (0.401) (0.280) T3: HH Training + CG * High intra-hh pressure BL -0.005 0.124 0.033 -0.019 0.412 0.297 (W) (above median) (0.070) (0.232) (0.225) (0.264) (0.318) (0.244) T4: Savings + CG * High intra-hh pressure BL (W) 0.066 0.129 0.168 -0.165 0.783** 0.491** (above median) (0.068) (0.240) (0.227) (0.288) (0.322) (0.249) High intra-hh pressure at BL (W) (above median) -0.004 0.001 -0.011 0.073 -0.225 -0.170 (0.047) (0.182) (0.172) (0.193) (0.221) (0.179) Combined effect: T1 + high intra-hh pressure at BL c 0.017 -0.058 0.017 -0.221 0.384 0.133 (0.063) (0.235) (0.226) (0.252) (0.295) (0.240) Combined effect: T2 + high intra-hh pressure at BL 0.022 0.053 0.218 0.063 -0.006 0.358 (0.068) (0.247) (0.235) (0.281) (0.329) (0.242) Combined effect: T3 + high intra-hh pressure at BL 0.105** 0.288 0.298* 0.221 0.553** 0.566*** (0.053) 0.189 (0.178) (0.213) -0.275 (0.211) Combined effect: T4 + high intra-hh pressure at BL 0.111** 0.172 0.238 -0.222 0.754*** 0.522** (0.053) (0.186) (0.176) (0.227) (0.266) (0.206) 34 Control group mean 0.003 7.891 7.063 6.789 6.557 5.514 Sample size 2,856 2,853 2,821 2,806 2,820 2,825 p-value: Treatment 1 = Treatment 2 0.290 0.776 0.700 0.506 0.173 0.079 p-value: Treatment 1 = Treatment 3 0.087 0.621 0.462 0.383 0.920 0.981 p-value: Treatment 1 = Treatment 4 0.537 0.893 0.740 0.832 0.426 0.209 p-value: Treatment 2 = Treatment 3 0.608 0.864 0.775 0.864 0.120 0.028 p-value: Treatment 2 = Treatment 4 0.508 0.626 0.424 0.226 0.416 0.391 p-value: Treatment 3 = Treatment 4 0.147 0.405 0.182 0.098 0.358 0.087 p-value: T1 + interaction = T2 + interaction 0.953 0.683 0.447 0.350 0.260 0.390 p-value: T1 + interaction = T3 + interaction 0.168 0.120 0.204 0.071 0.571 0.068 p-value: T1 + interaction = T4 + interaction 0.141 0.294 0.308 0.996 0.198 0.086 p-value: T2 + interaction = T3 + interaction 0.223 0.317 0.729 0.561 0.091 0.376 p-value: T2 + interaction = T4 + interaction 0.190 0.610 0.932 0.318 0.019 0.473 p-value: T3 + interaction= T4 + interaction 0.912 0.495 0.722 0.045 0.454 0.823 p-value test of equality: T1 = T2 = T3 = T4 0.107 0.794 0.384 0.433 0.538 0.089 p-value test of equality T1 + interaction = T2 + 0.136 0.432 0.419 0.245 0.028 0.037 interaction = T3 + interaction = T4 + interaction Notes: Specifications use the main follow-up survey and include strata dummies, a variable representing the initial outcome at baseline, and a variable indicating missing data at baseline. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. a Z-score includes different measures of investment including productive assets, inventory, working capital The variables are: 1- the total value of productive assets owned by the woman entrepreneur in her main business winsorized (top 1%) in USD, 2- inverse hyperbolic sine of the total value of productive assets owned by the woman entrepreneur in her main business in LCU, 3- the total value of inventory held in the woman entrepreneur’s business winsorized (top 1%) in USD, 4- inverse hyperbolic sine of the total value of inventory held in the woman entrepreneur’s business in LCU, 5- the total value of raw materials or inputs expenses in woman's business in the past month winsorized (top 1%) in USD, 6- inverse hyperbolic sine of the total value of raw materials or inputs expenses in woman's business in the past month in LCU. b The total investment is the sum of the total value of productive assets and the total value of inventory in LCU, both stock variables, and then we applied the inverse hyperbolic sine transformation. c Combined effects are the coefficient related to the sum of the treatment and the corresponding interaction variable of the treatment with the high intra-household pressure variable. 35 Appendix Table A1: Attrition Only women Only husbands Business Attrition Business Closure Attrition Closure Follow-up 1 Follow-up 2 Follow-up 1 Follow-up 2 Follow-up 2 Treatment 1: Cash grant (W) 0.015 -0.011 -0.003 -0.007 -0.047** -0.006 (0.023) (0.016) (0.007) (0.013) (0.022) (0.016) Treatment 2: Cash grant (M) 0.010 -0.025* -0.005 -0.003 -0.064*** -0.033** (0.020) (0.015) (0.005) (0.013) (0.020) (0.013) Treatment 3: HH Training + CG 0.034 -0.007 -0.004 -0.011 -0.017 -0.020* (0.025) (0.013) (0.006) (0.011) (0.018) (0.012) Treatment 4: Savings + CG 0.019 -0.021 0.004 -0.003 -0.025 -0.012 (0.022) (0.013) (0.008) (0.011) (0.017) (0.012) Control group mean 0.070 0.078 0.007 0.053 0.176 0.064 Sample size 1,350 3,096 1,350 3,096 3,096 3,096 p-value: Treatment 1 = Treatment 2 0.848 0.376 0.710 0.820 0.450 0.084 p-value: Treatment 1 = Treatment 3 0.465 0.779 0.888 0.708 0.164 0.350 p-value: Treatment 1 = Treatment 4 0.842 0.526 0.379 0.753 0.298 0.685 p-value: Treatment 2 = Treatment 3 0.312 0.200 0.815 0.524 0.013 0.282 p-value: Treatment 2 = Treatment 4 0.663 0.753 0.157 0.955 0.034 0.089 p-value: Treatment 3 = Treatment 4 0.576 0.278 0.276 0.405 0.620 0.475 p-value test of equality 0.716 0.379 0.382 0.855 0.013 0.110 Notes: Specifications include randomization strata dummies. Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. Follow-up 1 only included one third of the sample of women entrepreneurs. 36 Table A2: Heterogeneity in impacts on female entrepreneurs’ firm performance by an index of self- control/time preferences Women's outcomes Z-score Z-score Z-score Z-score sales and Sales profits investment profitsa T1: Cash grant (W) -0.061 -0.022 -0.089 0.019 (0.059) (0.060) (0.067) (0.044) T2: Cash grant (M) 0.011 -0.023 0.050 0.066 (0.058) (0.055) (0.065) (0.048) T3: HH Training + Cash grant 0.012 0.012 0.012 0.125*** (0.049) (0.050) (0.053) (0.036) T4: Savings + Cash grant 0.059 0.056 0.071 0.076** (0.044) (0.045) (0.050) (0.037) T1 * entrepreneur has high discount rate (above median) BL 0.089 0.048 0.120 -0.032 (0.125) (0.121) (0.147) (0.090) T2 * entrepreneur has high discount rate (above median) at BL -0.153 -0.074 -0.224 -0.009 (0.140) (0.132) (0.159) (0.103) T3 * entrepreneur has high discount rate (above median) at BL -0.032 -0.012 -0.054 -0.070 (0.112) (0.112) (0.123) (0.085) T4 * entrepreneur has high discount rate (above median) at BL 0.025 0.062 -0.034 -0.049 (0.109) (0.112) (0.118) (0.099) Woman entrepreneur has high discount rate (above median) at BL -0.048 -0.079 -0.015 -0.038 (0.083) (0.084) (0.092) (0.065) Combined effect: T1 + high discount rate at BL 0.028 0.027 0.032 -0.014 (0.114) (0.111) (0.131) (0.082) Combined effect: T2 + high discount rate at BL -0.142 -0.0973 -0.174 0.0574 (0.131) (0.122) (0.150) (0.092) Combined effect: T3 + high discount rate at BL -0.020 0.001 -0.042 0.055 (0.106) (0.105) (0.116) (0.077) Combined effect: T4 + high discount rate at BL 0.0836 0.117 0.037 0.0261 (0.105) (0.106) (0.115) (0.088) Control group mean 0.001 0.000 0.000 0.003 Sample size 2,872 2,858 2,868 2,856 p-value: Treatment 1 = Treatment 2 0.274 0.979 0.071 0.395 p-value: Treatment 1 = Treatment 3 0.222 0.565 0.135 0.022 p-value: Treatment 1 = Treatment 4 0.031 0.159 0.014 0.215 p-value: Treatment 2 = Treatment 3 0.993 0.508 0.560 0.249 37 p-value: Treatment 2 = Treatment 4 0.374 0.112 0.737 0.855 p-value: Treatment 3 = Treatment 4 0.288 0.335 0.230 0.191 p-value: T1 + interaction = T2 + interaction 0.196 0.303 0.185 0.442 p-value: T1 + interaction = T3 + interaction 0.658 0.806 0.557 0.368 p-value: T1 + interaction = T4 + interaction 0.604 0.400 0.969 0.653 p-value: T2 + interaction = T3 + interaction 0.339 0.406 0.361 0.974 p-value: T2 + interaction = T4 + interaction 0.076 0.076 0.143 0.746 p-value: T3 + interaction = T4 + interaction 0.292 0.248 0.464 0.731 p-value test of equality: T1 = T2 = T3 = T 4 0.270 0.438 0.143 0.011 p-value test of equality T1 + interaction = T2 + interaction = T3 + 0.488 0.492 0.631 0.882 interaction = T4 + interaction Notes: Specifications use the main follow-up survey and include strata dummies, a variable representing the initial outcome at baseline, and a variable indicating missing data at baseline. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. aZ-score includes different measures of sales and profits including weekly and monthly outcomes. Z-score for sales, for profits, and for investment are described in Table 4. Woman entrepreneur has high discount rate when the discount score is above the median. The discount score is computed as the sum of five indicators of time preferences that take the value of 1 when the amount today is preferred, and of 0 when the woman prefers to wait a month an received a higher amount. The questions are: “Would you prefer receiving GHc 100 today or receiving (i) GHc 105 in a month from now? (ii) GHc 120 in a month from now? (iii) GHc 135 in a month from now? (iv) GHc 150 in a month from now? and (v) GHc median of the minimum to wait in a month from now? The level in this last question is the median of the minimum amount women request to receive in one month to wait. 38 Table A3: Heterogeneity in impacts on female entrepreneurs’ firm performance by above/below median sales at baseline Women's outcomes Z-score sales Z-score Z-score Z-score and profitsa Sales profits investment T1: Cash grant (W) -0.009 -0.002 -0.005 0.048 (0.064) (0.065) (0.071) (0.044) T2: Cash grant (M) -0.057 -0.093 -0.016 0.042 (0.071) (0.070) (0.076) (0.049) T3: HH Training + Cash grant -0.035 -0.054 -0.016 0.053 (0.055) (0.056) (0.058) (0.044) T4: Savings + Cash grant -0.033 -0.046 -0.010 0.004 (0.052) (0.053) (0.055) (0.044) T1 * Sales above median at BL -0.079 -0.029 -0.128 -0.070 (0.096) (0.100) (0.108) (0.077) T2 * Sales above median at BL 0.053 0.081 0.024 0.042 (0.098) (0.100) (0.110) (0.075) T3 * Sales above median at BL 0.071 0.115 0.028 0.112* (0.084) (0.087) (0.091) (0.064) T4 * Sales above median at BL 0.202** 0.238*** 0.159* 0.144** (0.085) (0.091) (0.089) (0.066) Sales above median at BL 0.177** 0.155** 0.173** 0.076* (0.069) (0.073) (0.071) (0.042) Combined effect: T1 + Sales above median at BL -0.087 -0.031 -0.133 -0.022 (0.079) (0.082) (0.089) (0.065) Combined effect: T2 + Sales above median at BL -0.004 -0.011 0.007 0.084 (0.076) (0.072) (0.090) (0.064) Combined effect: T3 + Sales above median at BL 0.036 0.062 0.012 0.165*** (0.070) (0.071) (0.077) (0.048) Combined effect: T4 + Sales above median at BL 0.169** 0.192*** 0.149** 0.147*** (0.067) (0.070) (0.074) (0.048) Control group mean 0.001 0.000 0.000 0.003 Sample size 2,872 2,858 2,868 2,856 p-value: Treatment 1 = Treatment 2 0.529 0.229 0.888 0.907 p-value: Treatment 1 = Treatment 3 0.673 0.414 0.867 0.915 p-value: Treatment 1 = Treatment 4 0.691 0.471 0.938 0.350 39 p-value: Treatment 2 = Treatment 3 0.752 0.558 1.000 0.830 p-value: Treatment 2 = Treatment 4 0.720 0.476 0.928 0.471 p-value: Treatment 3 = Treatment 4 0.960 0.879 0.904 0.301 p-value: T1 + interaction = T2 + interaction 0.321 0.811 0.154 0.189 p-value: T1 + interaction = T3 + interaction 0.124 0.268 0.094 0.006 p-value: T1 + interaction = T4 + interaction 0.001 0.008 0.001 0.013 p-value: T2 + interaction = T3 + interaction 0.603 0.327 0.958 0.226 p-value: T2 + interaction = T4 + interaction 0.020 0.007 0.100 0.343 p-value: T3 + interaction = T4 + interaction 0.048 0.072 0.051 0.721 p-value test of equality: T1 = T2 = T3 = T4 0.921 0.648 0.999 0.643 p-value test of equality T1 + interaction = T2 + interaction = 0.012 0.024 0.019 0.001 T3 + interaction = T4 + interaction Notes: Specifications use the main follow-up survey and include strata dummies, a variable representing the initial outcome at baseline, and a variable indicating missing data at baseline. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. aZ-score includes different measures of sales and profits including weekly and monthly outcomes. Z-score for sales, for profits, and for investment are described in Table 4. 40 Table A4: Heterogeneity in impacts on female entrepreneurs’ firm performance by yes/no to saves in bank account at baseline Women's outcomes Z-score sales Z-score Z-score Z-score and profitsa Sales profits investment T1: Cash grant (W) -0.010 0.014 -0.022 0.040 (0.061) (0.062) (0.067) (0.049) T2: Cash grant (M) 0.006 -0.018 0.035 0.049 (0.061) (0.059) (0.066) (0.050) T3: HH Training + Cash grant 0.059 0.052 0.061 0.101*** (0.056) (0.055) (0.060) (0.038) T4: Savings + Cash grant 0.063 0.067 0.056 0.065* (0.052) (0.052) (0.058) (0.038) T1 * entrepreneur has savings in bank account at BL -0.089 -0.069 -0.119 -0.079 (0.108) (0.113) (0.118) (0.089) T2 * entrepreneur has savings in bank account at BL -0.081 -0.069 -0.091 0.074 (0.092) (0.090) (0.102) (0.085) T3 * entrepreneur has savings in bank account at BL -0.167* -0.131 -0.181* 0.032 (0.096) (0.100) (0.101) (0.083) T4 * entrepreneur has savings in bank account at BL 0.032 0.026 0.068 0.032 (0.094) (0.101) (0.098) (0.071) Woman entrepreneur has savings in bank account at BL 0.180*** 0.138** 0.206*** 0.103* (0.067) (0.070) (0.071) (0.053) Combined effect: T1 + bank savings at BL -0.099 -0.055 -0.141 -0.039 (0.094) (0.098) (0.103) (0.072) Combined effect: T2 + bank savings at BL -0.075 -0.086 -0.056 0.123* (0.085) (0.078) (0.099) (0.070) Combined effect: T3 + bank savings at BL -0.108 -0.080 -0.120 0.134* (0.080) (0.086) (0.084) (0.070) Combined effect: T4 + bank savings at BL 0.095 0.093 0.124 0.097 (0.076) (0.084) (0.081) (0.061) Control group mean 0.001 0.000 0.000 0.003 Sample size 2,872 2,858 2,868 2,856 p-value: Treatment 1 = Treatment 2 0.806 0.623 0.436 0.877 p-value: Treatment 1 = Treatment 3 0.260 0.547 0.222 0.217 p-value: Treatment 1 = Treatment 4 0.210 0.380 0.238 0.611 41 p-value: Treatment 2 = Treatment 3 0.375 0.235 0.686 0.305 p-value: Treatment 2 = Treatment 4 0.323 0.138 0.739 0.752 p-value: Treatment 3 = Treatment 4 0.938 0.769 0.929 0.334 p-value: T1 + interaction = T2 + interaction 0.803 0.733 0.449 0.047 p-value: T1 + interaction = T3 + interaction 0.929 0.805 0.837 0.035 p-value: T1 + interaction = T4 + interaction 0.038 0.127 0.008 0.067 p-value: T2 + interaction = T3 + interaction 0.703 0.935 0.514 0.895 p-value: T2 + interaction = T4 + interaction 0.044 0.025 0.059 0.708 p-value: T3 + interaction = T4 + interaction 0.012 0.050 0.003 0.605 p-value test of equality: T1 = T2 = T3 = T4 0.556 0.510 0.663 0.127 p-value test of equality T1 + interaction = T2 + interaction = T3 0.074 0.192 0.018 0.077 + interaction = T4 + interaction Notes: Specifications use the main follow-up survey and include strata dummies, a variable representing the initial outcome at baseline, and a variable indicating missing data at baseline. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. aZ-score includes different measures of sales and profits including weekly and monthly outcomes. Z-score for sales, for profits, and for investment are described in Table 4. 42 Table A5: Household level impacts for female entrepreneurs Z-score of Z-score of woman's Woman's Hours worked household contribution to level of in the week earningsa household happiness expensesb Treatment 1: Cash grant (W) -0.052 -0.015 0.033 2.805 (0.054) (0.044) (0.022) (2.401) Treatment 2: Cash grant (M) 0.016 -0.031 0.005 1.956 (0.055) (0.056) (0.025) (2.417) Treatment 3: HH Training + Cash Grant 0.036 -0.117*** 0.051*** 7.133*** (0.048) (0.042) (0.019) (1.983) Treatment 4: Savings + Cash Grant 0.077* -0.101** 0.025 1.618 (0.046) (0.042) (0.020) (1.783) Control group mean -0.011 -0.001 0.840 64.850 Sample size 2,830 2,887 2,855 2,842 p-value: Treatment 1 = Treatment 2 0.252 0.775 0.296 0.771 p-value: Treatment 1 = Treatment 3 0.111 0.018 0.386 0.098 p-value: Treatment 1 = Treatment 4 0.018 0.041 0.692 0.628 p-value: Treatment 2 = Treatment 3 0.714 0.118 0.054 0.048 p-value: Treatment 2 = Treatment 4 0.250 0.194 0.443 0.888 p-value: Treatment 3 = Treatment 4 0.377 0.688 0.150 0.005 p-value test of equality 0.180 0.020 0.067 0.009 Notes: Specifications use the main follow-up survey and include strata dummies, a variable representing the initial outcome at baseline, and a variable indicating missing data at baseline. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. aZ-score of household earnings includes different measures of earnings including 1- earnings of the woman entrepreneur in the past 7 days winsorized (top 1%) in USD, 2- inverse hyperbolic sine of the earnings of the woman entrepreneur in the past 7 days in LCU, 3- earnings of the woman entrepreneur in the past 30 days winsorized (top 1%) in USD, 4- inverse hyperbolic sine of the earnings of the woman entrepreneur in the past 30 days in LCU, 5- total household earnings in the past 30 days (adding all family members' earnings reported by her) winsorized (top 1%) in USD, 6- inverse hyperbolic sine of the total household earnings in the past 30 days (adding all family members' earnings reported by her) in LCU. bZ-score of woman's contribution to household expenses includes 6 variables. Those are: 1- the total value of the woman entrepreneur’s contribution to the household’s weekly expenditures winsorized (top 1%) in USD, 2- inverse hyperbolic sine of the total value of the woman entrepreneur’s contribution to the household’s weekly expenditures in LCU, 3- woman entrepreneur’s share on the household’s weekly expenditure, 4- the total value of the woman entrepreneur’s contribution to the household’s quarterly expenditures winsorized (top 1%) in USD, 5- inverse hyperbolic sine of the total value of the woman entrepreneur’s contribution to the household’s quarterly expenditures in LCU, 6- woman entrepreneur’s share on the household’s quarterly expenditure . 43 Table A6: Impact on male business performance and total income Monthly Monthly Monthly Z-score Z-score Profits Total Income sales (US$) Salesa profitsb (US$) (US$) winsorized winsorized winsorized Treatment 1: Cash grant (W) 0.050 0.025 24.029 -1.001 10.071 (0.064) (0.069) (38.792) (9.964) (11.027) Treatment 2: Cash grant (M) 0.074 0.033 -11.691 -7.095 1.066 (0.055) (0.061) (32.700) (9.138) (11.230) Treatment 3: HH Training + Cash Grant 0.065 0.065 36.195 11.172 27.343*** (0.054) (0.056) (32.768) (7.750) (7.937) Treatment 4: Savings + Cash Grant 0.043 0.046 24.049 11.242 19.348** (0.051) (0.056) (28.481) (8.029) (8.465) Control group mean 0.002 0.005 266.7 106.3 147.6 Sample size 2521 2577 2511 2566 2567 p-value: Treatment 1 = Treatment 2 0.723 0.919 0.398 0.583 0.506 p-value: Treatment 1 = Treatment 3 0.818 0.572 0.776 0.224 0.117 p-value: Treatment 1 = Treatment 4 0.914 0.771 1.000 0.238 0.426 p-value: Treatment 2 = Treatment 3 0.881 0.588 0.179 0.039 0.018 p-value: Treatment 2 = Treatment 4 0.572 0.830 0.271 0.056 0.120 p-value: Treatment 3 = Treatment 4 0.670 0.726 0.695 0.993 0.328 p-value test of equality 0.690 0.839 0.636 0.179 0.007 Notes: Specifications use the main follow-up survey and include strata dummies, a variable representing the initial outcome at baseline, and a variable indicating missing data at baseline. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. a Z-score of sales includes 4 variables. Those are: 1- revenues generated by the husband’s business in the past 7 days winsorized (top 1%) in USD, 2- inverse hyperbolic sine of the revenues generated by the husband’s business in the past 7 days in LCU, 3- revenues generated by the husband’s business in the past month winsorized (top 1%) in USD, 4- inverse hyperbolic sine of the revenues generated by the husband’s business in the past month in LCU. b Z-score of profits includes 4 variables. Those are: 1- profits generated by the husband’s business in the past 7 days winsorized (top and bottom 1%) in USD, 2- inverse hyperbolic sine of the profits generated by the husband’s business in the past 7 days in LCU, 3- profits generated by the husband’s business in the past month winsorized (top and bottom 1%) in USD, 4- inverse hyperbolic sine of the profits generated by the husband’s business in the past month in LCU. Husband's monthly total income (having or not business) generated in the past month winsorized (top 1%) in USD. 44 Table A7: Heterogeneity in treatment effects on men’s outcomes by baseline level of women’s vulnerability to expropriation Men's outcomes Monthly Monthly Monthly Total Z-score Z-score sales Profits Income Salesa profitsb (US$) (US$) (US$) winsorized winsorized winsorized T1: Cash grant (W) 0.057 0.010 -0.278 -5.160 2.934 (0.091) (0.086) (61.122) (13.349) (14.592) T2: Cash grant (M) 0.002 -0.037 -31.251 -13.637 -0.383 (0.073) (0.083) (51.684) (13.182) (15.640) T3: HH Training + Cash grant 0.013 0.018 -9.019 8.740 23.629** (0.069) (0.074) (47.058) (11.223) (10.658) T4: Savings + Cash grant -0.011 -0.023 -1.842 7.520 13.524 (0.069) (0.074) (45.075) (11.390) (10.389) T1 * woman entrepreneur has high sharing score at BL 0.009 0.061 53.164 13.017 22.889 (0.110) (0.115) (85.989) (18.970) (23.241) T2 * woman entrepreneur has high sharing score at BL 0.148 0.145 28.222 14.605 6.633 (0.101) (0.102) (67.550) (15.540) (17.952) T3 * woman entrepreneur has high sharing score at BL 0.133 0.135 101.404 8.176 11.943 (0.096) (0.097) (68.189) (14.889) (16.516) T4 * woman entrepreneur has high sharing score at BL 0.107 0.148 48.633 8.115 15.463 (0.082) (0.093) (59.560) (15.486) (16.808) Woman entrepreneur has high sharing score at BL -0.127** -0.162** -71.112 -19.142* -26.785** (0.060) (0.063) (45.235) (10.305) (10.816) Combined effect: T1 + high sharing score at BLc 0.0657 0.0709 52.89 7.857 25.82 (0.076) (0.093) (56.040) (14.070) (17.220) Combined effect: T2 + high sharing score at BL 0.149* 0.109 -3.028 0.969 6.25 (0.082) (0.079) (41.820) (10.750) (12.210) Combined effect: T3 + high sharing score at BL 0.146* 0.153** 92.38* 16.92 35.57*** (0.077) (0.076) (49.140) (10.480) (12.380) Combined effect: T4 + high sharing score at BL 0.0961 0.125* 46.79 15.63 28.99** 45 (0.064) (0.073) (37.910) (11.220) (13.220) Control group mean 0.003 0.001 270.2 105.7 145.7 Sample size 2,435 2,490 2,425 2,479 2,480 p-value: Treatment 1 = Treatment 2 0.548 0.618 0.614 0.567 0.865 p-value: Treatment 1 = Treatment 3 0.628 0.930 0.884 0.289 0.189 p-value: Treatment 1 = Treatment 4 0.443 0.704 0.978 0.343 0.500 p-value: Treatment 2 = Treatment 3 0.876 0.511 0.645 0.081 0.150 p-value: Treatment 2 = Treatment 4 0.861 0.869 0.511 0.105 0.397 p-value: Treatment 3 = Treatment 4 0.725 0.588 0.855 0.911 0.392 p-value: T1 + interaction = T2 + interaction 0.356 0.702 0.361 0.637 0.245 p-value: T1 + interaction = T3 + interaction 0.353 0.393 0.556 0.539 0.568 p-value: T1 + interaction = T4 + interaction 0.678 0.560 0.916 0.608 0.857 p-value: T2 + interaction = T3 + interaction 0.968 0.578 0.074 0.139 0.013 p-value: T2 + interaction = T4 + interaction 0.499 0.832 0.261 0.224 0.074 p-value: T3 + interaction = T4 + interaction 0.495 0.710 0.355 0.910 0.593 p-value test of equality: T1 = T2 = T3 = T4 0.958 0.962 0.970 0.404 0.215 p-value test of equality T1 + interaction = T2 + interaction = 0.264 0.302 0.284 0.388 0.021 T3 + interaction = T4 + interaction Notes: Specifications use the main follow-up survey and include strata dummies, a variable representing the initial outcome at baseline, and a variable indicating missing data at baseline. The Z-score index is constructed following Kling, Liebman, and Katz (2007). Clustered standard errors by enumeration areas in parentheses. *, ** and *** denote significant at the 10%, 5%, and 1% levels, respectively. a Z-score of sales includes 4 variables. Those are: 1- revenues generated by the husband’s business in the past 7 days winsorized (top 1%) in USD, 2- inverse hyperbolic sine of the revenues generated by the husband’s business in the past 7 days in LCU, 3- revenues generated by the husband’s business in the past month winsorized (top 1%) in USD, 4- inverse hyperbolic sine of the revenues generated by the husband’s business in the past month in LCU. b Z-score of profits includes 4 variables. Those are: 1- profits generated by the husband’s business in the past 7 days winsorized (top and bottom 1%) in USD, 2- inverse hyperbolic sine of the profits generated by the husband’s business in the past 7 days in LCU, 3- profits generated by the husband’s business in the past month winsorized (top and bottom 1%) in USD, 4- inverse hyperbolic sine of the profits generated by the husband’s business in the past month in LCU. Husband's monthly total income (having or not business) generated in the past month winsorized (top 1%) in USD. c Combined effects are the coefficient related to the sum of the treatment and the corresponding interaction variable of the treatment with the high intra-household pressure variable for women. 46