Policy Research Working Paper 10503 Psychology, Skills, or Cash? Evidence on Complementary Investments for Anti-Poverty Programs Megan Lang Edward Soule Catherine H. Tinsley Development Economics Development Research Group June 2023 Policy Research Working Paper 10503 Abstract Growing evidence on the links between poverty and psy- focuses on promoting self-confidence, sense of value and chology has prompted increased interest in the psychosocial self-worth, and perceived social status. The second targets impacts of economic interventions and the economic specific skills: goal setting, public speaking, and networking. impacts of psychologically motivated interventions. In prac- Both program-based investments cost around USD \$35 tice, psychologically motivated programs typically comprise per participant, motivating a benchmark, cost-equivalent one of many components in multifaceted poverty alleviation unconditional cash transfer. The findings show that the psy- programs. This paper asks, what are the benefits of allocat- chologically-targeted intervention significantly improves ing complementary, marginal investments in anti-poverty psychosocial outcomes but shows no economic gains relative programs towards skills development or psychologically-tar- to cash, while the skills-based program improves economic geted interventions versus direct economic assistance? The outcomes with few effects on psychosocial outcomes. The paper benchmarks two program-based investments against results illustrate that low-cost psychologically-targeted and an unconditional cash transfer by randomly assigning skills-based interventions can be effective marginal invest- participants in an existing anti-poverty program to one ments relative to a small cash transfer, but their benefits of three groups. The first is psychologically-targeted. It may accrue in different domains. This paper is a product of the Development Research Group, Development Economics. 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 mlang@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 Psychology, Skills, or Cash? Evidence on Complementary Investments for Anti-Poverty Programs ∗ † Megan Lang, Edward Soule, and Catherine H. Tinsley‡ KEYWORDS: poverty and psychology, cash transfers, psychosocial wellbeing. JEL Codes: O12, O15, D91. ∗ World Bank Development Research Group. mlang@worldbank.org. † The McDonough School of Business, Georgetown University ‡ The McDonough School of Business, Georgetown University. This project was supported by grants from The McDonough School of Business at Georgetown University. The opinions and conclusions expressed herein are solely those of the authors and should not be construed as representing the opinions or policies of the sponsoring agencies. Thanks to The Raffini Term Profes- sorship at Georgetown University. The authors gratefully thank anonymous reviewers, participants at the 2021 Psychology and Economics of Poverty Convening, the 2022 Special Meeting of the Economic Science Association, the 2022 Pacific Development Conference, William Jack, Kelsey Jack, Claire Duquennois, Karl Dunkle-Werner, Gabe Englander, Carolyn Fischer, Kate Pennington, Eeshani Kandpal, and Jeremy Ma- gruder for their suggestions. Thanks to Norette Turimuci, Claire Uwineza, and the rest of the team at Resonate for their collaboration and assistance with implementation. This study is pre-registered with the AEA registry (AEARCTR-0004846). It has IRB approval from U.C. Berkeley CPHS 2018-01-10656 and government approval from Nyaruguru District, Rwanda. 1 Introduction A growing understanding of the links between poverty and psychology has prompted in- creased interest in the psychosocial impacts of economic interventions and the economic impacts of psychologically motivated interventions.1 In practice, psychologically motivated programs are rarely implemented alone. They typically comprise one of many components in multifaceted poverty alleviation programs.2 Given this reality, we directly compare three cost-equivalent investments made on the margins of an existing anti-poverty program: a psychologically-targeted intervention, a skills-based, economically-targeted intervention, and a cost-equivalent unconditional cash transfer. Which of these produces the largest impacts on economic and psychosocial outcomes? We randomly assign women in rural Rwanda to one of three interventions.3 The one tar- geting psychology is a two-day workshop called “Storytelling for Leadership” (SFL). In SFL, a workshop facilitator begins by defining leadership as “being proactive in the face of a chal- lenge.” Over the course of two days, the facilitator works with participants to identify their values, think of times when their actions demonstrated these values, and shape those expe- riences into compelling stories. The workshop encourages participants to view themselves as agents for positive change. The other workshop is called “Professional Development” (PD), which targets skills in interpersonal communication and goal setting. Participants learn how to set SMART goals (Doran (1981)), how to network with others to seek out opportunities, and principles of effective communication. Both workshops cost approximately USD 35 per participant: the value of the unconditional cash transfer that we distribute to women in the third group and around 27% of total monthly consumption expenditures for the median 1 See Ridley et al. (2020) and Haushofer and Fehr (2014) for an overview of the psychosocial impacts of economic interventions. McKelway (2021), Ghosal et al. (2020), Lybbert, Rojas Valdez, and Wydick (2022), Baranov et al. (2020), Blattman, Jamison, and Sheridan (2017), among others, speak to the economic impacts of psychologically motivated interventions. 2 See for example Banerjee et al. (2015), Blattman et al. (2016), Bandiera et al. (2017). 3 Rwanda is well-known for having the highest proportion of women in parliament in the world; however, debate continues about whether this has led to meaningful improvements for average women in Rwanda (e.g., Abbott and Malunda (2016), Kagaba (2015), Burnet (2011)). 1 rural household in Rwanda (National Institute of Statistics of Rwanda (December 2017)). A social enterprise called Resonate developed all workshop content. Resonate has been delivering these workshops to women and girls throughout East Africa since 2013. The work- shops are designed to be implemented on the margins of programs already run by Resonate’s partner organizations, which in our study is CARE International Rwanda. The women in our study have benefited from financial literacy training, workshops on preventing and respond- ing to gender-based violence, and the formation of cooperatives and savings groups. At the time of our intervention, the women in our sample are only participating in cooperatives and savings groups and not receiving any additional support from CARE. Studying the marginal effects of cash versus psychologically-targeted and skills-based interventions speaks to real world concerns about whether organizations should allocate marginal investments towards additional programming, which is inherently paternalistic, or simply expand direct economic assistance to their beneficiaries.4 We draw from theories in psychology and economics to discuss plausible causal chains between each of our treatments and psychosocial and economic outcomes. While established literatures link cash transfers and the skills taught in PD to improved outcomes, the links between a psychologically-based intervention like SFL and improved outcomes are less clear. We start from the observation that participants in our sample have low self-images and low senses of personal agency at baseline, and propose misattribution of successes and failures as a plausible explanation. We identify specific activities in SFL that reduce misattribution, then go on to posit that the workshop initiates a cycle of new behaviors with positive reinforcement that lead to persistent changes in attitudes, beliefs, and behavior. Thirteen months after participating in SFL, participants have significantly higher psy- chosocial wellbeing compared to the cash control group on six of the pre-specified outcomes 4 For instance, Mali’s social safety net program includes a quarterly cash transfer of around USD 48 for 2 years alongside programming on a range of social issues (Hidrobo et al. (2020)). Similarly, Niger has implemented a regular monthly transfer equivalent to around 15% of the poverty line (Premand and Stoeffler (2020)). Policymakers implementing such programs face trade-offs between paying to develop and implement additional programs versus increasing the size of the cash transfers. 2 we test: self-value, problem-solving skills, connectedness to communities, subjective social status (SSS), current evaluations of prior SSS, and identifying as a leader. While SFL leads to significantly higher treatment effects on specific psychosocial outcomes than PD, in gen- eral we cannot reject that the two interventions have the same psychosocial effects. By contrast, Resonate’s PD workshop leads to improvements in income, participation in paid work, and achieving a stated goal relative to cash and SFL, although our overall index of economic outcomes shows no significant improvement over cash. There continue to be open questions about which types of non-economic interventions best complement traditional poverty alleviation programs. Our design allows us to isolate the effects of the psychologically-targeted SFL workshop and the skills taught in PD relative to cash, providing direct evidence on the marginal benefits of such interventions. We view this as complementary to studies rigorously evaluating graduation from poverty programs that include a host of services such as asset transfers, consumption support, life coaching, and other services in a single, bundled intervention (Banerjee et al. (2015), Blattman et al. (2016), Bandiera et al. (2017)). Banerjee et al. (2022) find that asset transfers and saving services alone cannot generate the same benefits as a bundled program, hinting at the benefits of services like life coaching. We join Bossuroy et al. (2021) in identifying the relative impacts of adding a psychosocial intervention versus a cash transfer to an existing program and find similarly positive results to theirs, though the costs of our programs are significantly lower. Our work enhances our understanding of the psychology of poverty by considering the marginal benefits of low-cost, positive psychological interventions in low-income countries.5 As such, our work on SFL is most closely related to studies like Baranov, Haushofer, and Jang (2020), Ghosal et al. (2020), McKelway (2021), Bossuroy et al. (2021), and John and Orkin (forthcoming), which use relatively light-touch interventions.6 More broadly, SFL is 5 Note that Bolier et al. (2013) and Sin and Lyubomirsky (2009) perform meta-analyses on positive psychological interventions but most take place in high-income countries and are implemented as stand- alone interventions. Our interventions are less targeted than those in Baranov et al. (2020), Haushofer, Mudida, and Shapiro (2020), Rahman et al. (2008), Bolton et al. (2003), and Angelucci and Bennett (2020) and less intensive than those in Heller et al. (2017) and Blattman, Jamison, and Sheridan (2017). 6 These are distinct from aspirations interventions (e.g., Bernard et al. (2014), Lybbert, Rojas Valdez, 3 similar to self-affirmation interventions that have been widely studied in psychology, albeit primarily in high-income countries (Cohen and Sherman (2014)). PD bears similarities to planning and “implementation intention” interventions, although the two-day workshop format is more intensive than many other planning interventions in the literature (e.g., Stadler, Oettingen, and Gollwitzer (2009), Milkman et al. (2011), Ekers et al. (2014), John and Orkin (forthcoming)). By comparing three cost-equivalent interventions on the same set of outcomes, we can clearly establish the relative marginal benefits of psychologically- targeted versus economic interventions in different domains. Finally, we contribute to the literature on the impacts of unconditional cash transfers versus program-based interventions on economic and psychosocial outcomes. Few studies di- rectly benchmark against a cost-equivalent cash transfer. Notable exceptions are McIntosh and Zeitlin (2020) and McIntosh and Zeitlin (2021), but both speak to the overall effective- ness of comprehensive programs relative to cash while we focus on impacts on the margin of a large program. Similarly, Ridley et al. (2020) perform a meta-analysis of the impacts of cash transfers on various measures of depression, anxiety, and psychological wellbeing and find that, on average, cash transfers lead to a 0.067 standard deviation increase in psychological wellbeing.7 We find that SFL produces comparable psychosocial results at a fraction of the cost, although the associated economic impacts are not comparable.8 This suggests that tar- geted, low-cost interventions to enhance psychosocial wellbeing may generate higher returns in terms of improvements to psychosocial wellbeing per dollar spent than cash transfers. Our results speak to the potential for marginal investments in program-based interven- tions to improve outcomes. Many programs do not have the budget or the mandate for large cash transfers. Our work suggests that smaller, directed investments in programs can and Wydick (2022), Riley (2021)). 7 The studies Ridley et al. (2020) include in their analysis use varying measures of mental health including the Short General Health Questionnaire (GHQ-12), the Center for Epidemiologic Studies Depression Scale (CES-D), and the Acholi Psychosocial Assessment (APAI-R), among others. 8 Hussam et al. (2021) find similar results when they benchmark employment against a small cash transfer among refugees in Bangladesh: employment leads to substantially higher psychosocial impacts than cash. However, their population is highly socially isolated, raising questions about the generalizability of the results. 4 be effective at generating improvements on the margins of larger, well-established programs, particularly for psychosocial outcomes. 2 Methodology Starting in August 2019, we surveyed 456 women who were part of CARE International Rwanda cooperatives in Nyaruguru district. After the baseline survey, Resonate held a series of SFL and PD workshops throughout September and early October. We randomly assigned approximately one-third (n= 153) of surveyed participants to attend SFL and one- third (n=148) to attend PD. The remaining participants (n=155) received cash equal to the approximate per participant cost of delivering the workshops: 32,000 FRW (approximately USD 35).9 Table 1 shows that our groups are generally balanced on age, household size, and our pre-registered primary psychosocial and economic outcomes.10 The women in our sample range in age from 23–75 at baseline, with a median age of 40. Among the women in our sample, 88% report their primary occupation as working on a family farm. In addition, 13% report having received no formal schooling, 74% report their highest level of education as some primary school or completing primary school, and 13% report some secondary or vocational school. The median household size is six. While we do not directly elicit marital status, 90% of the women in our sample report living with at least one man in their household. Among the respondent, 40% report earning no individual income at baseline. Mean baseline income is FRW 15,300 and median income is FRW 3,000.11 Of women who report earning any income at baseline, mean income is FRW 25,600 and median income is FRW 10,000. From the perspective of the individual earnings of the women in our sample, the cash transfer represents more than a tripling of median monthly income. While it is significantly 9 Concerns about fairness and the long-term reputations of our implementing partners with the commu- nities in our study prevented us from including a pure control group that would receive nothing. 10 Table A1 shows that our three treatment groups are also balanced on our pre-registered family of secondary psychosocial outcomes. 11 Figure A7 shows the distribution of income at baseline. 5 smaller than the amounts distributed in studies like Haushofer and Shapiro (2016), it is around 27% of total monthly consumption expenditures for the median rural household and 40% of total monthly consumption expenditures for households below the first quartile (National Institute of Statistics of Rwanda (December 2017)). Table 1: Balance Cash Control PD SFL p-value Age 40.53 40.62 39.67 0.45 (7.88) (6.81) (6.84) HH Size 6.04 6.22 6.03 0.6 (1.89) (1.8) (1.95) Self-Advocacy 3.16 3.08 3.15 0.54 (0.74) (0.69) (0.71) Problem Solving 3.21 3.06 3.18 0.26 (0.86) (0.86) (0.81) No. Leadership Positions 0.96 0.93 1.03 0.33 (0.58) (0.56) (0.55) Connectedness to Home/Village 3.43 3.32 3.41 0.29 (0.69) (0.68) (0.67) Connectedness to Co-op 3.41 3.34 3.49 0.2 (0.67) (0.71) (0.79) Communication 3.27 3.29 3.36 0.48 (0.71) (0.7) (0.66) Any Leadership Position 0.85 0.83 0.89 0.33 (0.36) (0.38) (0.32) IHS(Income) 5.6 5.79 6.54 0.22 (5.06) (4.99) (4.95) Earn Any Income 0.57 0.59 0.65 0.22 (0.5) (0.49) (0.48) Marginal Utility of Expenditure 0.17 0.26 0 0.07 (1.01) (0.99) (0.99) N 155 148 152 Notes: Mean baseline covariates by treatment group. Standard deviations are in parentheses. Column 4 reports p-values associated with F-tests of joint equality between the three groups. Self-advocacy, problem solving, and communication are pre-registered index variables, described in detail in the appendix. IHS(Income) is the inverse hyperbolic sine of income. Earn Any Income is a binary variable equal to one if the woman reports earning any income. The Marginal Utility of Expenditure is a measure based on consumption expenditures proposed by Ligon (2019). Connectedness to home/village and connectedness to co-op are questions with visual scales, described in detail in the appendix. 6 We present all results relative to this cash benchmark because it is the policy relevant comparison. A fundamental question for organizations or policymakers implementing in- tensive anti-poverty programs is whether additional program-based content generates more impact than simply distributing cash to participants on the margin. For instance, in propor- tional terms our cash transfer would be the equivalent of extending social safety net programs like those in Hidrobo et al. (2020) and Premand and Stoeffler (2020) by a few months or marginally increasing the size of the repeated transfers in those programs. Determining whether investing in programs like SFL or PD generates higher returns than expanding existing direct economic assistance is central to evidence-based policy design. We arrived at our cash benchmark by considering the specific cost of the workshops con- ducted during the RCT and Resonate’s general costing framework. During the RCT, supplies and transport cost USD 14.70 per participant and facilitator logistics (transport, food, and accommodation) were USD 5 per participant. Resonate’s costing template estimates USD 16 per participant for staff time. As such, our USD 35 cash benchmark represents the direct costs of implementing the workshops. We conducted all randomization at the level of individual women, and workshop leaders carefully took attendance to ensure compliance. If women in the workshops shared the content with women assigned to the control group, our estimates are lower bounds on the true treatment effects. We encouraged women in the control group to keep the cash transfer private. Even if women in the workshops learned of the cash transfer, it would bias our estimated effects towards zero if it caused participants to view the workshops less favorably. The two primary concerns are negative spillovers from the women in the workshops to women in the cash transfer group (e.g., if women receiving the cash transfer become discouraged at not being selected for the workshop) and positive spillovers from women in the cash transfer group to women in the workshops (e.g., directly sharing funds or general equilibrium effects). We attempted to limit negative spillovers by communicating multiple times that assignment was completely random. Anecdotally, the women in the cash group were highly enthusiastic 7 about the transfer, leading us to believe that negative spillovers are unlikely. We view direct positive spillovers between the cash transfer group and the workshop groups unlikely given that women receiving cash are more likely to face redistributive pressure from family than from other members of their cooperative. Given the relatively small size of the cash transfer and the limited number of women we distribute it to, we do not anticipate general equilibrium effects of the kind documented in Egger et al. (2019), who employ a substantially larger transfer and distribute it to a greater proportion of households. The same team of facilitators from Resonate conducted each workshop. To ensure that workshops exactly matched those typically delivered by Resonate, facilitators invited ap- proximately fifty women at a time to participate, for a total of three SFL and three PD workshops. After all six workshops were complete, we invited the women in the cash control group to a central gathering point to distribute the cash transfer. We originally planned to survey participants after six months and one year, but COVID- 19 regulations prevented us from carrying out the six-month follow-up survey. We present results based on the baseline data and data collected in November 2020, 13 months after the workshops and receipt of cash. 2.1 Data We group our dependent variables into three families: primary and secondary psychosocial outcomes and economic outcomes. Primary psychosocial outcomes are indicators of behav- ioral shifts that we hypothesize will change as a result of the workshops, while secondary outcomes are those that may change as a result of the workshop, but which also may be lagging indicators that take longer to change. Our pre-specified primary psychosocial outcomes are self-advocacy, leadership positions (holding any and the number held), connectedness to cooperatives, connectedness to home and village, problem solving, and effective communication. We measure all of these except leadership positions using an Anderson (2008) index of two or more survey questions. Our 8 family of secondary psychosocial outcomes includes subjective social status (SSS) evaluated at present, five years in the past, and five years in the future using the MacArthur scale (Adler et al. (2000)), connectedness to the workshop group, self-evaluation of value, resilience, identifying as a leader, and five measures of aspirations. We discuss measurement of all outcomes in Appendix A. Our economic outcomes are limited to three measures. First, we ask participants how much money they earned, either from being paid or selling their own production, in the past month. We use this as a continuous measure of income and a binary measure for generating any positive amount of income. Given that women in our sample earn very little income, we trained enumerators to ask explicitly about any money earned from working for others or from selling anything that they had produced. However, this provides at best a rough estimate given that most of the women in our sample primarily work on family farms that only generate income intermittently. We also use the marginal utility of expenditure (MUE) to aggregate data on consump- tion expenditures on eighteen food items. We use the MUE because it does not require the universe of disaggregated consumption expenditures to compute (Ligon (2019)). The MUE identifies the common part of residuals across multiple regressions of consumption expendi- tures for specific goods on household characteristics.12 We find that our measure of MUE is significantly and negatively correlated with income, as theory would predict, but the two measures appear to capture different aspects of economic wellbeing.13 Given that we are limited to self-reported data, researcher demand effects are a natural concern. The direction for researcher demand effects is not clear. Answers perceived as being desirable to the researchers may be more obvious to women who participated in SFL than those in the other two groups, but given the intensive programming already undertaken by CARE for all of the women in our sample such effects may be small. Conversely, women in the cash control group may be more prone to choosing what they perceive to be desirable 12 For complete details on constructing MUE, see Appendix A. 13 See appendix Figure A3. 9 answers since they received a substantial windfall from study participation. We examine differences between baseline and follow-up responses to see if there is any obvious evidence of researcher demand effects. If participants in one group are simply choos- ing desirable answers more than those in other groups, we may see higher responses for one group across our self-reported measures in the follow-up survey relative to the baseline survey. We see two patterns that help ameliorate concerns about researcher demand. First, respondents choose lower answers at follow-up than they do at baseline for twelve of our twenty-one outcomes across all three groups.14 Second, the pattern of responses is not uni- form across different psychosocial outcomes. For instance, participants report higher SSS in the follow-up survey than in the baseline survey on average, but the opposite is true for self- value.15 While these patterns cannot fully rule out researcher demand effects, they provide some reassurance. 3 Hypothesized Mechanisms In this section, we discuss plausible causal chains that may link each treatment to im- provements in psychosocial and economic outcomes. We expect cash transfers to lead to improvements in economic outcomes through direct consumption support or small produc- tive investments (Kondylis and Loeser (2021)). The literature on the psychology of poverty suggests that cash transfers may also lead to psychosocial benefits if economic impacts re- duce the stress associated with poverty (Ridley et al. (2020)). The skills taught in PD are also relatively well-studied (e.g., Latham (2001) and Locke and Latham (1990) on effective goal setting and Wanberg et al. (2020) on networking). While we study these skills in a new 14 The difference between follow-up and baseline responses is negative for self-advocacy, problem solving, number of leadership positions held, communication skills, connectedness to the co-op, holding any leadership position, self-value, having a goal, connectedness to the workshop group, income, and earning any income. Responses are higher for time to be like the role model, which also indicates “negative” movement between baseline and follow-up. MUE is higher at follow-up relative to baseline for women in the cash control group and SFL. 15 We speculate different patterns in self-value and SSS may reflect the relative nature of SSS versus the absolute nature of self-value. 10 context, the theory of change is the same. Developing professional skills enables participants to pursue their economic goals more effectively, leading to gains in economic outcomes which may improve psychosocial outcomes through the same stress-reduction mechanism at play for the cash transfer. PD could impact psychosocial outcomes directly if it builds confidence and an enhanced sense of agency, but it does not explicitly target such outcomes to the same extent as SFL. The mechanism that could lead to impacts for SFL is less clear. SFL is designed to address internal constraints such as poor self-image or a generally low sense of agency. We find that participants’ initial beliefs about themselves are low at the baseline survey: 58% say that the statement “I matter” is true half of the time or less, and 64% say that the statement “there are good things about me” is true half of the time or less. The goal of our framework is to propose a mechanism that (1) generates systematically low attitudes and beliefs about the self and (2) would be effectively counteracted by participation in SFL. Scholars have studied one such mechanism: imposter phenomenon (Bravata et al. (2019)). In the context of rural Rwanda, the fear of being recognized as an impostor is likely less important than the cognitive patterns associated with impostor phenomenon: externalizing successes and “overgeneralization of a failure experience to. . . overall self-concept” (Sakulku and Alexander (2011)). These cognitive patterns stem from misattribution: if participants misattribute challenges or failures to internal traits, then they may develop a low self-image or a poor sense of agency. Importantly, learning cannot correct such beliefs when the process of learning is flawed by misattribution.16 We draw on psychological theories of self-affirmation to show how SFL may persistently reduce misattribution in participants’ self-beliefs. Extensive empirical evidence suggests that exercises like writing about values can lead to large and persistent improvements in education and health outcomes, particularly for marginalized groups (Cohen and Sherman (2014)). Self-affirmation theory asserts that affirmed individuals gain the ability to view stressors in 16 Economists model similar problems by considering learning in stochastic environments (e.g., Hoel et al. (2021)). 11 a larger context, thereby reducing the impact of adverse events on psychological wellbeing. Such shifts in perceptions can trigger self-perpetuating cycles, leading to persistent changes. We posit that SFL draws on similar principles of self-affirmation to reduce misattribution and initiate a self-reinforcing cycle. SFL starts by defining leadership as being “proactive in the face of a challenge,” providing a metric of success that is difficult to misattribute. Next, participants identify their values and think of times when their actions reflected those values, which closely matches exercises in the self-affirmation literature (e.g., Crocker, Niiya, and Mischkowski (2008), Shnabel et al. (2013)). Such exercises offer “unconditional sources of integrity, often from social relationships” (Cohen and Sherman (2014)). Reflecting on values may further reduce misattribution by focusing on internal measures of success. For instance, one participant (not in our study sample) left her husband but then struggled to feed her children. She says, “Before the workshop, I didn’t even realize that I had anything to be proud of. As the session progressed, I became aware of my accomplishments and I came to value them. It had taken courage to leave my husband and that was a huge achievement.”17 This demonstrates focus on an internal metric and suggests that SFL can lead participants to re-evaluate past events, directly addressing past misattribution. After the initial set of activities, facilitators teach the structure of a compelling personal story and participants start to develop their own. The facilitators create a supportive envi- ronment: small groups provide positive feedback and facilitators help participants improve their delivery, fostering greater confidence. This process establishes a novel and agentic be- havior (speaking up and sharing personal stories) and positively reinforces it. As participants gain confidence, they share their stories with the entire workshop group and learn how to use stories to promote collective action and problem-solving. We posit that recalling and re-framing a past action as showing integrity and leadership, telling a personal narrative in a public setting, and receiving positive feedback establishes a cycle of positive self-image reinforcement. We hypothesize that SFL shifts perceptions 17 https://resonateworkshops.org/blog/2018/7/12/discovering-my-worth 12 through its focus on values (Sherman (2013)), then triggers a cycle in which participants speak up, become more confident from the positive comments they receive, and grow more confident and more likely to speak up in the future (Cohen et al. (2009)). Combined, these may lead to sustained improvements in self-image and a higher sense of agency. These changes may translate into improved economic outcomes if they encourage participants to pursue economic opportunities. Considering the posited causal chains, SFL likely has the most direct impact on psy- chosocial outcomes, although PD may also have some direct impacts. Conversely, the cash transfer is most closely tied to economic outcomes, followed by PD. In the next section, we show that our results largely align with these hypothesized chains. 4 Results For each of our pre-registered outcomes except achieving the goal set at baseline, our pre- ferred specification is Outcomei,post = α + β1 SF Li + β2 P Di + δOutcomei,pre + ϵi , (1) where SF Li is an indicator equal to one if participant i was assigned to SFL, P Di is an indicator equal to one if participant i was assigned to PD, and Outcomei,pre controls for participant i’s baseline level of the relevant outcome (McKenzie (2012)).18 We report naive p-values and q-values that control the false discovery rate within each family of outcomes using the adaptive step-up procedure from Benjamini, Krieger, and Yekutieli (2006). Though not pre-registered, we also present an Anderson (2008) index of all outcomes within a family. Figure 1 shows treatment effects on our pre-registered family of primary psychosocial 18 Our primary specification deviates from our pre-analysis plan because we were only able to collect two rounds of data instead of the originally planned three. We present robustness checks in Appendix B where we omit baseline controls. Ex post power calculations show that we are powered to detect effects of 0.27 standard deviations for our index of primary psychosocial outcomes, 0.3 for our index of secondary psychosocial outcomes, and 0.32 for economic outcomes. 13 Figure 1: Treatment Effects on Pre-Registered Family of Primary Psychosocial Outcomes Note: Light blue dots denote treatment effects for SFL relative to the unconditional cash transfer. Dark blue dots denote treatment effects for PD relative to the unconditional cash transfer. Bold lines show naive 95% confidence intervals, while the narrower lines show confidence intervals that control the false discovery rate within this family of outcomes using the adaptive step-up procedure from Benjamini, Krieger, and Yekutieli (2006). PPS Index is a Anderson (2008) index of all variables within the family, which we did not pre-specify. We normalize all results by the standard deviation of the control group at the follow-up survey. outcomes, with all treatment effects normalized by the standard deviation of the control group at the follow-up survey. SFL causes significant and positive effects relative to the cash control group on our measures of problem-solving ability and connectedness to home/village. Magnitudes are modest: most are between 3%–8% of the control group mean, although the number of leadership positions held and connectedness to others in the cooperative are even smaller. While many of the variables in this family of outcomes exhibit positive effects, our Anderson (2008) index of all variables shows an insignificant increase of 0.16 standard deviations. We find no statistically significant differences between the effects for SFL and PD on any of our primary psychosocial outcomes.19 Figure 2 shows large and significant effects from SFL on subjective social status (SSS) evaluated currently and five years in the past, both of which survive our multiple inference correction at the 5% level of significance. SSS increases by 19% and past SSS by 66% of the control group mean, with both increases being significantly higher than the effects from 19 See Table A2 for ATEs in levels, control group means, and p-values from tests of equality between SFL and PD. 14 Figure 2: Treatment Effects on Pre-Registered Family of Secondary Psychosocial Outcomes Note: Light blue dots denote treatment effects for SFL relative to the unconditional cash transfer. Dark blue dots denote treatment effects for PD relative to the unconditional cash transfer. Bold lines show naive 95% confidence intervals, while the narrower lines show confidence intervals that control the false discovery rate within this family of outcomes using the adaptive step-up procedure from Benjamini, Krieger, and Yekutieli (2006). SPS Index is a Anderson (2008) index of all variables within the family, which we did not pre-specify. We normalize all results by the standard deviation of the control group at the follow-up survey. PD.20 This may directly reflect the activities that take place during SFL: reflecting on past events appears to cause participants to positively update their assessments of past SSS by re-evaluating past experiences against internally driven criteria. Our results on SSS are in line with our estimated impact from SFL on self-value scores, which is also positive and remains significant at the 5% level after controlling the false discov- ery rate, though more modest in magnitude at 8%. As expected, we find that SFL increases how closely participants identify as a leader by 20%, which also survives our multiple infer- ence correction at the 5% level of significance. Our results on individual outcomes combine to yield an average treatment effect of 0.38 standard deviations on our index of secondary psychosocial outcomes, which is significant at the 5% level but not significantly different between SFL and PD.21 20 See Table A3 for ATEs in levels, control group means, and p-values associated with tests of equality between SFL and PD. 21 As an exploratory analysis, we check for differences in role model characteristics: gender, age relative to the respondent, whether the role model is a family member of the respondent or resides in the same village, and whether the respondent personally knows their role model. We find few differences, although women in SFL are 10pp more likely to choose a female role model than women in the cash control group, a 20% 15 For PD, the only treatment effects within the family of secondary psychosocial outcomes that are positive and significant relative to cash are those on identifying as a leader and achieving the goal set at baseline. The effect on achieving the goal set at baseline does not survive our multiple inference correction at conventional levels of significance, but is significantly higher than the effect for SFL. Our results combine to yield a significant increase of 0.22 standard deviations on our index of secondary psychosocial outcomes.22 Our results point to meaningful impacts from SFL on psychosocial outcomes, but it is not clear whether or how such mindset shifts lead to concrete changes in behavior. While not pre-specified, we present evidence on whether participants have spoken up in a group situation where they did not know everyone present at any time during the two weeks before each survey. We find that women who participated in SFL are significantly more likely to report speaking up than those in the cash control group: 24% of women in SFL compared to only 13.9% in the cash control group.23 When we ask women where they speak up, the most common answers are helping with sensitization and enforcement of COVID-19 regulations in local markets, dispute resolution meetings, and savings group meetings. These examples support the idea that SFL works by introducing and reinforcing a new behavior: speaking up. They also lend support to our theory that SFL starts a self-reinforcing cycle that leads to the persistent impacts we observe on a range of psychosocial outcomes. Figure 3 shows that SFL has no significant effects on any economic outcome relative to the cash control group. PD positively impacts the level of income earned and the likelihood that participants earn any income. Our multiple inference correction yields slightly smaller q-values for the MUE, leading PD to exhibit positive effects on all three measures of economic outcomes that are significant at the 10% level after controlling the false discovery rate. The increase that is significant at the 5% level. 22 We present exploratory treatment effects on a limited number of psychosocial outcomes measured im- mediately post workshop relative to the cash control group at baseline in Figure A8 in Appendix C. PD has positive and significant effects on nearly all of the psychosocial outcomes that we measure immediately post workshop. We interpret the difference in the persistence of these effects for SFL as evidence that the psychosocial aspect of PD is more limited. 23 Figure A9 visualizes results on speaking up. 16 Figure 3: Treatment Effects on Pre-Registered Family of Economic Outcomes Note: Light blue dots denote treatment effects for SFL relative to the unconditional cash transfer. Dark blue dots denote treatment effects for PD relative to the unconditional cash transfer. Bold lines show naive 95% confidence intervals, while the narrower lines show confidence intervals that control the false discovery rate within this family of outcomes using the adaptive step-up procedure from Benjamini, Krieger, and Yekutieli (2006). Note that adjusted confidence intervals are smaller than naive confidence intervals for the marginal utility of expenditure. Econ Index is a Anderson (2008) index of all variables within the family, which we did not pre-specify. We normalize all results by the standard deviation of the control group at the follow-up survey. impacts of PD are large in proportional terms: an 87% reduction in the marginal utility of expenditure, a 78% increase in cash earnings, and a 34% increase in the likelihood of earning any cash relative to the control group.24 However, these impacts occur off a low base as only 35% of women in the control group report earning any income in the month prior to the follow-up survey. Estimating impacts only for women earning positive income yields null results, indicating that our results are driven by the extensive margin of women beginning to earn any income rather than women increasing earnings. Our test of equality between SFL and PD on our index of all economic outcomes shows that PD yields effects that are significantly higher than those for SFL.25 By extension, we find that PD has a positive impact on our index of economic outcomes of 0.18 standard deviations, but it is not statistically significant. Taken together, our results suggest that SFL has benefits that primarily manifest in 24 See Table A4 for economic impacts in levels, control means, and tests for equality between SFL and PD. 25 The MUE from Ligon (2019) is meant to be interpreted as a true marginal utility, so lower values correspond to higher levels of economic wellbeing. 17 psychosocial outcomes relative to the cash control group, while the PD workshop has little impact on psychosocial outcomes but does improve economic outcomes and participants’ ability to achieve goals. These patterns are broadly consistent with our hypothesized causal chains, although it is somewhat surprising that PD outperforms cash on specific economic outcomes. 5 Conclusion The results of our RCT illustrate that low-cost psychologically-targeted and skills-based interventions can both be effective marginal investments relative to a small cash transfer, but their benefits may accrue in different domains. SFL has meaningfully positive and significant impacts on a range of psychosocial outcomes along with measures of peer leadership relative to our cash transfer. Conversely, PD has large and significant impacts on economic outcomes and participants’ ability to achieve their stated goals relative to the cash transfer. It is important to place our results in the context of the economic and social disruption caused by COVID-19. Skills-based interventions like PD may be more effective at improving economic outcomes when there are more economic opportunities available, and SFL may lead to more meaningful changes in behavior when there are more opportunities for group interactions. Conversely, the disruptions caused by COVID-19 may have eroded any short- term gains from the cash transfer, which we cannot observe. Our work points to the need for additional research on the cost effectiveness of cash trans- fers relative to alternative investments. Kondylis and Loeser (2021) find that both increasing the size of cash transfers as well as expanding the scope to provide additional services leads to reduced cost effectiveness. Our results suggest that program-based investments may be more effective than cash on the margins of large programs, particularly if policymakers wish to improve psychosocial outcomes. However, there are important limitations to our study. For one, cash transfers may be substantially less costly to implement at scale than programs like SFL or PD. While comparing direct costs is useful in understanding efficacy, accounting for 18 differences in fixed versus variable costs is central to evaluating effectiveness at scale. Better understanding the contexts in which cash transfers are the most cost-effective investment is critically important for policymakers and practitioners engaged in program design. Our study suggests that there are unique roles for psychologically-targeted and skills- based interventions within the context of traditional anti-poverty programs. We show that SFL and PD both offer improvements over cash on a range of outcomes when undertaken on the margin of conventional programs. Consistent with the findings in Banerjee et al. (2015), Blattman et al. (2016), and Bandiera et al. (2017), our results suggest that program participants may face a range of constraints that can hamper their economic progress such as a lack of certain skills and a poor foundation of psychological well-being. Designing programs that tackle multiple constraints offers a promising avenue to enhance the impacts of economic interventions. 19 References Abbott, Pamela and Dixon Malunda. 2016. “The Promise and the Reality: Women’s Rights in Rwanda.” African Journal of International and Comparative Law 24 (4):561–581. Adler, N. E., E. S. Epel, G. Castellazzo, and J. R. Ickovics. 2000. “Relationship of subjective and objective social status with psychological and physiological functioning: Preliminary data in healthy, White women.” Health Psychology 19 (6):586–592. Anderson, Michael. 2008. “Multiple Inference and Gender Differences in the Effects of Early Intervention: A Reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects.” Journal of the American Statistical Association 103 (484):1481–1495. Angelucci, Manuela and Daniel Bennett. 2020. “The Economic Impact of Mental Health Care in India.” Bandiera, Oriana, Robin Burgess, Narayan Das, Selim Gulesci, Imran Rasul, and Munshi Sulaiman. 2017. “Labor Markets and Poverty in Village Economies.” The Quarterly Journal of Economics 132 (2):811–870. Banerjee, Abhijit, Esther Duflo, Nathanael Goldberg, Dean Karlan, Robert Osei, William e, Jeremy Shapiro, Bram Thuysbaert, and Christopher Udry. 2015. “A multifaceted Parient´ program causes lasting progress for the very poor: Evidence from six countries.” Science 348 (6236). Banerjee, Abhijit, Dean Karlan, Robert Osei, Hannah Trachtman, and Christopher Udry. 2022. “Unpacking a multi-faceted program to build sustainable income for the very poor.” Journal of Development Economics 155:10278. Baranov, Victoria, Sonia Bhalotra, Pietro Biroli, and Joanna Maselko. 2020. “Maternal De- pression, Women’s Empowerment, and Parental Investment: Evidence from a Randomized Controlled Trial.” American Economic Review 110 (3):824–859. 20 Baranov, Victoria, Johannes Haushofer, and Chaning Jang. 2020. “Can Positive Psychology Improve Psychological Well-Being and Economic Decision-Making? Experimental Evi- dence from Kenya.” Economic Development and Cultural Change 68 (4):1345–1376. Benjamini, Yoav, Abba M. Krieger, and Daniel Yekutieli. 2006. “Adaptive linear step-up procedures that control the false discovery rate.” Biometrika 93 (3):491–507. Bernard, Tanguy, Stefan Dercon, Kate Orkin, and Alemayehu Seyoum Taffesse. 2014. “The Future in Mind: Aspirations and Forward-Looking Behaviour in Rural Ethiopia.” CSAE Working Paper Series 2014-16. Blattman, Christopher, Eric P. Green, Julian Jamison, M. Christian Lehmann, and Jeannie Annan. 2016. “The Returns to Microenterprise Support among the Ultrapoor: A Field Experiment in Postwar Uganda.” American Economic Journal: Applied Economics 8 (2):35–64. Blattman, Christopher, Julian C. Jamison, and Margaret Sheridan. 2017. “Reducing Crime and Violence: Experimental Evidence from Cognitive Behavioral Therapy in Liberia.” American Economic Review 107. Bolier, Linda, Merel Haverman, Gerben J Westerhof, Heleen Riper, Filip Smit, and Ernst Bohlmeijer. 2013. “Positive psychology interventions: a meta-analysis of randomized con- trolled studies.” BMC Public Health 13 (119). Bolton, Paul, Judith Bass, Richard Neugebauer, Helen Verdeli, Kathleen F Clougherty, Priya Wickramaratne, Liesbeth Speelman, Lincoln Ndogoni, and Myrna Weissman. 2003. “Group interpersonal psychotherapy for depression in rural Uganda: a randomized con- trolled trial.” JAMA 289 (23):3117–24. e, Bossuroy, Thomas, Markus Goldstein, Dean Karlan, Harounan Kazianga, William Parient´ Patrick Premand, Catherine Thomas, Christopher Udry, Julia Vaillant, and Kelsey Wright. 21 2021. “Pathways out of Extreme Poverty: Tackling Psychosocial and Capital Constraints with a Multi-Faceted Social Protection Program in Niger.” Bravata, Dena M., Sharon A. Watts, Autumn L. Keefer, Divya K. Madhusudhan, Katie T. Taylor, Dani M. Clark, Ross S. Nelson, Kevin O. Cokley, and Heather K. Hagg. 2019. “Prevalence, Predictors, and Treatment of Impostor Syndrome: a Systematic Review.” Journal of General Internal Medicine 35:1252–1275. Burnet, Jennie E. 2011. “Women Have Found Respect: Gender Quotas, Symbolic Repre- sentation, and Female Empowerment in Rwanda.” Politics & Gender 7 (3):303–334. Cohen, Geoffrey L., Julio Garcia, Valerie Purdie-Vaughns, Nancy Apfel, and Patricia Brzus- toski. 2009. “Recursive processes in self-affirmation: intervening to close the minority achievement gap.” Science (New York, N.Y.) 324 (5925):400–403. Cohen, Geoffrey L. and David K. Sherman. 2014. “The Psychology of Change: Self- Affirmation and Social Psychological Intervention.” Annual Review of Psychology 65 (1):333–371. Crocker, Jennifer, Yu Niiya, and Dominik Mischkowski. 2008. “Why does writing about important values reduce defensiveness? Self-affirmation and the role of positive other- directed feelings.” Psychological Science 19 (7):740–747. Doran, G.T. 1981. “There’s a S.M.A.R.T. Way to Write Management Goals and Objectives.” Management Review . Egger, Dennis, Johannes Haushofer, Edward Miguel, Paul Niehaus, and Michael W Walker. 2019. “General Equilibrium Effects of Cash Transfers: Experimental Evidence from Kenya.” Working Paper 26600, National Bureau of Economic Research. Ekers, David, Lisa Webster, Annemieke Van Straten, Pim Cuijpers, David Richards, and 22 Simon Gilbody. 2014. “Behavioural Activation for Depression; an Update of Meta-Analysis of Effectiveness and Sub Group Analysis.” PloS One . Ghosal, Sayantan, Smarajit Jana, Anandi Mani, Sandip Mitra, and Sanchari Roy. 2020. “Sex workers, Stigma and Self-Image: Evidence from Kolkata Brothels.” Review of Economics and Statistics :1–45. Haushofer, Johannes and Ernst Fehr. 2014. “On the psychology of poverty.” Science :862– 867. Haushofer, Johannes, Robert Mudida, and Jeremy Shapiro. 2020. “The Comparative Impact of Cash Transfers and a Psychotherapy Program on Psychological and Economic Well- being.” Haushofer, Johannes and Jeremy Shapiro. 2016. “The Short-term Impact of Unconditional Cash Transfers to the Poor: Experimental Evidence from Kenya.” Quarterly Journal of Economics 131 (4):1973–2042. Heller, Sara B., Anuj K. Shah, Jonathan Guryan, Jens Ludwig, Sendhil Mullainathan, and Harold A. Pollack. 2017. “Thinking, Fast and Slow? Some Field Experiments to Reduce Crime and Dropout in Chicago.” The Quarterly Journal of Economics 132 (1):1–54. Hidrobo, Melissa, Lieven Huybregts, Noreen Karachiwalla, and Shalini Roy. 2020. “Multi- Component Cash Transfer Programs: Evidence from Mali’s Social Safety Net Program.” Project note, International Food Policy Research Institute. Hoel, Jessica B., Hope Michelson, Ben Norton, and Victor Manyong. 2021. “Misattribution and uncertainty about beliefs prevent learning.” Hussam, Reshmaan, Erin M. Kelley, Gregory Lane, and Fatima Zahra. 2021. “The psychoso- cial value of employment.” 23 John, Anett and Kate Orkin. forthcoming. “Can Simple Psychological Interventions Increase Preventive Health Investment?” Journal of the European Economic Association . Kagaba, Mediatrice. 2015. “Women’s experiences of gender equality laws in rural Rwanda: the case of Kamonyi District.” Journal of Eastern African Studies 9 (4):574–592. Kondylis, Florence and John Loeser. 2021. “Intervention Size and Persistence.” Latham, Gary P. 2001. “The Reciprocal Effects of Science on Practice: Insights from the Practice and Science of Goal Setting.” Canadian Psychology 42 (1):1–11. Ligon, Ethan. 2019. “Estimating Household Welfare from Disaggregate Expenditures.” Locke, Edwin A. and Gary P. Latham. 1990. A Theory of Goal Setting and Task Performance. Prentice Hall. Lybbert, Travis J., Ruben Irvin Rojas Valdez, and Bruce Wydick. 2022. “Can Hope Elevate Microfinance? Evidence from Oxaca, Mexico.” Oxford Economic Papers 74 (1):236–264. McIntosh, Craig and Andrew Zeitlin. 2020. “Using Household Grants to Benchmark the Cost Effectiveness of a USAID Workforce Readiness Program.” ———. 2021. “Cash versus Kind: Benchmarking a Child Nutrition Program against Un- conditional Cash Transfers in Rwanda.” McKelway, Madeline. 2021. “Women’s Agency and Women’s Employment: How Women’s Sense of Agency Affects Their Labor Supply.” McKenzie, David. 2012. “Beyond baseline and follow-up: The case for more T in experi- ments.” Journal of Development Economics 99 (2):210–221. Milkman, Katherine L., Beshears. John, James J. Choi, David Laibson, and Brigitte C. Madrian. 2011. “Using implementation intentions prompts to enhance influenza vaccina- tion rates.” Proceedings of the National Academy of Sciences 108 (26), pages =. 24 National Institute of Statistics of Rwanda. December 2017. “Integrated Household Living Conditions Survey(2016-2017), Version 0.1 of the public use dataset.” URL http:// microdata.statistics.gov.rw. Premand, Patrick and Quentin Stoeffler. 2020. “Do Cash Transfers Foster Resilience? Evi- dence from Rural Niger.” Working Paper 9473, World Bank Group. Rahman, Atif, Abid Malik, Siham Sikander, Christopher Roberts, and Francis Creed. 2008. “Cognitive behaviour therapy-based intervention by community health workers for mothers with depression and their infants in rural Pakistan: a cluster-randomised controlled trial.” Lancet 372 (9642):902–909. Ridley, Matthew, Gautam Rao, Frank Schilbach, and Vikram Patel. 2020. “Poverty, depres- sion, and anxiety: Causal evidence and mechanisms.” Science 370 (1289). Riley, Emma. 2021. “Role models in movies: the impact of Queen of Katwe on students’ educational attainment.” Review of Economics and Statistics Forthcoming. Sakulku, Jaruwan and James Alexander. 2011. “The Impostor Phenomenon.” International Journal of Behavioral Science 6 (1):75–95. Sherman, David K. 2013. “Self-Affirmation: Understanding the Effects.” Social and Per- sonality Psychology Compass 7 (11):834–845. Shnabel, Nurit, Valerie Purdie-Vaughns, Jonathan E. Cook, Julio Garcia, and Geoffrey L. Cohen. 2013. “Demystifying values-affirmation interventions: writing about social belong- ing is a key to buffering against identity threat.” Personality & Social Psychology Bulletin 39 (5):663–676. Sin, Nancy L. and Sonja Lyubomirsky. 2009. “Enhancing well-being and alleviating depres- sive symptoms with positive psychology interventions: a practice-friendly meta-analysis.” Journal of Clinical Psychology 65 (5):467–487. 25 Stadler, Gertraud, Gabriele Oettingen, and Peter Gollwitzer. 2009. “Physical Activity inWomen: Effects of a Self-Regulation Intervention.” American Journal of Preventive Medicine 36:29–34. Wanberg, Connie R., Edwin A. J. Hooft, Songqi Liu, and Borbala Csillag. 2020. “Can Job Seekers Achieve More through Networking? The Role of Networking Intensity, Self- efficacy, and Proximal Benefits.” Personnel Psychology 73 (4):559–585. 26 A Appendix A: Construction of Index Variables Throughout the survey, we often use a five-point frequency scale where participants select one of almost never true, true less than half the time, true about half the time, true more than half the time, and almost always true. Primary psychosocial outcomes 1. Self-advocacy. Participants rate on a five-point frequency scale the statements “I ask others to support me”, “I seek out new opportunities”, and “I ask others to connect me to new opportunities.” We also ask, “Suppose you learn about a job that pays RWF 1000 more each day than you currently earn. It is not guaranteed that you will get the job, and it will cost you 2 days wages if you apply. How inclined would you be to apply for the job?” We combine these questions into an equally weighted index to estimate changes (Cronbach’s alpha of 0.58 at baseline, 0.54 at follow-up). 2. Leadership positions. We estimate changes in the number of leadership positions held as well changes in the likelihood that a participant holds any leadership position. 3. Connectedness within cooperatives. We have five statements to measure connections with others in the cooperative: ”people are likely to come to me for advice”, ”I help others”, ”I have people with whom I feel completely secure”, ”there are people who will stand by me during difficult times”, and ”people know a lot about me.” We measure all five outcomes on our five-point frequency scale. We combine these outcomes into an index, equally weighting each measure, to estimate changes in how connected women feel with others in their cooperative (Cronbach’s alpha of 0.72 at baseline, 0.78 at follow-up). 4. Connectedness within the home and village. We elicit the same five measures of con- nection with others in participants’ homes and villages as we elicit for others in par- ticipants’ cooperatives. Again, we combine these outcomes into an index (Cronbach’s 27 alpha of 0.71 at baseline, 0.73 at follow-up). 5. Problem solving. We have two measures of problem solving, both measured on our five-point frequency scale: ”I can solve problems” and ”if I am struggling to solve a problem, there are others whom I can go to for help.” We combine these into an index, equally weighting each measure (Cronbach’s alpha of 0.43 at baseline, 0.58 at follow-up). 6. Effective communication. We have six measures of effective communication that we combine into an index (Cronbach’s alpha of 0.77 at baseline, 0.83 at follow-up). Par- ticipants rate the following five statements on our five-point frequency scale: ”I make suggestions to others”, ”if I have a problem or a new idea that would affect or benefit my community I raise it to others in my community”, ”people understand me when I make a suggestion for how to accomplish a task”, ”people understand me when I give them feedback”, and ”people understand what I am saying when I ask them to do something.” We also ask whether participants have spoken up in a group situation where they did not know everyone present in the past two weeks. Secondary psychosocial outcomes 1. Subjective Social Status. We measure Subjective Social Status (SSS) in present, five years in the past, and five years in the past using the MacArthur scale. We pre- registered four measures of changes in self-perceived social status between the baseline and follow-up surveys: (1) current self-perceived social status; (2) self-perceived social status five years from now; (3) difference between current measure and five years in the past; (4) difference between five years from now and five years in the past. After collecting follow-up data, it became clear that SSS five years in the past had actually changed substantially for women in SFL. These treatment effects rendered our differ- ence measures difficult to interpret, so instead we simply present average treatment effects on SSS currently, five years in the past, and five years in the future. 28 Figure A1: Visual Scale to Measure Connectedness with Workshop Group 2. Connectedness to workshop group. We measure how connected and together partici- pants feel with one another using a visual six-point scale, reproduced in figure A1. 3. Self-evaluation of value. We have two measures eliciting self-value: ”I matter” and ”there are good things about me.” Both are measured on a four point scale” not at all true, a little true, somewhat true, and very true. We combine these into an index, equally weighting each measure (Cronbach’s alpha of 0.49 at baseline, 0.63 at follow- up). 4. Resilience. We have two measures of resilience, both measured on our five-point fre- quency scale: ”I am able to adapt to change” and ”I tend to bounce back after illness or hardship.” We combine these into an index, equally weighting each measure (Cron- bach’s alpha of 0.02 at baseline, 0.08 at follow-up). 5. Identifying as a leader. We ask how closely participants identify as being a leader on 29 Figure A2: Visual Scale to Measure Identifying as a Leader a six-point scale, reproduced in figure A2. 6. Aspirations for the future. We have five measures of aspirations for the future. First, does the participant have a role model? Second, does the participant believe that they can ever be as successful as their role model? Third, conditional on answering yes to the second question, how long does the participant think it will take to be as successful as their role model? Fourth, does the participant have a goal they are working toward? Fifth, has the participant achieved the goal they stated for themselves at the start of the study? We estimate effects for all five dimensions separately. Economic indicators 1. Estimated income. We ask for an estimate of income earned by the participant over the past month. If a participant cannot provide an estimate, we provide intervals of RWF 5,000 and ask the participant to choose the one that seems closest to what they think they earned. In the case that a participant chooses an interval, we use the midpoint of the interval as the estimate of income. 30 Figure A3: Correlation between MUE and Income Note: The vertical axis shows the marginal utility of expenditure at baseline. The horizontal axis shows the inverse hyperbolic sine of income at baseline. The blue line shows the linear relationship between the two. 2. Likelihood of generating any income. We ask participants to describe their current work situation and code one for participants who are earning any income and zero otherwise. 3. Marginal utility of expenditure. We calculate each participant’s marginal utility of expenditure using information about household composition and consumption expen- ditures on a set of eighteen food items following Ligon (2019). The method in Ligon (2019) involves first regressing disaggregated expenditures on a set of basic household level controls: total household size and the number of boys, girls, women, and men in the household. Next, we take the residuals from the disaggregated regressions and per- form a singular value decomposition. Ligon (2019) shows that the resulting parameter is an estimate of marginal utility of expenditure, up to a multiplicative constant. We predict that the marginal utility of expenditure will be lower for workshop participants than for participants in the cash control group, and that SFL participants will have a lower marginal utility of expenditures than PD participants. 31 B Appendix B: Robustness In this appendix section we present results using the alternative specification Outcomei = α + β1 SF Li + β2 P Di ϵi , (2) which omits the control for baseline levels of each outcome. Our results are largely robust to omitting baseline controls. As we show in Figure A4, all of our estimated average treatment effects in our family of primary psychosocial outcomes are a similar magnitude to those in our primary specification, with only small losses in precision. Similarly, Figure A5 shows that most of the outcomes in our family of secondary psychosocial outcomes are similar in magnitude to our primary specification. The single exception is time to be as successful as the role model, which has a point estimate that turns negative when we omit baseline controls but remains statistically indistinguishable from zero. Finally, Figure A6 shows that our results are similar for income and working, but omitting baseline controls flips the sign on the marginal utility of expenditure from positive to negative. Again, the result remains statistically insignificant across both specifications. 32 Figure A4: Treatment effects for Primary Psychosocial Outcomes, Omitting Baseline Control Note: Light blue dots denote treatment effects for SFL relative to the unconditional cash transfer. Dark blue dots denote treatment effects for PD relative to the unconditional cash transfer. Lines show naive 95% confidence intervals. PPS Index is a Anderson (2008) index combining all outcomes within this family. We normalized all results by the standard deviation of the control group at the follow-up survey. Figure A5: Treatment effects for Secondary Psychosocial Outcomes, Omitting Baseline Con- trol Note: Light blue dots denote treatment effects for SFL relative to the unconditional cash transfer. Dark blue dots denote treatment effects for PD relative to the unconditional cash transfer. Lines show naive 95% confidence intervals. SPS Index is a Anderson (2008) index combining all outcomes in this family except aspiring to be the role model and time to be like the role model. We normalized all results by the standard deviation of the control group at the follow-up survey. 33 Figure A6: Treatment effects for Economic Outcomes, Omitting Baseline Control Note: Light blue dots denote treatment effects for SFL relative to the unconditional cash transfer. Dark blue dots denote treatment effects for PD relative to the unconditional cash transfer. Lines show naive 95% confidence intervals. Econ Index is a Anderson (2008) index that combines all outcomes in this family. We normalized all results by the standard deviation of the control group at the follow-up survey. C Appendix C: Additional Context and Results First, we show balance across the three treatment arms on our pre-registered family of sec- ondary psychosocial outcomes. Table A1 shows that the three groups are generally balanced on these measures at baseline, with one exception: SFL participants are slightly more likely to have a role model than those in PD. Given that we test a large number of covariates for balance and that we control for baseline levels of all outcomes in our primary specification, we are not concerned that this slight imbalance invalidates our results. 34 Table A1: Balance: Pre-Registered Secondary Psychosocial Outcomes Cash Control PD SFL p-value SSS - Current 3.06 3.01 2.88 0.6 (1.8) (1.54) (1.64) SSS - Past 2.07 1.8 1.86 0.17 (1.44) (1.14) (1.33) SSS - Future 4.79 4.84 4.92 0.88 (2.24) (2.15) (2.27) Self-Value 3.34 3.3 3.28 0.63 (0.5) (0.51) (0.49) Resilience 3.23 3.15 3.31 0.14 (0.76) (0.65) (0.65) Identify as Leader 3.81 3.73 4 0.34 (1.57) (1.78) (1.53) Connectedness to Workshop Group 5.3 5.33 5.23 0.71 (1.01) (0.94) (1.12) Has Role Model 0.86 0.8 0.9 0.05 (0.34) (0.4) (0.31) Aspires to be Role Model 0.9 0.89 0.92 0.84 (0.29) (0.31) (0.28) Time to be Role Model 5.12 4.91 4.5 0.27 (2.92) (2.9) (2.88) Has a goal 0.93 0.97 0.95 0.21 (0.26) (0.16) (0.22) N 155 148 152 Notes: Mean baseline covariates by treatment group. Standard deviations are in parentheses. Column 4 reports p-values associated with F-tests of joint equality between the three groups. 35 For further context, Figure A7 shows the full distribution of income among all partici- pants at baseline, with lines denoting mean and median income as well as the value of the unconditional cash transfer. Figure A7: Distribution of Participant Income at Baseline Note: The solid grey line denotes mean income among all participants at baseline (15,300 FRW) while the dashed grey line denotes median income (3,000 FRW). The black line denotes the per participant cost of each workshop (32,000 FRW), which is the amount distributed to the control group as an unconditional cash transfer. Table A2, Table A3, and Table A4 show our results on primary psychosocial outcomes, secondary psychosocial outcomes, and economic outcomes in table form to provide additional context for the magnitudes of our estimated effects along with formal tests for equality between the effects of SFL and PD. 36 Table A2: Primary Psychosocial Outcomes Self- Problem Leadership Connection Connection CommunicationAny Leader- Index Advocacy Solving Positions Home/Village Co-op Skills ship SFL 0.11 0.25 0.04 0.15 0.03 0.11 0.04 0.12 (0.07) (0.08) (0.08) (0.07) (0.08) (0.08) (0.06) (0.07) (0.09) (0.11) (0.05) (0.09) (0.04) (0.09) (0.05) PD 0.12 0.13 -0.03 0.13 0.16 0.02 -0.04 0.1 (0.07) (0.09) (0.08) (0.07) (0.08) (0.08) (0.06) (0.07) (0.11) (0.12) (0.05) (0.12) (0.14) (0.03) (0.05) Baseline 0.34 0.22 0.42 0.31 0.32 0.4 0.24 0.51 Level (0.04) (0.05) (0.06) (0.04) (0.05) (0.05) (0.06) (0.04) Control 3.14 3.18 3.88 3.73 3.25 3.26 0.49 0.01 Mean N 438 438 437 438 438 438 437 437 37 Adj. R2 0.14 0.06 0.11 0.12 0.1 0.15 0.03 0.24 SFL=PD 0.9 0.18 0.35 0.79 0.13 0.22 0.18 0.79 (p-value) Notes: Average treatment effects for our family of pre-registered, primary psychosocial outcomes. Self-advocacy, problem-solving, connection to home/village, connection to co-op, and communication skills are pre-registered indices, typically with all questions on 5-point Likert scales. Leadership positions is the number of leadership positions currently held and leadership position is an indicator variable equal to one if the respondent currently holds any leadership positions. Index is an index of all seven outcomes (note that we did not pre-register this index). We show naive White robust standard errors in the first set of parentheses. The second set of parentheses shows standard errors adjusted to control the false discovery rate within this family of outcomes according to the step-up procedure in Benjamini, Kreiger, and Yekutieli (2006). The bottom row shows the p-values associated with hypothesis tests of equality between the estimated average treatment effect for Storytelling for Leadership and Professional Development. Table A3: Secondary Psychosocial Outcomes SSS Past Future CWG Self- Resilient ID Role Aspire Time Has Achieve Index SSS SSS value Lead Model RM RM Goal Goal SFL 0.64 1.44 0.2 -0.01 0.2 0.06 0.79 0.05 0.04 0.32 0.01 -0.09 0.19 (0.24) (0.35) (0.29) (0.11) (0.08) (0.08) (0.17) (0.03) (0.03) (0.41) (0.02) (0.06) (0.07) (0.28) (0.43) (0.25) (0.06) (0.09) (0.08) (0.24) (0.04) (0.04) (0.4) (0.01) (0.07) PD 0.03 0.18 0.65 0.17 0.15 0.01 0.52 0.01 0.04 -0.11 -0.01 0.13 0.16 (0.19) (0.16) (0.37) (0.1) (0.08) (0.08) (0.18) (0.04) (0.04) (0.45) (0.02) (0.06) (0.08) (0.74) (0.22) (0.53) (0.14) (0.13) (0.27) (0.28) (0.31) (0.05) (2.96) (0.21) (0.1) Baseline 0.31 0.32 0.48 0.15 0.36 0.13 0.35 0.24 0.24 0.3 0.02 0.55 Level (0.06) (0.11) (0.07) (0.04) (0.06) (0.05) (0.04) (0.06) (0.09) (0.06) (0.05) (NA) (0.07) Control 3.39 2.19 4.96 5.06 2.61 3.27 3.88 0.88 0.91 5.47 0.97 0.49 0.11 Mean N 438 438 432 398 438 438 434 436 307 253 433 423 247 38 Adj. 0.07 0.07 0.11 0.04 0.07 0.02 0.17 0.09 0.08 0.08 0 0.03 0.2 R2 SFL=PD 0.01 0 0.2 0.08 0.57 0.48 0.11 0.19 1 0.32 0.43 0 0.69 Notes: Average treatment effects for our family of pre-registered, secondary psychosocial outcomes. SSS is current subjective social status. SSS Past is SSS evaluated 5 years in the past and SSS Future is expected SSS five years in the future. CWG (connection to workshop group), self-value, and resilience are pre-registered indices of questions. ID lead measures how closely the respondent identifies with being a leader. Role model is an indicator variable equal to one if the respondent has a role model. Aspire RM is an indicator equal to one if the respondent aspires to be their role model. Time RM is the time (in years) that the respondent thinks it will take them to be like their role model. Goal is an indicator equal to one if the respondent has a goal at follow-up. Achieve goal is an indicator equal to one if the respondent reports that they achieved the goal set at baseline by the time of the follow-up survey. Index is an index of all outcomes except aspiring to be the role model and the time to take to be like the role model, to avoid losses in power (note that we did not pre-register this index). We show naive White robust standard errors in the first set of parentheses. The second set of parentheses shows standard errors adjust to control the false discovery rate within this family of outcomes according to the step-up procedure in Benjamini, Kreiger, and Yekutieli (2006). The bottom row shows the p-values associated with hypothesis tests of equality between the estimated average treatment effect for Storytelling for Leadership and Professional Development. Table A4: Economic Outcomes Marginal Utility IHS(Income) Earn Any In- Index of Expenditure come SFL 0.03 0.69 0.07 -0.11 (0.11) (0.55) (0.06) (0.1) (0.04) (0.95) (0.1) PD -0.13 1.1 0.12 0.14 (0.1) (0.55) (0.06) (0.11) (0.07) (0.62) (0.07) Baseline Level 0.6 0.25 0.23 -0.03 (0.04) (0.05) (0.05) (0.06) Control Mean 0.15 3.39 0.35 -0.01 N 437 438 438 419 Adj. R2 0.31 0.07 0.06 0.01 39 SFL=PD (p-value) 0.11 0.46 0.36 0.02 Notes: Average treatment effects for our family of pre-registered economic outcomes. Index is an index of the three individual outcomes (note that we did not pre-register this index). We show naive White robust standard errors in the first set of parentheses. The second set of parentheses shows standard errors adjust to control the false discovery rate within this family of outcomes according to the step-up procedure in Benjamini, Kreiger, and Yekutieli (2006). The bottom row shows the p-values associated with hypothesis tests of equality between the estimated average treatment effect for Storytelling for Leadership and Professional Development. Figure A8: Post Workshop Treatment Effects Note: Light blue dots denote treatment effects for SFL at the post workshop survey relative to the unconditional cash transfer at baseline. Dark blue dots denote treatment effects for PD at the post workshop survey relative to the unconditional cash transfer at baseline. Lines show naive 95% confidence intervals. We normalized all results by the standard deviation of the control group at the baseline survey. We also present results from a smaller survey conducted immediately following each work- shop (Figure A8). Here, we compare post workshop outcomes for women who participated in SFL and PD to baseline outcomes for women in the cash control group. We did not pre-specify this analysis. Finally, we present our estimated effects on speaking up in a group situation (an outcome which we did not pre-register but which speaks to potential behavioral changes as a result of the workshop). Figure A9 shows effects for SFL and PD on speaking up. C.1 Heterogeneity by Age The type of psychologically motivated intervention that we study with SFL may be more impactful for younger participants whose sense of self and ideas about their place in the community are less fixed. Conversely, since age bestows a degree of social status, it may be easier for older women in our sample to leverage the skills they learn in PD to achieve their goals. While not pre-specified, we take advantage of our sample stratification to explore differential impacts for women who are over versus under the median age of forty at the time of the intervention. 40 Figure A9: Treatment Effects on Speaking Up in Group Situations Note: Estimated treatment effects on the probability that participants have spoken up in a situation where they did not know everyone present in the past 2 weeks. Light blue dots denote treatment effects for SFL relative to the unconditional cash transfer. Dark blue dots denote treatment effects for PD relative to the unconditional cash transfer. Lines show naive 95% confidence intervals. We normalized all results by the standard deviation of the control group at the follow-up survey. Note that this estimate was not pre-specified. Figure A10 shows heterogeneous treatment effects on our family of primary psychosocial outcomes, with the upper panel showing estimates for SFL and the lower panel showing results for PD. None of our primary psychosocial outcomes show differences for participants who are over versus under forty that are significant at the 5% level. Figure A11 does show that women under forty qualitatively appear to be driving the positive treatment effects of SFL on speaking up, suggesting that younger women may be using the workshop material more in the context of peer leadership. Figure A12 shows heterogeneous average treatment effects by age group for our family of secondary psychosocial outcomes. For SFL, the likelihood of achieving the goal set at baseline is statistically significantly higher for women under forty than women over forty. The general pattern of effects mirrors what we find in our family of primary psychosocial outcomes for SFL: point estimates of average treatment effects for women under forty tend to be greater than or equal to those for women over forty. By contrast, self-value scores are higher for women over forty in the PD group, again with age differences significant at the 10% level. Beyond these two outcomes, it is difficult to detect any systematic patterns in 41 Figure A10: Heterogeneity by Age for Primary Psychosocial Outcomes Note: The upper panel shows treatment effects for SFL relative to the cash transfer group, with effects for women over 40 in dark blue and effects for women under 40 in light blue. The bottom panel shows analogous effects for PD, with effects for women over 40 in dark red and effects for women under 40 in pink. Lines show naive 95% confidence intervals. We normalized all results by the standard deviation of the control group at the follow-up survey. Figure A11: Heterogeneity by Age for Speaking Up in Group Situations Note: Treatment effects on speaking up in a group situation, with effects for women over 40 in dark blue and effects for women under 40 in light blue. Lines show naive 95% confidence intervals. We normalized all results by the standard deviation of the control group at the follow-up survey. 42 our estimated heterogeneous treatment effects for PD. Finally, we examine heterogeneous effects by age on our family of economic outcomes in Figure A13. Here we see no suggestive or statistically significant differences between age groups for SFL. PD exhibits higher treatment effects on all economic outcomes for women who are over forty except the marginal utility of expenditure, where we see the opposite pattern. While only the difference in the index of all economic outcomes is significant at the 10% level, the magnitude of the differences in income earned and earning any income are economically meaningful: women over 40 in PD earn around 150% higher incomes and are 17.6 percentage points more likely to earn any income than those who are under 40. Combined with our result on the likelihood of achieving the goal set at baseline, these results suggest that women over forty may be driving the economic impacts we estimate for PD. 43 Figure A12: Heterogeneity by Age for Secondary Psychosocial Outcomes Note: The upper panel shows treatment effects for SFL relative to the cash transfer group, with effects for women over 40 in dark blue and effects for women under 40 in light blue. The bottom panel shows analogous effects for PD, with effects for women over 40 in dark red and effects for women under 40 in pink. Lines show naive 95% confidence intervals. We normalized all results by the standard deviation of the control group at the follow-up survey. 44 Figure A13: Heterogeneity by Age for Economic Outcomes Note: The upper panel shows treatment effects for SFL relative to the cash transfer group, with effects for women over 40 in dark blue and effects for women under 40 in light blue. The bottom panel shows analogous effects for PD, with effects for women over 40 in dark red and effects for women under 40 in pink. Lines show naive 95% confidence intervals. We normalized all results by the standard deviation of the control group at the follow-up survey. 45