Policy Research Working Paper 9850 Helping Families Help Themselves? Heterogeneous Effects of a Digital Parenting Program Sofia Amaral Lelys Dinarte Patricio Dominguez Santiago M. Perez-Vincent Development Economics Development Research Group November 2021 Policy Research Working Paper 9850 Abstract Parenting practices are crucial for the development of chil- stress and anxiety and lowered caregiver-child interactions. dren’s brains and social skills. However, parenting styles The effect on males was concentrated among the poorer may be far from ideal, particularly those of caregivers with and those residing with a partner. In contrast, women’s high stress levels. Using an individual-level experiment mental health was not impacted. Yet, their use of physical with male and female caregivers of young children in El violence toward children decreased by 18 percent. These Salvador, this paper evaluates the impact of a free digital results align with theories linking economic deprivation stress management and positive parenting intervention. The and family structure to caregivers’ cognitive overload and results indicate that, for males, the intervention increased mental health. 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 ldinartediaz@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 Helping Families Help Themselves? Heterogeneous Effects of a Digital Parenting Program∗ Sofia Amaral† Lelys Dinarte Diaz‡ Patricio Dominguez§ Santiago M. Perez-Vincent¶ Keywords: Mental health, positive parenting, parental stress, child maltreatment JEL Codes: J13, J22, I24, I12, J12, J16 ∗We appreciate the valuable feedback from Kathleen Beegle, Sven Resnjanskij, Victoria Endl-Geyer, Andrew Friedson, Selim Gulesci, Florencia Lopez Boo, Barbara Petrongolo, and participants at the NBER Winter Development Conference, ifo Institute, IZA, and Nordic Development Economics Conference. We also thank Miguel Paniagua for his invaluable fieldwork coordination and Stef- fanny Romero for her superb work as research assistant. We thank Glasswing International, our implementation partner, for their support in the development and implementation of this program, as well as their trust and commitment to generate rigorous evidence. This work was supported by the IDB-COVID-19 Call for Research Projects, the World Bank Research Support Budget, and the COVID-19 Emergency Call Window from the World Bank Early Learning Partnership. This research project’s protocol was reviewedand approved by the Institutional Review Board (IRB) at the Universidad Francisco Gavidia in El Salvador in April 2020 with approval ID No. 003-2020. A preanalysis plan was registered at the American Economic Association RCT registry - AEARCTR- 0007096. The authors have no conflicts of interest to report. The findings, interpretations, and conclusions expressed in this report are entirely thoseof the authors. They do not necessarily represent the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent; nor those of the World Bank and its affiliated organizations, its Executive Directors, or the governments they represent. †ifo Institute at the Ludwig Maximilian University of Munich and CESifo. Email: amaral@ifo.de ‡Development Research Group. The World Bank. Email: ldinartediaz@worldbank.org §Pontificia Universidad Catolica, Chile. Email: pdomingr@ing.puc.cl ¶Inter-American Development Bank. Email: santiagoper@iadb.org 1 Introduction Parenting styles and decisions are crucial in driving human capital accumulation (Olivetti and Petrongolo, 2017; Doepke et al., 2019; Attanasio et al., 2020b). Parental inputs during early life stages influence the development of children’s cognitive and socioemotional skills, which subsequently influence their health, school performance, and labor market outcomes in adulthood (Carneiro et al., 2019; Attanasio et al., 2020b,a; Baranov et al., 2020).1 Despite the importance of parenting to the development of children’s brains and social skills, parenting styles and practices are far from ideal. UNICEF estimates that nearly three in four children regularly suffer physical or psychological violence by their caregivers (UNICEF, 2017). This early exposureto violence can have long-lasting effects. For instance, children raised in a violent or stern environment are more likely to participate in criminal activities (Doyle Jr and Aizer, 2018; Sviatschi, 2018) and exhibitrisky behaviors as teenagers (Hamby et al., 2011). Overall, the high prevalence of child maltreatment and its potential long- term impact on children’s well-being call for innovative and effective strategies. We exploitthe association between caregivers’ mental health and quality of interactions between parents and children (Cluver et al., 2020; Renzetti, 2009; UNICEF, 2020, 2017) as a potential avenue to address the prevalence of child maltreatment in the context of a highly violent country. In this paper, we evaluate the impact of a digital parenting skills intervention that provided caregivers with stress-management and positive parenting techniques, using a large-scale, individual-level, randomized controlled trial (RCT) with 3,103 caregivers2 of children aged 0 to 8 years old in El Salvador3 duringthe second half of 2020—that is, in the context of the COVID-19 pandemic and stay-at-home orders. We randomly assigned caregivers to a treatment or control group with equal probability. Caregivers in the treatment group received up to four SMS/WhatsApp messages per week over eight consecutive weeks.4 Before the intervention, we collected data on caregivers’ mental health, impulsiveness, caregiver -child interactions, attitudes towards violent parenting, violence perpetration, and other sociodemographic characteristics. We combined direct questions and vignettes to reduce potential social desirability bias in responses to sensi- tive questions, such as those on attitudes toward violent parenting practices. We resurveyed caregivers one month after the completion of the intervention to measure the short-run impact of the program. 1Growing evidence in medicine and psychology shows that the emotional development of children is shaped early in life through interactions with caregivers. Emotional development impacts the architecture of the developing brain and lays the foundations for sound mental health and life experiences (Thompson and Lagattuta, 2006). 2In our setting, a caregiver can be a parent or a non-parent adult. Throughout the paper we use caregivers to mean individuals such as mothers, fathers, grandmothers, and others who provide a child’s primary care. 3Violence against children is a major socioeconomic problem in El Salvador. A recent nationally repre- sentative survey shows that 55% of boys and 50% of girls (1–14 years old) experienced violent discipline in the past month (WHO, 2020). 4Overall, the intervention consisted of 27 messages containing information, videos, infographics and exercises on parental stress-management and positive parenting techniques. Section 2 provides a detailed description of the intervention. Figure A2 in the Appendix shows examples of the materials received. 2 We document three main results. First, we show that the large majority of caregivers opened the SMS/ WhatsApp messages. Importantly, by using knowledge incorporation surveys, we document that caregivers in the treatment group were more likely to know about stress and parenting techniques than those in the control group. Second, we find the intervention had an overall negative impact on caregiver mental health. Contrary to what we had hypothesized in our pre-analysis plan (PAP), our estimations indicate that the mental health of treated caregivers worsened by 0.057 standard deviation (sd) relative to that of caregivers in the control group. This effect was driven by a negative impact on stress (0.072 sd). Third, despite the impact on mental health, we find no evidence that the intervention changed caregiver impulsiveness, the quantity of caregiver-child interactions, caregiver attitudes towards violent parenting, or children’s behaviorsor socio- emotional development. Moreover, we find heterogeneous impacts of the intervention depending on caregiver sex. The overall unintended effects of the intervention were concentrated among men: treated male caregivers’ stress and anxiety levels increased relative to those in the control group by 0.108 sd and 0.095 sd, respectively. We observe no significant impact on women’s mental health. We also observe that the intervention led to fewer interactions with children among male caregivers (0.137 sd) and had no impact among women. Finally, we find that the intervention reduced the reported use of physical violence among female caregivers (0.098 sd) and had no effect among men. This reduction in the use of physical violence amounts to a decrease of 18 percent. The often observed differential roles and involvement in parenting activities (Hupkau and Petron- golo, 2020; Olivetti and Petrongolo, 2017) and mental health status (Offer and Schneider, 2011) betweenmale and female caregivers presaged a heterogeneous impact of the intervention between these groups. 5 Overall, our results verify the empirical importance of caregiver sex in moderating the impact of a parenting intervention. We also show that the intervention improved the quality of parenting for female caregivers by reducing the use of physical violence against children.6 In contrast, for males, the intervention led to more mental distress and a retraction from parenthood. To further understand the differential impacts by gender, we show that living in more economically deprived households and cohabiting with a partner were associated with a greater negative impact of the intervention on mental health, especially among male caregivers. For example, among male caregivers in the most economically deprived households, the intervention had a 0.36 sd impact in mental distress (vs. a non-significant -0.032 sd impact among those in other less-deprived households). These results align with the literature linking economic deprivation with increased cognitive load (Mullainathan and Shafir, 2013; Mani et al., 2013; Asadullah et al., 2021; Schilbach et al., 2016; Ridley et al., 2020) and suggest that, for caregivers burdened by economic deprivation, the intervention may have added to the cognitive overload 5Our PAP contemplated the estimation of the intervention’s impact by caregiver sex. 6In the U.S., females are generally the main perpetrators of child maltreatment (Lindo et al., 2018). This is also the case in our setting – with women having a higher rate of perpetration of violence when comparedto men. This outcome may be more easily impacted with the intervention for females. 3 and been a source of stress. Also, we find that the negative impact on the frequency of caregiver-child interactions is concentrated among male caregivers living with a partner. This finding suggests that, when living as a couple, parenting interventions may make men even more reliant on their partner for parenting but with a cost for their own mental health.7 We conduct a variety of robustness tests. First, we show that attrition is low in our study and that there is no selective attrition between treatment and control groups. Second, following Asadullah et al. (2021) and Aguero and Frisancho (2021), we test the robustness of our results to the inclusion of a social desirability index that captures individual-level propensity to misreport sensitive items as an additional control variable. Finally, we verify the robustness of our results to the exclusion of control variables and to the use of randomization inference to estimate standard p-values. In all cases, our results remain similar in magnitude and statistical significance. Our paper contributes to four strands of the literature. First, it contributes to the emerging literature measuring the impact of digital intervention and low implementation cost parenting programs. Within this literature, programs that involve group sessions (Carneiro et al., 2019) or text messages (Barrera et al., 2020) have shown mixed results. On the one hand, Carneiro et al. (2019) find that group sessions improve caregiver-child interactions and child development, and lower negative parenting attitudes. On the other hand, Barrera et al. (2020) do not detect effects on child development but show evidence of positive impacts on parenting skills. Our study adds to this work by offering evidence of the impact of a similar interventionin content by identifying specific characteristics of the recipient population that might moderate interventions’ impact – i.e, gender – that is also delivered digitally and during a turbulent period for families. Our findings illustrate the empirical relevance of targeting and tailoring these interventions based on caregiverand household characteristics. Theoretically, the impact of parenting policies or interventions like ours may have an ambiguous impact on children and families. Yet, there is limited understanding of such effects and the literature suggests such policies may have a heterogeneous effect depending on various aspects, suchas family structure, as we show (Olivetti and Petrongolo, 2017). Second, we contribute to the growing and varied evidence on the effect that different parenting-related policies have on men’s participation in parental responsibilities and other household outcomes. Evidence from high-income countries shows mixed results of the effects of fathers’ parental leave on parent-child ´ and Gonza interactions and mothers’ well-being (Farre ´ lez, 2019; Ekberg et al., 2013), while showing positive impacts on mother’s labor force participation (Bartel et al., 2018).8 Evidence from low-income countries 7For instance, it is likely that by realizing the importance of being a good parent, males residing with a partner may have become more distressed by feeling guilty or simply by becoming aware of their own lack of skills. An increase in mental distress is also plausible if men felt more pressured to perform in other tasks – presumably more male-oriented tasks - within the household. 8 Farre ´ and Gonza ´ lez (2019) show that Spain’s two weeks of paid paternity leave raised fathers’ interac- tions with children but lowered fathers’ desired fertility. Also, Ekberg et al. (2013) show that Swedish fathers 4 also paints a mixed picture; some studies show that fathers’ engagement in chi ldcare can reduce violence toward mothers and children (Doyle et al., 2018), while other evidence depicts null effects on caregiver- child interactions (Maselko et al., 2019; Justino et al., 2020). Our paper aligns with Rossin-Slater (2017), who shows that parental leave for fathers in the United States had zero to negative effects on interactions between fathers and children, with some positive effects among unmarried fathers. Third, we add to a small but growing economics literature on mental health and its link with parenting practices. Emotions such as stress, anxiety, and frustration affect children’s health and development and the quality of interactions between parents and children, potentially leading to violent or harmful interactions (Persson and Rossin-Slater, 2018; Cluver et al., 2020; Renzetti, 2009; Bendini and Dinarte, 2020; UNICEF, 2020, 2017). Interventions that help parents cope with these emotions can improve the mental health of mothers (Persson and Rossin-Slater, 2019; Baranov et al., 2020) and the relationship between parents and children (Knerr et al., 2013; Cluver et al., 2018), as well as reduce the risk of child maltreatment (Doyle et al., 2018). We add to this body of research by testing the effects of a digital intervention that, leveraging on previous findings in this literature, seeks to promote better parenting through improved caregiver mental health. Our results reveal some challenges for this type of intervention and identify contextual variables to consider in their design. Finally, we contribute to the evidence on the impact of digital health and education interventions—a policy- relevant issue worldwide (Crawfurd et al., 2021; Berlinski et al., 2021; Campion et al., 2020; Health, 2021). In low-income settings, the widespread availability of mobile phones, high social inequalities in access to health care, and the stigma associated with mental health problems make such tools attractive, but their effectiveness is not yet well understood (Naslund et al., 2017; Kola, 2020). We provide evidence thatthe net benefits of health light-touch and low-cost interventions may depend critically on the characteristics of the recipient population and, in our context, on the family structure. This result contrasts with findings from education interventions showing positive impacts on learning as a result of nudges and light digital support to students (Berlinski et al., 2021; Crawfurd et al., 2021; Lichand and Christen, 2020). The remainder of the paper proceeds as follows. Section 2 documents the intervention we study and its implementation process. In Section 3, we discuss the design of the experiment and its validity. Section 4 presents the different data categories we collected. Section 5 presents the identification strategy and Section 6 discusses our main findings. Section 7 shows the robustness checks. Section 8 provides possible mechanisms and Section 9 presents the concluding remarks. who benefit from parental leave do not improve mothers’ labor supply and fathers are not more likely to take days off work to assume childcare responsibilities. Moreover, Persson and Rossin-Slater (2019) document that work flexibility for fathers in Sweden lowers the cost of childbearing for mothers as it lowers the risk of poor maternal mental health. 5 2 The Intervention: Digital Stress-Management and Positive Parenting Program in El Salvador 2.1 Intervention structure We study a program that provided stress management and positive parenting techniques to caregivers of children in El Salvador. This program was developed by psychologists and early childhood development experts from the NGO Glasswing International, based on material from its existing in-person programs.9 After the onset of the pandemic, the NGO adapted this material to a digital format compatible with stay-at- home orders and social distancing requirements. The anticipation of the pandemic’s effects on caregivers’ mental health and child maltreatment, together with the restricted availability of traditional care services, underscored the importance of an intervention providing caregivers with strategies to improve their mental health and parenting skills during the pandemic. This motivation is in line with recent evidence showing that stress, anxiety, and frustration can affect the quality of interactions between parents and children (Cluver et al., 2020; Renzetti, 2009; UNICEF, 2020, 2017) and confirms the potential of parenting interventions to improve caregivers’ relationships with their children in low and middle-income countries (Knerr et al., 2013; Cluver et al., 2018). The intervention sought to help caregivers recognize their own emotions, use coping strategies to lower mental distress, promote positive parenting practices and, ultimately, reduce violent relationships between caregivers and children (Humphreys et al., 2020; Szabo et al., 2020). We partnered with Glasswing International to evaluate the effectiveness of this intervention. While it was the pandemic that prompted the redesign of the intervention, its new digital nature makes it an innovative alternative to complement other more costly parenting policy initiatives, such as home visitations, cognitive behavioral therapy, or parental leave programs, beyond this context. Its low cost, easy accessibility and scalability make it an attractive in- tervention for governments willing to expand their supply of social services. Generating evidence on the effectiveness of this type of policy is essential to determine the real value of these alternatives and to inform their design and implementation. The intervention has two intertwined components: (A) stress-management skills development for caregivers, and (B) positive caregiving techniques. The first component includes materials that show how to identify and manage physical, cognitive, emotional, and behavioral stress manifestations. It aims to help participants effectively manage the stressors of daily life and improve their mental well-being. The materials 9Glasswing International was founded in El Salvador in 2007 and currently works in seven Central Amer- ica and Caribbean countries on education, health, and community empowerment, as well as an employment and entrepreneurship program. Glasswing International’s website (https://glasswing.org/) provides more information on its programs. 6 emphasized the topics that lockdown intensified. The second component includes positive parenting practices and basic concepts of family life. The intervention’s ultimate objective is to reduce violent relationships between caregivers and children. As shown in Figure 1, the core of its theory of change is that informationon stress- management and positive parenting techniques can help caregivers to (i) identify stressors and their effects on themselves and children in the household, (ii) use the provided strategies to better copewith these stressors and reduce mental distress, (iii) understand that some actions and attitudes are formsof negative parenting and learn alternative disciplinary tools, and (iv) adopt positive parenting techniques (Glasswing, 2020). These changes could reduce violent caregiver interactions with children through two channels. First, better stress-management strategies and improved mental well-being can help caregivers reduce impulsive violent reactions toward children. Second, increased knowledge on positive parenting techniques can increase awareness of the harmful effects of violent interactions, and give caregivers alter- native disciplinary tools. We hypothesized that the program would improve caregivers’ mental distress, change their views about child maltreatment, and improve the quality of interactions through the use of positive parenting techniques. These changes would reduce the likelihood that they would abuse and mal- treat their children. It is also worth mentioning that both components are delivered jointly since the frame of the materials was purposely developed considering the context of family relations. As a result, both components cannot be separated in different individual treatments. Below we describe the details of each component of the intervention. A. Stress-management skills development This component highlights different ways to manage stress, such as stress-inoculation training and self- control techniques.10 Existing empirical evidence shows that individualized interventions focused on stress management—such as mindfulness, meditation, psychosocial programs, and others —can improve individuals’ well-being by reducing their stress levels (Dinarte et al., 2021; Holman et al., 2018). This first component includes 12 topics with exercises and information aimed at helping participants understand stress and its effects and learn coping strategies to reduce mental distress. It offers exercises and techniques for meditation, breathing, self-control, stretching, and emotional freedom. These topics are grouped into two categories. The first one consists of the identification of stressors. It focuses on helping participants to identify the types of stressors they experience, including disturbances in the family balance that come from external factors (e.g., unemployment) or internal problems (e.g., the illness of a family member); intraindividual transmission of stress from a specific domain (e.g., the workplace) to another (e.g., family life); and interindividual transmission of stress from one family member to another. The second category 10Stress inoculation aims to make an individual capable of addressing future stressful situations by ana- lyzing stressful circumstances and learning specific coping skills (Guarino, 2013). 7 consists of relaxation and breathing techniques and includes activities that bolster well-being, producing a ´ ndez et al., 2012). The relaxation techniques help participants to gradually reduce sense of tranquility (Ferna their stress and anxiety, contributing to problem solving and producing health benefits and mental balance ´ ndez et al., 2012). The main goal of breathing techniques is to improve people’s mental well-being by (Ferna showing how to voluntarily control their breathing and helping their minds to become calm and focus on the present. B. Positive parenting techniques Positive discipline principles posit that caregivers can learn to interact with children through positive actions and mutual respect. Positive parenting requires the capacity to direct a child’s behavior through thefrequent use of anticipation, negotiation, and perspective (Nelsen, 2007). This capacity helps caregivers to avoid using punishment to control children’s conduct and provides a constructive framework to teach children how to regulate their emotions. This component combines three main elements developed by Durrant (2013) in 15 topics, which first aim to promote caregivers’ understanding of how children think, feel, and behave, as well as the ir needs at each development stage. Next, the intervention materials seek to make caregivers understand that some actions and attitudes are forms of negative parenting and become aware of the need for warmth and structureduring interactions with their children.11 Finally, caregivers receive techniques to address children’s needs and behaviors through warm and structured practices, based on their development stages. The materials include activities that promote positive management of children’s emotions, better communication, family coexistence, self-control, and recommendations for an optimal family environment.12 2.2 Intervention delivery and piloting The 27 topics (12 for stress-management skills development and 15 for positive parenting techniques) were delivered through three or four SMS/WhatsApp messages per week for two months. These messages explained the main concepts and exercises. The messages contained the links to videos, animations, audio messages, quizzes and short essays with recommendations for practicing specific techniques. The messages also offered a link to a web page (a blog) wherein participants could find additional content for every topic, as shown in Figure A2 in the Appendix.13 The intervention was free to participants. To do so, we partnered with Tigo, the 11The definition of warmness in the theory of change refers to emotional security, verbal and physical affect, respect, sensibility, and empathy to respond to children’s needs. The theory of change defines structureas the ability to state clear instructions to better guide children’s behavior. 12The complete set of intervention materials can be found at this link. 13Access to the blog was restricted to participants in the treatment group. 8 largest mobile phone provider in El Salvador, which gave each participant free access to all content. We ran a small pilot project to validate understanding and wording of materials. Glasswing International invited 12 adults who met the eligibility criteria to participate in the pilot. Once they agreed, the NGO sent these caregivers the materials, asked them to review and implement the program at home, and invited them to a virtual focus group. During this event, caregivers discussed the materials and highlighted potential concerns and recommendations. The program was then adjusted based on the information fromthis process. Pilot participants are not part of the study sample. 3 Experimental Design 3.1 Recruitment and enrollment of participants We enrolled participants using three alternatives: (i) dissemination through Facebook, (ii) enrollment through Glasswing International’s communities network, and (iii) dissemination through SMS/WhatsApp messages sent to customers of Tigo. The main recruitment of participants was done through Tigo’s clients database (iii). As we show in Table A1, this group constitutes the largest share of our sample. To enroll participants, we sent a link to an enrollment survey through these three channels. This survey included questions that helped us to identify our target group. We specified the following eligibility criteria: caregivers had to be 45 years or younger, live in the same house with at least one child eight years old or younger, and provide their consent to receive digital messages and to participate in the study. We enrolled a total of 4,718 individuals who met the eligibility criteria and provided consent to participate in the study and intervention.14 We collected baseline data from 3,103 individuals (66 percent).15 3.2 Randomization We randomly assigned all enrolled caregivers who met the eligibility criteria and completed the baseline survey to either the treatment or the control group with equal probability. We stratified the group of 3,103 individuals based on two criteria: gender and enrollment modality. Given usual differences in mental health and parenting involvement between men and women, we wanted the treatment and control groups to be balanced in this dimension. In addition, given the different roles women and men take in caregiving, 14The enrollment survey questions and consent form are available at this link. 15The remaining 1,615 (34 percent) enrolled individuals did not complete the baseline survey for several reasons, including: they did not provide a correct phone number, we were not able to reach them after the maximum number of attempts agreed in the ethics agreement, they changed their mind and decidednot to participate in the study, among others. 9 we also aimed to assess whether the intervention had heterogeneous effects depending on the gender of the caregiver.16 Similarly, since individuals contacted through the three enrollment channels could differ in characteristics potentially relevant to our study, such being participants of Glasswing programs, we wanted to ensure that the treatment and control groups had an equal proportion of each characteristic. Table A1 in the Appendix provides the size of each stratum in the sample. We implemented the random assignmentright after concluding the baseline data collection. 4 Data To measure our outcomes of interest and gather additional information from caregivers in our study, we conducted three data-collection processes and used several survey instruments, as we describe below. 4.1 Data-collection stages Baseline data collection: We contacted all caregivers who provided consent to enroll in the study and were eligible for the intervention. Each caregiver received a link to an online baseline survey through SMS/WhatsApp. Participants also received reminders to complete the online survey if they had not done so within approximately two weeks.17 All data collected at baseline were self-reported. To reduce the risk of respondent fatigue, we limited the instrument length to approximately 30 minutes. Participants also received a small monetary incentive to complete the survey.18 We collected data from 3,103 respondents at baseline. Following information protection protocols, collected data were stored on a private server of the survey firm. Access was restricted to the project staff and researchers. The baseline survey included modules on socioeconomic condition of the caregiver and the household; employment status; mental health; child-parenting interactions; and child maltreatment. Section 4.2 de- scribes the information collected in more detail. Viewership rates and knowledge incorporation surveys: The second set of data-collection activities was carried out during the implementation of the intervention. First, we gathered information on the number of 16This research question and analysis was pre-specified in the PAP. 17The survey instrument was pretested in a pilot phase. Since self-completed surveys were faster and cheaper to implement, this was the strategy adopted for the baseline, which we complemented with phone reminders. Staff responsible for these reminders were trained in two topics: the content and structure ofthe baseline instrument and the protocol to protect victims of violence. The second topic prepared them to follow ethical guidelines outlined in the IRB protocol. 18Each participant received US$2.50 in Tigo ”money” as an incentive to complete each survey. In addition, we motivated participants to remain in the study with raffles of iPads after its completion. 10 opened SMS/WhatsApp messages sent to caregivers in the treatment group. This data was collected by Tigo and shared with the researchers at the aggregate level. We use this data to monitor the rate of viewership of the intervention.19 Second, we also conducted four knowledge incorporation survey rounds with a random subset of enrolled individuals. These surveys evaluated if individuals in the treatment group were more likely to self- report knowing concepts taught in the intervention, similar to the data collection in (Carneiro et al., 2021).We surveyed a total of 659 participants —26% of the sample at baseline— split into four survey rounds. Each survey round contained only four questions and was conducted by phone. In order to avoid saturation of respondents, we selected a different random sample of participants for each survey round. Thus, our data constitutes a cross-section of a sub-sample of the study participants.20 In each wave, we surveyed respondents on how often they had used different stress-management and positive parenting techniquesin the previous week. The timeline of events was such that participants would be surveyed with a lag oftwo to three weeks with respect to the contents of the messages in each survey round.21 Table A2 in the appendix shows the topics of each survey round. We use this information to assess if respondents’ take-upof the intervention’s material was positive. Endline data collection: The intervention implementation finished by the end of November 2020. We initiated the endline data collection at the end of December 2020. The timing of follow-up was designed to testthe short-term effects of the intervention and to minimize attrition. All 3,103 respondents of the baseline survey were contacted to complete the endline survey. To minimize the risks of attrition and low response rates, the endline survey was conducted over the phone. In a companion paper, we study an experiment that was embedded in the baseline data collection.22 In that paper, we show that the phone survey in comparison to the SMS survey increased the rate of survey completion by 40 percentage points. For this reason, at endline all participants were surveyed by phone interviews. As at baseline, all enumerators were trained in the content and structure of the follow-up instrument and in the guidelines of the ethics protocol. The structure of the endline survey was similar to that of the baseline. In addition, we also included a module on observed child behaviors and demand for another parenting intervention by Glasswing International. We collected follow-up data from 2,280 caregivers. 19Since the data is at the aggregate level we are unable to compute a treatment-on-the-treated estimator using the rate of messages opened as a proxy for take-up. Instead, we will use the knowledge incorporation sub- sample. 20Overall, only 14% of the 659 respondents were included in more than one round. 21For example, in week one treated participants saw a video regarding the turtle technique. Two weeks after this first content, study participants were surveyed regarding this technique which they could haveonly learned from watching the video. 22The aim of this experiment was to inform us about what would be the best method of data collection in order to minimize attrition. 11 4.2 Survey instruments and outcomes Based on the intervention’s theory of change, and as specified in our PAP, we focus our analysis on the following main outcomes: mental health, impulsiveness, and the quality of interactions with children. Appendix A1 presents a detailed description of all outcomes and the survey instruments used to measure them. Mental health: We use the Depression, Anxiety, and Stress Scale (DASS-21) instrument to measure mental distress (Lovibond and Lovibond, 1996). The DASS-21 module includes 21 items organized in three sub- scales: depression, anxiety, and stress.23 One major advantage of the DASS-21 is that it includes questions on physiological responses consistent with poor mental health rather than only on perceptions of mental distress, which are more prone to reporting bias and subjectivity. Our main outcome of interest is the aggregate index (comprising the three subcategories), but we also separately consider levels of depression, anxiety, and stress. On average, caregivers in our sample display low levels of mental health: about 34% exhibit an above- normal level of stress, 54% show high levels of anxiety, and 34% have depressive symptoms(See panel D in Table 1). Women in our sample exhibit worse mental health when compared to men in thethree subscales (see panel D in Table 2 in the Appendix). These rates are high, but similar to those observed in other populations—for example, in Turkey (Altindag et al., 2020), Spain (Rodr´ıguez-Rey et al., 2020), and Ireland (Burke et al., 2020). Impulsiveness: We measure caregiver impulsiveness using the Barratt Impulsiveness Scale BIS-11 (Patton et al., 1995).24 The outcome is the sum of the 15 items in the instrument, with a higher level reflecting a greater degree of impulsiveness. We use a standardized value of this index as our measure of impulsiveness. On average, respondents exhibit a low level of impulsiveness with only 2% of caregivers exhibiting above- normal levels, as shown in Table 1. Relative to men, female caregivers are 5 percentage points more likely to exhibit high impulsiveness levels (Table 2). Caregiver-child interactions: We use the 10 items related to support for learning/stimulating environment and setting limits domains from the Family Care Indicators instrument developed by UNICEF (Kariger 23We report the instrument in section A1.1 in the Appendix. Following Lovibond and Lovibond (1996), depression is characterized by hopelessness, devaluation of life, and lack of interest. Anxiety is characterized by a state of intense, excessive, and persistent worry and nervousness. Stress is the degree to whichan individual shows nervous arousal, difficulty in relaxing, impatience, and ease in becoming agitated or irritable. 24The instrument is included in section A1.2 in the Appendix. Impulsiveness broadly refers to an individual’s tendency to act suddenly without careful thought about the consequences of her actions. The psychology literature links this tendency to difficulties in planning, thinking carefully, enjoying challenging mental tasks, and focusing on the tasks at hand, as well as inclinations to act on the spur of the moment and having racing thoughts (Patton et al., 1995). Recent studies have found suggestive evidence on how behavioral interventions to reduce automatic responses —a form of impulsive behavior—can reduce violent behaviors (Heller et al., 2017; Dinarte and Egana-delSol, 2019). 12 et al., 2012) to measure the frequency of caregiver-child interactions.25 The items ask about the frequency of different interactions with the child in the previous week. Our outcome of interest is a standardized index of these items, constructed following Anderson (2008). On average, caregivers in our study completed 9 out of 10 activities with their children in a week (Table 1). Female caregivers completed more activities with children when compared to men in our sample (Table 2, standardized index of positive caregiver-child interactions). We also measure the quality of caregiver-child interactions through three additional measures. First, wemeasure tolerance towards violent parenting. Recall that challenging behaviors associated with violent parenting were also an important component of the intervention. We would expect that such norms of tolerance of punishment as a tool of parenting to be lower after parents learn of possible alternatives. We measure tolerance towards violence parenting practices using responses from both direct questions and vignettes to attenuate potential social desirability bias in the responses. 26 Second, we measure observed child behavior using the internalizing/externalizing behaviors sections of the parent/caregiver report survey developed by the World Bank. Third, we measure violence perpetration by a caregiver. This measure captures both physical and psychological violence towards children. Addressing this form of abuse and maltreatment was a main priority of the intervention and as such we would expect that the main perpetrator to be directly impacted by the intervention. To measure this outcome, we use a shortened version of the International Society for Prevention of Child Abuse and Neglect’s screening tool for caregivers (ICAST-P) (Meinck et al., 2020). We also combine this measure with an elicitation of perpetration of violence through vignettes. We create an index of the responses of perpetration of violence using the responses of both tools following the literature on violence (Cluver et al., 2018; Aguero and Frisancho, 2021). Finally, we measure child socioemotional development as reported by the caregiver using items from the parent/caregiverreport survey developed by the World Bank.27 Sociodemographic information and additional covariates: In addition to the outcome measures described above, we collected information on caregivers’ education, employment status (before and during the lockdown), marital status, household size and structure (including whether the caregiver lived with a partneror not), time spent with the children, and number of children in their care. We also gathered information on economic deprivation by asking if, during the pandemic, the household did not have enough money for food, health and education, home services, and clothes and gifts. At endline, we collected information on individual-level propensity to misreport sensitive items (social desirability index), following Crowne and 25The 10 items are listed in section A1.3 in the Appendix. 26Section A1.4 in the Appendix describes the instruments used in detail. 27We provide more details of the child behavior instrument in the appendix section A1.5, violence perpetration instrument in appendix section A1.6, and section A1.7 in the appendix describes the child socioemotional instrument used in this study. 13 Marlowe (1960).28 As shown in Table 1, on average, caregivers in our sample are 32 years old, are mostly female (60%), have a child enrolled in school, and reside in households with 4 people. Most caregivers (59%) in our sample do not have tertiary education. They exhibit a high level of economic deprivation (with 25% worrying about not having enough money for food). We also show that 29% of the sample was either unemployed or has lost their job during the pandemic. In terms of parenting, we find a high rate of violence and tolerance of it – 78% of caregivers use at least one form of physical or psychological punishment towards their children, and 29% tolerates of others using violence as a form of punishment towards children. At the same time, while parents use punitive parenting tools, this is combined with the use of other positive caregiver-child interactions and parenting techniques. 5 Empirical Strategy 5.1 Econometric model We rely on the random allocation of participants to treatment or control groups to identify the causal effect of the intervention on the set of outcomes described above. The main identification assumption is that, had there been no intervention, our outcomes of interest would be, on average, statistically equal between caregivers assigned to the treatment and control groups. The intervention’s intention-to-treat (ITT) effect is estimated as the difference in the post-treatment means of the outcome variables between treatment and control individuals. Formally, we estimate the following linear regression model by Ordinary Least Squares (OLS) using information at the caregiver level i: = + ∑ + + (1) where Yi is an outcome variable. Di is an indicator variable that takes value 1 if the caregiver was assigned to the treatment group and zero otherwise. Xi is a vector of covariates measured at baseline, including the pre-intervention outcome value for those outcomes collected in our baseline survey. 29 Si corresponds to the strata fixed effects. is an idiosyncratic error term. Since the randomization occurred at the individual level and we do not expect clusters in our sample, we report heteroskedasticity-robust standard errors. δ is our coefficient of interest, which provides the estimate of the intervention’s intention- 28More details in section A1.8 in the Appendix. 29In our main specification, we include the following variables as covariates: age (in years), educational attainment (primary, secondary and tertiary), number of girls and boys at home cared for the respondent. Following McKenzie (2012), if a respondent has a missing value for a covariate, we impute this value with the respective mean of the variable. Missing values never exceed 5% of the sample. 14 to-treat impact on the outcome (Yi).30 We also assess how the intervention’s impact varied by the gender of caregivers. To estimate the heterogeneous impact of the intervention, we estimate the following model: = 1 + 2 × + ∑ + + (2) where Malei is an indicator variable taking value one if the caregiver is a male and zero if female. We separately estimate this equation for two other different variables: an indicator of extreme economic deprivation, and an indicator variable taking value of one for caregivers living together with a partner. In specification 2, δ1 delivers the ITT estimate for female caregivers, and δ2 provides the difference in the ITT effect between sexes. The ITT estimate for men is given by the sum of δ1 and δ2. The interpretation of the coefficients is analogous for the other two indicator variables. We estimate equation 2 by OLS, re- port heteroskedasticity-robust standard errors and we also present randomization inference p-values as an additional inference test. 5.2 Balance on covariates and pre-intervention outcomes To support the validity of our identification assumption, we compare caregivers in the treatment and control groups in terms of sociodemographic characteristics and baseline levels of the outcome variables. Table 1 reports the mean value for these variables for caregivers in the treatment group (column 6) and caregivers in the control group (column 7), and the p-value of the difference between the two means (columns 8). There are almost no differences across groups in terms of caregiver, child, and household characteristics. The only significant differences we detect at baseline pertain to time caregiver spends with the child, which is larger in the treatment group but does not make an economically meaningful difference, and to economic deprivation, which is greater in the control group. In the case of the baseline level of outcome variables, we found no significant differences in all our primary outcomes (mental health —overall, stress, anxiety, and depression—impulsiveness, and positive caregiver-child interactions). We also found no significant differences in tolerance of violent parenting and physical violence. We do observe differences that are significant at conventional levels for the indexes of positive parenting practices, which is larger in the treatment group; negative parenting practices, more negative in the treatment group; and psychological violence perpetration, lower in the treatment group. These differences are also not economically meaningful as can be seen by the absolute number of items in each question. In Figure A3 in Appendix, we also present kernel density estimations and show 30We do not observe the intervention take-up at the individual level for the full-sample. Therefore, we estimate an ITT. In Appendix section - Table A23, we present an alternative estimation using a TOT analysis which we discuss in the robustness section. 15 that the main outcomes are balanced across treatment arms at baseline. Importantly, as we discuss in the empirical strategy section, whenever possible, we include the baseline level of the dependent variable in all regression estimates. In Table 2, we show that indeed male and female caregivers are substantially different – see columns 1 to 3. Male caregivers in comparison to female are older, are more likely to be employed, spend less time with children, and have better mental health. When it comes to parenting, males are less likely to tolerate violence but are less likely to perpetrate it. We show that 79 percent of females and 76 percent of males useat least one tool of violence. This parenting context is consistent with the literature discussed in Section 1 and highlights the importance of our design – of gender stratification. We also show –in columns 4 to 9 of Table 2–that the observed characteristics of caregivers in treatment and control at baseline are balanced within gender. 5.3 Endline survey attrition We collected information at endline from 2,280 individuals of the 3,103 original caregivers included in the study. We examine if differential attrition confounds our estimates and report results in Table A3. In column (1), we examine whether our treatment variable can predict the probability that the endline survey is missing for a given caregiver. We observe that treatment status is not significant, which indicates that participation in the program did not affect the probability of completing the endline survey. To further analyze the issue of differential attrition, we regress an indicator variable taking a value of one if the endline survey is missing on the treatment status and the treatment status interacted with our key baseline covariates and outcome variables. Column (2) in Table A3 shows the results of this regression. The objective of this regression is to assess if baseline characteristics differentially explain the decision to complete the endline survey across groups. We find that the p-value for the joint significance test for all interactions is 0.947, suggesting no differential attrition (in terms of baseline characteristics) between the treatment and control groups. 5.4 Power calculation and minimum detectable effect sizes Our final sample of caregivers who answered both the baseline and endline surveys comprises 2,280 individuals. Most of our outcome variables are based on standardized measures as observed in our final sample following Anderson (2008). Based on this sample size and using a level of significance (α) of 0.05 and conventional power (β) level of 0.8, we estimate that the minimum detectable effect (MDE) size of our survey design ranges from 0.083 to 0.114 sd.31 31We calculate this by following Raudenbush and Liu’s (2000) guidelines on nondichotomic outcomes where MDEs depend on the value used for the variation explained by covariates (R2) 0.05 (small) and 0.50 16 6 Results This section presents the results of the estimation of our available measures of take-up and the intervention’s impact on our outcomes of interest. It also assesses the extent to which this impact varies by sex.32 6.1 Compliance: Opening of SMS/WhatsApp and Viewership of Digital Content For the intervention to be effective we need to first understand its take-up rate. We identify two necessary conditions that need to be met: i) that participants open the SMS/WhatsApp messages, and ii) that thedifferent digital content in each message is consumed.33 We assess this in two ways. First, we show in Figure A1 in Appendix that the rate of message viewership by caregivers assigned to the treatment groupis high. Overall, we observe that 72% of caregivers in the treatment group opened the SMS/WhatsAppmessages sent to their mobile phones.34 We also show in Figure A1 that the trend in the viewership rate remained stable for every SMS/WhatsApp message sent. Second, using the knowledge incorporation surveys, we test our hypothesis that caregivers in the treat- ment group were more likely to report knowing about techniques they could only have learned through the intensity of participation. To test this hypothesis, we estimate the following: Recalli = β0 + δDti + Xi + γw(i) + (3) where Recalliw is the dependent variable of interest measured for every participant i responding to survey wave w. We have four outcomes of interest: the share of stress-management techniques known by respondents in the past week, the share of positive parenting techniques know by respondents in the past week,35. Di is a dummy variable indicating that the participant is in the treatment group and δ is the OLS estimator. To improve the precision of the estimates, we also control for socioeconomic variables measuredat baseline (Xi). These variables include age, gender, education level, and the number of female and male dependent children. We (large), respectively. 32This section follows the analysis proposed in our PAP. 33This is essential since it is possible that individuals never open SMS messages or they may even block the sender. 34Caregivers in the control group did not receive any message so this rate is always zero. 35We generate the share of techniques known by round from categorical variables of the usage of each technique – a dummy variable taking the value one if respondents say ”Yes” and zero otherwise. This share was created by summing up the total techniques that respondents reported knowing and dividing this number by the total techniques that were asked about in a survey round regarding stress managementand positive parenting. 17 also include as a control variable the number of times an individual responded to a monitoring survey. To account for the fact that each of the four survey rounds was conducted with different questions, we also account for survey-round fixed effects (γw(i)). corresponds to error term. We impose robust standard errors. We present the summary statistics of the survey in Table A4. On average, participants report using 50% and 85% of the stress-management and parenting techniques, respectively. Table A5 in Appendix shows the results from Equation 3. We find that individuals in the treatment group were 20.6 percentage points more likely to report using stress-management techniques (column (1)) and 5.7% more likely to use positive parenting techniques (column (3)). The effect sizes of the estimates described are not impacted by the inclusion of controls (as expected) given the randomized nature of the intervention. Overall, during the intervention, we find that caregivers randomly assigned to the treatment group are more likely to incorporate knowledge of the digital intervention compared to those in the control group. This result, along with the evidence displayed in Figure A1 in appendix, shows that the intervention had a first-stage impact on information assimilation and adoption of content. 6.2 Impact of the program on caregiver’s mental health and caregiver- child interactions We first estimate the intervention’s impact on the primary outcomes that the program targeted most directly: mental distress (linked to anxiety, stress, and depression); impulsiveness; and positive caregiver-child interactions. Table 3 shows these results. Contrary to our hypothesis, we find that the intervention increased mental distress. On average, treated caregivers reported worse mental health than those in the control group. The difference is equal to 0.057 sd and is statistically significant at the 10% level. This detrimental effect on mental health is mainly driven by a 0.072 sd impact on stress (statistically significant at the 5% level). We find no statistically significant effects on anxiety and depression (the two other components of the aggregate mental health measure), but the coefficients are also positive. This points to a relatively worse status among treated caregivers. We find no significant impact on impulsiveness and positive caregiver-child interactions. We next assess whether the intervention’s impact differs by caregiver gender. Table 4 shows these results. We document two main findings. First, we find that the intervention’s negative impact on mental health is mainly explained by male caregivers. The program led to an increase of 0.093 sd in treated male caregivers’ mental distress relative to those in the control group. This effect is driven by a 0.108 sd relative increase in 18 stress and a 0.095 sd relative increase in anxiety.36 Yet, we find that the intervention had no impacton mental health among female caregivers. Although differences in treatment effects are not statistically different across genders, estimates of the intervention’s impact among women are small—always smaller in absolute value than those for men—and not statistically different from zero. Second, we find that the intervention had differential effects by sex on positive caregiver-child interactions.While we do not observe any significant effect on caregiver-child interactions among female caregivers, treated men reported 0.137 sd fewer positive interactions with their children than those in the control group. The difference in treatment effects across genders is equal to 0.154 sd. Overall, these results show the intervention had an unintended impact on mental distress and caregiver- child interactions, which was concentrated among male caregivers. The intervention had no significant effect on primary outcomes among female caregivers. 6.3 Impact of the program on caregiver use of violence, attitudes toward child maltreatment, and observed child behavior This intervention sought to improve norms regarding tolerance of violent parenting practices and perpetration of violence against children, and to foster healthier child behavior. These expected effects relied on thefact that poor mental health is a major trigger for violence (Spencer et al., 2019). Therefore, by improving the awareness about mental distress and how to address it, reducing impulsiveness (primary outcomes), and providing information about positive parenting techniques, the intervention could change attitudes about child maltreatment and promote the use of alternative behavior (secondary outcomes). This change in parenting quality outcomes could have taken place on its own - since the intervention targets it directly - but also, as part of a chain of alterations that would take place as a result of changes in mental health. Table 5 reports the results of the estimation of the intervention’s impact on violence and other child specific outcomes that reflect additional domains of parental quality. First, Table 5 reports these results for violence perpetration, norms and observed child behavior. On average, we find no statistically significant effects on any of these outcomes. Similarly, we show that the intervention had no significant effect on caregiver attitudes about corporal punishment and children’s observed positive behavior. The absence of significant effects on these outcomes might result from a combination of the possible negative effects ofthe unintended impact on mental health and the positive effects of increased information or awarenessabout positive parenting practices. Overall, the intervention does not appear to have been strong enough to change caregiver norms regarding violence against children or child behavior. However, as we present in Table 6, there is an interesting result when considering the intervention’s 36Table A6 reports these results. 19 impact among male and female caregivers separately. We find a 0.098 sd decrease in the use of physical violence against children among treated female caregivers relative to women in the control group. 37 This impact is particularly relevant in the Salvadoran context, where female caregivers spend more time with the children and tend to exert more violence against their children, as we show in Table 2 in the Appendix. 7 Robustness Checks This section addresses some potential concerns with the results presented in Section 6. First, the use of self-reported measures to assess mental health status (as opposed to, for example, interviews or clinical assessments) might confound changes in mental distress with changes in respondents’ awareness of their mental health status. Therefore, the intervention’s estimated negative impact on mental health may simply reflect that it increased caregiver awareness of mental distress. Although it is not possible to fully ruleout this alternative explanation, our choice of a widely validated survey instrument —namely, the DASS- 21— that asks about specific physiological responses (such as mouth dryness and trembling) and not aboutself- perceptions of stress, anxiety, or depression should attenuate this potential concern (Lovibond and Lovi- bond, 1996). 38 We therefore believe that the observed results on mental health likely at least partially reflect actual changes in distress. Second, the use of self-reported measures can also be problematic when assessing attitudes and perpetration of violence. In sensitive topics like this one, responses might be affected by social desirability bias.This is especially the case for interventions that target those outcomes.39 To address this potential concern, we use two complementary approaches. First, we complement direct questions with vignettes to indirectly elicit respondents’ views regarding the use of violent parenting practices. 40 Second, following Asadullah et al. (2021) and Aguero and Frisancho (2021), we test the robustness of all our results to the inclusion of a social desirability index that captures individual-level propensity to misreport sensitive items as an additional control variable. All results remain similar in magnitude and statistical significance (Tables A10– A13 in the Appendix). It is also worth noticing that the treatment itself did not impact social desirability - seeTable A14 in Appendix. 37As we show in the robustness section, this effect is highly robust to various specifications, the inclusion of controls, and after controlling for a social desirability bias index. This result persists when considering alternative measures of physical violence perpetration, such as the probability of reporting physical abuse (- 0.038 sd) and the acceptance of hitting with the knuckle or back of the hand (-0.039 sd), as shown in Table A9. 38The American Psychiatric Association also suggests that to diagnose anxiety and depression symptoms, specialists should include physiological symptoms instead of psychological signs. 39See the meta-analysis conducted by Sugarman and Hotaling (1997). The authors find a low-to- moderate negative association between social desirability bias and intimate-partner violence (IPV). Also, reports ofIPV perpetration were more strongly correlated with social desirability scores than were reports of beingvictimized. 40For a detailed description of each vignette and the index construction, see the outcomes list and descrip- tion in Appendix A1. 20 Third, in the PAP, we pre-specified the inclusion of some control variables in the regression model to increase the precision of our estimations. We show that the exclusion of these covariates does not change our results. As presented in Tables A15–A18 in the Appendix, all results of the estimations remain similar in magnitude and statistical significance after excluding the control variables. Fourth, our findings related to heterogeneity by gender remain unchanged when we conduct the analysis by sample split – these results are presented in Tables A19 and A20 in Appendix. Next, As mentioned before, the statistical significance of our results is also unchanged when we calculate p-values through randomization inference – these results are shown in Tables A19 and A20 in Appendix with the respective p-values shown in squared brackets. Finally, using the smaller sample of individuals who responded to the KIs (N=659), we conduct a treated- on-the-treated (TOT) analysis and we show that our results remain stable – see Table A23 in Appendix. Due to the small sample we can no longer detect significant effects, however, the fact that the first-stage shows a strong take-up effect, and that the direction of effects is consistent is reassuring.41 8 Mechanisms This section explores two mechanisms that could explain the detrimental effects of the intervention on caregivers’ mental health and interactions with their children, and the heterogeneous results by caregiver gender: economic deprivation and interactions between partners.42 8.1 Limited attention and economic deprivation The pandemic affected households’ economic security and created a highly stressful environment (Robillard et al., 2020; Salari et al., 2020; Lakhan et al., 2020). One of the largest comparative studies to examine the pandemic’s economic impact found that, in nine developing countries, households reported declines in employment and income in all settings; the median share of households experiencing an income drop was 41Since we did not want to conduct the KIs with a larger sample – to avoid fatigue of the respondents – we did not contemplate a TOT analysis in the pre-analysis since we could not detect effects given the sample size of the KIs. Nonetheless, we display the TOT analysis results in the Appendix. 42The analysis below was not pre-specified in the PAP but it is deemed essential to understanding the main effects detected. We explored alternative hypotheses, but we did not find consistent evidence of potential alternative mechanisms that could explain our results. Alternative explanations included the total number of children, sex ratio of children residing in the household, age of the child, etc. None of these could explain our results. 21 68% (Egger et al., 2021).43 Economic deprivation can limit cognitive space and bandwidth (Mullainathan and Shafir, 2013; Mani et al., 2013; Haushofer and Fehr, 2014; Schilbach et al., 2016; Ridley et al., 2020). In this extremely challenging context, the addition of even a relatively minor task –such as reading the intervention program’s messages– might have increased stress. Although participants could drop out of the intervention at any moment or simply not open the messages, their interest in the material or their desire to improve parenting practices might have prevented them from doing so —thus creating an additional cognitive burden. We assess the empirical importance of this mechanism by examining if the intervention’s impact varied depending on the extent of caregivers’ economic deprivation. To do so, we construct a dummy variable indicating whether at baseline the respondent replied that, during the pandemic, the household did not have enough money for food, health and education, home services, and clothes and gifts.44 In our regression model, we interact this dummy variable with the treatment indicator to estimate the difference in the treatment effect between those with the highest level of economic deprivation and the rest (as shown in Equation 2). Table 7 reports the results. We observe that the treatment led to a significantly higher increase in mental distress among those who reported the highest level of economic deprivation. The difference is large (0.248 sd) and statistically significant (at the 1% level). While the intervention had no effect on mental distress among the least deprived (the coefficient is actually negative, but not statistically different from zero), it ledto a 0.217 sd increase for those most deprived (significant at the 1% level). 8.2 Interactions between partners The presence of a partner in the house might also have moderated the intervention’s impact on our out- comes of interest. Partner quality is an important determinant of caregiver mental health (Bendini andDinarte, 2020; Pico-Alfonso, 2005; Taylor et al., 2009), and it influences the relationship between a caregiver and their children. For example, Bendini and Dinarte (2020) show that the presence of a partner in thehouse (heavy drinkers, particularly) worsens the effect of maternal mental health on child development. The intervention’s focus on positive parenting, which stresses the importance of sharing quality time with children and being aware of children’s emotions, might have changed preferences or expectations about the partner’s 43Moreover, these authors document that household coping strategies and government responses were generally insufficient, leading to generalized food insecurity and deteriorating economic conditions eventhree months into the crisis. 44Table 1 shows descriptive statistics for each of the questions considered to create this indicator variable. At baseline, caregivers were asked if, during the pandemic, they had enough money for different household expenses. Among this group, 75% responded that they did not have enough money for food, 66% indicated not having enough money for health and education, 38% reported not having enough money for basic services, and 90% said they lacked money for clothes and gifts. Thirty percent of households indicated not having enough money for all four categories. 22 role in child-rearing and the distribution of parenting duties. If treated caregivers were unable to communicate these revised expectations effectively, the intervention could have created a wedge between partners, increased negative interactions within the household, and affected mental health. In light of this hypothesis, we assess if the intervention’s impact varied depending on whether the treated caregivers lived with a partner or not. Table 8 presents the results of this analysis. We find that the intervention’s impact on caregiver-child interactions was significantly different for caregivers who lived with a partner and those who did not. The difference in the treatment effect was -0.167 sd (statistically significant at the 10% level). Although the intervention did not have a clear impact on caregiver-child interactions among caregivers who did not live with a partner, it significantly reduced caregiver-child interactions among those who lived with one (-0.091 sd, significant at the 5% level). Moreover, we also find that the negative impact on caregiver mental health was mainly concentrated among caregivers living with a partner. Among these caregivers, the intervention increased mental distress by 0.094 sd (statistically significant at the 5% level). We cannot directly assess the mechanism that explains these heterogeneous effects, but the results emphasize the importance of the presence of a partner in terms of the intervention’s impact, and they align with the hypothesis of an increase in negative interactions within the household due to changes in caregivers’ expectations about parenting roles and the distribution of duties. 8.3 What can explain the differential effects by sex? The analysis presented in Section 6 shows that the intervention’s impacts on mental distress, positive caregiver-child interactions and violence were mostly driven by the impact among male caregivers. Subsections 8.1 and 8.2 show that the extent of economic deprivation and cohabitation with a partner also moderate the effect of the intervention. We now assess how the differential effect across genders relates to these two channels. First, we assess how the intervention’s impact differed across genders and levels of economic depr ivation (at baseline) by estimating a triple-difference regression model. We interact the treatment status dummy with a variable indicating if the caregiver is a male and with an indicator of economic deprivation,and add a triple interaction term (treatment-male-economic deprivation). This specification allows us to compare the intervention’s impact across four different groups of caregivers: men in the most economically deprived households, women in the most economically deprived households, men in less economically deprived households, and women in less economically deprived households. Figure 2 presents a summary ofthe results and Table A21 shows all estimated coefficients. Considering the intervention’s impact on mental distress, results show that among both men and women, the intervention’s impact was worse for caregivers in the most economically deprived households. The difference in the intervention’s impact in the most economically deprived households versus the other households is 0.158 sd (not statistically significant at standard levels) for women and 0.388 sd (significant at the 1% level) for men. The detrimental impact of 23 the intervention on mental health was particularly observed among men in the most economically deprived households. When considering households with lower levels of economic deprivation, the intervention had no significant impact on the mental health of either male or female caregivers. The estimated impact is -0.032 sd (that is, suggestive of lower distress among treated caregivers) for both men and women. However, in these households, the intervention led to significantly fewer positive caregiver-child interactions, especially for male caregivers. Then, we assess how the differential impact of the intervention varied depending on whether the caregiver was living with a partner or not. We estimate an analogous triple-difference model, replacing the economic deprivation indicator variable with a dummy indicating if the caregiver lived with a partner. This specification allows us to compare the intervention’s impact across four different caregiver groups: menliving with a partner, women living with a partner, men not living with a partner, and women not livingwith a partner. Figure 3 presents a summary of the results and Table A22 shows all estimated coefficients. Results show that, for female caregivers, the intervention’s impact was very similar for those who livedwith a partner and those who did not. Instead, for male caregivers, results differ greatly depending on whether they lived with a partner or not. The intervention led to a 0.157 sd increase (significant at the 10% level) in mental distress among men living with a partner and to a 0.21 sd decrease (significant at the 10% level) among men not living with a partner. The difference between treatment effects is 0.368 sd (significant at the 1% level). We also observe that the intervention led to significantly fewer positive caregiver-child interactions only among men who lived with a partner (-0.177 sd, significant at the 10% level). Overall, the results show that characteristics such as caregiver gender, household economic situation, and cohabitation with a partner shaped the impact of the intervention. We observe that the degree to which economic and family circumstances moderate the intervention’s impact depends on caregiver gender. The intervention’s impact on male caregivers varies greatly depending both on the extent of economic deprivation and whether they live with a partner, even leading to opposite effects. In women, the differences in the intervention’s impact among different groups (based on economic deprivation and cohabitation with apartner) are less pronounced. 9 Discussion and Concluding Remarks Parenting is challenging. Much of the learning about parenting styles and decisions occurs informally through social and family ties, which can perpetuate suboptimal and violent parenting practices, especially in violent contexts. The use of these improper practices can have broad negative effects on child development, as emphasized by a growing literature on the importance of parental investments (Cunha and Heckman, 2007). This issue has become a central focus of development policy. Decision-making on this issue requires a better understanding of the effects of early-childhood and parenting interventions that seek 24 to improve the quality of parental interactions (Britto et al., 2017; Doepke and Zilibotti, 2017, 2019). We study the impact of a program to promote positive parenting and parental stress management de- livered remotely via SMS/WhatsApp messages to caregivers in El Salvador, and containing a variety of digital content akin to that of other successful interventions (Cluver et al., 2018; Carneiro et al., 2019). This mode of delivery offers an attractive option to improve the well-being of caregivers and children given the widespread availability of mobile phones (even in low-income settings). This is an attractive alternative in a severely constrained setting such as the COVID-19 lockdown and in non-crisis periods, because of its scalability and low implementation costs. The program was developed by an NGO with vast experience administering mental health programs in El Salvador and Central America. Contrary to what we expected, we find that, on average, the program increased the mental distress of caregivers. In spite of that, we find no evidence that this deterioration translated to their children or their relationship with them. We document no significant impact on caregiver-child interactions or child maltreat- ment, overall. However, we find that the effect of the program differed substantially by sex. We find that the increase in mental distress was concentrated among male caregivers, and observe no significant change among females. Furthermore, we find that female caregivers lowered their use of physical violence toward children. Finally, we document that family structure and the extent of economic deprivation greatly moderated the impact of the intervention, especially among male caregivers. We interpret our average results on mental health as a backlash reaction in the context of limited attention and disrupted family dynamics. As observed in the literature, familial and cultural context can greatly moderate the impact of shocks and interventions, and lead to unintended consequences.45 In our case, the backlash reaction is associated with caregiver gender, family structure, and economic deprivation, consistent with prevailing theories linking the latter to caregivers’ cognitive overload and mental health (Mullainathan and Shafir, 2013; Mani et al., 2013; Asadullah et al., 2021; Schilbach et al., 2016; Ridley et al., 2020). This result provides relevant insights for future research and policy interventions. There is a major evidence gap in knowledge about parenting policies to address child maltreatment (Pundir et al., 2020). We show that addressing this problem with a one-size-fits-all program can have unintended consequences, but these can largely be avoided by tailoring the program to the particular context and the specific needs of the beneficiary group. Caregiver characteristics can substantially shape the effect of the intervention, so they should be carefully considered in the program design. 45See, for example, Dahl et al. (2020) and Tur-Prats (2021). 25 References Aguero, J. and Frisancho, V. (2021). Measuring violence against women with experimental methods. Eco- nomic Development and Cultural Change. Altindag, O., Erten, B., and Keskin, P. (2020). Mental health costs of lockdowns: Evidence from age-specific curfews in Turkey. Available at SSRN 3765838. Anderson, M. L. (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. Asadullah, M. N., De Cao, E., Khatoon, F. Z., and Siddique, Z. (2021). Measuring gender attitudes using list experiments. Journal of Population Economics, 34(2):367–400. Attanasio, O., Cattan, S., Fitzsimons, E., Meghir, C., and Rubio-Codina, M. (2020a). Estimating the pro- duction function for human capital: Results from a randomized controlled trial in Colombia. American Economic Review, 110(1):48–85. Attanasio, O., Meghir, C., and Nix, E. (2020b). Human capital development and parental investment in India. The Review of Economic Studies, 87(6):2511–2541. Baranov, V., Bhalotra, S., Biroli, P., and Maselko, J. (2020). Maternal depression, women’s empowerment, and parental investment: Evidence from a randomized controlled trial. American Economic Review, 110(3):824– 59. Barrera, O., Macours, K., Premand, P., and Vakis, R. (2020). Texting parents about early child development. World Bank Policy Research Working Paper, (9492). Bartel, A. P., Rossin-Slater, M., Ruhm, C. J., Stearns, J., and Waldfogel, J. (2018). Paid family leave, fathers’ leave-taking, and leave-sharing in dual-earner households. Journal of Policy Analysis and Management , 37(1):10–37. Bendini, M. and Dinarte, L. (2020). Does maternal depression undermine childhood cognitive development? Evidence from the Young Lives Survey in Peru. International Journal of Environmental Research and Public Health, 17(19):7248. Berlinski, S., Busso, M., Dinkelman, T., and Mart´ınez, C. (2021). Reducing parent -school information gaps and improving education outcomes: Evidence from high-frequency text messages. Technical report, National Bureau of Economic Research. 26 Britto, P. R., Lye, S. J., Proulx, K., Yousafzai, A. K., Matthews, S. G., Vaivada, T., Perez-Escamilla, R., Rao, N., Ip, P., Fernald, L. C., et al. (2017). Nurturing care: Promoting early childhood development. The Lancet, 389(10064):91–102. Burke, T., Berry, A., Taylor, L. K., Stafford, O., Murphy, E., Shevlin, M., McHugh, L., and Carr, A. (2020). Increased psychological distress during Covid-19 and quarantine in Ireland: A national survey. Journal of Clinical Medicine, 9(11):3481. Campion, J., Javed, A., Sartorius, N., and Marmot, M. (2020). Addressing the public mental health challenge of Covid-19. The Lancet Psychiatry, 7(8):657–659. Carneiro, P., Kraftman, L., Mason, G., Moore, L., Rasul, I., and Scott, M. (2021). The impacts of a multifaceted prenatal intervention on human capital accumulation in early life. American Economic Review, 111(8):2506– 49. Carneiro, P. M., Galasso, E., Lopez Garcia, I. X., Bedregal, P., and Cordero, M. (2019). Parental beliefs, investments, and child development: Evidence from a large-scale experiment. World Bank Policy Research Working Paper, (8743). Cluver, L., Lachman, J. M., Sherr, L., Wessels, I., Krug, E., Rakotomalala, S., Blight, S., Hillis, S., Bachmand, G., Green, O., et al. (2020). Parenting in a time of Covid-19. Lancet, 395(10231). Cluver, L. D., Meinck, F., Steinert, J. I., Shenderovich, Y., Doubt, J., Romero, R. H., Lombard, C. J., Redfern, A., Ward, C. L., Tsoanyane, S., et al. (2018). Parenting for lifelong health: A pragmatic cluster randomised controlled trial of a non-commercialised parenting programme for adolescents and their families in South Africa. BMJ Global Health, 3(1):1–15. Crawfurd, L., Evans, D. K., Hares, S., and Sandefur, J. (2021). Teaching and testing by phone in a pandemic. Crowne, D. P. and Marlowe, D. (1960). A new scale of social desirability independent of psychopathology. Journal of Consulting Psychology, 24(4):349. Cunha, F. and Heckman, J. (2007). The technology of skill formation. American Economic Review, 97(2):31–47. Dahl, G. B., Felfe, C., Frijters, P., and Rainer, H. (2020). Caught between cultures: Unintended consequences of improving opportunity for immigrant girls. Technical report, National Bureau of Economic Research. Dhar, D., Jain, T., and Jayachandran, S. (2018). Reshaping adolescents’ gender attitudes: Evidence from a school-based experiment in india. Technical report, National Bureau of Economic Research. Dinarte, L., Egana, P., and Martinez, C. (2021). Non-cognitive skills development and school-based violence reduction in Central America. Unpublished Manuscript. 27 Dinarte, L. and Egana-delSol, P. (2019). Preventing violence in the most violent contexts: Behavioral and neurophysiological evidence. World Bank Policy Research Working Paper Series. Doepke, M., Sorrenti, G., and Zilibotti, F. (2019). The economics of parenting. Annual Review of Economics, 11:55–84. Doepke, M. and Zilibotti, F. (2017). Parenting with style: Altruism and paternalism in intergenerational preference transmission. Econometrica, 85(5):1331–1371. Doepke, M. and Zilibotti, F. (2019). Love, money, and parenting: How economics explains the way we raise our kids. Princenton, New Jersey: Princeton University Press. Doyle, K., Levtov, R. G., Barker, G., Bastian, G. G., Bingenheimer, J. B., Kazimbaya, S., Nzabonimpa, A., Pulerwitz, J., Sayinzoga, F., Sharma, V., et al. (2018). Gender-transformative bandebereho couples’ inter- vention to promote male engagement in reproductive and maternal health and violence prevention in Rwanda: Findings from a randomized controlled trial. PLOS One, 13(4):e0192756. Doyle Jr, J. J. and Aizer, A. (2018). Economics of child protection: Maltreatment, foster care, and intimate partner violence. Annual review of economics, 10:87–108. Durrant, J. (2013). Disciplina positiva en la crianza cotidiana. Sweden: Save the children. Egger, D., Miguel, E., Warren, S. S., Shenoy, A., Collins, E., Karlan, D., Parkerson, D., Mobarak, A. M., Fink, G., Udry, C., et al. (2021). Falling living standards during the Covid-19 crisis: Quantitative evidence from nine developing countries. Science Advances, 7(6):eabe0997. Ekberg, J., Eriksson, R., and Friebel, G. (2013). Parental leave-A policy evaluation of the Swedish ‘Daddy - Month’ reform. Journal of Public Economics, 97:131–143. ´ , L. and Gonza Farre ´ lez, L. (2019). Does paternity leave reduce fertility? Journal of Public Economics, 172:52–66. Ferna ´ . R., Gar´ ´ ndez, M. A ıa, M. I. D., and Crespo, A. V. (2012). Manual de te ´ n cognitivo ´cnicas de intervencio ´ e de Brouwer. conductuales. Bilbao: Descle Glasswing (2020). Design and implementation report of a parenting program and stress coping strategies for caregivers in El Salvador. Technical (internal) report. ´ n moderada, sensibilidad emocional y afrontamiento en el contexto del estre Guarino, L. (2013). Mediacio ´ s. ´ gico, 11(1):141–159. Pensamiento Psicolo Hamby, S., Finkelhor, D., Turner, H., and Ormrod, R. (2011). Children’s Exposure to Intimate Partner Violence and Other Family Violence. Washington, DC: U.S. Department of Justice. 28 Haushofer, J. and Fehr, E. (2014). On the psychology of poverty. Science, 344(6186):862–867. Health, T. L. D. (2021). Digital tools for mental health in a crisis. Lancet Digit Health, 3(4):e204. Heller, S. B., Shah, A. K., Guryan, J., Ludwig, J., Mullainathan, S., and Pollack, H. A. (2017). Thinking, fast and slow? Some field experiments to reduce crime and dropout in Chicago. The Quarterly Journal of Economics, 132(1):1–54. Holman, D., Johnson, S., and O’Connor, E. (2018). Stress management interventions: Improving subjective psychological well-being in the workplace. Handbook of Well-Being. Salt Lake City, UT: DEF Publishers. DOI: nobascholar. com. Humphreys, K. L., Myint, M. T., and Zeanah, C. H. (2020). Increased risk for family violence during the Covid- 19 pandemic. Pediatrics, 145(4):e20200982. Hupkau, C. and Petrongolo, B. (2020). Work, care and gender during the Covid-19 crisis. Fiscal Studies, 41(3):623–651. Justino, P., Leone, M., Rolla, P., Abimpaye, M., Dusabe, C., Uwamahoro, D., and Germond, R. (2020). Im- proving parenting practices for early child development: Experimental evidence from rwanda. WIDER Working Paper. Kariger, P., Frongillo, E. A., Engle, P., Britto, P. M. R., Sywulka, S. M., and Menon, P. (2012). Indicators of family care for development for use in multicountry surveys. Journal of Health, Population, and Nutrition, 30(4):472. Knerr, W., Gardner, F., and Cluver, L. (2013). Improving positive parenting skills and reducing harsh and abusive parenting in low-and middle-income countries: A systematic review. Prevention Science , 14(4):352–363. Kola, L. (2020). Global mental health and Covid-19. The Lancet Psychiatry, 7(8):655–657. Lakhan, R., Agrawal, A., and Sharma, M. (2020). Prevalence of depression, anxiety, and stress during Covid- 19 pandemic. Journal of Neurosciences in Rural Practice, 11(4):519. Lichand, G. and Christen, J. (2020). Using nudges to prevent student dropouts in the pandemic. arXivpreprint arXiv:2009.04767. Lindo, J. M., Schaller, J., and Hansen, B. (2018). Caution! men not at work: Gender-specific labor market conditions and child maltreatment. Journal of Public Economics, 163:77–98. Lovibond, S. H. and Lovibond, P. F. (1996). Manual for the depression anxiety stress scales. Psychology Foun- dation of Australia. 29 Mani, A., Mullainathan, S., Shafir, E., and Zhao, J. (2013). Poverty impedes cognitive function. Science, 341(6149):976–980. Maselko, J., Hagaman, A. K., Bates, L. M., Bhalotra, S., Biroli, P., Gallis, J. A., O’Donnell, K., Sikander, S., Turner, E. L., and Rahman, A. (2019). Father involvement in the first year of life: Associations with maternal mental health and child development outcomes in rural Pakistan. Social Science & Medicine, 237:112421. McKenzie, D. (2012). Beyond baseline and follow-up: The case for more t in experiments. Journal of Devel- opment Economics, 99(2):210–221. Meinck, F., Murray, A. L., Dunne, M. P., Schmidt, P., Nikolaidis, G., Petroulaki, K., Zarokosta, F., Tsirigoti, A., Hazizaj, A., Cenko, E., et al. (2020). Measuring violence against children: The adequacy of the Interna- tional Society for the Prevention of Child Abuse and Neglect (ISPCAN) child abuse screening tool-child version in 9 Balkan countries. Child Abuse & Neglect, 108:104636. Mullainathan, S. and Shafir, E. (2013). Scarcity: Why having too little means so much. New York: Macmillan. ¨ tzer, J., Patel, V., and Bartels, S. J. (2017). Naslund, J. A., Aschbrenner, K. A., Araya, R., Marsch, L. A., Unu Digital technology for treating and preventing mental disorders in low-income and middle-income coun- tries: A narrative review of the literature. The Lancet Psychiatry, 4(6):486–500. ˜ o, disciplina positiva. Barcelona: Medici. Nelsen, J. (2007). Como educar con firmeza y carin Offer, S. and Schneider, B. (2011). Revisiting the gender gap in time-use patterns: Multitasking and well- being among mothers and fathers in dual-earner families. American Sociological Review, 76(6):809–833. Olivetti, C. and Petrongolo, B. (2017). The economic consequences of family policies: lessons from a century of legislation in high-income countries. Journal of Economic Perspectives, 31(1):205–30. Patton, J. H., Stanford, M. S., and Barratt, E. S. (1995). Factor structure of the Barratt Impulsiveness Scale. Journal of Clinical Psychology, 51(6):768–774. Persson, P. and Rossin-Slater, M. (2018). Family ruptures, stress, and the mental health of the next genera- tion. American Economic Review, 108(4-5):1214–52. Persson, P. and Rossin-Slater, M. (2019). When dad can stay home: Fathers’ workplace flexibility and ma- ternal health. Technical report, National Bureau of Economic Research. Pico-Alfonso, M. A. (2005). Psychological intimate partner violence: The major predictor of posttraumatic stress disorder in abused women. Neuroscience & Biobehavioral Reviews, 29(1):181–193. 30 Pundir, P., Saran, A., White, H., Subrahmanian, R., and Adona, J. (2020). Interventions for reducing violence against children in low-and middle-income countries: An evidence and gap map. Campbell Systematic Reviews, 16(4):e1120. Raudenbush, S. W. and Liu, X. (2000). Statistical power and optimal design for multisite randomized trials. Psychological Methods, 5(2):199. Renzetti, C. M. (2009). Economic stress and domestic violence. CRVAW Faculty Research Reports and Papers. Ridley, M., Rao, G., Schilbach, F., and Patel, V. (2020). Poverty, depression, and anxiety: Causal evidence and mechanisms. Science, 370(6522). ` re, S. P. L., Quilty, L., Robillard, R., Saad, M., Edwards, J., Solomonova, E., Pennestri, M.-H., Daros, A., Veissie Dion, K., Nixon, A., et al. (2020). Social, financial and psychological stress during an emerging pandemic: Observations from a population survey in the acute phase of Covid-19. BMJ Open, 10(12):e043805. Rodriguez-Rey, R., Garrido-Hernansaiz, H., and Collado, S. (2020). Psychological impact and associated factors during the initial stage of the coronavirus (Covid-19) pandemic among the general population in Spain. Frontiers in Psychology, 11:1540. Rossin-Slater, M. (2017). Signing up new fathers: Do paternity establishment initiatives increase marriage, parental investment, and child well-being? American Economic Journal: Applied Economics, 9(2):93–130. Salari, N., Hosseinian-Far, A., Jalali, R., Vaisi-Raygani, A., Rasoulpoor, S., Mohammadi, M., Rasoulpoor, S., and Khaledi-Paveh, B. (2020). Prevalence of stress, anxiety, depression among the general population during the Covid-19 pandemic: A systematic review and meta-analysis. Globalization and Health, 16(1):1– 11. Schilbach, F., Schofield, H., and Mullainathan, S. (2016). The psychological lives of the poor. American Economic Review, 106(5):435–40. Spencer, C., Mallory, A. B., Cafferky, B. M., Kimmes, J. G., Beck, A. R., and Stith, S. M. (2019). Mental health factors and intimate partner violence perpetration and victimization: A meta-analysis. Psychology of Violence, 9(1):1. Sugarman, D. B. and Hotaling, G. T. (1997). Intimate violence and social desirability: A meta-analytic review. Journal of Interpersonal Violence, 12(2):275–290. Sviatschi, M. M. (2018). Making a narco: Childhood exposure to illegal labor markets and criminal life paths. Working Manuscript, Dept. Econ., Princeton Univ. https://rpds. Princeton. edu/sites/rpds/files/sviatschi making- a-narco march2018. pdf 31 Szabo, T. G., Richling, S., Embry, D. D., Biglan, A., and Wilson, K. G. (2020). From helpless to hero: Promoting values-based behavior and positive family interaction in the midst of Covid-19. Behavior Analysis in Practice. Taylor, C. A., Guterman, N. B., Lee, S. J., and Rathouz, P. J. (2009). Intimate partner violence, maternal stress, nativity, and risk for maternal maltreatment of young children. American Journal of Public Health, 99(1):175–183. Thompson, R. A. and Lagattuta, K. H. (2006). Feeling and understanding: Early emotional development. Blackwell handbook of early childhood development. Tur-Prats, A. (2021). Unemployment and intimate partner violence: A cultural approach. Journal of Economic Behavior & Organization, 185:27–49. UNICEF (2017). A familiar face: Violence in the lives of children and adolescents. New York, NY: UNICEF. UNICEF (2020). Policy brief: The impact of Covid-19 on children. New York: United Nations Children’s Fund. WHO (2020). Global status report on preventing violence against children 2020. World Health Organization. 32 Tables and Figures Table 1: Summary Statistics and Balance Tests Variable Obs. Mean Std Min Max Mean Mean P-val. Dev control treatment Difference (1) (2) (3) (4) (5) (6) (7) (8) Panel A. Caregiver characteristics Age (years) 3103 32.08 6.34 19 50 32.196 31.968 (0.302) Female (%) 3103 0.60 0.49 0 1 0.599 0.599 (0.994) Education level Basic (1-9 grades, %) 3103 0.20 0.40 0 1 0.205 0.205 (0.988) High school (10-12 grades, %) 3103 0.38 0.49 0 1 0.381 0.386 (0.766) Bachelor or higher (%) 3103 0.41 0.49 0 1 0.415 0.410 (0.756) Employment status pre and post pandemic Always unemployed (%) 3076 0.19 0.40 0 1 0.191 0.197 (0.625) Always employed (%) 3076 0.48 0.50 0 1 0.482 0.483 (0.963) Lost job (%) 3076 0.10 0.31 0 1 0.106 0.104 (0.887) Found job (%) 3076 0.22 0.41 0 1 0.221 0.215 (0.686) Girls under 9 years cared (N) 3054 0.76 0.66 0 5 0.752 0.765 (0.582) Boys under 9 years cared (N) 3054 0.77 0.65 0 4 0.782 0.756 (0.265) Time caregiver spends with the child (hours) 3103 9.64 3.64 0 12 9.803 9.480 (0.012) Living with a partner (%) 3103 0.70 0.46 0 1 0.699 0.696 (0.833) Panel B. Child characteristics Oldest child under 9 is female (%) 3103 0.50 0.50 0 1 0.492 0.503 (0.553) Child enrolled in school (%) 3103 0.66 0.48 0 1 0.659 0.652 (0.676) Child video/screen time for fun (hours) 3103 3.64 3.76 0 24 3.735 3.542 (0.151) Child video/screen time for homework (hours) 3103 3.00 3.45 0 24 3.007 2.986 (0.863) Panel C. Household characteristics Household size (N of members) 3102 4.15 1.97 1 20 4.158 4.148 (0.892) Intergenerational household (%) 3103 0.38 0.49 0 1 0.392 0.373 (0.284) Income vulnerability index 2953 1.31 1.18 0 4 1.278 1.337 (0.171) Money for food (%) 3057 0.25 0.44 0 1 0.247 0.260 (0.413) Money for health and education (%) 3041 0.34 0.47 0 1 0.339 0.344 (0.748) Money for home services (%) 3058 0.62 0.49 0 1 0.607 0.631 (0.181) Money for clothes and gifts (%) 3051 0.10 0.30 0 1 0.096 0.106 (0.367) Economic deprivation (%) 2953 0.33 0.47 0 1 0.343 0.307 (0.032) Panel D. Outcomes Mental health index, std. 3103 0.00 1.01 -2 4 0.000 -0.008 (0.817) Anxiety (high, %) 3103 0.54 0.50 0 1 0.539 0.547 (0.680) Stress (high, %) 3103 0.34 0.47 0 1 0.336 0.336 (0.999) Depression (high, %) 3103 0.34 0.47 0 1 0.339 0.336 (0.847) Impulsiveness index, std. 3103 0.00 1.02 -4 5 0.000 0.006 (0.871) Positive caregiver-child interactions index, std. 3103 -0.02 1.01 -5 1 0.000 -0.048 (0.182) Positive caregiver-child interactions (N) 3103 9.20 1.31 1 10 9.239 9.168 (0.131) Tolerance of violent parenting index, std. 2938 -0.03 0.99 -1 7 -0.027 -0.037 (0.794) Tolerate at least one item of maltreatment (%) 2991 0.29 0.45 0 1 0.282 0.289 (0.657) Positive parenting index, std. 3038 0.06 0.96 -6 1 0.025 0.102 (0.027) Positive practices (N) 3038 9.99 1.19 2 11 9.958 10.026 (0.110) Negative parenting index, std. 2925 -0.03 0.92 -1 9 0.000 -0.067 (0.049) Negative practices (N) 2925 2.32 2.22 0 12 2.369 2.276 (0.262) Physical violence index, std. 3053 0.00 0.98 -1 4 0.005 -0.005 (0.801) Psychological violence index, std. 2950 -0.03 0.92 -1 8 0.006 -0.069 (0.027) This table shows baseline characteristics of the sample and the balance of baseline measures, comparing control and treatment groups. It shows means of each group and p-value for the difference in means. In a test for differences in gender, we exclude strata fixed effects due to collinearity (gender is the stratification variable). Income vulnerability index is the sum of all subcategories. 33 Table 2: Differences in means by gender (baseline) Mean Mean P-val. Mean Mean P-val. Mean Mean P-val. Variable All Male All Female Diff Control Male Treated Male Diff Control Female Treated Female Diff (1) (2) (3) (4) (5) (6) (7) (8) (9) Panel A. Caregiver characteristics Age (years) 32.636 31.711 (0.000)*** 32.921 32.352 (0.111) 31.712 31.711 (0.997) Education level Basic (1-9 grades, %) 0.200 0.208 (0.000)*** 0.182 0.217 (0.122) 0.220 0.197 (0.218) High school (10-12 grades, %) 0.402 0.370 (0.033)** 0.422 0.383 (0.160) 0.353 0.388 (0.118) Bachelor or higher (%) 0.398 0.422 (0.000)*** 0.396 0.400 (0.882) 0.427 0.416 (0.599) Employment status pre and post pandemic Always unemployed (%) 0.081 0.270 (0.000)*** 0.083 0.079 (0.802) 0.262 0.277 (0.500) Always employed (%) 0.597 0.406 (0.000)*** 0.597 0.598 (0.994) 0.406 0.407 (0.956) Lost job (%) 0.080 0.121 (0.100)* 0.088 0.071 (0.274) 0.117 0.126 (0.570) Found job (%) 0.242 0.203 (0.533) 0.232 0.252 (0.399) 0.215 0.191 (0.204) Girls under 9 years cared (N) 0.748 0.766 (0.000)*** 0.742 0.755 (0.735) 0.759 0.772 (0.664) Boys under 9 years cared (N) 0.790 0.755 (0.361) 0.806 0.774 (0.396) 0.766 0.744 (0.455) Time caregiver spends with the child (hours) 8.865 10.160 (0.000)*** 9.055 8.675 (0.077)* 10.302 10.018 (0.076)* Living with a partner (%) 0.818 0.617 (0.000)*** 0.818 0.818 (1.000) 0.620 0.614 (0.802) Panel B. Child characteristics Oldest child under 9 is female (%) 0.471 0.515 (0.071)* 0.457 0.486 (0.319) 0.516 0.515 (0.962) Child enrolled in school (%) 0.641 0.665 (0.778) 0.643 0.640 (0.919) 0.670 0.660 (0.647) Child video/screen time for fun (hours) 3.798 3.532 (0.919) 3.944 3.653 (0.179) 3.596 3.467 (0.455) Child video/screen time for homework (hours) 3.208 2.855 (0.630) 3.122 3.294 (0.423) 2.930 2.781 (0.322) Panel C. Household characteristics Household size (N of members) 4.105 4.185 (0.130) 4.166 4.045 (0.265) 4.153 4.217 (0.493) Intergenerational household (%) 0.338 0.412 (0.022)** 0.354 0.322 (0.228) 0.417 0.407 (0.678) Income vulnerability index 1.343 1.284 (0.510) 1.315 1.371 (0.412) 1.253 1.314 (0.272) Money for food (%) 0.258 0.250 (0.087)* 0.259 0.257 (0.954) 0.239 0.262 (0.265) Money for health and education (%) 0.352 0.334 (0.031)** 0.348 0.357 (0.736) 0.333 0.336 (0.890) Money for home services (%) 0.623 0.617 (0.843) 0.608 0.637 (0.289) 0.607 0.626 (0.389) Money for clothes and gifts (%) 0.124 0.086 (0.591) 0.123 0.126 (0.862) 0.079 0.093 (0.279) Economic deprivation (%) 0.324 0.326 (0.878) 0.343 0.304 (0.140) 0.343 0.309 (0.119) Panel D. Outcomes Mental health index, std. -0.210 0.133 (0.000)*** -0.210 -0.210 (0.998) 0.140 0.126 (0.773) Anxiety (high, %) 0.468 0.593 (0.003)*** 0.465 0.471 (0.841) 0.589 0.597 (0.712) Stress (high, %) 0.254 0.391 (0.019)** 0.245 0.264 (0.442) 0.397 0.385 (0.574) Depression (high, %) 0.260 0.390 (0.000)*** 0.258 0.262 (0.860) 0.394 0.386 (0.712) Impulsiveness index, std. 0.032 -0.017 (0.198) -0.009 0.073 (0.144) 0.006 -0.039 (0.350) Positive caregiver-child interactions index, std. -0.155 0.063 (0.002)*** -0.160 -0.150 (0.877) 0.107 0.020 (0.046)** Positive caregiver-child interactions (N) 9.144 9.244 (0.987) 9.134 9.154 (0.788) 9.310 9.177 (0.024)** Tolerance of violent parenting index, std. 0.017 -0.067 (0.112) 0.034 0.001 (0.602) -0.070 -0.063 (0.908) Tolerate at least one item of maltreatment (%) 0.301 0.274 (0.066)* 0.300 0.302 (0.910) 0.269 0.280 (0.627) Positive parenting index, std. -0.007 0.111 (0.096)* -0.050 0.035 (0.138) 0.076 0.148 (0.101) Positive practices (N) 9.940 10.027 (0.274) 9.895 9.985 (0.205) 10.000 10.054 (0.313) Negative parenting index, std. -0.070 -0.008 (0.004)*** -0.054 -0.087 (0.553) 0.037 -0.054 (0.040)** Negative practices (N) 2.288 2.347 (0.026)** 2.315 2.260 (0.689) 2.406 2.287 (0.256) Physical violence index, std. 0.016 -0.011 (0.014)** 0.025 0.006 (0.754) -0.010 -0.012 (0.954) Psychological violence index, std. -0.068 -0.006 (0.016)** -0.051 -0.086 (0.503) 0.045 -0.057 (0.022)** Observations 1,243 1,860 621 622 929 931 Balance table of baseline measures, comparing male and female groups. It shows means of each group and p-value for the difference in means. Income vulnerability index is the sum of all subcategories. *** p<0.01, ** p<0.05, * p<0.1. 34 Table 3: Effect of the treatment on primary outcomes (1) (2) (3) (4) (5) (6) Mental Health Distress Index Anxiety Stress Depression Impulsiveness Positive (A + S + D) (A) (S) (D) caregiver-child interactions Treatment 0.057* 0.039 0.072** 0.029 -0.007 -0.043 (0.033) (0.034) (0.035) (0.035) (0.039) (0.039) [0.088] [0.269] [0.037] [0.451] [0.864] [0.295] Mean of Dep.Var (control) 0.000 0.000 0.000 0.000 0.000 0.000 R-squared 0.348 0.318 0.286 0.267 0.107 0.182 Observations 2,280 2,280 2,280 2,280 2,280 2,280 Strata FE Yes Yes Yes Yes Yes Yes Baseline Controls Yes Yes Yes Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Yes Yes Yes This table shows the estimated impacts of the intervention on primary outcomes. Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized average of the self-report scales of the emotional states of anxiety, stress, and depression (all items of DASS-21); in column (2) is the standardized average of the self-report scales of the emotional state of anxiety (7 items of DASS-21); in column (3) is the standardized average of the self-report scales of the emotional state of stress (7 items of DASS-21); in column (4) is the standardized average of the self-report scales of the emotional state of depression (7 items of DASS-21); in column (5) is the standardized sum of the self-report instrument of the Barratt Impulsiveness Scale BIS-11; and in column (6) is the standardized index of the responses to the 10 questions related to support for learning/stimulating environment and setting limits domain. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 35 Table 4: Heterogeneous effects across genders: Primary outcomes (1) (2) (3) Mental Health Impulsiveness Positive Distress caregiver-child interactions i. Treatment 0.034 0.037 0.018 (0.045) (0.051) (0.047) ii. Male x Treatment 0.059 -0.111 -0.154* (0.067) (0.080) (0.082) iii. Total effect on men ([i] + [ii]) 0.093* -0.074 -0.137** (0.050) (0.062) (0.067) Mean of Dep. Var (control) 0.000 0.000 0.000 R-squared 0.348 0.107 0.183 Observations 2,280 2,280 2,280 Strata FE Yes Yes Yes Baseline Controls Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes This table shows the differential impacts of the intervention on primary outcomes by gender. Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized average of the self-report scales of the emotional states of anxiety, stress, and depression (all items of DASS-21); in column (2) is the standardized sum of the self-report instrument of the Barratt Impulsiveness Scale BIS-11; and in column (3) is the standardized index of the responses to the 10 questions related to support for learning/stimulating environment and setting limits domain. All the regressions include interaction between male variable and treatment status. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 36 Table 5: Effect of the treatment on parental quality outcomes (1) (2) (3) (4) (5) Total Physical Psychological Tolerance of Child violence violence violence violent parenting behavior Treatment 0.007 -0.050 0.027 -0.020 0.066 (0.039) (0.038) (0.040) (0.039) (0.042) [0.860] [0.199] [0.513] [0.607] [0.123] Mean of Dep. Var (control) -0.008 0.002 -0.007 -0.005 0.005 R-squared 0.103 0.131 0.078 0.164 0.006 Observations 2,245 2,270 2,251 2,234 2,226 Strata FE Yes Yes Yes Yes Yes Baseline Controls Yes Yes Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Yes No Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized index of nine items of negative parenting techniques (physical and psychological violence); in column (2) is the standardized index of the two items of physical abuses; in column (3) is the standardized index of the seven items of psychological violence (emotional abuses); in column (4) is the standardized index of the responses in the ICAST-P module and the vignettes; and in column (5) is the standardized index of the seven items that are reported by the caregiver from the internalizing/externalizing behaviors sections of the Parent/Caregiver Report Survey. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 37 Table 6: Heterogeneous effects across genders: Parental quality outcomes (1) (2) (3) (4) (5) Total Physical Psychological Tolerance of Child violence violence violence violent parenting behavior i. Treatment -0.005 -0.098** 0.032 -0.020 0.063 (0.049) (0.049) (0.050) (0.048) (0.056) ii. Male x Treatment 0.029 0.123 -0.011 0.001 0.008 (0.082) (0.078) (0.084) (0.081) (0.086) iii. Total effect on men ([i] + [ii]) 0.025 0.025 0.020 -0.019 0.071 (0.065) (0.061) (0.067) (0.066) (0.065) Mean of Dep. Var (control) -0.008 0.002 -0.007 -0.005 0.005 R-squared 0.103 0.132 0.078 0.164 0.006 Observations 2,245 2,270 2,251 2,234 2,226 Strata FE Yes Yes Yes Yes Yes Baseline Controls Yes Yes Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Yes No Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized index of nine items of negative parenting techniques (physical and psychological violence); in column (2) is the standardized index of the two items of physical abuses; in column (3) is the standardized index of the seven items of psychological violence (emotional abuses); in column (4) is the standardized index of the responses in the ICAST-P module and the vignettes; and in column (5) is the standardized index of the seven items that are reported by the caregiver from the internalizing/externalizing behaviors sections of the Parent/Caregiver Report Survey. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 38 Table 7: Heterogeneous effects by economic deprivation: Primary outcomes (1) (2) (3) Mental Health Impulsiveness Positive Distress caregiver-child interactions i. Treatment -0.032 0.005 -0.048 (0.040) (0.049) (0.049) ii. Economic deprivation -0.032 0.075 0.071 (0.053) (0.060) (0.058) iii. Economic deprivation x Treatment 0.248*** -0.034 -0.049 (0.077) (0.086) (0.086) iv. Total effect on most deprived ([i] + [iii]) 0.217*** -0.029 -0.097 (0.065) (0.071) (0.071) Mean of Dep. Var (control) 0.003 -0.004 0.010 R-squared 0.352 0.109 0.182 Observations 2,172 2,172 2,172 Strata FE Yes Yes Yes Baseline Controls Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized average of the self-report scales of the emotional states of anxi-ety, stress, and depression (all items of DASS-21); in column (2) is the standardized sum of the self-report instrument of the Barratt Impulsiveness Scale BIS-11. In column (3) is the standardized index of the responses to the 10 questions related to support for learn- ing/stimulating environment and setting limits domain. The variable Economic deprivation is a dummy equal to one if the individual responded that from the beginning of the pandemic in their house, there was not enough money for food, home services, education,and others. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 39 Table 8: Heterogeneous effects by partner (1) (2) (3) Mental Health Impulsiveness Positive Distress caregiver-child interactions i. Treatment -0.032 -0.037 0.076 (0.629) (0.605) (0.297) ii. Living with a partner -0.117** -0.094 0.158** (0.039) (0.138) (0.010) iii. Living with a partner x Treatment 0.125 0.043 -0.167* (0.100) (0.620) (0.052) iv. Total effect on caregivers living with 0.094** 0.005 -0.091** a partner ([i] + [iii]) (0.039) (0.047) (0.046) Mean of Dep. Var (control) 0.000 0.000 0.000 R-squared 0.349 0.108 0.185 Observations 2,280 2,280 2,280 Strata FE Yes Yes Yes Baseline Controls Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized average of the self-report scales of the emotional states of anxi- ety, stress, and depression (all items of DASS-21); in column (2) is the standardized sum of the self-report instrument of the Barratt Impulsiveness Scale BIS-11; and in column (3) is the standardized index of the responses to the 10 questions related to support for learning/stimulating environment and setting limits domain. All the regressions include interaction between male variable and treat- ment status. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 40 Figure 1: Intervention Theory of Change Source: Author’s translation from Glasswing (2020) 41 Figure 2: Heterogeneous effects by sex and economic deprivation Figure 3: Heterogeneous effects by sex and partner 42 Appendix A1 Outcomes List and Description Main Outcomes A1.1 Mental health We use the Depression, Anxiety and Stress Scale (DASS-21) instrument to measure mental distress (Lovi- bond and Lovibond, 1996). This instrument includes 21 items organized in three sets of questions measur- ing the emotional states of depression, anxiety, and stress, respectively. Each set of questions contains seven items. Each item is measured on a scale of 0 –3 points (Never, Rarely, Almost Always or Always). We com- pute an index for each set of questions, constructed as the standardized average of the seven items. We also compute and aggregate index with a standardized average of all 21 items. We use the mean and standard deviation of the control group for the standardization. The DASS-21 are the following: No. Item Emotion Original version 1 Le costo´ mucho relajarse Stress I found it hard to wind down 2 Se dio cuenta que ten´ıa la boca seca Anxiety I was aware of dryness of my mouth 3 No pod´ ´ n sentimiento ıa tener ningu Depression I could not seem to experience any positive feeling at all positivo 4 ´ respirar Le costo Anxiety I experienced breathing difficulty (e.g., excessively rapid breathing, breathlessness in the absence of physical exer- tion) 5 ´ tomar la iniciativa para Le costo Depression I found it difficult to work up the initiative to do things hacer cosas 6 Reacciono ´ de forma exagerada en Stress I tended to over-react to situations ciertas situaciones 7 Sintio´ que las manos le temblaban Anxiety I experienced trembling (e.g., in the hands) 8 Sintio´ que estaba muy nervioso/a Stress I felt that I was using a lot of nervous energy 9 Estaba preocupado/a por situa- Anxiety I was worried about situations in which I might panic and ciones en las cuales podr´ıa hacer el make a fool of myself rid´ıculo 10 Sintio´ que no ten´ıa ninguna razo´n Depression I felt that I had nothing to look forward to para vivir 11 Noto´ que estaba muy agitado/a Stress I found myself getting agitated 12 Le costo´ mucho calmarse Stress I found it difficult to relax 13 Se sentio´ triste o deprimido/a Depression I felt down-hearted and blue 14 Fue intolarante con las cosas que le Stress I was intolerant of anything that kept me from getting on distra´ıan o que le desconcentraban with what I was doing 15 ´ que iba a tener un ataque de Sintio Anxiety I felt I was close to panic miedo 16 ´ que no pod´ Sintio ıa entusiasmarse Depression I was unable to become enthusiastic about anything por nada 17 ´ que no val´ Sintio ıa mucho como Depression I felt I was not worth much as a person persona 18 ´ que estaba muy irritable Sintio Stress I felt that I was rather touchy 19 ´ que ten´ Sintio ´n ıa agitado el corazo Anxiety I was aware of the action of my heart in the absence of a pesar de no haber hecho ningu ´n physical exertion (e.g., sense of heart rate increase, heart esfuerzo f´ısico missing a beat) 43 20 Tuvo miedo sin razo´ n aparente Anxiety I felt scared without any good reason 21 Sintio ´ que la vida no ten´ ´n ıa ningu Depression I felt that life was meaningless sentido 44 A1.2 Impulsiveness We use the Barratt Impulsiveness Scale BIS-11 (Patton et al., 1995) to measure impulsiveness. It includes 15 items, each measured on a scale of 1–4 points (Never, Rarely, Almost Always or Always). We construct as index as the standardized sum of the 15 items. We use the mean and standard deviation of the control group for the standardization. The BIS-11 items are the following: No. Item Original version 1 Actue ´ impulsivamente I act on impulse 2 Hice las cosas en el momento que se me ocurrieron I act on the spur of the moment 3 Hice las cosas sin pensarlas I do things without thinking 4 Dije las cosas sin pensarlas I say things without thinking 5 Compre ´ cosas impulsivamente I buy things on impulse 6 Busque ´ un mejor trabajo I change jobs 7 Hice planes para el futuro I am future oriented 8 Ahorre ´ con regularidad I save regularly 9 Planifique ´ mis tareas con cuidado I plan tasks carefully 10 Pense´ las cosas cuidadosamente I am a careful thinker 11 Me sent´ı muy inquieto/a al tener que escuchar hablar a alguien I am restless at the theater or lectures 12 Se me hizo dif´ıcil estar quieto/a por largos periodos de tiempo I squirm at plays or lectures 13 Me concentre ´ facilmente I concentrate easily 14 Pude enfocar mi mente en una sola cosa por mucho tiempo I can only think about one thing at a time 15 Me aburrio ´ pensar en algo por demasiado tiempo I get easily bored when solving thought problems A1.3 Positive caregiver-child interactions We use the 10 questions related to support for learning/stimulating environment and setting limits domains from the Family Care Indicators instrument developed by UNICEF (Kariger et al., 2012) to measure posi- tive caregiver-child interactions. We construct an standardized index using inverse covariance weighting, following Anderson (2008). We use the mean and standard deviation of the control group for the standard- ization. The items used to measure this outcome are the following (response categories are never, one day, two or three days and four or more days): No. Item Original version 1 Leyo´ libros o ima ´ genes en libros last week someone read books/showed images to kid 2 Conto ´ cuentos last week someone told stories/tales to kid 3 Le canto ´ canciones last week someone sang songs to kid 4 Jugaron last week someone played with kid 5 Usaron tiempo en actividades de aprendizaje, como contar, nombrar objetos o dibujar last week someone shared learning time with kid 6 Hicieron labores de la casa como cocinar, limpiar, cuidar a los animales, u otras similares last week someone did chores with kid 7 Le ensen ˜o´ lecciones de la iglesia, leyeron la biblia, etc. last week someone gave biblical lessons/read bible to kid 8 Se sento´ con el nin ˜ a a comer juntos durante el almuerzo o cena ˜ o o nin last week someone sat to have meal with kid 9 Alimento ˜o ´ al nin last week someone fed the kid 10 Hablaron durante las comidas last week someone spoke with kid during meals Secondary outcomes 45 A1.4 Tolerance towards violent parenting We measure tolerance towards violence parenting practices using responses from two sets of questions. First, we elicited tolerance through two fictional stories about typical violent practices that could take place in the studied context. After each story, we asked respondents two questions about how justifiable they think the behavior of the fictional caregiver was (the rating scale in each vignette is: inadequate, little adequate, neutral, slightly adequate, and adequate). Second, we asked direct questions about tolerance of child abuse perpetrated by fathers, mothers, and teachers, and whether respondents think physical punishment is an effective disciplinary tool (in each question response options are yes or no). Our main outcome is a standardized index using inverse covariance weighting, following (Anderson, 2008). We used the mean and standard deviation of the control group for the standardization. The items used to measure this trait are the following: 46 No. Item Original version Vignettes baseline 1 Es un d´ıa normal de cuarentena y la familia se prepara para Considerando lo que paso ´ en este How do you consider Roberto’s re- almorzar. Todos esta ´ n sentados a la mesa excepto Rebecca ´mo considera que fue la relato ¿Co action to the behavior of his daugh- quien tiene 8 an ˜ os. Roberto su padre, quien ha estado de- ´n de Roberto ante el compor- reaccio ter Rebecca? sempleado desde hace un par de meses, la llama varias ve- tamiento de su hija Rebecca? ces, pero ella no le escucha o decide no obedecer, por lo que no responde a sus llamadas durante casi 30 minutos para ir al comedor y almorzar. Roberto se altera mucho y, como cree que la actitud de su hija debe ser corregida, le da una palmada y la lleva al comedor para hacerle entender que su desobediencia no puede repetirse. 2 Marcos y Stephanie son hermanos y esta ´ n jugando dentro Considerando lo que paso´ en este How do you consider Ana’s reac - de la casa. Sus padres esta ´ n cansados e irritables por el ´mo considera que fue la relato ¿Co tion to the behavior of her children ruido que hacen los nin ˜ os, ya que la casa es pequen ˜ a, pero ´n de Ana ante el compor- reaccio Marcos and Stephanie? los aguantan ya que prefieren que sus hijos este ´ n dentro tamiento de sus hijos Marcos y y no fuera de la casa. De repente el juego se sale de con- Stephanie? trol y Marcos le pega con la pelota al televisor, lo bota y lo arruina. Su madre, Ana, se enfurecio ´ much´ ısimo, ya que recie ıan comprado el televisor y todav´ ´ n hab´ ıa lo segu´ ıan pagando en cuotas. Ana se molesto ´ tanto que perdio ´ el control y empezo ´ a gritarle a sus hijos, dicie ´ ndoles que ya no los soportaba, los agarro ´ de los brazos y los tiro ´ al piso. Vignette endline 1 Ya es la hora de cenar en la casa de Paco. Paco estuvo traba- Considerando lo que paso ´ en este How do you consider Paco’s reac - jando todo el d´ıa y Mar´ıa su esposa estuvo todo el d´ıa con ´mo considera que fue la relato ¿Co tion to the behavior of his daughter los nin˜ os. Todos esta´ n sentados a la mesa excepto Ana, ´n de Paco ante el compor- reaccio Ana? la hija de Paco y Mar´ ıa. Ana tiene 10 an˜ os y a veces no tamiento de su hija Ana? hace todo lo que sus padres le piden. Maria y Paco esta ´n cansados por todo lo que esta ´ pasando con la pandemia, y hoy Mar´ ´ algo rico para que todos pudieran re- ıa cocino lajarse un poco. Ana se ha quedado viendo la tele y no se acerca a comer, a pesar de que sus padres le han estado llamando. Paco se comienza a alterar mucho, ya que esta ´ muy cansado. El cree que la actitud de su hija no esta´ bien, por lo que le grita, le da una palmada y la lleva a la mesa para hacerle entender que no debe ser desobediente. Direct questions 1 ¿Cree usted que es aceptable que Do you think it is acceptable for a un padre, madre o encargado(a), caregiver to use violence to disci- cuidador(a) castigue f´ısicamente a pline a kid? un nin ˜ o(a) cuando e´ l(ella) se porta mal? 2 ¿Cree usted que es aceptable que Do you think it is acceptable for a el(la) docente castigue f´ısicamente a teacher to use violence to discipline un nin ˜ o(a) cuando e´ l(ella) se porta a kid? mal? 3 ¿Piensa que el castigo f´ısico es un Do you think physical punishment me´ todo de correccio´n efectivo? is an effective method to discipline children? A1.5 Child behavior We measure child behavior using the internalizing/externalizing behaviors sections of the parent/caregiver report survey developed by the World Bank. This module includes seven items reported by the caregiver, in each question response options are yes or no. The questions relate to negative behaviors (i.e. an high outcome indicates bad child behavior), were framed with respect to the last three months, and were only administered to parents with children aged three or older following the recommendations of the module developers. This module was collected at endline only. We construct a standardized index (we used the mean and standard deviation of the control) using inverse covariance weighting, as in Anderson (2008). The items included are the following: 47 No. Item Original version 1 Llorando mucho Crying a lot 2 Hablando con mayor dificultad de la habitual Speaking with greater difficulty than usual 3 Aislado(a) o muy callado(a) Isolated or very quiet 4 Irritable (se ha molestado o enojado fa´ cilmente por cosas que le pasan) Irritable (easily upset or angry about things that happen to you) 5 Rebelde (no respeta las reglas de la casa) Rebel (does not respect the house rules) 6 Destruyendo o dan ˜ ando cosas Destroying or damaging things 7 Tranquilo/a. No nota ningu ´ n cambio de comportamiento (Revertido) Quiet. You do not notice any change in behavior (reversed) Exploratory outcomes A1.6 Parenting practices and violence perpetration We use an adapted version of the International Society for Prevention of Child Abuse and Neglect’s screen- ing tool for caregivers (ICAST-P) to measure child abuse.1 The original instrument contains 38 items of violence. We shortened this instrument to eight items. To improve our measure of child abuse, we com- plemented this self-reported instrument with responses to hypothetical perpetration of violence elicited through vignettes. We presented caregivers two fictional stories about regular caregiver-child interactions. We use the same vignettes as in the module on tolerance towards violent parenting practices. After each story, we surveyed respondents on eight items of abuse that they might use with the child in their carein that hypothetical scenario. Each item of abuse included in this module was also selected following the ICAST- P. The scale of responses is a five-point Likert scale. We explore four outcomes in this line. First, weestimate a standardized index of the seven items of positive parenting techniques. Second, a standardizedindex of nine items of negative parenting techniques including physical and psychological violence. We an- alyze negative parenting in two separate aspects. Third, a standardized index of the two items of physical abuse, and finally a standardized index of the seven items of psychological violence (emotional abuse). All the outcomes using inverse covariance weighting, following Anderson (2008). The items included are the following: 1This is an internationally validated tool to measure child abuse (Meinck et al., 2020). When this module was developed, the scientific understanding of the best approaches to surveying parents remotely on sucha sensitive topic were limited. Due to this uncertainty, we include this item as an exploratory outcome. The module follows the Conflict Tactics Scale methodology, and therefore the coding of each item is a dummy that equals one if the respondent uses an item of violence. 48 No. Item Form of Parenting Original version Direct Caregiver Questions 1 Prohibirle que se mueva de lugar Psychological violence Forced him or her to hold a position that caused pain or humiliated him or her as a means of punishment? 2 Pegarle en la mano cuando toca algo Physical violence Hit on head with knuckle or back of the hand? que no debe 3 Decirle ”no” y explicarle por que Positive parenting Explained to him/her why something s/he did was wrong? 4 Hacer que se siente o mandarlo Psychological violence Locked him or her up in a small place or in a dark room? a otro cuarto para que tenga un P37 a tiempo a solas 5 Gritarle Psychological violence Shouted, yelled, or screamed at him/her very loud and ag- gressively? 6 Quitarle los objetos y ponerlos en un Positive parenting Took away his/her pocket money or other privileges? For- lugar donde no los alcance bade him/her of something he/she liked? Gave him/her something else to do in order to distract his/her attention? 7 Distraerlo con otras actividades Positive parenting Took away his/her pocket money or other privileges? For- bade him/her of something he/she liked? Gave him/her something else to do in order to distract his/her attention? 8 Ofenderlo(a) o humillarlo(a) Psychological violence Insulted him/her by calling him/her dumb, lazy, or other names like that? Vignettes 9 Quitar los juguetes o cualquier cosa Positive parenting Took away his/her pocket money or other privileges? For- que la nina o el nino le guste bade him/her of something he/she liked? Gave him/her something else to do in order to distract his/her attention? 10 Gritarle para que obedezca Psychological violence Shouted, yelled, or screamed at him/her very loud and ag- gressively? 11 Llamarla ” desobediente” ,” malcri- Psychological violence Insulted him/her by calling him/her dumb, lazy or other ada” o ” torpe” names like that? 12 Pegarle con la palma de la mano o Physical violence Hit elsewhere (not buttocks) with an object such as a stick, un objeto broom, cane, or belt? 13 Encerrarla o dejarla sin alimentos Psychological violence Did not get enough to eat (went hungry) and/or drink por un tiempo (were thirsty) even though there was enough for everyone, as a means of punishment? 14 Explicarle por que el compor- Positive parenting Explained to him/her why something s/he did was wrong? tamiento no es el adecuado 15 Pedirle que no lo haga nuevamente Positive parenting Told her/him to start or stop doing something ? 16 Preguntar por que la nina actuo de Positive parenting Explained to him/her why something him/her did was esa manera wrong? A1.7 Child socioemotional development To measure child behavior and socioemotional development, we follow the parent/caregiver report survey developed by the World Bank. The data collection was framed with respect to the last three months and was only asked of parents with children aged three or more. 2 We construct an index of the frequency with which caregivers observe certain behaviors in their child. We use nine items, each item is measured on a scale of 49 2 The socioemotional development variable was not included initially in the PAP, and therefore is consid- ered an exploratory outcome. 50 1–4 points (Never, Rarely, Almost Always or Always). The outcome is a standardized index (we used the mean and standard deviation of the control) using inverse covariance weighting, as in (Anderson, 2008) ofthe responses to the following items of typical socioemotional development. An high outcome indicatesbetter socioemotional development. No. Item Original version 1 ¿Tuvo una buena concentracio ´ n? Did the child have good concentration? 2 ¿Recordo ´ todas las instrucciones que recibio ´ para hacer varias cosas? Did the child remember all the instructions he received to do various things? 3 ¿Planifico´ lo que ten´ıa que hacer antes de hacerlo? Did the child plan what he had to do before doing it? 4 ¿Dejo´ de hacer algo cuando se le pidio ´ que lo hiciera? Did the child stop doing something when asked to do it? 5 ¿Continuo ´ trabajando en una actividad hasta que la termino ´? Did the child keep working on an activity until he finished it? 6 ¿Asumio ´ la responsabilidad de sus actos? Did the child take responsibility for his actions? 7 ´ bien con otros nin ¿Se llevo ˜ os? Did the child get along with other children? 8 ¿Se ajusto ´ cilmente a las transiciones? (Por ejemplo, ir al doctor, cambiarse de cuarto, pasar jugar a descansar) ´ fa Did the child adjust easily to transitions? 9 ¿Se tranquilizo ´ despue´ s de haber estado muy activo/a? Did the child calm down after being very active? A1.8 Social desirability We use the short form of 13 questions to measure the individual-level propensity to misreport sensitive items, following Crowne and Marlowe (1960) and Dhar et al. (2018). We included the social desirability module at endline (due to time limitations in the baseline survey instrument). Each question allowed two possible answers: yes or no. The outcome is the sum of positive answers to all the following items: No. Item Original version 1 A veces se me hace dif´ıcil ponerme a trabajar sin que me pidan que lo haga (Revertido) It is sometimes hard for me to go on with my work if I am not encouraged 2 A veces me siento frustrado(a) o triste porque las cosas no salen como yo quiero (Revertido) I sometimes feel resentful when I don’t get my way 3 En alguna ocasiones, he dejado de intentar hacer algo porque he pensado que soy poco capaz de hacerlo (Revertido) On a few occasions, I have given up doing something because I thought too little of my ability 4 ´ n (Revertido) En ocasiones quiero llevarle la contraria a la gente con autoridad, aunque sepa que tienen razo There have been times when I felt like rebelling against people in authority even though I knew they were right 5 Sin importar con quien este ´ hablando siempre escucho con atencio ´n No matter who I’m talking to, I’m always a good listener 6 Han habido ocasiones en que me he aprovechado de alguna persona (Revertido) There have been occasions when I took advantage of someone 7 Siempre estoy dispuesto/a a aceptar cuando cometo un error I’m always willing to admit it when I make a mistake 8 En ocasiones trato de desquitarme o vengarme en lugar de perdonar u olvidar (Revertido) I sometimes try to get even rather than forgive and forget 9 Siempre soy amable, aun con la gente que no es tan agradable I am always courteous, even to people who are disagreeable 10 Nunca me molesto cuando la gente tiene ideas que son muy distintas a las m´ıas I have never been irked when people expressed ideas very different from my own 11 A veces he sentido muchos celos de la buena suerte de otras personas (Revertido) There have been times when I was quite jealous of the good fortune of others 12 Algunas veces me irrito por que la gente me pida favores (Revertido) I am sometimes irritated by people who ask favors of me 13 De forma consciente he dicho cosas que han herido los sentimientos de otra persona (Revertido) I have deliberately said something that hurt someone’s feelings 51 A1.9 Sociodemographic information and additional covariates Categorias de respuesta Fecha de nacimiento Dia/Mes/Ano Sexo Mujer Hombre ´ s alto que ha alcanzado Nivel educativo ma Kinder (4-5) Preparatoria 1er grado 2do grado 3er grado 4to grado 5to grado 6to grado 7mo grado 8vo grado 9no grado 1er ano bachto 2do ano bachto 3er ano bachto Tecnico superior incompleto Tecnico superior completo Universitario incompleto Universitario completo ´ ltimos 6 meses, ha desempen Durante los u ´ n trabajo, ya sea como ˜ ado algu S´ı, empleado S´ı, empleado por cuenta propia No No sabe empleado o por cuenta propia Realizaba algu´ n trabajo/ocupacin ´ /oficio remunerado al momento que Si No No sabe No responde el gobierno declaro ´ la cuarentena obligatoria por COVID-19 el 22 de Marzo ˜ os que estan bajo su cuidado o responsabilidad Nin Nin˜ as ´ mero Nu Nin˜ os ´ mero Nu Horas en promedio que el nin ˜ a pasa con usted ahora que las clases ˜ o o nin ´ n suspendidas presenciales esta ´ ero (0-12) Nm ´ s edad que tenga 8 an Sexo del hijo/hija con ma ˜ os o menos ˜ o/nin Sexo del nin ˜a ˜ o/a esta El nin ´ matriculado/a en alguna escuela o colegio Si No Horas al d´ ´ n o videos ıa que su hijo o hija vio programas en la televisio en el celular para divertirse ´ mero (0-24) Nu Horas al d´ıa que su hijo o hija vio programas en la television o videos en el celular para estudiar/hacer tareas ´ mero (0-24) Nu Personas que viven en su casa ´ mero Nu ´ mero de personas al lado del parentesco que aplique para las Escriba el nu personas que viven en su hogar Esposo/a o companero/a de vida ´ mero Nu Hijas o hijos solo de su pareja ´ mero Nu Hija o hijo ´ mero Nu Papa ´ mero Nu Mama ´ mero Nu Suegro ´ mero Nu Suegra ´ mero Nu Hermana o ´ mero Nu Nuera / yerno ´ mero Nu Nieta o ´ mero Nu Sobrina a ´ mero Nu Otros ´ mero Nu Desde el inicio de la cuarentena, considera que en su casa hay dinero suficiente para: Comida/alimentacio ´n Si No No sabe No responde Servicios importantes como salud, gastos educativos Si No No sabe No responde Servicios ba ´ sicos como agua, electricidad, gas Si No No sabe No responde Otros bienes o servicios como ropa, recreacio ´ n, regalos Si No No sabe No responde 52 A2 Appendix Tables and Figures Figure A1: Message Viewership Notes: Calculation of user-opened messages by message received. Viewership rates calculated for the treat- ment group only (N=2,663). Table A1: Summary of stratification sample Mode of Data Collection Total Facebook Glasswing Communities SMS/WhatsApp Female Caregiver 747 78 1035 1860 Male Caregiver 312 11 920 1243 Total 1,059 89 1,955 3,103 This table provides the size of each stratum in the sample recruitment by mode and gender. 53 Table A2: Recall and Knowledge Incorporation Survey Rounds Stress Management Positive Parenting N Wave 1 Short breathing exercise Storytelling Turtle technique 204 Deep breathing exercise Wave 2 Drop technique Listen attentively Deep breathing exercise Rapport Child 204 Wave 3 Lemon technique Conflict management Short breathing exercise 100 Sleep well Wave 4 Relaxation technique Let win Manual activities 151 This table shows an overview of the techniques surveyed by knowledge incorporation (KI) rounds. 54 Table A3: Individual characteristics and attrition (1) (2) Attrition Treatment 0.020 0.081 (0.016) (0.126) Age x Treatment -0.002 (0.003) Female x Treatment 0.024 (0.038) High school (10-12 grades) x Treatment -0.046 (0.051) Bachelor or higher x Treatment -0.038 (0.052) Always employed x Treatment 0.038 (0.050) Lost job x Treatment -0.000 (0.069) Found job x Treatment 0.028 (0.056) Girls under 9 years cared x Treatment -0.014 (0.036) Boys under 9 years cared x Treatment -0.039 (0.035) Oldest child under 9 is female x Treatment 0.002 (0.050) Child enrolled in school x Treatment 0.071* (0.040) Child video/screen time for fun x Treatment -0.001 (0.006) Child video/screen time for homework x Treatment -0.002 (0.006) Household members x Treatment 0.001 (0.010) Intergenerational household x Treatment 0.000 (0.040) Income vulnerability index x Treatment 0.006 (0.015) Mental health index, std. x Treatment 0.001 (0.019) Impulsiveness index, std. x Treatment -0.021 (0.019) Positive caregiver-child interactions, std. x Treatment -0.002 (0.018) Tolerance norms index, std. x Treatment -0.011 (0.021) Parenting positive index, std. x Treatment -0.017 (0.019) Parenting negative index, std. x Treatment 0.303* (0.169) Physical violence, std. x Treatment -0.064 (0.046) Psychological violence, std. x Treatment -0.290* (0.155) Individual controls No Yes Strata fixed effects Yes Yes Number of observations 3103 2589 P-value for F-test (interactions) 0.947 The dependent variable in columns (1)–(2) is a dummy equal to one if the individual did not respond to the follow-up survey. The regression in column (2) includes the interactions between individual characteristics/ outcomes from the baseline survey and treatment (all the individual variables are included, coefficients are omitted. The reduction in sample size in column two is because it contains missing values). We use the F-statistics (interactions) to test the hypothesis of nondifferential attrition between treatment and control groups. 55 Table A4: Summary Statistics - Knowledge Incorporation Surveys Variable N Mean Std Dev Min Max Pooled Waves % known stress-management techniques 659 0.50 0.40 0 1 Known stress-management technique (dummy) 659 0.69 0.46 0 1 % known parenting techniques 659 0.81 0.37 0 1 Known parenting technique (dummy) 659 0.85 0.36 0 1 =1 if stressed or very stressed, =0 if little or not stressed 659 0.51 0.50 0 1 =1 if quality of interaction is very easy or easy, =0 if hard or very hard 659 0.62 0.49 0 1 Treatment Group 659 0.47 0.50 0 1 No. of monitor surveys 659 1.15 0.39 1 3 Wave 1 % known stress-management techniques 204 0.57 0.37 0 1 Known stress-management technique (dummy) 204 0.77 0.42 0 1 % known parenting techniques 204 0.63 0.48 0 1 Known parenting technique (dummy) 204 0.63 0.48 0 1 =1 if stressed or very stressed, =0 if little or not stressed 204 0.57 0.50 0 1 =1 if quality of interaction is very easy or easy, =0 if hard or very hard 204 0.79 0.41 0 1 Treatment Group 204 0.48 0.50 0 1 No. of monitor surveys 204 1.00 0.00 1 1 Wave 2 % known stress-management techniques 204 0.58 0.36 0 1 Known stress-management technique (dummy) 204 0.81 0.39 0 1 % known parenting techniques 204 0.95 0.15 1 1 Known parenting technique (dummy) 204 1.00 0.00 1 1 =1 if stressed or very stressed, =0 if little or not stressed 204 0.45 0.50 0 1 =1 if quality of interaction is very easy or easy, =0 if hard or very hard 204 0.10 0.30 0 1 Treatment Group 204 0.48 0.50 0 1 No. of monitor surveys 204 1.09 0.28 1 2 Wave 3 % known stress-management techniques 100 0.59 0.31 0 1 Known stress-management technique (dummy) 100 0.92 0.27 0 1 % known parenting techniques 100 0.76 0.43 0 1 Known parenting technique (dummy) 100 0.76 0.43 0 1 =1 if stressed or very stressed, =0 if little or not stressed 100 0.57 0.50 0 1 =1 if quality of interaction is very easy or easy, =0 if hard or very hard 100 0.93 0.26 0 1 Treatment Group 100 0.45 0.50 0 1 No. of monitor surveys 100 1.28 0.51 1 3 Wave 4 % known stress-management techniques 151 0.25 0.44 0 1 Known stress-management technique (dummy) 151 0.25 0.44 0 1 % known parenting techniques 151 0.91 0.21 0 1 Known parenting technique (dummy) 151 0.99 0.11 0 1 =1 if stressed or very stressed, =0 if little or not stressed 151 0.46 0.50 0 1 =1 if quality of interaction is very easy or easy, =0 if hard or very hard 151 0.88 0.33 0 1 Treatment Group 151 0.45 0.50 0 1 No. of monitor surveys 151 1.34 0.54 1 3 This table shows variables of the four knowledge incorporation (KI) survey rounds with a subset of individuals who enrolled in the study at baseline. 56 Table A5: Effect of the treatment on caregiver’s knowledge incorporation (1) (2) (3) (4) Stress Techniques Parenting Techniques Treatment 0.206*** 0.161*** 0.057** 0.094** (0.029) (0.047) (0.026) (0.039) Treatment x Female 0.078 -0.063 (0.059) (0.053) Female -0.000 -0.037 -0.004 0.026 (0.030) (0.038) (0.027) (0.039) Mean of Dep. Var 0.5 0.81 N 659 658 659 659 Adjusted R-squared 0.181 0.182 0.140 0.182 Wave FE Yes Yes Yes Yes No. of Surveys Taken Yes Yes Yes Yes Baseline Controls Yes Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Yes This table presents results from Equation 3. Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in columns (1)-(2) is the share of stress-management techniques used by the caregiver in the last week; and in columns (3)-(4) is the share of positive parenting techniques used by the caregiver in the last week. All regressions include wave fixed effects. The controls include the number of surveys taken by a respondent, age in years, household-size dummies for average, big or small household, a dummy for intergenerational household, sex of the respondent, and the educational level of the respondent for primary, high school—the omitted category—or tertiary education. We also control for the baseline level of mental health using the standardized index of the DASS score in columns (1)-(4) and the standardized index of the quality of caregiver child interactions. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 57 Table A6: Heterogeneous effects across genders: Mental health outcomes (1) (2) (3) Anxiety Stress Depression (A) (S) (D) i. Treatment 0.004 0.049 0.022 (0.046) (0.046) (0.047) ii. Male x Treatment 0.091 0.059 0.016 (0.069) (0.071) (0.070) iii. Total effect on men ([i] + [ii]) 0.095* 0.108** 0.039 (0.051) (0.053) (0.052) Mean of Dep. Var (control) 0.000 0.000 0.000 Adjusted R-squared 0.315 0.282 0.263 Observations 2,280 2,280 2,280 Strata FE Yes Yes Yes Baseline Controls Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized average of the self-report scales of the emotional state of anxiety (7 items of DASS-21); in column (2) is the standardized average of the self-report scales of the emotional state of stress (7 items of DASS- 21); and in column (3) is the standardized average of the self-report scales of the emotional state of depression (7 items of DASS-21).All the regressions include interaction between male variable and treatment status. The controls include age in years, girls and boyscared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 58 Table A7: Effect of the treatment on additional outcomes (1) (2) (3) Positive Interest in Child parenting additional materials socioemotional development Treatment 0.004 -0.011 -0.088 (0.041) (0.015) (0.059) [0.931] [0.454] [0.133] Mean of Dep. Var (control) 0.007 0.841 3.779 R-squared 0.043 0.028 0.020 Observations 2,265 2,280 1,094 Strata FE Yes Yes Yes Baseline Controls Yes Yes Yes Dep. Var Baseline Level Yes No No Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized index of the seven items of positive parenting techniques; in column (2) is a dummy equal to one if the individual is interested in additional materials; and in column (3) is the standardized index of the nine items of typical socioemotional development. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 59 Table A8: Heterogeneous effects across genders: Additional outcomes (1) (2) (3) Positive Interest in Child parenting additional materials socioemotional development i. Treatment -0.008 -0.022 -0.089 (0.052) (0.019) (0.075) ii. Male x Treatment 0.031 0.028 0.004 (0.083) (0.033) (0.121) iii. Total effect on men ([i] + [ii]) 0.023 0.006 -0.085 (0.065) (0.027) (0.095) Mean of Dep. Var (control) 0.007 0.841 3.779 R-squared 0.043 0.028 0.020 Observations 2,265 2,280 1,094 Strata FE Yes Yes Yes Baseline Controls Yes Yes Yes Dep. Var Baseline Level Yes No No Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized index of the seven items of positive parenting techniques; in column (2) is a dummy equal to one if the individual is interested in additional materials; and in column (3) is the standardized index of the nine items of typical socioemotional development. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 60 Table A9: Robustness check: Physical results (1) (2) (3) (4) (5) Physical violence Social desirability Physical violence Hit Hit index controlling dummy (perpetration) (vignettes) i. Treatment -0.098** -0.094* -0.038* -0.039* -0.013 (0.049) (0.048) (0.023) (0.020) (0.019) ii. Male x Treatment 0.123 0.116 0.044 0.038 0.027 (0.078) (0.077) (0.037) (0.032) (0.030) iii. Social desirability score -0.073*** (0.013) iv. Total effect on men ([i] + [ii]) 0.025 0.022 0.007 -0.002 0.014 (0.061) (0.061) (0.029) (0.025) (0.024) Mean of Dep. Var (control) 0.002 0.002 0.305 0.222 0.154 Adjusted R-squared 0.127 0.145 0.145 0.163 0.061 Observations 2,270 2,270 2,270 2,270 2,280 Strata FE Yes Yes Yes Yes Yes Baseline Controls Yes Yes Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Yes Yes Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in columns (1)-(2) is the standardized index of the two items of physical abuses; in column (3) isa dummy equal to one if the individual reports at least one item of physical abuse; in column (4) is a dummy equal to one if theindividual reports the Hit with knuckle or back of the hand item; and in column (5) is a dummy equal to one if the individual reportsthe Hit with knuckle or back of the hand or with an object item. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Social desirability score captures individual-level propensity to misreport sensitive items as an additional control variable in column (2).Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 61 Table A10: Effect of the treatment on primary outcomes including the SDB as control (1) (2) (3) (4) (5) (6) Mental Health Distress Index Anxiety Stress Depression Impulsiveness Positive (A + S + D) (A) (S) (D) caregiver-child interactions Treatment 0.060* 0.042 0.074** 0.032 -0.005 -0.045 (0.057) (0.208) (0.026) (0.335) (0.893) (0.246) Social desirability score -0.146*** -0.115*** -0.154*** -0.134*** -0.054*** 0.073*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Mean of Dep. Var (control) 0.000 0.000 0.000 0.000 0.000 0.000 Adjusted R-squared 0.407 0.355 0.351 0.318 0.111 0.193 Observations 2,280 2,280 2,280 2,280 2,280 2,280 Strata FE Yes Yes Yes Yes Yes Yes Baseline Controls Yes Yes Yes Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Yes Yes Yes Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized average of the self-report scales of the emotional state of anxiety (7 items of DASS-21); in column (2) is the standardized average of the self-report scales of the emotional state of stress (7 items of DASS-21); in column (3) is the standardized average of the self-report scales of the emotional state of depression (7 items of DASS-21); in column (4) is the standardized average of the self-report scales of the emotional states of anxiety, stress, and depression (all items of DASS-21); in column (5) is the standardized sum of the self-report instrument of the Barratt Impulsiveness Scale BIS-11; and in column (6) is the standardized index of the responses to the 10 questions related to support for learning/stimulating environment and setting limits domain. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Social desirability score captures the individual-level propensity to misreport sensitive items as an additional control variable. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 62 Table A11: Heterogeneous effects across gender including the SDB as control: Primary outcomes (1) (2) (3) Mental Health Impulsiveness Positive Distress caregiver-child interactions i. Treatment 0.044 0.040 0.012 (0.298) (0.429) (0.793) ii. Male x Treatment 0.041 -0.115 -0.146* (0.516) (0.147) (0.072) iii. Social desirability -0.146*** -0.054*** 0.073*** (0.000) (0.000) (0.000) iv. Total effect on men ([i] + [ii]) 0.085* -0.075 -0.134** (0.047) (0.062) (0.066) Mean of Dep. Var (control) 0.000 0.000 0.000 Adjusted R-squared 0.407 0.111 0.194 Observations 2,280 2,280 2,280 Strata FE Yes Yes Yes Baseline Controls Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if to the control group. The dependent variable in column (1) is the standardized average of the self-report scales the emotional states of anxiety, stress and depression (all items of DASS-21); in column (2) is the standardized sum of the self-report instrument of Barratt Impulsiveness Scale BIS-11; and in column (3) is the standardized index of the responses to the ten questions related to support for learning/stimulating environment and setting limits domain. The controls include age in years, girls and boys cared for by the respondent and the educa- tional level of the respondent for primary, high-school or tertiary education -the omitted category-. Social desirability score capturing individual-level propensity to misreport sensitive items as an additional control variable. Robust standard errors-. *** p<0.01, ** p<0.05, * p<0.1. 63 Table A12: Effect of the treatment on secondary outcomes including the SDB as control (1) (2) Tolerance of Child violent parenting behavior Treatment -0.020 0.068* (0.038) (0.041) Social desirability score -0.073*** -0.153*** (0.012) (0.012) Mean of Dep. Var (control) -0.005 0.005 Adjusted R-squared 0.176 0.077 Observations 2,234 2,226 Strata FE Yes Yes Baseline Controls Yes Yes Dep. Var Baseline Level Yes No Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized index of the responses in the ICAST-P module and the vignettes. The dependent variable in column (2) is the standardized index of the seven items that are reported by the caregiver from the internaliz- ing/externalizing behaviors sections of the Parent/Caregiver Report Survey. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Social desirability score captures the individual-level propensity to misreport sensitive items as an additional control variable. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 64 Table A13: Effect of the treatment on exploratory outcomes including the SDB as control (1) (2) (3) (4) (5) (6) Positive Negative Physical Psychological Interest in Child parenting parenting violence violence additional socioemotional materials development Treatment 0.004 0.008 -0.048 0.028 -0.011 -0.106* (0.040) (0.038) (0.038) (0.039) (0.015) (0.057) Social desirability score -0.043*** -0.121*** -0.074*** -0.116*** -0.001 0.141*** (0.011) (0.015) (0.013) (0.015) (0.004) (0.017) Observations 2,265 2,245 2,270 2,251 2,280 1,094 Adjusted R-squared 0.043 0.146 0.144 0.115 0.023 0.077 Mean of Dep. Var (control) 0.007 -0.008 0.002 -0.007 0.841 3.779 Strata FE Yes Yes Yes Yes Yes Yes Baseline Controls Yes Yes Yes Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Yes No No Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized index of the seven items of positive parenting techniques; in column (2) is the standardized index of nine items of negative parenting techniques (physical and psychological violence); in column (3) is the standardized index of the two items of physical abuses; in column (4) is the standardized index of the seven items of psychological violence (emotional abuses); in column (5) is a dummy equal to one if the individual is interested in additional materials; and in column (6) is the standardized index of the nine items of typical socioemotional development. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Social desirability score captures the individual-level propensity to misreport sensitive items as an additional control variable. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 65 Table A14: Effect of the treatment on social desirability (1) Social desirability score Treatment 0.019 (0.076) Mean of Dep. Var (control) 0.005 Adjusted R-squared 0.009 Observations 2,280 Strata FE Yes Baseline Controls Yes Dep. Var Baseline Level No Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) captures the individual-level propensity to misreport sensitive items. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 66 Table A15: Effect of the treatment on primary outcomes without individual controls (1) (2) (3) (4) (5) (6) Mental Health Distress Index Anxiety Stress Depression Impulsiveness Positive (A + S + D) (A) (S) (D) caregiver-child interactions Treatment 0.054 0.036 0.070** 0.025 -0.005 -0.042 (0.033) (0.034) (0.035) (0.035) (0.039) (0.039) Mean of Dep. Var (control) 0.000 0.000 0.000 0.000 0.000 0.000 Adjusted R-squared 0.341 0.312 0.281 0.258 0.101 0.177 Observations 2,280 2,280 2,280 2,280 2,280 2,280 Strata FE Yes Yes Yes Yes Yes Yes Baseline Controls No No No No No No Dep. Var Baseline Level Yes Yes Yes Yes Yes Yes This table shows the estimated impacts of the intervention on primary outcomes. Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized average of the self-report scales of the emotional states of anxiety, stress, and depression (all items of DASS-21); in column (2) is the standardized average of the self-report scales of the emotional state of anxiety (7 items of DASS-21); in column (3) is the standardized average of the self-report scales of the emotional state of stress (7 items of DASS-21); in column (4) is the standardized average of the self-report scales of the emotional state of depression (7 items of DASS-21); in column (5) is the standardized sum of the self-report instrument of the Barratt Impulsiveness Scale BIS-11; and in column (6) is the standardized index of the responses to the 10 questions related to support for learning/stimulating environment and setting limits domain. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 67 Table A16: Heterogeneous effects across genders without individual controls (1) (2) (3) Mental Health Impulsiveness Positive Distress caregiver-child interactions i. Treatment 0.031 0.040 0.015 (0.045) (0.051) (0.048) ii. Male x Treatment 0.058 -0.114 -0.145* (0.067) (0.080) (0.082) iii. Total effect on men ([i] + [ii]) 0.089* -0.074 -0.130* (0.050) (0.062) (0.067) Mean of Dep. Var (control) 0.000 0.000 0.000 Adjusted R-squared 0.341 0.101 0.178 Observations 2,280 2,280 2,280 Strata FE Yes Yes Yes Baseline Controls No No No Dep. Var Baseline Level Yes Yes Yes This table shows the differential impacts of the intervention on primary outcomes by gender. Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized average of the self-report scales of the emotional states of anxiety, stress, and depression (all items of DASS-21); in column (2) is the standardized sum of the self-report instrument of the Barratt Impulsiveness Scale BIS-11; and in column (3) is the standardized index of the responses to the 10 questions related to support for learning/stimulating environment and setting limits domain. All the regressions include interaction between male variable and treatment status. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 68 Table A17: Effect of the treatment on secondary outcomes without individual controls (1) (2) Tolerance of Child violent parenting behavior Treatment -0.021 0.067 (0.039) (0.042) Mean of Dep. Var (control) -0.005 0.005 Adjusted R-squared 0.157 0.002 Observations 2,234 2,226 Strata FE Yes Yes Baseline Controls No No Dep. Var Baseline Level Yes No This table shows the impacts of the intervention on secondary outcomes. Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized index of the responses in the ICAST-P module and the vignettes. The dependent variable in column (2) is the standardized index of the seven items that are reported by the caregiver from the internalizing/externalizing behaviors sections of the Parent/Caregiver Report Survey. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 69 Table A18: Effect of the treatment on exploratory outcomes without individual controls (1) (2) (3) (4) (5) (6) Positive Negative Physical Psychological Interest in Child parenting parenting violence violence additional socioemotional materials development Treatment 0.005 0.007 -0.050 0.027 -0.013 -0.100* (0.041) (0.039) (0.038) (0.040) (0.016) (0.059) Mean of Dep. Var (control) 0.007 -0.008 0.002 -0.007 0.841 3.779 Adjusted R-squared 0.038 0.099 0.127 0.073 0.023 0.010 Observations 2,265 2,245 2,270 2,251 2,280 1,094 Strata FE Yes Yes Yes Yes Yes Yes Baseline Controls No No No No No No Dep. Var Baseline Level Yes Yes Yes Yes No No This table shows the impacts of the intervention on secondary outcomes. Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized index of the seven items of positive parenting techniques; in column (2) is the standardized index of nine items of negative parenting techniques (physical and psychological violence); in column (3) is the standardized index of the two items of physical abuses; in column (4) is the standardized index of the seven items of psychological violence (emotional abuses); in column (5)is a dummy equal to one if the individual is interested in additional materials; and in column (6) is the standardized index of the nine items of typical socioemotional development. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 70 Table A19: Heterogeneous effects across genders: Primary outcomes (1) (2) (3) (4) (5) (6) Mental Health Distress Index Anxiety Stress Depression Impulsiveness Positive (A + S + D) (A) (S) (D) caregiver-child interactions Panel A: Male observations Treatment 0.089* 0.091* 0.102* 0.035 -0.071 -0.129* (0.050) (0.051) (0.053) (0.052) (0.062) (0.067) [0.084] [0.076] [0.067] [0.514] [0.261] [0.071] Mean of Dep. Var (control) -0.209 -0.181 -0.199 -0.169 0.020 -0.078 R-squared 0.288 0.250 0.216 0.229 0.090 0.175 Observations 892 892 892 892 892 892 Panel B: Female observations Treatment 0.032 0.002 0.046 0.022 0.034 0.020 (0.045) (0.046) (0.046) (0.047) (0.051) (0.047) [0.500] [0.962] [0.305] [0.657] [0.514] [0.665] Mean of Dep. Var (control) 0.133 0.115 0.126 0.108 -0.013 0.050 R-squared 0.354 0.339 0.299 0.266 0.122 0.185 Observations 1,388 1,388 1,388 1,388 1,388 1,388 This table shows the estimated impacts of the intervention on primary outcomes. Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized average of the self-report scales of the emotional states of anxiety, stress, and depression (all items of DASS-21); in column (2) is the standardized average of the self-report scales of the emotional state of anxiety (7 items of DASS-21); in column (3) is the standardized average of the self-report scales of the emotional state of stress (7 items of DASS-21); in column (4) is the standardized average of the self-report scales of the emotional state of depression (7 items of DASS-21); in column (5) is the standardized sum of the self-report instrument of the Barratt Impulsiveness Scale BIS-11; and in column (6) is the standardized index of the responses to the 10 questions related to support for learning/stimulating environment and setting limits domain. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 71 Table A20: Effect of the treatment on parental quality outcomes (1) (2) (3) (4) (5) Total Physical Psychological Tolerance of Child violence violence violence violent parenting behavior Panel A: Male observations Treatment 0.019 0.023 0.015 -0.015 0.063 (0.065) (0.061) (0.067) (0.066) (0.065) [0.780] [0.708] [0.832] [0.792] [0.302] Mean of Dep. Var (control) -0.073 -0.047 -0.061 0.070 -0.052 R-squared 0.097 0.142 0.071 0.172 0.009 Observations 878 886 881 877 866 Panel B: Female observations Treatment -0.008 -0.095* 0.027 -0.018 0.060 (0.049) (0.049) (0.050) (0.048) (0.056) [0.890] [0.061] [0.602] [0.696] [0.290] Mean of Dep. Var (control) 0.034 0.034 0.028 -0.053 0.041 R-squared 0.108 0.131 0.083 0.155 0.007 Observations 1,367 1,384 1,370 1,357 1,360 Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized index of nine items of negative parenting techniques (physical and psychological violence); in column (2) is the standardized index of the two items of physical abuses; in column (3) is the standardized index of the seven items of psychological violence (emotional abuses); in column (4) is the standardized index of the responses in the ICAST-P module and the vignettes; and in column (5) is the standardized index of the seven items that are reported by the caregiver from the internalizing/externalizing behaviors sections of the Parent/Caregiver Report Survey. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. 72 Table A21: Heterogeneous effects by economic deprivation and gender: Primary out- comes (1) (2) (3) Mental Health Impulsiveness Positive Distress caregiver-child interactions Economic deprivation 0.035 0.066 0.190*** (0.072) (0.077) (0.068) Treatment -0.032 0.050 0.036 (0.053) (0.063) (0.060) Male x Treatment -0.000 -0.114 -0.215** (0.081) (0.101) (0.103) Economic deprivation x Male -0.174* 0.024 -0.304** (0.102) (0.122) (0.121) Economic deprivation x Treatment 0.158 -0.025 -0.124 (0.104) (0.113) (0.104) Economic deprivation x Male x Treatment 0.230 -0.022 0.197 (0.152) (0.174) (0.181) Treatment effects Without economic deprivation women -0.032 0.050 0.036 With economic deprivation women 0.126 0.025 -0.088 Without economic deprivation men -0.032 -0.064 -0.178** With economic deprivation men 0.356* -0.111 -0.106 Differences in treatment effects Men - Women, Without economic deprivation -0.000 -0.114 -0.215** Men - Women, With economic deprivation 0.230* -0.136 -0.018 With economic deprivation - Without economic deprivation, Women 0.158 -0.025 -0.124 With economic deprivation - Without economic deprivation, Men 0.388*** -0.047 0.072 Mean of Dep. Var (control) 0.003 -0.004 0.010 Adjusted R-squared 0.348 0.103 0.179 Observations 2,172 2,172 2,172 Strata FE Yes Yes Yes Baseline Controls Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if assigned to the control group. The dependent variable in column (1) is the standardized average of the self-report scales of the emotional states of anxi-ety, stress, and depression (all items of DASS-21); in column (2) is the standardized sum of the self-report instrument of the Barratt Impulsiveness Scale BIS-11. In column (3) is the standardized index of the responses to the 10 questions related to support for learn- ing/stimulating environment and setting limits domain. The variable Economic deprivation is a dummy equal to one if the individual responded that from the beginning of the pandemic in their house, there was not enough money for food, home services, education,and others. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category. Robust standard errors-. *** p<0.01, ** p<0.05, * p<0.1. 73 Table A22: Heterogeneous effects by partner: Primary outcomes (1) (2) (3) Mental Health Impulsiveness Positive Distress caregiver-child interactions Living with a partner -0.049 -0.071 0.083 (0.068) (0.075) (0.069) Treatment 0.022 0.048 0.082 (0.077) (0.082) (0.079) Male x Treatment -0.232* -0.379** -0.034 (0.141) (0.174) (0.191) Living with a partner x Male -0.226* -0.127 0.219 (0.121) (0.142) (0.153) Living with a partner x Treatment 0.019 -0.021 -0.102 (0.094) (0.104) (0.099) Living with a partner x Male x Treatment 0.349** 0.332* -0.123 (0.161) (0.198) (0.212) Treatment effects Women not living with a partner 0.022 0.048 0.082 Women living with a partner 0.040 0.027 -0.020 Men not living with a partner -0.210* -0.331** 0.048 Men living with a partner 0.157* -0.020 -0.177* Differences in treatment effects Men - Women, Not living with a partner -0.232 -0.379** -0.034 Men - Women, Living with a partner 0.117 -0.047 -0.157* Living with a partner - Not living with a partner, Women 0.019 -0.021 -0.102 Living with a partner - Not living with a partner, Men 0.368*** 0.311* -0.225 Adjusted R-squared 0.346 0.103 0.180 Observations 2,280 2,280 2,280 Strata FE Yes Yes Yes Baseline Controls Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Treatment is a dummy equal to one if the individual is randomly assigned to the treatment group and zero if to the control group. The dependent variable in column (1) is the standardized average of the self-report scales the emotional states of anxiety (7 items of DASS-21); in column (2) is the standardized average of the self-report scales the emotional states of stress (7 items of DASS-21); in column (3) is the standardized average of the self-report scales the emotional states of depression (7 items of DASS-21). The controls include age in years, girls and boys cared for by the respondent and the educational level of the respondent for primary, high-school or tertiary education -the omitted category-. Robust standard errors-. *** p<0.01, ** p<0.05, * p<0.1. 74 Table A23: Instrumental variable regression (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Mental Health Distress Index Anxiety Stress Depression Impulsiveness Positive Total Physical Psychological Tolerance of Child (A + S + D) (A) (S) (D) caregiver-child violence violence violence violent behavior interactions parenting % known management techniques 0.480 0.519 0.881 -0.545 -0.725 -0.006 -1.006 -1.645** -0.610 -2.389*** 0.686 (0.558) (0.573) (0.584) (0.589) (0.687) (0.615) (0.672) (0.663) (0.701) (0.744) (0.627) Treatment for fist-stage 0.126*** 0.124*** 0.127*** 0.127*** 0.129*** 0.132*** 0.128*** 0.130*** 0.129*** 0.124*** 0.136*** (0.023) (0.023) (0.023) (0.023) (0.023) (0.023) (0.023) (0.023) (0.023) (0.024) (0.023) Kleibergen-Paap rk Wald F statistic 29.88 29.27 30.32 30.17 31.40 33.72 31.19 32.22 31.72 27.75 35.77 R-squared 0.328 0.347 0.243 0.177 0.072 0.221 0.073 0.052 0.059 -0.068 -0.042 Observations 529 529 529 529 529 529 523 526 525 516 517 Strata FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Baseline Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Dep. Var Baseline Level Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes No The dependent variable in column (1) is the standardized average of the self-report scales of the emotional states of anxiety, stress, and depression (all items of DASS-21); in column (2) is the standardized average of the self-report scales of the emotional state of anxiety (7 items of DASS-21); in column (3) is the standardized average of the self-report scales of the emotional state of stress (7 items of DASS-21); in column (4) is the standardized average of the self-report scales of the emotional state of depression (7 items of DASS-21); in column (5) is the standardized sum of the self-report instrument of the Barratt Impulsiveness Scale BIS-11; in column (6) is the standardized index of the responses to the 10 questions related to support for learning/stimulating environment and setting limits domain; in column (7) is the standardized index of nine items of negative parenting techniques (physical and psychological violence); in column (8) is the standardized index of the two items of physical abuses; in column (9) is the standardized index of the seven items of psychological violence (emotional abuses); in column (10) is the standardized index of the responses in the ICAST-P module and the vignettes; and in column (11) is the standardized index of the seven items that are reported by the caregiver from the internalizing/externalizing behaviors sections of the Parent/Caregiver Report Survey. The controls include age in years, girls and boys cared for by the respondent, and the educational level of the respondent for primary, high school, or tertiary education—the omitted category.*** p<0.01, ** p<0.05, * p<0.1. 75 Figure A2: Example SMS/WhatsApp Messages 76 Figure A3: Variation of primary outcomes (Baseline) 77 Figure A4: Variation of parental quality outcomes (Baseline) 78