WPS6257 Policy Research Working Paper 6257 Impact Evaluation Series No. 74 Promoting Handwashing Behavior in Peru The Effect of Large-Scale Mass-Media and Community Level Interventions Sebastian Galiani Paul Gertler Alexandra Orsola-Vidal The World Bank Sustainable Development Network Water and Sanitation Program November 2012 Policy Research Working Paper 6257 Abstract This paper analyzes a randomized experiment that communications campaign, was successful in reaching uses novel strategies to promote handwashing with the target audience with handwashing promotion soap at critical times in Peru. It evaluates a large-scale messages and in improving the knowledge of the treated intervention that includes a mass media provincial population on appropriate handwashing behavior. Those campaign and a district-level community component. improvements translated into higher self-reported and The analysis finds that the mass media intervention alone observed handwashing with soap at critical junctures. had no significant effect on exposure to the handwashing However, no significant improvements in the health of promotion campaign messages, and therefore no effect children under the age of five were observed. The results on handwashing knowledge or handwashing behavior. are consistent with earlier literature, which indicates that In contrast, the community-level intervention, a more substantively changing behavior to improve health is a comprehensive intervention that included several complex task requiring intensive and more personalized community and school activities in addition to the interventions. This paper is a product of the Water and Sanitation Program (WSP), Sustainable Development Network. WSP is a multi- donor partnership created in 1978 and administered by the World Bank to support poor people in obtaining affordable, safe, and sustainable access to water and sanitation services. 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://econ.worldbank.org. For more information, please email Sebastian Galiani at galiani@econ.umd.edu or Alexandra Orsola-Vidal at aorsolavidal@worldbank.org. The Impact Evaluation Series has been established in recognition of the importance of impact evaluation studies for World Bank operations and for development in general. The series serves as a vehicle for the dissemination of findings of those studies. Papers in this series are part of the Bank’s Policy Research Working Paper Series. 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 Promoting Handwashing Behavior in Peru: The Effect of Large-Scale Mass-Media and Community Level Interventions† Sebastian Galiani 1 Paul Gertler 2 Alexandra Orsola-Vidal 3 JEL codes: I Key words: hygiene, handwashing, behavior change, randomized evaluation 1 University of Maryland. 2 University of California, Berkeley. 3 World Bank, WSP. † This trial is registered, number NCT01465204. Sebastian Galiani is Professor of Economics at the University of Maryland and Alexandra Orsola- Vidal is Senior Impact Evaluation Specialist at the World Bank. They served as country investigators for the Peru Global Scaling Up Handwashing Project impact evaluation. This evaluation is part of a cross-country study led by Bertha Briceño (in its early stages led by Jack Molyneaux) together with Alexandra Orsola-Vidal and Claire Chase. Professor Paul Gertler has provided guidance and advice throughout the project. Global advisors also include Sebastian Galiani, Jack Colford, Ben Arnold, Pavani Ram, Lia Fernald, and Patricia Kariger. The authors are grateful to Rocio Florez, the project task team leader in Peru, Glenn Pearce-Oroz, Regional Team Leader for WSP-Latin America and Caribbean region, and to Eduardo Perez, overall task team leader. We also thank Nicolas Ajzenman, Andres Drenik, Laura Jaitman and María Lucía Yanguas for their excellent research assistance. Generous financial support was provided by the Bill and Melinda Gates Foundation. 1 Introduction Hygiene is essential to the public health mission of reducing the transmission and consequences of disease. The two leading causes of childhood mortality worldwide are diarrheal disease and acute respiratory infections,4 accounting for two-thirds of the deaths of children under age five.5 Both of these categories of illness are closely associated with inadequate hygiene. In addition, chronic parasitic infections and diarrhea can lead to anemia, which further hinders children’s development.6 The provision of safe water and sanitation, and improved hygienic behaviors more generally, has the potential to alleviate the proximate causes of these illnesses and thereby improve health.7 Medical evidence suggests that the hands are the main transmitters of diarrhea and respiratory infections. As such, they constitute disease vectors carrying respiratory microorganisms and fecal material into the domestic environment of the susceptible child. 8 Health experts recommend handwashing with soap as a critical action in protecting public health because it is a mainstay in infection control. 9 Yet, rates of handwashing with soap at critical times remain low throughout the world, even when both soap and water are available.10 In a sample of developing countries, the observed rates of handwashing with soap 4 Thirty-five percent of these deaths are due to diarrheal disease and 32 percent are due to acute respiratory infections. It was estimated that 21 percent of deaths in the 42 countries with the highest mortality are due to pneumonia (Black et al. 2003). 5 WHO 2002. 6 Curtis and Cairncross 2003. 7 See, for example, Esrey et al. 1991 and Galiani et al. 2005. 8 Hendley et al. 1973, WHO 2003. 9 Most parasites live and breed in feces and are transmitted to humans when ingested or through the hands. Parasitic infestations pose serious threats to young children and are a cause of child mortality (World Bank 2005). Among those who survive, parasitic infestations are associated with diarrhea and micronutrient malnutrition, which often leads to iron-deficiency anemia, protein-energy malnutrition, and enlargement of the liver and spleen (see, among others, Anderson and May 1991; and Hesham et al. 2004). Currently, anemia, which leads to slow cognitive development, is a widespread global health problem (see, among others, Nokes et al. 1992, and Pollit 1990). In addition, some respiratory tract infections, including the SARS-causing corona virus, are transmitted via the fecal-oral route or simply on hands (World Bank 2005). Hence, adequate handwashing is a primary barrier to the transmission of enteric pathogens, as hands can be cleansed of viruses and bacteria by washing with soap (Curtis and Cairncross 2003), (Faix 1987; Ansari et al. 1989; Luby et al. 2001; Gibson et al. 2002; Montville et al. 2002; Larson et al. 2003). 10 Scott et al. 2003. 2 range between 0 and 34 percent after defecation and 3 and 37 percent after cleaning up a child. 11 This paper focuses on handwashing promotion intervention, which attempts to improve child health by changing this key hygiene-related behavior rather than by implementing large and costly infrastructure investments in water supply or sanitation. In particular, we analyze the Global Scaling Up Handwashing Project in Peru, a large-scale intervention that aims to generate and sustain handwashing with soap behavior at critical junctures among mothers, caregivers, and children up to 12 years old in rural households. By trying to change the underlying factors that determine handwashing behavior (such as knowledge, beliefs, and accessibility of soap and water) among the target population, which is expected to result in improved handwashing behavior, the program tries to disrupt the causal chain that links poor hygiene with parasitic and microbiological disease transmission and thus ill health in children. Thereby, the study focuses on the following relevant outcomes: exposure to the intervention, changes in determinants of handwashing behavior (knowledge, beliefs, and access to/placement of soap and water), handwashing behavior (self-reported and observed), environmental contamination, and child health. To measure these behaviors and outcomes, we combined different quantitative and qualitative data collection techniques, including detailed questionnaires, structured observations, microbiological analysis of samples of children’s stools, capillary blood, and drinking water collected in the field, and anthropometric measures performed by health experts. The Global Scaling Up Handwashing Project in Peru intervention consisted of two main components: • a province-level mass media campaign; and • a more comprehensive district-level community treatment that included, in addition to the media campaign, training of community agents; capacity building of mothers, caregivers, and children; and the inclusion of handwashing promotion as part of primary school curricula. 11 World Bank 2005. 3 The project also included supportive activities such as partnership building and policy reform that aimed to create an enabling environment capable of inducing and sustaining appropriate handwashing behavior. This study does not evaluate these enabling environment activities, but focuses exclusively on the effect of the mass media campaign and community treatment. The results of this evaluation show that the mass media intervention alone was not effective in reaching the targeted population with the handwashing messages and therefore failed to improve knowledge and beliefs of mothers and caregivers regarding appropriate handwashing, or to generate a behavior change that could improve child health. However, the district-level interventions also analyzed in this paper proved to be effective in reaching the targeted audience with handwashing promotional messages. The community-level treated group was 19.4 percent more likely to report having received the intervention messages through at least one communication channel than the control group. Handwashing campaigns and promotional events at the community level and one-to-one activities seemed to have successfully transmitted the importance of handwashing with soap. Not only did the message reach the treated population, but it also improved caregivers’ knowledge about the best way to wash hands by 6 percent, and increased the availability of soap and water in the household by 8.4 percent. These improvements led to a statistically significant increase in self-reported and observed handwashing with soap before food contact as compared to the control groups. However, this behavior change did not translate into better child health. No significant impact was found on parasite and bacterial prevalence in stools and drinking water (respectively), which is consistent with the fact that there were no treatment effects on diarrhea, nutrition, or anemia. These results are consistent with previous studies in the literature of randomized handwashing promotion campaigns, which typically find that handwashing does reduce diarrhea in children under five years old, but those campaigns usually require intensive and controlled interventions. For example, Ejemot et al. 12 reviewed 14 randomized trials, concluding that handwashing programs resulted in a 39 percent reduction in diarrhea 12 Ejemot et al. 2009. 4 episodes in children residing in institutions in high-income countries and a 32 percent reduction in such episodes in children living in communities in low- or middle-income countries. The authors suggest that the significant reduction is comparable to the effect of providing clean water in low-income areas. However, the community or institutional interventions studied required a high cost of monitoring and implementing and hence these authors (Ejemot et al. (2009)) conclude that larger scale and less demanding pilots should be performed in developing countries. Luby et al. 13 also showed that handwashing with soap reduces the incidence of acute respiratory tract infections, as well diarrhea, as a result of implementing an intensive and small-scale community-level intervention.14 Others have studied school-level interventions. Bowen et al.15 evaluated a school program in China, in which 87 Chinese schools were randomized to a handwashing program that included training for teachers, in-class sessions, encouraging handwashing at school, and a pack for the children’s families that included soap; or to an expanded intervention (handwashing program, soap for school sinks, and peer hygiene monitors). 16 They found that 13 Luby et al. 2005. 14Luby et al. (2005) studied the causal impact of handwashing with soap in child health through a randomized control trial in Pakistan. Fieldworkers visited households weekly for a year to encourage handwashing with soap by residents and to record symptoms in all households. The authors found that children younger than 5 in households that received soap and handwashing promotion had a 50 percent lower incidence of pneumonia, and children younger than 15 had a 53 percent lower incidence of diarrhea than the children in control households. 15 Bowen et al. 2007. 16 Bowen et al. (2007) evaluated the Procter & Gamble’s Safeguard promotion program. Teachers presented the program to first-grade children during a single 40-minute classroom session. Children were instructed in handwashing behavior and asked to wash their hands before meals and after using the toilet. The program included a single two-hour training session for each first-grade teacher by Procter & Gamble staff and provision of a teacher’s pack, animated videotape for classroom use, and a take-home pack for each student. The teacher’s pack contained a guidebook outlining five handwashing steps (wet hands, lather fingers, lather palms and backs of hands, rinse, and dry with a clean towel) and basic information about infectious disease transmission. It also contained five posters describing handwashing procedures and five wall charts designed for classroom hygiene competitions. The student take-home pack included a hygiene board game, a parents’ booklet about handwashing, and a 50-gram bar of Safeguard soap. Although no significant differences in symptoms were found among in-class illnesses, absence incidence decreased in 44 percent for the standard intervention and in 42 percent for the extended treatment. The decrease in days absent was statistically significant only for the extended treatment (decrease of 52 percent). Syndrome-specific absence incidence also differed between groups: students in the standard intervention group were less likely than control students to be absent due to fever and students in the expanded intervention group were less likely to be absent due to headaches or stomachaches than the control students. 5 the expanded intervention significantly reduced syndrome-specific absence incidence (absence due to stomachaches or headaches). To the best of our knowledge, the studies that so far have found handwashing programs to have significant effects on child health have focused on interventions that impose controlled conditions in small populations over short time periods. These studies are akin to efficacy trials in drug development, which evaluate the impact of a specific intervention under ideal conditions. In most published handwashing studies, the participants are visited each week over a period of months. This style of intense promotion can cause important behavior changes that we would not necessarily expect under non-study conditions where interaction with handwashing promoters is less frequent. Thus, although intensive handwashing interventions have proven effective in reducing diarrhea and acute lower respiratory infections (ALRIs), it has not been proven that similar results could be obtained if those interventions were implemented at scale. This paper, however, studies the effectiveness of a national handwashing campaign to learn the impacts of large-scale handwashing interventions in a real-world context. Thus, this paper is the first to assess the effect of a large-scale handwashing intervention on a wide range of health indicators. Furthermore, to the best of our knowledge, we are also the first to study other intermediate outcomes, such as the campaigns’ effectiveness and behavior change, which provides important insights on the full theoretical causal chain of disease transmission and ill health.17 The rest of the paper is organized as follows. Section 2 details the program components. In section 3 we explain the experimental design. Section 4 describes relevant issues concerning the data and sampling procedure and presents some descriptive statistics. Section 5 describes the results of the baseline balance checks and provides an analysis of the panel sample attrition. Section 6 describes the methodological framework. In section 7 we show and interpret the main results of the interventions. Section 8 concludes. See Cattaneo et al. (2009) for another example of a study of the causal path of an intervention on child health. 17 In this case, the authors studied the effect of improving the floor of houses with cement. 6 2 Background and Description of the Program In response to the preventable threats posed by poor sanitation and hygiene, the Water and Sanitation Program (WSP) launched two large-scale projects in December 2006—the Global Scaling Up Handwashing Project and the Global Scaling Up Rural Sanitation Project—to improve child health and welfare outcomes of rural households around the world. These projects were implemented by local, regional, and national governments, with technical support from WSP. The Global Scaling Up Handwashing Project implemented in Peru, Vietnam, Tanzania, and Senegal, borrowed from commercial and social marketing to promote better hygiene. Communication campaigns and messages developed for this project were designed and strategically delivered across multiple integrated channels and in various settings in order to “surround� target audiences with handwashing promotion. Formative research conducted during 2007 with mothers and caregivers revealed that soap was not available for handwashing in most households in Peru, that there was a common belief that washing hands with water was sufficient, and that people did not know the critical times to wash hands with soap. In fact, the results of the structured observations in the baseline showed that soap was used in only 16 percent of the events in which it would have been necessary, that 20 percent of people were observed to wash their hands after fecal contact, 18 and that 25 percent of people washed their hands before eating.19 Thus, the programs studied had sufficient scope to improve the hygiene habits of the treated households. These rates of soap use are lower than the observed rates in more developed peri-urban areas (47 percent after cleaning up a child in Northern England20) and higher than the rates observed in poorer countries (10 percent after cleaning up a child in Nigeria, 3 percent after defecation in Ghana, or 1 percent in after toilet use in urban Burkina Faso 21). 18Fecal contact includes defecation, toileting of any kind, and cleaning a child who has defecated. 19At the baseline, structured observations of handwashing behavior were conducted in Peru as part of this study, in a subsample of 159 households (see Galiani and Orsola-Vidal 2010). 20 Curtis et al. 2003. 21 World Bank 2005. 7 In Peru, the Global Scaling Up Handwashing project was a national intervention implemented in a total of approximately 800 districts randomly selected (in 104 provinces). The project’s primary target audience consisted of mothers of reproductive age (15 to 49 years), caregivers of children under five years old, and children up to 12 years old. The project’s main objective was to improve handwashing behavior among the target audience in order to better the health of children under five. Children under five are the most susceptible to serious consequences from diarrhea and respiratory infection. These infections are usually transferred from dirty hands to food or water sources, or by direct contact with the mouth. Diarrheal disease and respiratory infection among children under five can be prevented by their mothers/caregivers washing their hands with soap at critical times, such as before feeding a child, cooking, or eating, and after using the toilet or changing a child. The intervention comprised two main components that were delivered at different administrative levels: a mass media plus direct consumer contact treatment at province level, and a community treatment at the district level. The activities included under each component of the project are as follows. 2.1 Province-Level Intervention: Mass Media plus Direct Consumer Contact Treatment A mass media plus direct consumer contact (DCC) communication campaign was implemented at the provincial level. Because research conducted before project implantation revealed that most people did not use soap when washing hands and that many mothers and caregivers thought using soap was not necessary, the communication strategy focused on the issue of nonuse. The campaign emphasized the importance of the availability and use of soap for handwashing, and the need to wash hands with soap immediately before cooking or eating and after fecal contact (going to the bathroom and changing a baby). The mass media plus DCC campaign targeted mothers, caregivers, and children, and the main communication channel was broadcast radio. Radio spots, lasting between 30 and 50 seconds each, were aired from five to nine times daily during the months of September to December 8 2009, April to June 2010, and August to November 2010. 22 The campaign also included print materials such as posters with reminders of key junctures in which to wash hands with soap, comic books, and brochures featuring a superhero cartoon character (Super Jaboncin) created especially for the campaign. Additionally, promotional events such as street parades, games, and local theater performances were conducted in public spaces, where Super Jaboncin came to life to promote handwashing and the use of soap. The jingles developed for the radio spots were used as the events’ background music, ensuring that the target audience received the same messages from multiple channels. On average, the events lasted from two to three hours and had audiences of 100 to 1,000 women and children. 2.2 District-Level Intervention: Community Treatment The community (or community and school) intervention was conducted at the district level and consisted of: • a mass-media plus DCC campaign; • training of trainers of community-based agents of change such as teachers, medical professionals, and community leaders; • capacity building and provision of educational handwashing sessions for mothers, caregivers, and children; and • handwashing curricula in select primary schools. With all these activities, the community intervention aimed to achieve an integral and sustainable change through different actors and channels. The communications campaign included broadcast radio advertisements with the same messages and frequencies described in the province-level treatment, and print materials such as posters with reminders of key junctures in which to wash hands with soap. It also included print materials featuring Super Jaboncin, the superhero cartoon character developed for the campaign. As part of the communications strategy, the community activities were intentionally tied to the handwashing promotional events. The events took place in different 22According to the baseline survey held for this study, 79 percent of the households had a radio, CD, or cassette player (see Appendix 1 for more information on household assets). 9 venues in the district and the audience might have been different each time or some people might have attended several events. In addition to the communication campaign, the community treatment included handwashing education sessions with groups of mothers, caregivers, and children. During these handwashing sessions, community-based agents, trained for such a purpose, demonstrated how to properly wash hands with soap, explained the critical junctures in which we must wash our hands with soap, and provided information on the extent to which improved handwashing behavior impacts infant health and welfare. The specialists in charge of conducting the handwashing promotional sessions included schoolteachers, health promoters, and local leaders who were trained as part of the community treatment to play a mediating role in influencing mothers’, caregivers’, and children’s handwashing behavior. In the districts that received the community treatment, a school-level treatment was delivered to the main primary schools in each district. In these schools, handwashing behavior was introduced as part of the school curricula. The activities included designating a place in the classroom for soap, performing regular handwashing practices in groups each day, weekly handwashing promotion classes, and other children’s activities such as singing songs and drawing posters. This school component of the community intervention tried to transmit the handwashing message to households of children attending the treated schools by including handwashing education in their formal studies. These components were implemented by national, regional, and local governments. The Water and Sanitation Program (WSP) provided technical assistance, but the intervention was mainly conducted by public and private partners who integrated these activities into the governments’ ongoing projects.23 23The Global Scaling Up Handwashing Project also included additional activities that were not evaluated as part of this study. These activities involved partnership building between public and private agents, capacity building of government entities, and policy reforms. These activities aimed to create an enabling environment that facilitated and sustained handwashing behavior with soap. These activities were assessed separately and the results published in a different study (Favin 2011). 10 3 Evaluation Study Design Estimating the causal relationship between the treatment and the outcomes of interest requires the construction of an accurate counterfactual—that is, a comparison group that shows what would have happened to the treatment group in the absence of the intervention. To account for factors external to the intervention, counterfactuals are created using comparison groups (control) that are equivalent to the treatment group on every dimension (observed and unobserved) but the treatment. Because a good counterfactual approximates what would have happened to the treatment group in the absence of the treatment, any differences in the average outcome measurements of the treatment and control groups following the implementation can be understood as the causal effect of the intervention. The randomization process, by which a random selection of communities receives the treatment and the remaining serve as controls, generates an appropriate counterfactual for the purposes of the impact evaluation. Thus, to assess the causal impact of each of the project components on a set of relevant variables, we conducted a controlled randomized trial comprising the province-level treatment as well as the district-level intervention, taking into consideration the general community treatment and its school component. 11 Figure 1: Design of the Experiment Universe: 193 Provinces (total of 195 Provinces excl. Pisco and Lima) Randomization 80 Provinces Randomization 40 Provinces 40 Provinces Randomization Randomization 41 Districts 44 Districts 40 Districts Control Group for "Mass District Level Intervention: Province Level Media -DCC alone" & Community Treatment Intervention: Community Treatments "Mass Media -DCC alone" Treatment Main Primary School: Main Primary School: Control Group for "School "School Component" Component" treatment The targeted areas were districts with populations ranging from 1,500 to 100,000 inhabitants. Figure 1 shows the experiment’s design. First, Pisco and Lima were excluded from the 195 Peruvian provinces. From the 193 remaining provinces, 80 provinces were 24 randomly selected, with 40 assigned to a first group and 40 to a second. From the first group of 40 provinces, 40 districts (with between 1,500 and 100,000 inhabitants) were randomly chosen to receive the mass media province-level treatment. From the second group of provinces, 85 districts (with between 1,500 and 100,000 inhabitants) were randomly selected, with 44 randomly assigned to receive the district-level community treatment and the other 41 randomly assigned to serve as control group for the mass media and community treatments. 25 In addition, for the 44 districts that were randomly assigned to the community treatment, the main school in each district received the school 24The province of Pisco was excluded because an earthquake had just hit the area. The province of Lima was excluded because most of its districts were too large for this type of intervention (more than 100,000 inhabitants), and its inhabitants were relatively wealthy. 25 Note that because every district had the same probability of being selected for the evaluation, small, poor districts were overrepresented in the sample. This was desired, however, given that the program was targeted toward the poor. 12 treatment. To create a counterfactual for the subgroup of households with children attending the main school in the treated districts, we also chose a subsample clustered around the main schools in the districts of the control group to serve as a control group for the school intervention. In what follows we assess the impact of the two treatments: mass media and community on a wide range of outcomes and we also study separately the school level component in the community intervention. We will refer to the households employed to explore the community treatment as community sample and to the households used to investigate the school component of the community treatment as the school sample. 4 Data Collection We conducted baseline and follow-up surveys. The baseline survey was conducted from May through August 2008 in a total of 3,576 households. The follow-up took place four months after the project activities ended—from March through June 2011. We located 2,847 of the original households interviewed for the baseline survey during the follow-up round. We replaced the households we could not find with new households that fit the following criteria: the family had lived in the dwelling for at least the two previous years (e.g., when the intervention took place) and had a child under two years of age at the time of the baseline survey. For the household-level sampling, in each of the 125 districts allocated to treatment and control groups, we randomly selected between 15 and 20 households from a census conducted prior to our baseline survey that listed all households with children under two years of age. From the 44 districts assigned to the community treatment, we randomly selected an additional set of 15 to 20 households with children under two years old and with at least one sibling attending the main treated schools in each district to assess the effect of the school subcomponent of the district-level intervention. To serve as counterfactual for this last group, in each of the 41 districts allocated to the control group, we randomly chose a second set of 15 to 20 households with children under two years old 13 and with at least one sibling attending a school similar to the main treated school in the treatment districts. Imasen, a local survey firm, conducted data collection together with a local institution specializing in nutrition (Instituto de Investigation Nutritional). The surveys were performed by more than a dozen teams, each consisting of a field supervisor, anthropometric and biometric technicians, household enumerators, and an observer for structured observations. All field team members received extensive training and followed standardized protocols supervised by the project’s investigators to ensure accurate and consistent data collection. 4.1 Variables of Interest and Measurement Data were collected on a wide range of variables and from different sources. First, a comprehensive household survey inquired about dwelling characteristics, household assets, education, income, labor outcomes, water sources, sanitation, and toilet and handwashing facilities, among other relevant variables (see Appendix 1 for the complete list). Also included were questions on exposure to handwashing campaigns, determinants for handwashing behavior—handwashing knowledge, beliefs, and access to and placement of soap and water—and caregiver self-reported handwashing behavior. To complement the handwashing behavior module, structured observations, during which the family’s handwashing behavior within the household was observed for four to five hours, were conducted on another day following the survey. During that period, the interviewer noted in detail any handwashing event involving adults and children during her stay. 26 In addition, to measure environmental contamination, drinking water microbiology samples were collected inside the households at the time of the survey and analyzed in the field using a membrane filtration technique. The samples were analyzed to determine the presence of Escherichia coli and other types of coliforms. To assess the household situation with regard to children, caregivers were given a detailed questionnaire with questions about the child’s environment, maternal depression, the 26 Structured observations were only carried out for the treatment and control samples of the school treatment. 14 relationship between family and school, and child health. We were mostly interested in symptoms directly or indirectly related to poor handwashing habits. We asked about diarrhea and ALRI prevalence in the last 48 hours and in the last week. Both illnesses are mainly transmitted through the hands—that is, the hands constitute disease vectors carrying respiratory microorganisms and fecal material. 27 We also performed anthropometric measurements of children following standard international procedures, because the synergistic relationship between malnutrition and infection is usually exacerbated in diarrheal episodes given that children tend to eat less during episodes and their ability to absorb nutrients is reduced. 28 These measurements helped us detect the children’s nutritional situation. A more detailed assessment of the presence of intestinal parasitic infestations in children was achieved by the laboratory analysis of children’s stools. Stool samples 29 were also collected during the interviews in duplicates and sent to Lima for parasitical analysis, using quantitative eggs per gram estimates of soil transmitted helminthes, quantified using Kato-Katz microscopy tests. We registered the prevalence of parasites and the parasite count in stool samples of children under five years of age. Parasitic infestations pose serious threats to young children’s health because the associated diarrhea and micronutrient malnutrition often leads to iron-deficiency anemia. 30 We therefore also measured anemia in children between six months and five years of age. Health specialists analyzed hemoglobin concentrations from the children’s capillary blood using the HemoCue Hb 201 photometer, a portable device that allows for immediate and reliable quantitative results. 27 Hendley et al. 1973; WHO 2003. 28 WHO 2003. 29 Stool samples were collected only for children under five years of age in the treatment and control groups of the school treatment. 30 See, for example, Hesham et al. 2004. 15 4.2 Descriptive Statistics Peru is a middle-income country with an estimated per capita income of US$5,195 in 2010.31 Peru has a high human development index (HDI) score of 0.725 based on 2011 data (the index for Latin America and the Caribbean is 0.731 and for the world is 0.682).One-third of the population is poor (31.3 percent) and 9.8 percent is extremely poor. 32 The wealthiest region of the country is Lima, Peru’s capital city, which is excluded from this study. The mean years of schooling for adults older than 25 is 8.9 years and is increasing over time. Most of the children attend school (93 percent primary school and 76 percent high school) according to 2010 data. 33 According to the baseline survey, the households analyzed averaged 5.3 members, with 1.4 children under age five.34 Around one-half of household heads had attained secondary education and most (95 percent) were employed. The average monthly household income per capita was 140 Peruvian nuevos soles (around US$ 54). The population included in the study was poorer on average than the population of Peru, given that poverty is concentrated in the small districts and that we sampled families with at least one child under two years old. 35 Three-quarters of the households had access to improved sources of drinking water, whereas half of the households had improved sanitation. More than 20 percent of all households had no sanitation facilities of any type. Handwashing with soap behavior was not prevalent among the interviewed households. Although almost all caregivers report having washed their hands with soap at least once during the previous 24 hours, fewer than half confirmed having done so at times of fecal contact (46 percent of caregivers associated handwashing with soap with toilet use and 42 percent with cleaning up children). Self-reported handwashing with soap was higher at times of cooking or food preparation (68 percent), but lower when feeding a child (34 percent). Structured observations on the use of soap were substantially lower than self-reported rates. 31 IMF 2011. 32 INEI 2012. 33 INEI 2012. 34 For the complete list of variables analyzed, see Appendix 1. 35 Indeed, the average household in the evaluation sample is poorer than the average Peruvian household (see Galiani and Orsola-Vidal 2010). 16 Handwashing with soap was only observed in 16 percent of the events that required it. For instance, handwashing with soap was observed in only 20 percent of fecal contact events, 25 percent of eating events, 6 percent of child feeding events, and 10 percent of food preparation events.36 In 64 percent of the households, a handwashing facility—a designated place for handwashing—stocked with soap and water was observed within the dwelling or the yard. The higher the income, the closer the handwashing station was to the toilet or kitchen facility. More than half of the caregivers (53 percent) appeared to have clean fingernails and approximately 67 percent had clean hands or finger pads, whereas 75 percent of the children appeared clean at the time of the interview but 47 percent had dirty fingernails. Households with access to improved sanitation and water source presented lower counts of bacteria in their drinking water. When accounting for income levels, there was a declining trend of Escherichia coli (E. coli) counts with increased income 37. Parasitological analysis showed that on average, parasites were detected in 12 percent of the stool samples collected from children under two (the most frequent were Giardia and Blastocystis). Prevalence of parasites was lower among households with access to improved sanitation (7 percent) and water (8 percent) than those with unimproved sanitation (18 percent) or unimproved water (25 percent). The lowest prevalence of parasites was found in households with a handwashing station stocked with soap and water (3 percent) and highest in those without (29 percent). Concerning child health, 10 percent of children under the age of five were reported by their caregivers to have had diarrhea symptoms in the previous 48 hours and 18.4 percent in the past seven days. Prevalence of diarrhea was higher in households with unimproved sanitation (12 percent) and lower for those with improved sanitation (8 percent); however, diarrhea prevalence was not significantly lower in households with access to a handwashing station 37For a subsample of 159 households, we also analyzed samples from sources other than drinking water. Households with access to improved sanitation also presented lower counts of bacteria in hand rinse and on sentinel objects. Water and caregivers’ hand rinse samples from households with a handwashing station with soap and water had lower bacteria counts (see Galiani and Orsola-Vidal 2010). 17 with soap and water nor in households with access to improved water sources, compared to those without access. Diarrhea prevalence appeared to be uncorrelated with income, but it varied noticeably by geographic location. On average, 4 percent of children had presented ALRI symptoms in the previous 48 hours, and 6 percent in the previous seven days. ALRI prevalence increased for children living in households with unimproved sanitation and those with unimproved water sources. As with diarrhea, similar percentages of households had presented ALRI symptoms in the previous seven days, irrespective of whether they had a handwashing station stocked with soap and water. In addition to these, three-quarters of the samples taken from children younger than two years old indicated the presence of anemia (the proportion was slightly lower for households with improved sanitation). Finally, the average child was breastfed for 12 months, although more than 60 percent of caregivers gave their children instant formula during the first three days of life. Vitamin A was given to 23 percent of the children and iron supplements to 22 percent. On average, children living in dwellings without improved sanitation, an improved water source, or soap and water at a handwashing station tended to have a lower average z-score for each anthropometric measure included in the analysis as well as for child development indicators (including communication, social-personal, and gross motor skills). 5 Balance and Attrition 5.1 Baseline Balance Random assignment of the study participants to treatment and control groups ensures that in probability the groups will be similar in their observable and unobservable characteristics. In this section we check that the randomization was performed adequately, resulting in comparable groups. Appendix 1 shows the mean comparison tests 38 across treatment/control groups for an exhaustive set of variables included in the baseline survey. We compare the characteristics 38 The standard errors used in those tests were clustered at the district level, allowing the possibility of intradistrict correlation. 18 of households allocated to each treatment with the households of the corresponding control groups. We test for differences in a wide range of variables including socio- economic characteristics, dwelling facilities and materials, household assets, children’s situation (education, health, nutrition, cognitive development, childcare, and the relationship between family and school), and many variables directly or indirectly associated with handwashing habits. We obtained the variables reported through questionnaires, microbiological analysis of samples (drinking water and child stools), anthropometric measurements, and capillary blood sample of children (for anemia). The province-level mass media treatment and control comparison used more than 242 variables; in 39 (16.1 percent) of them, we rejected the null hypothesis of mean equality at the 10 percent significance level. This proportion is slightly lower for the district-level intervention: for the community treatment sample, 30 of 242 (12.4 percent), and for the school treatment sample, 29 of 251 (11.5 percent) were unbalanced at conventional significance levels. 39 The larger proportion of statistically significant mean differences in the mass media treatment could have been expected given that the randomization was at province level for this intervention, unlike the district-level randomization within the same provinces for the community treatment samples. One would expect to find less significant differences among districts within the same group of provinces than among districts in different provinces. The differences between treatment and control groups go in either direction without a clear pattern. The only apparent difference is that the control group for the mass media province treatment appears to perform better in terms of improved sanitation than the treatment group. However, the opposite occurs when comparing the school treatment sample. 39 The proportion of unbalanced variables is close to the standard of 10 percent usually acceptable in randomized experiments. It is important to note, however, that the variables measured are not independent (probably neither the balanced nor the unbalanced ones). 19 5.2 Attrition Analysis An important concern for any impact evaluation is the fact that participants sometimes drop out of the study before its completion. Our baseline and endline survey records show an overall attrition rate of 20 percent after three years. This rate is in line with the attrition rates found in other similar randomized studies of this type. 40 In addition to the fact that sample size is reduced, when attrition is systematically related to the outcomes under study or to the program itself, it can bias the estimates of the causal effects of the interventions studied. We followed the standard procedures in the impact evaluation literature to address this potential problem. First, we explored whether the attrition rates differed between the treatment and comparison groups for each of the interventions studied. As Appendix 2 shows, the proportion of households that stayed in each group ranges from 0.72 to 0.84, and we cannot reject the null hypothesis of equal level of attrition in the control and treatment groups at conventional significance levels in any of the samples studied. The fact that there was no differential attrition in the treatment and comparison groups is consistent with the hypothesis that attrition is ignorable, and it suggests that the estimate of the treatment effects will not be biased unless different types of households dropped out of the sample in the treatment and the comparison groups. 41 For example, if households with healthier children were more likely to drop out when they were assigned to the treatment group, this could bias the estimator of the program effect on health outcomes downward. Fortunately, this does not seem to be the case in our study. Baseline balance was achieved for the entire initial sample in most of the variables (see Appendix1), and in Appendix 2 we can see that baseline balance is also high for a set of variables for the non-attriters’ sample. This implies that those remaining in the sample and the attriters were balanced in terms of observable characteristics. 40 For example, Banerjee et al. (2007) find attrition rates of 17 and 18 percent, respectively, in the comparison and treatment groups in Vadodara (India) in the first year when they evaluated Balsakhi education program in India. 41 Angrist 1996 20 To make up for the loss in sample size due to attrition, we included 688 new households in the follow-up survey to replace households that dropped out. To confirm that the characteristics of the replacement households were balanced between treatment and control groups, we selected 32 variables that were presumably independent of the treatment and affected by more structural patterns (e.g., the dwelling’s characteristics). In most of them, we could not reject the null hypothesis of equality of means at conventional significance levels (see Appendix 2). 6 Methodology The treatment assignment in this study was random and, as shown in section 5, the randomization produced comparable groups in terms of observables for all the groups studied. Furthermore, the results of the previous section suggest that attrition is unlikely to bias the experiment’s results. Therefore, in the context of this randomized experiment, we can obtain the average treatment effect on the outcomes of interest by estimating the following simple regression model: where i indexes households or individuals, Y is any of the outcomes under study, T is the dummy variable indicating treatment assignment (equal to one for units in treated groups), γ is the parameter that captures the causal effect of the intervention considered on the outcome of interest (Y), X is a vector of control variables, and ε is the error term. We estimate robust standard errors clustered at the district level in all the analysis. The vector X comprises the following control variables: gender and education of the head of household, dummy variables for children’s ages in months, gender of children, an indicator variable for mother living in the house, rainfall 42 (at district level) and geographical dummies for region (jungle, coast, and mountain). 42Rainfalls are measured as the maximum rainfalls per district, averaging January, February, March, and April 2011. 21 The results presented throughout the paper are estimates of model (1). The results are robust to two alternative models: dropping all the control variables and dropping only the geographical dummies. In Appendix 3 we present the results for the three models as robustness checks. 7 Effects of the Global Project in Peru In this section we present the results for both the province-level and district-level interventions. We present the results following the relevant outcomes of the hypothesized causal chain: exposure to handwashing promotion, effects on handwashing determinants (handwashing knowledge and beliefs, and access to and placement of soap and water), handwashing behavior (self-reported and observed handwashing and hand cleanliness), environmental contamination (bacteria prevalence in drinking water), and child health (prevalence of diarrhea, ALRI, anemia, parasites in stools, nutrition, and anthropometric measures). 43 In addition to assessing the impact of the mass media and community interventions on the mentioned outcomes, we are interested in studying the effect of the school handwashing curricula component, which is part of the community intervention, separately. Thus, in Tables 1–5, �community treatment sample� refers to the sample of households employed to explore the comprehensive community treatment, and �school component sample� refers to the households sampled to investigate the community treatment’s school component. In the remaining of the paper, we will use school component and school treatment interchangeably. 7.1 Exposure to Handwashing Promotion The Global Scaling Up Handwashing Project in Peru uses a behavior change approach; therefore, to expect any project impact, the campaign and its messages must reach the target population. Thus, we asked mothers and caregivers whether they had received handwashing 43We do not pursue the study of child development outcomes because the health results suggest that those could not be causally interpreted as a response to the interventions studied. 22 promotional messages during the past 12 months, through any of the three channels used by the intervention: radio and printed materials, promotional events, and educational sessions. As Table 1 illustrates, the province-level mass media plus DCC treatment did not increase the exposure of the treated households to the handwashing promotion message through any of the channels proposed. Thus, starting with this finding, we do not expect to find any other causal effect that could be attributed to the mass media province-level intervention. The district-level intervention (mass-media plus DCC, and community treatment) was considerably more successful in exposing the target audience to the handwashing with soap message. The random sample exposed to the community-level treatment was the most affected as a result of this intervention, as the proportion of mothers or caregivers that reported having received handwashing messages through at least one (two) of the communications channels was 15.7 percent (33.5 percent) higher in the treatment group than in the control group. 44 We are also interested in assessing whether the messages imparted at schools reached the attending children—even though those children and their mothers/caregivers might also have been exposed to the other community-level activities. Table 1 shows that the school component increased the probability of low exposure to the treatment (that is, receiving the messages through one channel) by 8.9 percent. To compare this result with that of the community treatment, notice that the households in this treatment group have at least one child attending the school. These households are located in districts also affected by the community treatment, but because they are clustered in a certain area (around the main primary school), their probability of receiving the handwashing message through the other promotional activities in the district might be lower than the probability of receiving them for a random household in the district. 44Note, however, that the control group also reports high levels of exposure to similar messages. Although this might partially reflect measurement error, it might also reflect the fact that other, less intensive campaigns could have been implemented elsewhere. 23 7.2 Treatment Effects on Handwashing Determinants Improving handwashing behavior requires changing the factors that motivate that behavior. These factors are known as handwashing determinants, and include knowledge about the best way to wash hands, beliefs about whether soap is needed for effective handwashing, and the availability and placement of soap and water. Caregivers were asked about their knowledge of effective handwashing habits as well as the availability of water or soap in the household. This would constitute the second link—after exposure to information—in understanding whether the project could lead to a behavior change. Table 2 shows that the mass media plus DCC treatment was ineffective in improving the knowledge of the survey respondents or in increasing the availability of soap and water in households. This result is expected given that respondents did not show a higher exposure to the handwashing promotion campaign. Table 2 also shows that the district-level interventions increased not only the exposure to the treatment, but also the knowledge of mothers and caregivers regarding handwashing practices. For instance, in the treated districts, there was a significant (around 5 percent) increase in the proportion of respondents answering that the best method for washing hands is using water and soap. In addition, 3.3 percent more households in the treated districts knew that inadequate handwashing is the main cause of diarrhea. It is important to note that the mean of the control group for these knowledge variables is high: 88 percent of the caregivers answered that the best method for washing hands is with water and soap and 94 percent of the interviewees claimed that inadequate handwashing is the main cause of diarrhea. In these variables, we observe a modest gap of 12 and 6 percentage points, respectively, to total knowledge. If we measure the effect of the community intervention in terms of closing the knowledge gap, we can see that the program reduced this gap by as much as 50 percent (the gap on knowledge regarding the best method for washing hands was reduced by 42 percent and the gap about how inadequate handwashing can cause diarrhea was reduced by 50 percent). Finally, the community treatment increased the availability of water and soap in households by 8.4 percent (among the school component sample). Again, it is worth mentioning that 24 availability of soap and water anywhere in the dwelling is also high in the control group (e.g., 77 percent of the households in the control group for the school component of the community treatment have water and soap somewhere in the dwelling). Thus, the treatment reduced the gap by 28 percent, which constitutes remarkable progress. 7.3 Treatment Effects on Handwashing Behavior The next step in the causal chain is assessing whether the improvements in handwashing knowledge and the availability of water and soap brought about by the community treatment translated into handwashing behavior changes. Measuring handwashing behavior is complex. The simplest and most affordable method for measuring handwashing with soap is to ask respondents to self-report their behavior. However, respondents tend to over report their behavior, especially if they know that handwashing with soap is the right thing to do. Ideally, multiple methods should be used to obtain a more reliable rate of true handwashing behavior. This study uses four different measurement methods and proxies that vary in validity, reliability, and cost to obtain accurate rates of handwashing behavior.45 These measures include observation of handwashing facilities in the households, cleanliness of caregivers’ hands, self-reported handwashing with soap behavior, and direct structured observations of handwashing. Table 3 summaries the results for these four measures. Observations of handwashing facilities were conducted in all households. The mother or caregiver was asked to identify any place in the household designated for handwashing after fecal contact or before food contact. 46 When the answer was positive, the enumerator noted all the details about the handwashing facility, including the facility type, its distance from the toilet or kitchen, the availability of water and soap, and the type of soap. At baseline, the proportion of households with at least one handwashing facility near the toilet or kitchen was quite high (66 percent). The community treatment increased the share of households with 45Ram 2010 46Fecal contact includes using the toilet and cleaning a child’s bottom; food contact includes food preparation, eating, or feeding a child. 25 handwashing facilities by 4.9 percent (9.2 percent for the school component sample), but the increase was not statistically significant at conventional levels of significance. The enumerators also observed the hands of mothers and caregivers and recorded the cleanliness of their nails, palms, and finger pads. These observations were used to create a Hand Cleanliness Index.47 The district-level intervention had a positive and significant effect on the hand cleanliness of mothers, as respondents in the community treatment group were 3.6 percent more likely to have clean hands than those in the control group. Self-reported handwashing behavior was measured by asking mothers and caregivers under what circumstances they had washed their hands with soap in the previous 24 hours. Respondents were asked to recall every time they had used soap during handwashing. Analysis of the responses focused on self-reported handwashing with soap behavior during any of the critical junctures. As Table 3 shows, self-reported handwashing with soap behavior improved in the district-level interventions in almost all the critical junctures, but it was statistically significant only in the school treatment. Among the school component sample, 23.6 percent more respondents in the treatment group reported having washed their hands with soap before eating and 27.8 percent more reported having washed with soap before feeding a child, relative to the control group. As mentioned earlier, self-reported handwashing behavior tends to be over reported, so in order to triangulate the results, the study included structured observations of handwashing in a subsample of 600 households. Direct observations were conducted among the school component and its control group during a 4-to-5-hour period on the day after the main questionnaire was administered. Enumerators in charge of the structured observations received extensive training on being discreet and not revealing that the visit’s main focus was to observe handwashing behavior. The enumerators observed events that should be followed by handwashing and recorded whether the caregiver had actually washed hands, the time of the event, whether water and soap were used, whether hands were dried, and what was used 47 The Hand Cleanliness Index comprises the following components: a) Nails: +1 if visibly dirty, +2 if apparently dirty, +3 if clean; b) Palms: +1 if visibly dirty, +2 if apparently dirty, +3 if clean; and c) Fingerpads: +1 if visibly dirty, +2 if apparently dirty, +3 if clean. The index ranges from 3 to 9, with higher scores indicating greater cleanliness. 26 for drying. Table 3 summaries the results of these structured observations. As the results show, the district-level intervention had a significant and large effect on observed handwashing in two of the four main critical junctures. In particular, 61 percent more households in the treatment group washed their hands with soap before eating than those in the control groups. Similarly, treated households were 69 percent more likely to wash hands before food preparation than nontreated households. These effects are encouraging with respect to the potential of the community intervention in achieving behavior change. There is a possible caveat regarding the results of the structured observations, as individuals may have changed their behavior because they were observed. For instance, Clasen et al. show that during a field trial in India where they measured the number of latrine events with a device hidden in the latrine, the presence of a human observer was associated with a statistically significant increase in the number of latrine events (i.e., the users modified their behavior in response to the observer).48 In the context of this handwashing campaign in Peru, that type of change in behavior would be true for both the control and the treatment group. However, rates of handwashing with soap as measured by structured observations are substantially lower than self-reported rates, so there does not seem to be a high overreaction to the observer. What could still bias our causal estimates is a situation in which the treated group would react differently in the observer’s presence (probably showing better practices given that they received the handwashing message) relative to what they would do in the control group. If this were the case, our results still would confirm that the treated group had received the handwashing messages and knew when to practice handwashing with soap. Nevertheless, structured observations continue to be the gold standard in measuring handwashing behavior, and it provides a more reliable measure of true handwashing practices than self-reported rates. Thus, the results show that the community intervention was successful in achieving behavior change. For the mass media provincial-level intervention, we found no negative effects, which are not likely to be attributed to the treatment, as its messages did not reach the targeted audience. 48 Clasen et al. 2012. 27 7.4 Treatment Effects on Environmental Contamination Medical evidence suggests that the transmission of diarrhea and respiratory infections is mainly through the hands, which constitute disease vectors carrying respiratory microorganisms and fecal material to the domestic environment of the susceptible child. The use of soap reduces germs and bacteria in hands, thus reducing the risk of becoming sick. It is expected that those practicing handwashing with soap at key junctures will have a lower presence of bacteria such as Escherichia Coli or total coliforms on their hands, thereby reducing the risk of contamination when handling, preparing, or serving drinking water at home. Thus, the study collected samples of drinking water49 from a subsample of 600 households among the school component sample and its control group. These samples were collected during the administration of the questionnaire, and analyzed to detect the presence of Escherichia Coli and total coliforms. Consistent with previous results, the microbiological analysis of drinking water samples summarized in Table 4 shows that the prevalence of Escherichia Coli and of total coliforms was lower in the treated households than in those of the control group, but none of the effects are statistically significant at conventional levels. 7.5 Treatment Effects on Child Health The last step in the causal chain is to assess whether the positive results found on exposure to the campaign—knowledge of appropriate handwashing practices, access to soap and water, and handwashing with soap behavior—resulted in health improvements. Table 5 summarizes the effects of the treatments in a wide range of indicators of child health and well-being, including symptoms of illness, nutrition, parasite infestations, and anemia. The results show that overall none of the treatments had a significant effect on those variables. Because the mass media plus DCC provincial-level treatment alone had no effects on previous outcomes, we did not expect to find any impact on child health. If there were significant estimates, they would probably not be causally attributed to the treatment given that it did not reach the target audience (see Table 1). This is the case of a reduction in the prevalence of Drinking water is not the water at source, but rather the water that the household drinks directly. Drinking 49 water can be boiled or treated and is usually stored in containers or jars inside the household. 28 ALRI. In the baseline, this variable was not balanced, as households in the control group were performing significantly worse than those in the treatment group. In the follow-up survey, the situation reverted but there are no grounds to think that it was the effect of the program. This suggests that in this specific health dimension the two groups might not be comparable and that respiratory diseases are influenced by regional temporary variability. For the district-level intervention (community and school component samples), the improvement in knowledge and behavior change do not seem to have resulted in better health for the children. There are no statistically significant differences between the prevalence of diarrhea or ALRI among children in treated households and those not treated. Nor do we observe any important improvement in the anthropometric measurement50 or anemia indicators. The prevalence of parasites and the parasite counts obtained from the stool samples (of children under 5 years old) are not significantly different in the control and treatment groups. These results are consistent with the fact that the treatments had no effect on diarrhea, micronutrient malnutrition, or anemia, as these health problems are usually related to parasitic infestations. 51 8 Conclusion Some simple technologies, such as vaccines and mosquito bed nets, are cheap and effective preventative measures that can save lives at minimal costs. 52 Handwashing with soap at critical times is another of these technologies: soap is generally easy to use, is relatively cheap, and has the potential of substantially improving the living standards of households. Despite these potential benefits, however, only 3 to 34 percent of people in developing countries routinely wash their hands with soap at critical junctures during the day. 53 50The z-scores for the anthropometric measures were calculated discarding the lower and upper 1 percent of the distribution. 51 Unfortunately, a disaggregation of the prevalence of different parasites shows extremely large variability, suggesting that making inference at that level of disaggregation could be unreliable. Nevertheless, we report finding large drops in the treatment group in the prevalence of any Ascaris Lumbricoides infections and any Trichuris Trichuria infections, no differences in Hookworm infections, but a significant rise in the treatment group in any other parasites infection. 52 Banerjee and Duflo 2011 53 World Bank 2005 29 Although medical evidence about the benefits of handwashing with soap is clear, changing handwashing behavior is a complex phenomenon related to personal habits that has multiple determinants (economic, cultural, social, etc.). 54 These characteristics of handwashing result in important difficulties in changing and sustaining these hygienic practices. This paper studies a large-scale handwashing promotion intervention that took place in Peru between 2008 and 2010, and that introduced an innovative mix of communication strategies to induce handwashing with soap at critical times. In particular, the study examined the effect of two different treatments—a province-level mass media campaign and a district-level community treatment—on several outcomes: exposure to the program, handwashing determinants, handwashing behavior, environmental contamination, and child health. We combined different quantitative and qualitative data collection techniques to measure these variables, including detailed questionnaires; structured observations; microbiological analysis of samples of child stools, capillary blood, and drinking water collected in the field; and anthropometric measures performed by health experts. This paper is the first to assess the effect of a large-scale handwashing with soap intervention using such a wide range of indicators. Previous studies in the literature of randomized handwashing experiments focus on intensive (in terms of labor, inputs and/or monitoring) and controlled treatments, showing that they are effective in reducing diarrhea and pneumonia incidence in children.55 This study, in contrast, examines a significantly less intensive intervention at scale under real-world conditions. In addition, we study a full set of intermediate outcomes, such as the effectiveness of the campaigns and behavior change, in order to better understand the results in the context of the causal path of disease transmission identified in the medical and public health literature. To the best of our knowledge, ours is the first study of handwashing to focus on all components of this causal chain. The results of this evaluation show that the mass media intervention alone was not effective in reaching the targeted audience with the handwashing message and therefore failed to improve the handwashing knowledge of mothers and caregivers and to generate a behavior 54 Chapman 2010 55 Luby et al. 2005 30 change that could improve child health. The null results are in line with most of the existing evidence on health promotion campaigns, which typically show that they are ineffective in impacting behavior or that they, at best, enhance the knowledge about the promoted topic. (See, for example, Babor et al.’s work on alcohol prevention campaigns 56; Madajewicz et al.’s work on arsenic in water in Bangladesh57; Banerjee and Duflo’s work on vaccination campaigns in Zambia 58; and Summerbell et al.’s work on obesity prevention campaigns 59). In contrast, the community intervention, which combined the mass media campaign with more intense training and promotional activities delivered at the district level, proved to be more effective in reaching the targeted audience with the handwashing promotion message, significantly increasing the proportion of mothers and caregivers that reported receiving it. The community treated group reported receiving the message at least through one communication channel more than 15 percent more often than the control group. This more comprehensive treatment seemed to have successfully transmitted the key messages related to handwashing with soap. Increased exposure to the campaign and educational sessions translated into observable learning about best handwashing practices. For example, the proportion of mothers and caregivers that responded correctly that the best method for washing hands is using water and soap increased significantly (around 5 percent) as compared to the control group. Considering the generally good levels of knowledge about this among sample households, this 5 percent change means that almost half the remaining gap in 56 Babor et al. 2003 57 Madajewicz et al. 2007. In this paper, the authors also compared two types of information campaigns and found that the more personalized treatment was more effective. People exposed to a media information campaign obtained as much information as those exposed to a door-to-door campaign, but they did not change their behavior regarding checking whether the dwelling where they obtained water was safe and moving to another dwelling if it was not. The authors also studied a personalized campaign that provided information specific to the individual (whether her/his well is safe and which wells in the vicinity are safe), achieving behavior changes. 58 Banerjee and Duflo (2011) review, among others, the case of Zambia in which a big NGO distributed subsidized chlorine tablets widely, and as a result the campaign informed the population (98 percent identified it as a good way to clean drinking water). However, only 10 percent of families use the tablets. In addition, Banerjee et al. (2010) show that immunization campaigns in rural India are more effective when the poor are given small incentives apart from the vaccines (in this case, lentils and metal plates for completed immunization). 59 Summerbell et al. 2007. Two recent Cochrane reviews on childhood obesity (Campbell et al. 2001; Summerbell et al. 2005) find only modest evidence that diet education impacts the BMIs and consumption habits of schoolchildren, especially in the short term. 31 hygiene knowledge (between current levels and full awareness) has been closed. This improvement in knowledge led, in turn, to statistically significant behavior changes in key areas, such as an 8.4 percent increase in the availability of water and soap in the household, hand cleanliness, and observed and caregiver self-reported handwashing behavior before eating, feeding a child, and preparing food. In addition, observed handwashing with soap only increased among the treated households by 61% before eating and by 69% before preparing food, as compared to the control groups. However, these behavior changes did not translate into better child health. Overall, no impact was found on parasite and bacterial prevalence in stools and drinking water (respectively), which is consistent with the fact that there were no effects on diarrhea, micronutrient malnutrition, or anemia. The results of this study yield important insights for policymakers and practitioners. Although we did not observe treatment effects on health outcomes, the study successfully identified which component of the intervention was more effective in changing handwashing behavior. Handwashing with soap at critical junctures continues to be a preventive measure to improve child health, especially in developing countries with a high incidence of diarrhea. Thus, future efforts should be invested in identifying additional activities, delivery mechanisms, and/or more effective messages in order to ensure significant and sustainable changes in behavior that improve child health. 32 References Anderson, R., and R. May. 1992. Infectious Diseases of Humans: Dynamics and Control. Oxford, UK: Oxford University Press. Angrist, J. 1996. Conditioning on the Probability of Selection to Control Selection Bias. NBER Technical Paper: No. 181. Ansari, S., S. Sattar, V. Springthorpe, G. Wells, and W. Tostowaryk. 1989). In vivo protocol for testing efficacy of hand-washing agents against viruses and bacteria: Experiments with rotavirus and Escherichia coli. Applied and Environmental Microbiology 55, 3113–3118. Babor, T., R. Caetano, S. Casswell et al. 2003. Alcohol: No ordinary commodity: Research and public health. Geneva: Oxford University Press/World Health Organization. Banerjee, A., S. Cole, E. Duflo, and L. Linden. 2007. Remedying education: Evidence from two randomized experiments in India. The Quarterly Journal of Economics, Vol. 122 (3), pp. 1235– 1264. Banerjee, A., and E. Duflo. 2011. Poor Economics: A Radical Rethinking of the Way to Fight Global Poverty. Public Affairs, New York. Banerjee, A., R. Glennerster, and D. Kothari. 2010. Improving immunization coverage in rural India: A clustered randomized controlled evaluation of immunization campaigns with and without incentives. British Medical Journal, 2010; 340: c2220. Black, R., S. Morris, and J. Bryce. 2003. Where and why are 10 million children dying every year? Lancet 361, 2226–2234. Bowen, A., H. Ma, J. Ou, W. Billhimer, T. Long, and E. Mintz. 2007. A cluster-randomized controlled trial evaluating the effect of a handwashing-promotion program in Chinese primary schools. American Journal of Tropical Medicine and Hygiene 76(6): 1166–1173. Campbell, K., E. Water, S. O’Meara, and C. Summerbell. 2001. Interventions for preventing obesity in childhood: A systematic review. Obesity Reviews 2, 149–157 Cattaneo, M., S. Galiani, P. Gertler, S. Martinez, and R. Titiunik. 2009. Housing, health, and happiness. American Economic Journal: Economic Policy, American Economic Association, 1(1), pages 75–105, February. Chapman, S. 2010. Evaluating social marketing interventions. Evaluating Health Promotion: Practice and Methods, 3rd ed. Eds. Margaret Horogood and Yolande Coombes, 105–120. Oxford: Oxford University Press. Clasen, T., D. Fabini, S. Boisson, J. Taneja, J. Song, E. Aichinger, A. Bui, S. Dadashi, W. Schmidt, Z. Burt, and K. Nelson. 2012. Making sanitation count: Developing and testing a device for assessing latrine use in low-income settings. Environmental Science & Technology, forthcoming. 33 Curtis, V., S. Cairncross, and R. Yonli. 2000. Tropical Medicine and International Health 5 (1), 22–32, January. Curtis, V., and S. Cairncross. 2003. Effect of washing hands with soap on diarrhea risk in the community: a systematic review. The Lancet Infectious Diseases 3 (5), 275–281, May. Engle, P., M. Black, J. Behrman, M. Cabral de Mello, P. Gertler, L. Kapiriri, R. Martorell, and M. Young. 2007. Strategies to avoid the loss of developmental potential among 200 million children in the developing world. The Lancet 329, 229–240. Esrey, S., J. Potash, L. Roberts, and C. Shiff. 1991. Effects of improved water supply and sanitation on ascariasis, diarrhea, dracunculiasis, hookworm infection, schistosomiasis and trachoma. Bulletin of the World Health Organization 69, 609–621. Faix, R. 1987. Comparative efficacy of handwashing agents against cytomegalovirus. Infection Control 8, 158–162. Favin, M. 2011. Endline Assessment of the Enabling Environment in Peru. Water and Sanitation Program, World Bank, August 2010. Galiani, S., P. Gertler, and E. Schargrodsky. 2005. Water for life: The impact of the privatization of water services on child mortality. Journal of Political Economy 113(1), 83– 120, February. Galiani, S., and A. Orsola-Vidal. 2010. Scaling Up Handwashing Behavior: Findings from the Impact Evaluation Baseline Survey in Peru. Water and Sanitation Program, World Bank, August 2010. Gibson, L., J. Rose, C. Haas, C. Gerba, and P. Rusin. 2002. Quantitative assessment of risk reduction from hand washing with antibacterial soaps. Journal of Applied Microbiology 92, 136S–143S. Hendley, J., R. Wenzel, and J. Gwaltney. 1973. Transmission of rhinovirus colds by self- inoculation. The New England Journal of Medicine 288, 1361–1364. Hesham, M., A. Edariah, and M. Norhayati. 2004. Intestinal parasitic infections and micronutrients deficiency: A review. Medical Journal of Malaysia 59(2): 284–293. Instituto Nacional de Estadísticas e Informática (INEI). 2012. Condiciones de Vida en Peru, Octubre-Noviembre-Diciembre 2011. International Monetary Fund (IMF). 2011. World Economic Outlook. Luby, S., M. Agboatwalla, J. Paynter, A. Altaf, W. Billhimer, and R. Hoekstra. 2001. Effect of intensive handwashing promotion on childhood diarrhea in high-risk communities in Pakistan. JAMA, Journal of the American Medical Association 291(21), 2547–2554. 34 Luby, S., M. Agboatwalla, D. Feikin, J. Painter, W. Billhimer, A. Altaf, and R. Hoekstra. 2005. Effect of handwashing on child health: a randomized controlled trial. The Lancet 366, 225–233. Luby, S., A. Halder, T. Huda, L. Unicomb, and R. Johnston. 2011. The effect of handwashing at recommended times with water alone and with soap on child diarrhea in rural Bangladesh: An observational study. PLoS Med 8(6), e1001052; doi:10.1371/journal.pmed.1001052. Madajewicz, M., A. Pfaff, A. van Geen, J. Graziano, I. Hussein, H. Momotaj, R. Sylvi, and H. Ahsan. 2007. Can information alone change behavior? Response to arsenic contamination of groundwater in Bangladesh. Journal of Development Economics, 84(2), pp. 731–754. Montville, R., Y. Chen, and D. Schaffner. 2002.Risk assessment of hand washing efficacy using literature and experimental data. International Journal of Food Microbiology 73, 305– 313. Nokes, C., S. Grantham-McGregor, A. Sawyer, E. Cooper, and D. Bundy. 1992. Parasitic Helminth infection and cognitive function in schoolchildren. Proceedings of the Royal Society of London, Series B, Biological Sciences 247: 77–81. Pollit, E. 1990. Malnutrition and Infection in the Classroom. Paris, France: UNESCO. Ram, P. 2010. Guidance for Measuring Handwashing Behavior. The Water and Sanitation Program. Scott, B., V. Curtis, and T. Rabie. 2003. Protecting children from diarrhea and acute respiratory infections: the role of handwashing promotion in water and sanitation programmes. Regional Health Forum WHO South-East Asia Region 7,42–47. Summerbell, C., W. Waters, L. Edmunds, S. Kelly, T. Brown, and K. Campbell. 2007. Interventions for preventing obesity in children. Cochrane Database of Systematic Reviews, Issue 3 Art. No.: CD001871. DOI: 10.1002/14651858. CD001871.pub2. WHO. 2002. The World Health Report 2002: Reducing Risks, Promoting Healthy Life. World Health Organization, Geneva. WHO. 2003. Update 47—Studies of SARS Virus Survival, Situation in China. Available at: http://www.who.int/csr/sarsarchive/2003_05_05/en/ (accessed July 16, 2003). World Bank. 2005. The Handwashing Handbook: A Guide for Developing a Hygiene Promotion Program to Increase Handwashing with Soap. 35 Table 1: Effect of the Interventions on Exposure to Handwashing Promotion Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Effect of treatment Control mean Effect of treatment Control mean Effect of treatment + Exposure High exposure to the treatments 0.147 -0.0039 0.147 0.0551 0.2 -0.0275 (HW message through three channels) (0.037) (0.037) (0.041) Medium exposure to the treatments 0.432 -0.000929 0.432 0.145*** 0.499 0.065 (HW message through two channels) (0.058) (0.051) (0.050) [33.5%] Low exposure to the treatments 0.674 -0.065 0.674 0.106** 0.717 0.0637* (HW message through one channel) (0.051) (0.045) (0.042) [15.7%] [8.9%] + Channels: mass media (radio and/or printed materials), promotional events and personal training/educational sessions Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 Table 2: Effect of the Interventions on Handwashing Determinants Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Effect of treatment Control Mean Effect of treatment Control Mean Effect of treatment Knowledge Events that require handwashing (summary index) 0.225 0.02 0.225 0.009 0.222 0.008 (0.019) (0.020) (0.019) Best method to wash hands 0.879 -0.003 0.879 0.0512* 0.882 0.044** (0.021) (0.027) (0.021) [5.8%] [4.9%] Not washing hands with water and soup is main 0.94 -0.006 0.94 0.0313** 0.94 0.011 cause of diarrhea (0.014) (0.012) (0.014) [3.3%] Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 36 Table 3: Effect of the Interventions on Behavior Change Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Effect of treatment Control mean Effect of treatment Control mean Effect of treatment Handwashing facilities HW Facilities stocked with water and soap 0.653 -0.019 0.653 0.032 0.67 0.062 (0.057) (0.048) (0.043) Water and soap anywhere in the house 0.787 -0.025 0.787 0.0267 0.768 0.065** (0.040) (0.037) (0.032) [8.4%] HW hand cleanliness Hands Cleanliness Index 7.527 -0.178 7.527 0.278* 7.403 0.210 (0.166) (0.161) (0.151) [3.6%] Self-Reported handwashing behavior HW with soap and water previous to eat 0.415 -0.156*** 0.415 0.0252 0.402 0.095** (0.040) (0.046) (0.044) [-37.5%] [23.6%] HW with soap and water before food preparation 0.653 -0.007 0.653 0.027 0.694 -0.042 (0.037) (0.036) (0.039) HW with soap and water after fecal contact 0.654 -0.084** 0.654 0.0254 0.656 0.019 (0.040) (0.042) (0.043) [-12.8%] HW with soap and water before feeding child 0.206 0.037 0.206 0.004 0.194 0.054* (0.031) (0.034) (0.030) [27.8%] Structured observations Observed HW with soap and water previous to 0.189 0.115** eat - - - - (0.049) [60.8%] Observed HW with soap and water before food 0.099 0.068* preparation - - - - (0.038) [68.6%] Observed HW with soap and water after fecal 0.342 -0.022 contact - - - - (0.056) Observed HW with soap and water before feeding 0.076 0.04 baby - - - - (0.060) Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 Table 4: Effect of the Interventions on Environmental Contamination Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Effect of treatment Control mean Effect of treatment Control mean Effect of treatment Water analysis Prevalence of E.Coli (Percentage) - - - - 0.607 -0.041 (0.072) Prevalence of total coliforms (Percentage) - - - - 0.793 -0.044 (0.062) Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 37 Table 5: Effect of the Interventions on Child Health Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Effect of treatment Control mean Effect of treatment Control mean Effect of treatment Child health Diarrhea prevalence 48 hours 0.04 0.01 0.04 -0.002 0.033 0.0012 (0.013) (0.011) (0.009) Diarrhea prevalence 7 days 0.06 0.011 0.06 0.001 0.069 -0.005 (0.018) (0.015) (0.014) ALRI prevalence 48 hours 0.041 -0.039*** 0.041 -0.011 0.049 -0.0183 (0.014) (0.018) (0.020) [-96.5%] ALRI prevalence 7 days 0.051 -0.047*** 0.051 -0.016 0.056 -0.021 (0.018) (0.022) (0.023) [-89%] Weight-for-age z-score -0.69 0.106 -0.69 0.000 -0.82 0.107 (0.082) (0.085) (0.087) Length/Height-for-age z-score -1.453 0.037 -1.453 -0.056 -1.619 0.083 (0.080) (0.079) (0.088) Weight-for-length/height z-score 0.203 0.14* 0.203 0.076 0.165 0.106 (0.076) (0.078) (0.083) Anemia (Hb < 110 g/L) 0.283 -0.046 0.283 -0.026 0.278 -0.022 (0.033) (0.033) (0.029) Parasites in child stools Prevalence of parasites - - - - 0.227 -0.0508 (0.046) Parasite count - - - - 0.303 -0.0851 (0.062) Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 The z-scores for the anthropometric measures were calculated discarding the lower and upper 1 percent of the distribution. 38 Appendix 1: Balance Checks Individual Variables Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Average number of children under 5 per HH 707 1.459 717 1.398 -0.061 707 1.46 763 1.409 0.051 684 1.493 705 1.435 0.058 Average HH size 707 5 717 5.359 -0.359* 707 5 763 5.05 -0.05 684 6.26 705 6.04 0.22 Gender (M ale): HH head 707 0.908 717 0.898 -0.1 707 0.908 763 0.927 -0.019 684 0.905 705 0.929 -0.024 Age: HH head 706 35.39 713 37.96 -2.97* * * 706 35.392 762 35.757 -0.365 683 38.269 702 38.47 -0.201 Percentage of HH heads that ever attended school 706 0.954 710 0.952 -0.002 706 0.955 763 0.969 -0.014 682 0.959 704 0.97 -0.011 Highest educational level achieved: HH head Primary 672 0.382 674 0.445 -0.063 672 0.382 735 0.41 -0.028 649 0.473 677 0.411 0.062 Secondary 672 0.522 674 0.439 0.083* * 672 0.522 735 0.49 0.032 649 0.414 677 0.484 -0.07* Trade school 672 0 674 0 - 672 0.052 735 0.061 -0.009 649 0.049 677 0.04 0.009 University 672 0.052 674 0.051 0.001 672 0.043 735 0.039 0.004 649 0.062 677 0.065 -0.003 Gender (M ale): other HH members 2831 0.37 3126 0.383 -0.013 2831 0.371 3092 0.379 -0.008 3600 0.406 3555 0.416 -0.01 Age: other HH members 2827 14.18 3123 15 -0.82* 2827 14.184 3088 14.776 -0.592 3594 13.609 3549 13.689 -0.08 Percentage of other HH members that ever attended school 1783 0.933 2106 0.935 0.002 1783 0.934 1993 0.963 -0.029* * 2557 0.958 2515 0.957 0.001 Highest educational level achieved: other HH members Kindergarten 1652 0.078 1941 0.062 -0.016* 1652 0.078 1911 0.061 0.017* 2434 0.096 2391 0.081 0.015 Primary 1652 0.518 1941 0.536 -0.018 1652 0.518 1911 0.507 0.011 2434 0.605 2391 0.622 -0.017 Secondary 1652 0.35 1941 0.347 0.003 1652 0.35 1911 0.38 -0.03 2434 0.272 2391 0.269 0.003 Trade school 1652 0.039 1941 0.029 0.01 1652 0.039 1911 0.038 0.001 2434 0.014 2391 0.018 -0.004 University 1652 0.013 1941 0.024 0.011* * 1652 0.014 1911 0.015 -0.001 2434 0.014 2391 0.01 0.004 Percentage of teenagers that spent time on: 0 School 710 0.961 819 0.927 0.034 710 0.962 695 0.965 -0.003 1393 0.933 1375 0.974 -0.041 Studying 710 0.973 819 0.958 0.015 710 0.973 695 0.967 0.006 1393 0.953 1375 0.978 -0.025 Children care 710 0.728 819 0.71 0.018 711 0.729 695 0.683 0.046 1395 0.72 1375 0.723 -0.003 Homework 711 0.713 819 0.752 -0.039 711 0.713 695 0.722 -0.009 1395 0.691 1375 0.699 -0.008 Paid work 711 0.016 819 0.019 -0.003 711 0.017 695 0.01 0.007 1395 0.014 1375 0.007 0.007 Unpaid work 711 0.091 819 0.188 -0.097* * 711 0.091 695 0.056 0.035 1395 0.08 1375 0.094 -0.014 39 Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Treatment mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Percentage of employed HH heads 707 0.945 714 0.948 -0.003 707 0.945 762 0.963 -0.018 682 0.965 705 0.956 0.009 Last week activity: HH head Looking for work 39 0.231 37 0.135 0.096 39 0.231 28 0.286 -0.055 24 0.167 31 0.226 -0.059 Taking care of home 39 0.462 37 0.324 0.138 39 0.462 28 0.357 0.105 24 0.542 31 0.484 0.058 Not working and not looking for job 39 0.128 37 0.216 -0.088 39 0.128 28 0.25 -0.122 24 0.208 31 0.161 0.047 Other 39 0.179 37 0.324 -0.145 39 0.179 28 0.107 0.072 24 0.083 31 0.129 -0.046 Percentage of employed other HH members 1055 0.344 1273 0.346 -0.002 1055 0.344 1289 0.38 -0.036 1151 0.381 1119 0.35 0.031 Last week activity: other HH members Looking for work 692 0.01 832 0.006 0.004 692 0.01 799 0.005 0.005 713 0.011 727 0.003 0.008* Studying 692 0.136 832 0.137 -0.001 692 0.136 799 0.14 -0.004 713 0.181 727 0.169 0.012 Taking care of home 692 0.803 832 0.802 0.001 692 0.803 799 0.801 0.002 713 0.75 727 0.779 -0.029 Not working and not looking for job 692 0.04 832 0.038 0.002 692 0.04 799 0.038 0.002 713 0.046 727 0.03 0.016 Other 692 0.01 832 0.017 -0.007 692 0.01 799 0.016 -0.006 713 0.011 727 0.019 -0.008 Primary work: position Self-employed 1290 0.557 1421 0.55 0.007 1290 0.557 1462 0.538 0.019 1325 0.546 1294 0.59 -0.044 Employee 1290 0.302 1421 0.265 0.037 1290 0.302 1462 0.325 -0.023 1325 0.303 1294 0.268 0.035 Employer or boss 1290 0.003 1421 0.001 0.002 1290 0.003 1462 0.005 -0.002 1325 0.004 1294 0.003 0.001 Worker without remuneration 1290 0.123 1421 0.165 -0.042 1290 0.123 1462 0.126 -0.003 1325 0.137 1294 0.134 0.003 Day laborer 1290 0.015 1421 0.018 -0.003 1290 0.015 1462 0.005 0.01 1325 0.008 1294 0.005 0.003 Other 1290 0.001 1421 0.001 0 1290 0.001 1462 0 0.001 1325 0.002 1294 0 0.002 Monthly salary 1082 369.64 1148 334.22 34.42 1082 369.64 1245 391.42 -21.78 1101 351.01 1087 336.88 14.13 Hours per week 1287 42.933 1404 42.551 0.382 1287 42.933 1448 41.823 1.11 1322 40.735 1287 42.045 -1.31 Months worked in last 12 months 1279 9.177 1405 9.731 -0.554 1279 9.177 1446 9.577 -0.4 1307 9.431 1271 9.862 -0.431 40 Household Assets Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Average HHs non labor income 312 110.37 240 106.01 4.36 312 110.37 205 106.13 4.24 315 111.45 191 129.89 -18.44 HHs Assets Radio, CD, cassette 706 0.813 717 0.815 -0.002 706 0.813 762 0.745 0.068* 682 0.812 705 0.75 0.062* TV 707 0.532 717 0.513 0.019 707 0.532 762 0.552 -0.02 682 0.595 705 0.609 -0.014 VCR 707 0.248 717 0.197 0.051 707 0.248 762 0.241 0.007 682 0.224 705 0.254 -0.03 Computer 707 0.02 717 0.015 0.005 707 0.02 762 0.012 0.008 682 0.025 705 0.016 0.009 Bicycle 707 0.184 716 0.196 -0.012 707 0.184 762 0.213 -0.029 682 0.217 705 0.24 -0.023 M otorbike 707 0.025 717 0.032 -0.007 707 0.025 762 0.039 -0.014 682 0.032 705 0.035 -0.003 Car or Tractor 707 0.014 717 0.018 -0.004 707 0.014 762 0.021 -0.007 682 0.015 705 0.007 0.008 Refrigerator 707 0.107 717 0.075 0.032 707 0.107 762 0.079 0.028 682 0.087 705 0.118 -0.031 Gas Stove 707 0.383 717 0.329 0.054 707 0.383 762 0.419 -0.036 682 0.346 705 0.445 -0.099 Other type of stove 706 0.153 717 0.079 0.074* * 706 0.153 762 0.083 0.07* * 682 0.15 705 0.092 0.058* Blender 707 0.209 717 0.17 0.039 707 0.209 762 0.22 -0.011 682 0.214 705 0.248 -0.034 Toaster 707 0.004 717 0.013 -0.009 707 0.004 762 0.007 -0.003 682 0.007 705 0.01 -0.003 M icrowave 707 0.011 717 0.007 0.004 707 0.011 762 0.005 0.006 682 0.01 705 0.013 -0.003 Washing machine 707 0.006 717 0.003 0.003 707 0.006 762 0.007 -0.001 682 0.009 705 0.011 -0.002 Water boiler 707 0.028 717 0.018 0.01 707 0.028 762 0.016 0.012 682 0.018 705 0.026 -0.008 Other houses/properties 707 0.109 717 0.035 0.074* 707 0.109 762 0.171 -0.062 682 0.104 704 0.182 -0.078 M achinery, equipment for family business 707 0.023 717 0.031 -0.008 707 0.023 761 0.021 0.002 682 0.015 703 0.02 -0.005 Percentage of HHs having other piece of land 707 0.475 716 0.369 0.106 707 0.475 763 0.383 0.092 684 0.477 705 0.43 0.047 Percentage of HHs having farm equipment 707 0.201 716 0.26 -0.059 707 0.201 763 0.215 -0.014 684 0.2 705 0.214 -0.014 Percentage of HHs having animals 707 0.754 717 0.826 -0.072 707 0.754 763 0.742 0.012 684 0.775 705 0.729 0.046 Average number of livestock owned 707 2.337 717 2.787 -4.5 707 2.337 763 1.992 0.345 684 2.401 705 2.009 0.392 41 Dwelling Characteristics Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Dwelling ownership HH member, still paying 707 0.018 716 0.038 -0.02 707 0.018 762 0.026 -0.008 683 0.041 705 0.027 0.014 HH member, fully paid 707 0.454 716 0.536 -0.082* 707 0.454 762 0.459 -0.005 683 0.515 705 0.482 0.033 Rented 707 0.13 716 0.084 0.046 707 0.13 762 0.118 0.012 683 0.111 705 0.15 -0.039 Family/Friend Loan 707 0.198 716 0.226 -0.028 707 0.198 762 0.277 -0.079* * * 683 0.145 705 0.197 -0.052* Other 707 0.199 716 0.116 0.083* * 707 0.199 762 0.119 0.08* 683 0.187 705 0.143 0.044 Type of dwelling Detached house 699 0.93 715 0.959 -0.029 699 0.93 755 0.926 0.004 679 0.981 700 0.933 0.048* * * Room in a larger dwelling 699 0.027 715 0.027 0 699 0.027 755 0.05 -0.023 679 0.009 700 0.034 -0.025* Other 699 0.043 715 0.014 0.029* * 699 0.043 755 0.024 0.019 679 0.01 700 0.033 -0.023* Dwelling light source No lighting 703 0.001 714 0 0.001 703 0.001 761 0.011 -0.01* * 683 0.004 699 0.016 -0.012* Electricity 703 0.686 714 0.576 0.11 703 0.686 761 0.748 -0.062 683 0.717 699 0.785 -0.068 Kerosene 703 0.159 714 0.154 0.005 703 0.159 761 0.059 0.1* * 683 0.127 699 0.059 0.068* Candles 703 0.137 714 0.227 -0.090* 703 0.137 761 0.146 -0.009 683 0.138 699 0.112 0.026 Other 703 0.017 714 0.043 -0.026 703 0.017 761 0.037 -0.02 683 0.013 699 0.029 -0.016 Dwelling cooking fuel Gas 703 0.296 714 0.237 0.023 703 0.296 761 0.293 0.003 683 0.233 699 0.313 -0.08 Wood 703 0.587 714 0.718 -0.131 703 0.587 761 0.618 -0.031 683 0.672 699 0.568 0.104 Peat/M anure 703 0.09 714 0.001 0.089* * 703 0.09 761 0.045 0.045 683 0.073 699 0.06 0.013 Other 703 0.027 714 0.043 -0.016 703 0.027 761 0.045 -0.018 683 0.022 699 0.059 -0.037 Dwelling heat fuel Do not heat dwelling 706 0.969 717 0.897 0.072* * * 706 0.969 763 0.971 -0.002 683 0.968 705 0.989 -0.021* * Wood Stove 706 0.02 717 0.095 -0.07* * * 706 0.02 763 0.025 -0.005 683 0.023 705 0.006 0.017* * Other 706 0.011 717 0.008 0.003 706 0.011 763 0.004 0.007 683 0.009 705 0.006 0.003 42 Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Material of dwelling's walls Esteras 699 0.029 715 0.042 -0.013 699 0.029 755 0.032 -0.003 679 0.034 700 0.024 0.01 Brick 699 0.094 715 0.062 0.032 699 0.094 755 0.127 -0.033 679 0.09 700 0.123 -0.033 Concrete 699 0.054 715 0.014 0.040* * 699 0.054 755 0.057 -0.003 679 0.019 700 0.07 -0.051* * Unbaked brick, adobe 699 0.584 715 0.664 -0.08 699 0.584 755 0.576 0.008 679 0.58 700 0.543 0.037 Wood, logs 699 0.103 715 0.088 0.015 699 0.103 755 0.044 0.059 679 0.138 700 0.051 0.087 Other 699 0.136 715 0.13 0.006 699 0.136 755 0.164 -0.028 679 0.138 700 0.189 -0.051 Material of dwelling's roof Esteras 699 0.04 715 0.056 -0.016 699 0.04 755 0.045 -0.005 679 0.05 700 0.054 -0.004 Brick 699 0.023 715 0.038 -0.015 699 0.023 755 0.019 0.004 679 0.029 700 0.024 0.005 Concrete 699 0.052 715 0.017 0.037* 699 0.052 755 0.038 0.014 679 0.024 700 0.053 -0.029* Wood, logs 699 0.019 715 0.007 0.012 699 0.019 755 0.009 0.01 679 0.012 700 0.02 -0.008 Tin, zinc sheeting 699 0.534 715 0.47 0.102 699 0.534 755 0.668 -0.134* 679 0.571 700 0.636 -0.065 Bamboo 699 0.006 715 0.007 -0.001 699 0.006 755 0.028 -0.022 679 0.01 700 0.023 -0.013 Other 699 0.328 715 0.406 -0.078 699 0.328 755 0.193 0.135* * 679 0.303 700 0.19 0.113* Material of dwelling's floor Painted wood 699 0.009 713 0.004 0.005 699 0.009 753 0.009 0 679 0.013 700 0.017 -0.004 Concrete 699 0.156 713 0.111 0.045 699 0.156 753 0.159 -0.003 679 0.138 700 0.217 0.055 Clay, earthen floor 699 0.701 713 0.749 -0.048 699 0.701 753 0.699 0.002 679 0.698 700 0.636 0.062 Non polished concrete 699 0.076 713 0.093 -0.017 699 0.076 753 0.098 -0.022 679 0.082 700 0.091 -0.009 Other 699 0.059 713 0.043 0.016 699 0.059 753 0.035 0.024 679 0.068 700 0.039 0.029 HH has food not covered 685 0.225 651 0.257 -0.032 685 0.225 704 0.207 0.018 663 0.27 658 0.24 0.03 HH is clean 682 0.543 687 0.518 0.025 682 0.543 715 0.593 -0.05 663 0.508 669 0.538 -0.03 HH with garbage in kitchen or house 686 0.541 677 0.589 -0.048 686 0.541 717 0.488 0.053 668 0.581 666 0.568 0.013 43 Toilet Facilities and Water Sources Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Main toilet facility No facilities, bush, field 717 0.24 707 0.221 0.019 707 0.221 763 0.239 -0.018 683 0.218 704 0.203 0.015 Hanging toilet, latrine 717 0.01 707 0.017 -0.007 707 0.017 763 0.001 0.016 683 0.018 704 0.003 0.015 Flush, to piped sewer system 717 0.132 707 0.226 -0.094* 707 0.226 763 0.229 -0.003 683 0.217 704 0.339 -0.122* Flush, to other place 717 0.047 707 0.095 -0.048* * 707 0.095 763 0.101 -0.006 683 0.088 704 0.101 -0.013 Ventilated improved pit latrine 717 0.073 707 0.042 0.031 707 0.042 763 0.069 -0.027 683 0.028 704 0.061 -0.033 Pit latrine with slab 717 0.033 707 0.034 -0.001 707 0.034 763 0.06 -0.026 683 0.023 704 0.044 -0.021 Pit latrine without slab, open pit 717 0.411 707 0.325 0.086 707 0.325 763 0.263 0.062 683 0.381 704 0.217 0.164* * * Other 717 0.053 707 0.04 0.013 707 0.04 763 0.037 0.003 683 0.028 704 0.031 -0.003 % of toilet facilities that are public 544 0.121 548 0.144 -0.023 548 0.144 581 0.129 0.015 532 0.092 553 0.105 -0.013 Location of main toilet facility Inside household 717 0.156 707 0.202 -0.046 707 0.202 763 0.224 -0.022 684 0.211 705 0.278 -0.067 In household yard 717 0.392 707 0.382 0.01 707 0.382 763 0.391 -0.009 684 0.405 705 0.403 0.002 Less than 10 mins walk 717 0.34 707 0.291 0.049 707 0.291 763 0.248 0.043 684 0.269 705 0.213 0.056 M ore than 10 mins walk 717 0.071 707 0.088 -0.017 707 0.088 763 0.11 -0.022 684 0.076 705 0.077 -0.001 No designated area 717 0.038 707 0.035 0.003 707 0.035 763 0.025 0.01 684 0.037 705 0.026 0.011 Other 717 0.003 707 0.001 0.002 707 0.001 763 0.003 -0.002 684 0.003 705 0.004 -0.001 Percentage of shared toilet facility 717 0.254 707 0.263 -0.009 707 0.263 763 0.304 -0.041 684 0.23 705 0.271 -0.041 Percentage of safe toilet facilities during the night 715 0.738 707 0.745 -0.007 707 0.745 763 0.773 -0.028 683 0.761 704 0.781 -0.02 Disposal of child defecation Bushes, ground 717 0.279 707 0.337 -0.058 707 0.337 763 0.266 0.071 684 0.303 705 0.173 0.13* * * Pit, hole in the ground 717 0.1 707 0.092 0.008 707 0.092 763 0.087 0.005 684 0.076 705 0.096 -0.02 Open sewer, drain 717 0.025 707 0.048 -0.023 707 0.048 763 0.045 0.003 684 0.047 705 0.065 -0.018 Toilet, latrine 717 0.209 707 0.163 0.046 707 0.163 763 0.215 -0.052 684 0.209 705 0.237 -0.028 Garbage 717 0.301 707 0.301 0 707 0.301 763 0.307 -0.006 684 0.308 705 0.34 -0.032 River 717 0.121 707 0.12 0.001 707 0.12 763 0.092 0.028 684 0.11 705 0.098 0.012 Basin, sink 717 0.114 707 0.098 0.016 707 0.098 763 0.06 0.038 684 0.104 705 0.062 0.042 Other 717 0.064 707 0.071 -0.007 707 0.071 763 0.041 0.03* 684 0.064 705 0.045 0.019 44 Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Same sources along the year 707 0.97 717 0.974 -0.004 707 0.97 763 0.988 -0.018 683 0.99 705 0.989 0.001 Source of drinking water Tanker truck 707 0 717 0.026 -0.026* 707 0 763 0.004 -0.004 683 0.015 705 0.007 0.008 Surface water 707 0.055 717 0.081 -0.026 707 0.055 763 0.031 0.024 683 0.064 705 0.044 0.02 Piped water, into dwelling 707 0.223 717 0.233 0.142 707 0.223 763 0.249 -0.026 683 0.228 705 0.25 -0.022 Piped water, into yard, plot 707 0.16 717 0.201 -0.041 707 0.16 763 0.232 -0.072 683 0.17 705 0.247 -0.077 Piped water, public tap, standpipe 707 0.051 717 0.095 -0.044 707 0.051 763 0.043 0.008 683 0.053 705 0.045 0.008 Tube well, wore whole 707 0.024 717 0.013 0.011 707 0.024 763 0.01 0.014 683 0.004 705 0.013 -0.009 Dug well, protected 707 0.017 717 0.026 -0.009 707 0.017 763 0.008 0.009 683 0.006 705 0.01 -0.004 Dug well, unprotected 707 0.04 717 0.004 0.036 707 0.04 763 0.007 0.033 683 0.026 705 0.009 0.017 Spring water, protected 707 0.262 717 0.121 0.152* 707 0.262 763 0.257 0.005 683 0.261 705 0.23 0.031 Spring water, unprotected 707 0.041 717 0.077 -0.036 707 0.041 763 0.031 0.01 683 0.038 705 0.023 0.015 Other 707 0.127 717 0.123 0.004 707 0.127 763 0.127 0 683 0.135 705 0.123 0.012 Source location In own dwelling 436 0.128 406 0.074 0.054 436 0.128 396 0.278 -0.15 411 0.148 355 0.282 -0.134 In own yard, plot 436 0.431 406 0.355 0.076 436 0.431 396 0.313 0.118 411 0.372 355 0.287 0.085 Elsewhere 436 0.44 406 0.571 -0.131 436 0.44 396 0.409 0.031 411 0.479 355 0.431 0.048 Covered source Covered 433 0.607 404 0.597 0.01 433 0.607 393 0.687 -0.08 407 0.582 355 0.645 -0.063 Open 433 0.372 404 0.389 -0.017 433 0.372 393 0.303 0.069 407 0.388 355 0.352 0.036 Both covered and open 433 0.021 404 0.015 0.006 433 0.021 393 0.01 0.011 407 0.029 355 0.003 0.026 Who mainly collects water from this source Adult woman 436 0.846 405 0.847 -0.001 436 0.846 396 0.886 -0.04 411 0.82 355 0.865 -0.045 Adult man 436 0.115 405 0.114 0.001 436 0.115 396 0.086 0.029 411 0.117 355 0.101 0.016 Girl (< 15 years) 436 0.023 405 0.017 0.006 436 0.023 396 0.015 0.008 411 0.027 355 0.014 0.013 Boy (< 15 years) 436 0.014 405 0.02 -0.006 436 0.014 396 0.013 0.001 411 0.022 355 0.02 0.002 Other 436 0.002 405 0.002 0 436 0.002 396 0 0.002 411 0.015 355 0 -0.015* Satisfied with the quantity 704 0.724 715 0.734 -0.01 704 0.724 763 0.738 -0.014 681 0.686 702 0.708 -0.022 Does the household pay for the water 706 0.564 716 0.603 -0.039 706 0.564 761 0.662 -0.098 683 0.698 705 0.674 0.024 Fixed, limited quantity obtained for the payment 389 0.524 428 0.339 0.185* * 389 0.524 495 0.442 0.082 466 0.47 465 0.447 0.023 How water was prepared (last 7 days) Boil 649 0.948 603 0.954 -0.006 649 0.948 689 0.972 -0.024 618 0.963 635 0.98 -0.017 Chlorine 649 0.034 603 0.06 -0.026 649 0.034 689 0.026 0.008 618 0.026 635 0.02 0.006 Let it stand and settle 649 0.032 603 0.06 -0.028 649 0.032 689 0.022 0.01 618 0.026 635 0.022 0.004 Other 649 0.015 603 0.015 0 649 0.015 689 0 0.015 618 0.01 635 0.002 0.008 Improved water source 707 0.737 717 0.69 0.047 707 0.737 763 0.801 -0.064 683 0.722 705 0.804 -0.082 Improved sanitation 707 0.386 717 0.278 0.108* 707 0.386 763 0.461 -0.075 683 0.335 704 0.544 -0.209* * * Soap and water at HW station 707 0.588 717 0.562 0.026 707 0.588 763 0.598 -0.01 684 0.639 705 0.603 0.036 45 Handwashing Facilities Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Wash hands after going to toilet 698 0.986 707 0.993 -0.007 698 0.986 738 0.985 0.001 675 0.994 674 0.987 0.007 Place where usually wash hands after going to toilet 0 0 Inside toilet facility 683 0.104 685 0.058 0.046* 683 0.104 718 0.113 -0.009 667 0.1 659 0.118 -0.018 Inside kitchen, cooking place 683 0.105 685 0.067 0.038 683 0.105 718 0.131 -0.026 667 0.112 659 0.082 0.03 In yard less than 3 feet from toilet 683 0.193 685 0.194 -0.001 683 0.193 718 0.132 0.061* * 667 0.219 659 0.185 0.034 Between 10 feet and 3 feet from toilet 683 0.142 685 0.155 -0.013 683 0.142 718 0.117 0.025 667 0.148 659 0.141 0.007 M ore than 10 feet from toilet 683 0.335 685 0.352 -0.017 683 0.335 718 0.373 -0.038 667 0.315 659 0.34 -0.025 No specific place 683 0.12 685 0.174 -0.054 683 0.12 718 0.134 -0.014 667 0.105 659 0.134 -0.029 Handwashing device, toilet 0 0 Tap, faucet 600 0.643 563 0.584 0.059 600 0.643 620 0.655 -0.012 597 0.631 570 0.695 -0.064 Basin, bucket 600 0.335 563 0.352 -0.017 600 0.335 620 0.319 0.016 597 0.337 570 0.295 0.042 Other 600 0.022 563 0.064 -0.042* * 600 0.022 620 0.026 -0.004 597 0.032 570 0.011 0.021* * Water available at handwashing station 599 0.871 561 0.85 0.021 599 0.871 619 0.889 -0.018 594 0.877 570 0.854 0.023 Soaps available 0 0 M ultipurpose bar soap 601 0.095 567 0.254 -0.159* * * 601 0.095 622 0.127 -0.032 597 0.134 571 0.128 0.006 Beauty, toilet bar soap 601 0.245 567 0.224 0.021 601 0.245 622 0.241 0.004 597 0.214 571 0.254 -0.04 Powder soap, detergent 601 0.486 567 0.347 0.139* * 601 0.486 622 0.471 0.015 597 0.524 571 0.478 0.046 No soap observed 601 0.306 567 0.275 0.031 601 0.306 622 0.273 0.033 597 0.295 571 0.282 0.013 Ash, mud at handwashing station 0 0 Ash 595 0.012 553 0.016 -0.004 595 0.012 614 0.01 0.002 590 0.005 566 0.004 0.001 M ud 595 0.245 553 0.195 0.05 595 0.245 614 0.238 0.007 590 0.237 566 0.221 0.016 Ash and M ud 595 0.034 553 0.045 -0.011 595 0.034 614 0.024 0.01 590 0.049 566 0.037 0.012 Neither observed 595 0.709 553 0.743 -0.034 595 0.709 614 0.728 -0.019 590 0.708 566 0.739 -0.031 46 Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Wash hands before- after cooking, feeding a child 697 0.996 706 0.993 0.003 697 0.996 738 0.996 0 675 0.996 674 0.987 0.009 Place where usually wash hands Inside toilet facility 647 0.028 631 0.014 0.014 647 0.028 700 0.011 0.017 635 0.02 636 0.017 0.003 Inside kitchen, cooking place 647 0.396 631 0.361 0.035 647 0.396 700 0.434 -0.038 635 0.413 636 0.414 -0.001 In yard less than 3 feet from kitchen 647 0.207 631 0.195 0.012 647 0.207 700 0.164 0.043 635 0.17 636 0.165 0.005 Between 10 feet and 3 steps from kitchen 647 0.162 631 0.189 -0.027 647 0.162 700 0.154 0.008 635 0.191 636 0.173 0.018 M ore than 10 feet from kitchen 647 0.124 631 0.127 -0.003 647 0.124 700 0.139 -0.015 635 0.131 636 0.17 -0.039 No specific place 647 0.083 631 0.114 -0.031 647 0.083 700 0.097 -0.014 635 0.076 636 0.061 0.015 Handwashing device Tap, faucet 243 0.263 245 0.208 0.055 243 0.263 283 0.314 -0.051 237 0.283 288 0.299 -0.016 Container from which water is poured 243 0.716 245 0.78 -0.064 243 0.716 283 0.678 0.038 237 0.696 288 0.688 0.008 Other 243 0.021 245 0.012 0.009 243 0.021 283 0.007 0.014 237 0.021 288 0.014 0.007 Water available at handwashing station 244 0.75 246 0.785 -0.035 244 0.75 283 0.837 -0.087* 237 0.819 288 0.792 0.027 Soaps available M ultipurpose bar soap 244 0.07 246 0.138 -0.068 244 0.07 283 0.078 -0.008 237 0.097 288 0.049 0.048 Beauty, toilet soap 244 0.107 246 0.081 0.026 244 0.107 283 0.11 -0.003 237 0.114 288 0.101 0.013 Powder or laundry soap, detergent 244 0.561 246 0.472 0.089 244 0.561 283 0.509 0.052 237 0.612 288 0.573 0.039 No soap observed 244 0.352 246 0.341 0.011 244 0.352 283 0.385 -0.033 237 0.287 288 0.365 -0.078 Ash, mud at handwashing station Ash 240 0.038 241 0.008 0.03 240 0.038 282 0 0.038* 232 0.013 284 0.007 0.006 M ud 240 0.15 241 0.124 0.026 240 0.15 282 0.078 0.072* 232 0.168 284 0.123 0.045 Ash and M ud 240 0.058 241 0.087 -0.029 240 0.058 282 0.025 0.033 232 0.069 284 0.053 0.016 Nor Ash nor M ud 240 0.754 241 0.78 -0.026 240 0.754 282 0.897 -0.143* * 232 0.75 284 0.817 -0.067 47 Handwashing Behavior Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Percentage of caregivers that washed their hands with soap since yesterday 712 0.997 720 0.999 -0.002 712 0.997 765 0.996 0.001 693 0.999 707 0.994 0.005 Last moment of hand wash since yesterday Bathing a child 707 0.255 717 0.195 0.06 707 0.255 759 0.244 0.011 689 0.253 703 0.262 -0.009 Washing child's hands 707 0.123 717 0.11 0.013 707 0.123 759 0.083 0.04* 689 0.094 703 0.083 0.011 Cleaning dishes 707 0.41 717 0.459 -0.049 707 0.41 759 0.348 0.062 689 0.437 703 0.331 0.106* * * Doing laundry 707 0.436 717 0.445 -0.009 707 0.436 759 0.44 -0.004 689 0.498 703 0.511 -0.013 Because they look dirty 707 0.102 717 0.066 0.036* 707 0.102 759 0.045 0.057* * 689 0.115 703 0.053 0.062* * * Bathing oneself 707 0.204 717 0.153 0.051 707 0.204 759 0.219 -0.015 689 0.224 703 0.259 -0.035 Using toilet 707 0.426 717 0.389 0.037 707 0.426 759 0.397 0.029 689 0.356 703 0.385 -0.029 Cleaning baby bottom 707 0.334 717 0.424 -0.090* * 707 0.334 759 0.368 -0.034 689 0.327 703 0.356 -0.029 Cleaning latrine 707 0.021 717 0.01 0.011 707 0.021 759 0.008 0.013* 689 0.015 703 0.024 -0.009 Cleaning toilet 707 0.035 717 0.035 0 707 0.035 759 0.022 0.013 689 0.036 703 0.028 0.008 Returning home 707 0.123 717 0.13 -0.007 707 0.123 759 0.119 0.004 689 0.129 703 0.134 -0.005 Preparing food, cooking 707 0.717 717 0.763 -0.046 707 0.717 759 0.675 0.042 689 0.704 703 0.643 0.061 Feeding children 707 0.369 717 0.351 0.018 707 0.369 759 0.278 0.091* * * 689 0.335 703 0.282 0.053 Other 707 0.057 717 0.102 -0.045* * * 707 0.057 759 0.075 -0.018 689 0.038 703 0.051 -0.013 Best way to clean hands Wipe on cloth 713 0.011 718 0.011 0.989 713 0.011 765 0.013 -0.002 694 0.013 704 0.009 0.004 Wash with water alone 713 0.132 718 0.11 0.43 713 0.132 765 0.116 0.016 694 0.134 704 0.111 0.023 Wash with soap 713 0.847 718 0.864 0.591 713 0.847 765 0.854 -0.007 694 0.84 704 0.865 -0.025 Wash with ash, mud 713 0 718 0.003 0.157 713 0 765 0 694 0.003 704 0 0.003 Other 713 0.01 718 0.013 -0.003 713 0.01 765 0.017 -0.007 694 0.01 704 0.016 -0.006 Self-Reported Behavior: HW with soap and water before feeding child 710 0.367 719 0.35 0.017 710 0.367 762 0.276 0.091* * * 692 0.333 703 0.281 0.052 HW with soap and water before food prep. 710 0.714 719 0.760 -0.046 710 0.714 762 0.671 0.043 692 0.7 703 0.642 0.058 HW with soap and water after fecal contact 710 0.615 719 0.668 -0.053 710 0.615 762 0.628 -0.013 692 0.562 703 0.613 -0.051 Caregiver's fingernails are: Visible dirty 714 0.284 719 0.303 -0.019 714 0.284 767 0.21 0.074* 694 0.281 703 0.225 0.056 Unclean appearance 714 0.322 719 0.325 -0.003 714 0.322 767 0.286 0.036 694 0.336 703 0.297 0.039 Clean 714 0.394 719 0.371 0.023 714 0.394 767 0.505 -0.111* * 694 0.383 703 0.478 0.098* Caregiver's palms are: Visible dirty 714 0.225 719 0.22 0.005 714 0.225 767 0.154 0.071* 694 0.225 703 0.137 0.036* Unclean appearance 714 0.252 719 0.243 0.009 714 0.252 767 0.207 0.045 694 0.272 703 0.272 0 Clean 714 0.522 719 0.537 -0.015 714 0.522 767 0.639 -0.117* * 694 0.503 703 0.592 -0.089 Caregiver's finger pads are: Visible dirty 714 0.224 719 0.224 0 714 0.224 767 0.147 0.077* 694 0.22 702 0.14 0.07* * Unclean appearance 714 0.265 719 0.249 -0.016 714 0.265 767 0.21 0.055* 694 0.277 702 0.255 0.022 Clean 714 0.511 719 0.527 -0.016 714 0.511 767 0.643 -0.132* * 694 0.503 702 0.605 -0.102 *p < 10%, ** < 5%, *** < 1% 48 Mass Media Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Recalls of any handwashing campaign 714 0.245 722 0.253 -0.008 714 0.245 769 0.224 0.021 694 0.272 711 0.226 0.046 Family-School Relationship Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Participation in schools activities Parents association 296 0.176 389 0.288 -0.112* * 296 0.176 301 0.256 -0.08 688 0.196 702 0.249 -0.053 Speeches, conferences 296 0.267 389 0.342 -0.075 296 0.267 301 0.369 -0.102* 688 0.298 702 0.325 -0.027 Kermesses 296 0.139 389 0.111 0.028 296 0.139 301 0.096 0.043 688 0.122 702 0.094 0.028 APAFA 296 0.72 389 0.584 0.136* * 296 0.72 301 0.611 0.109* 688 0.67 702 0.598 0.072 Other 296 0.199 389 0.185 0.014 296 0.199 301 0.169 0.03 688 0.161 702 0.175 -0.014 Does not participate 296 0.034 389 0.09 -0.056* * 296 0.034 301 0.053 -0.019 688 0.078 702 0.078 0 Percentage of caregivers that recall any campaign on health and hygiene promoted by the school 296 0.338 389 0.298 0.04 296 0.338 301 0.336 0.002 688 0.39 702 0.392 -0.002 *p < 10%, ** < 5%, *** < 1% 49 Child Development Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference 581 0.06 557 0.051 0.009 581 0.06 601 -0.014 0.074 533 -0.043 554 -0.057 0.014 581 -0.011 556 -0.03 0.818 581 -0.011 601 0.001 -0.012 532 0.072 554 -0.027 0.099 581 -0.009 554 0.07 0.028 581 -0.009 599 -0.052 0.043 533 0.079 553 -0.08 0.159* *p < 10%, ** < 5%, *** < 1% 50 Child Care Situation 51 Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Average number of times child was left at the charge of another child 732 0.657 735 0.707 -0.05 732 0.657 786 0.641 0.016 717 1.024 729 1.06 -0.036 Average number of times was child left alone 734 0.327 735 0.405 -0.078 734 0.327 786 0.314 0.013 717 0.211 729 0.258 -0.047 Percentage of children with clean aspect 982 0.667 955 0.625 0.042 982 0.667 1027 0.689 -0.022 980 0.624 955 0.672 -0.048 Percentage of children with dirty hands 979 0.501 953 0.518 -0.017 979 0.501 1020 0.429 0.072 976 0.513 951 0.449 0.064 Percentage of children with g nails dirty finger 973 0.597 948 0.584 0.013 973 0.597 1018 0.476 0.121* * 969 0.608 945 0.525 0.083* dirty face g g 981 0.414 955 0.452 -0.038 981 0.414 1029 0.331 0.083* 979 0.441 955 0.365 0.076 clothes 976 0.411 955 0.439 -0.028 976 0.411 1028 0.371 0.04 980 0.476 954 0.399 0.077 Percentage of children with dirty clothes 980 0.989 955 0.988 0.001 980 0.989 1030 0.993 -0.004 981 0.989 956 0.985 0.004 Pot-belly 979 0.153 944 0.144 0.009 979 0.153 1019 0.106 0.047 977 0.159 951 0.139 0.02 Percentage of children wearing shoes (or shoes available) 986 0.853 956 0.844 0.009 986 0.853 1032 0.834 0.019 982 0.869 962 0.823 0.046* Percentage of children that play with household objects 733 0.673 734 0.58 0.093* * 733 0.673 784 0.626 0.047 718 0.636 729 0.647 -0.011 Percentage of children that play with toys 733 0.809 734 0.8 0.009 733 0.809 784 0.8 0.009 718 0.788 729 0.813 -0.025 Average number of children's books or pictures 736 0.292 736 0.255 0.037 736 0.292 788 0.208 0.084 718 0.23 730 0.225 0.005 attended early education programs 732 0.04 734 0.03 0.01 732 0.04 783 0.041 -0.001 718 0.033 726 0.032 0.001 Percentage of adults that read books with child 731 0.274 733 0.225 0.049 731 0.274 784 0.241 0.033 717 0.225 729 0.263 -0.038 Percentage of adults that tell stories to the child 732 0.265 731 0.197 0.068* * 732 0.265 784 0.227 0.038 717 0.247 729 0.254 -0.007 Percentage of adults that take the child outside the home 733 0.943 734 0.913 0.03 733 0.943 784 0.926 0.017 718 0.911 729 0.918 -0.007 Percentage of adults that play with the child 733 0.868 734 0.869 -0.001 733 0.868 784 0.857 0.011 718 0.831 729 0.842 -0.011 Average daily caring time 1028 5.038 988 4.924 0.114 1028 5.038 1062 5.704 -0.666 1015 4.858 1001 5.393 -0.535 *p < 10Percentage, ** < 5Percentage, *** < 1Percentage 52 Child Health Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference ALRI in prev. 48 hrs 1031 0.029 1003 0.103 -0.07* * * 1031 0.029 1074 0.049 -0.02 1017 0.031 1012 0.053 -0.022 ALRI in prev. 1 week 1031 0.04 1003 0.139 -0.09* * * 1031 0.04 1074 0.065 -0.025 1017 0.043 1012 0.073 -0.03 Diarrhea in prev. 48 hrs 1031 0.098 1003 0.082 0.016 1031 0.098 1074 0.101 -0.003 1017 0.077 1012 0.084 -0.007 Diarrhea in prev. 1 week 1031 0.167 1003 0.153 0.014 1031 0.167 1074 0.162 0.005 1017 0.14 1012 0.139 0.001 Household w/ lost hours due child illness 3534 0.014 3832 0.063 -0.04* * * 3534 0.014 3852 0.019 -0.005 4236 0.02 4232 0.01 0.01 Anemia (Hb < 110 g/L) - children <2 605 0.711 652 0.701 0.01 605 0.711 632 0.731 0.503 596 0.711 565 0.701 0.01 *p < 10%, ** < 5%, *** < 1% Anthropometric Measures Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference BMI-for-age z-score 709 0.361 719 0.463 -0.102 709 0.361 769 0.405 -0.044 692 0.455 709 0.471 -0.016 Head circumference-for-age z- score 707 -0.238 714 -0.332 0.094 707 -0.238 769 -0.31 0.072 690 -0.281 714 -0.261 -0.02 Arm circumference-for-age z- score 635 0.3 631 0.397 -0.097 635 0.3 684 0.35 -0.05 609 0.271 634 0.314 -0.043 Length/height-for-age z-score 695 -1.318 709 -1.404 0.086 695 -1.318 746 -1.333 0.015 679 -1.436 698 -1.399 -0.037 Weight-for-length/height z-score 692 0.279 692 0.398 -0.119 695 0.279 757 0.35 -0.071 685 0.389 687 0.363 0.026 Weight-for-age z-score 701 -0.528 710 -0.527 0.001 701 -0.528 758 -0.493 -0.035 684 -0.497 700 -0.498 0.001 53 Microbiology Analysis District level intervention School component sample Control mean Treatment mean N Avg. N Avg. Difference Prevalence of E.Coli in water 85 0.329 74 0.514 -0.184* Log10 E. coli, MPN/100ml, Child 86 0.584 74 0.463 0.121 Log10 E. coli, MPN/100ml, Mother 86 0.702 74 0.829 -0.127 Log10 E. coli, MPN/100ml, Object 82 0.595 72 0.621 -0.026 Stool sample, Ascaris detected 86 0.023 74 0.000 0.023 Stool sample, Blastocystis detected 86 0.105 74 0.108 -0.003 Stool sample, Giardia detected 86 0.105 74 0.041 0.064 Parasite count in stools 86 0.233 74 0.149 0.084 Prevalence of parasites in stools 86 0.186 74 0.122 0.064 *p < 10%, ** < 5%, *** < 1% 54 Appendix2: Attrition Household Attrition Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Percentage HHs Included in Group 707 0.7222 717 0.8131 -0.0909 707 0.722 763 0.7614 -0.039 684 0.799 705 0.835 -0.036 55 Baseline Comparison of Non-Attriters Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Average HH size 546 1.461 583 1.397 0.064 546 1.461 581 1.442 0.02 547 6.177 589 6.014 0.164 Percentage of HH heads that are male 546 0.91 583 0.907 0.003 546 0.91 581 0.9277 -0.02 547 0.909 589 0.930 -0.022 Percentage of other HH members that are male 2198 0.37 2563 0.382 -0.012 2198 0.37 2356 0.369 0.00 5030 0.389 5309 0.393 -0.004 Percentage of HH heads that ever attended school 546 0.961 577 0.949 0.012 546 0.961 581 0.967 -0.01 546 0.962 588 0.973 -0.011 Highest educational level achieved by HH head 0.00 0.000 Primary 524 0.374 546 0.457 -0.083 524 0.374 558 0.416 -0.04 522 0.469 568 0.394 0.075 Secondary 524 0.532 546 0.441 0.091* * 524 0.532 558 0.480 0.05 522 0.423 568 0.495 -0.071* Preparatory 524 0 546 0 - 524 0 558 0.000 0.00 522 0.002 568 0.000 0.002 Trade School 524 0.053 546 0.043 0.01 524 0.053 558 0.066 -0.01 522 0.048 568 0.042 0.006 University 524 0.04 546 0.056 -0.016 524 0.04 558 0.038 0.00 522 0.057 568 0.069 -0.011 Percentage of teenagers that spent time on: 0.00 0.000 School 552 0.969 673 0.937 0.032 552 0.969 553 0.966 0.00 1103 0.941 1149 0.977 -0.036 Studying 552 0.98 669 0.965 0.015 552 0.98 553 0.966 0.01 1103 0.961 1149 0.983 -0.022 Children care 553 0.726 669 0.696 0.03 553 0.726 553 0.682 0.04 1105 0.726 1149 0.730 -0.004 Homework 553 0.71 669 0.751 -0.041 553 0.71 553 0.731 -0.02 1105 0.692 1149 0.701 -0.008 Paid work 553 0.019 670 0.02 -0.001 553 0.019 553 0.007 0.01 1105 0.014 1149 0.005 0.009 Unpaid work 553 0.074 670 0.167 -0.009* * 553 0.074 553 0.045 0.03 1105 0.074 1149 0.096 -0.022 Percentage of employed HH heads 546 0.9395 581 0.955 -0.015 546 0.9395 581 0.962 -0.02 545 0.965 589 0.954 0.011 Primary salary 841 340.48 931 321.09 19.39 841 340.48 926 389.846 -49.37 867 339.348 902 341.755 -2.407 Primary work: hours per week 998 43.078 1151 42.513 0.565 998 43.078 1054 41.599 1.48 1044 40.173 1068 41.596 -1.422 Primary work: months worked in last 12 months 993 9.309 1151 9.807 -0.498 993 9.309 1052 9.711 -0.40 1028 9.346 1057 9.798 -0.452 Source of drinking water (Percentage HHS) Tanker truck 546 0 583 0.032 -0.032* 546 0 581 0.002 0.00 546 0.015 589 0.008 0.006 Surface water 546 0.034 583 0.082 -0.048* 546 0.034 581 0.033 0.00 546 0.055 589 0.036 0.019 Piped water, into dwelling 546 0.25 583 0.236 0.014 546 0.25 581 0.262 -0.01 546 0.233 589 0.267 -0.034 Piped water, into yard, plot 546 0.166 583 0.209 -0.043 546 0.166 581 0.236 -0.07 546 0.167 589 0.224 -0.057 Piped water, public tap, standpipe 546 0.045 583 0.087 -0.042 546 0.045 581 0.048 0.00 546 0.060 589 0.049 0.011 Tube well, worewhole 546 0.027 583 0.008 0.019 546 0.027 581 0.010 0.02 546 0.005 589 0.012 -0.006 Dug well, protected 546 0.016 583 0.025 -0.009 546 0.016 581 0.010 0.01 546 0.002 589 0.008 -0.007 Dug well, unprotected 546 0.04 583 0.003 0.037 546 0.04 581 0.007 0.03 546 0.026 589 0.008 0.017 Spring water, protected 546 0.23 583 0.126 0.104 546 0.23 581 0.236 -0.01 546 0.253 589 0.241 0.012 Spring water, unprotected 546 0.043 583 0.07 -0.027 546 0.043 581 0.022 0.02 546 0.040 589 0.022 0.018 Other 546 0.142 583 0.116 0.026 546 0.142 581 0.134 0.01 546 0.145 589 0.124 0.021 Diarrhea prevalence (Percentage of children under 5) 2 days 797 0.102 815 0.078 0.024 797 0.102 838 0.095 0.01 816 0.077 840 0.080 -0.003 7 days 797 0.174 815 0.154 0.02 797 0.174 838 0.156 0.02 816 0.145 840 0.140 0.004 *** p<0.01 ** p<0.05 * p<0.1 56 Comparison of Replacement Households Province level intervention District level intervention Mass media treatment Community treatment sample School component sample Control mean Treatment mean Control mean Treatment mean Control mean Treatment mean N Avg. N Avg. Difference N Avg. N Avg. Difference N Avg. N Avg. Difference Average HH size 153 4.588 132 4.719 -0.131 153 4.588 179 4.587 0.002 122 4.852 115 5.130 -0.278 Percentage of HH heads that are male 152 0.921 131 0.931 -0.01 152 0.921 179 0.860 0.061 122 0.877 114 0.860 0.017 Percentage of other HH members that are male 550 0.383 492 0.369 0.014 550 0.384 642 0.399 -0.015 470 0.402 476 0.391 0.011 Percentage of HH heads that ever attended school 150 0.946 131 0.961 -0.015 153 1.216 179 1.268 -0.052 122 1.320 115 1.209 0.111* * Highest educational level achieved by HH head 0.000 0.000 Primary 119 0.184 101 0.207 -0.023 150 0.947 179 0.961 -0.014 122 0.975 114 0.921 0.054 Secondary 119 0.1 101 0.099 0.001 141 0.333 171 0.380 -0.047 117 0.282 103 0.398 -0.116* * Preparatory 119 0.512 101 0.455 0.057 141 0.553 171 0.497 0.056 117 0.547 103 0.534 0.013 Trade school 119 0.193 101 0.237 -0.044 141 0.092 171 0.064 0.028 117 0.120 103 0.039 0.081 University 119 0.008 101 0 0.008 141 0.021 171 0.058 -0.037 117 0.051 103 0.029 0.022 % of employed HH heads 152 0.96 131 0.938 0.022 152 0.961 179 0.939 0.022 122 0.893 114 0.921 -0.028 Source of drinking water 0.000 0.000 (% HHS) Tanker truck 153 0 132 0 - 153 0.000 179 0.000 0.000 122 0.008 115 0.000 0.008 Water vendor 153 0 132 0 - 153 0.000 179 0.000 0.000 122 0.016 115 0.000 0.016 Surface water 153 0.019 132 0.022 -0.003 153 0.020 179 0.039 -0.019 122 0.016 115 0.043 -0.027 Bottled water 153 0.013 132 0 0.013 153 0.013 179 0.000 0.013 122 0.025 115 0.000 0.025 Piped water, into dwelling 153 0.281 132 0.333 -0.052 153 0.281 179 0.296 -0.015 122 0.320 115 0.426 -0.106 Piped water, into yard, plot 153 0.464 132 0.356 0.108 153 0.464 179 0.486 -0.022 122 0.393 115 0.357 0.037 Piped water, public tap, stand 153 0.071 132 0.136 -0.065 153 0.072 179 0.056 0.016 122 0.066 115 0.017 0.048 Dug well, protected 153 0.065 132 0 0.065 153 0.065 179 0.011 0.054 122 0.033 115 0.000 0.033 Dug well, unprotected 153 0 132 0.015 -0.015 153 0.000 179 0.000 0.000 122 0.000 115 0.017 -0.017 Spring water, protected 153 0.032 132 0.068 -0.036 153 0.033 179 0.017 0.016 122 0.008 115 0.043 -0.035 Spring water, unprotected 153 0.019 132 0.037 -0.018 153 0.020 179 0.011 0.008 122 0.008 115 0.009 0.000 Rainwater 153 0 132 0 - 153 0.000 179 0.006 -0.006 122 0.000 115 0.000 0.000 Other 153 0.026 132 0.015 0.011 153 0.026 179 0.050 -0.024 122 0.107 115 0.070 0.037 *** p<0.01 ** p<0.05 * p<0.1 *** p<0.01 ** p<0.05 * p<0.1 57 Appendix 3: Robustness Checks Mass Media Treatment Control mean (1) (2) (3) Exposure+ High exposure to the treatments 0.147 0.0266 0.0145 0.0039 (HW message through three channels) (0.040) (0.037) (0.037) Medium exposure to the treatments 0.432 0.0308 0.0145 -0.00929 (HW message through two channels) (0.061) (0.057) (0.058) Low exposure to the treatments 0.674 -0.0343 -0.0457 -0.0613 (HW message through one channel) (0.054) (0.054) (0.056) Controls NO YES YES District dummies NO NO YES + Channels: mass media (radio and/or printed materials), promotional events and personal training/educational sessions Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 Control mean (1) (2) (3) Knowledge Events that require handwashing (summary index) 0.225 0.0221 0.0215 0.02 (0.018) (0.019) (0.019) Best method to wash hands is using water and soap 0.879 0.005 0.0115 -0.003 (Percentage) (0.028) (0.025) (0.021) 0.94 -0.002 -0.001 -0.006 Main cause of diarrhea is not washing hands with water (0.016) (0.016) (0.014) and soap (Percentage) Controls NO YES YES District dummies NO NO YES Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 58 Control mean (1) (2) (3) Handwashing facilities HW Facilities stocked with water and soap (=1) 0.653 -0.029 -0.031 -0.019 (0.057) (0.058) (0.057) Water and soap anywhere in the house (=1) 0.787 -0.028 -0.029 -0.025 (0.040) (0.040) (0.040) Hand cleanliness Hands Cleanliness Index 7.527 -0.327 -0.294* -0.178 (0.204) (0.177) (0.166) [3.9%] Self-Reported handwashing behaviour HW with soap and water previous to eat 0.415 -0.166*** -0.162*** -0.156*** (0.040) (0.038) (0.040) [-40%] [-39%] [-37.5%] HW with soap and water before food prep. 0.653 0.0054 0.009 -0.007 (0.037) (0.037) (0.037) HW with soap and water after fecal contact 0.654 -0.088** -0.092** -0.084** (0.037) (0.037) (0.040) [-13.4%] [-12.8%] HW with soap and water before feeding child 0.206 0.036 0.032 0.037 (0.029) (0.030) (0.031) Controls NO YES YES District dummies NO NO YES Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 59 Control mean (1) (2) (3) Child health (children < 5 years old) Diarrhea prevalence 48 hours 0.04 0.014 0.013 0.01 (0.014) (0.013) (0.013) Diarrhea prevalence 7 days 0.06 0.019 0.017 0.011 (0.019) (0.018) (0.018) ALRI prevalence 48 hours 0.041 -0.035*** -0.035** -0.039*** (0.013) (0.014) (0.014) [-96%] [-96%] [-96.5%] ALRI prevalence 7 days 0.051 -0.042** -0.041** -0.047** (0.017) (0.016) (0.018) [-82.3%] [-80.3%] [-89%] Weight-for-age z-score -0.69 0.0616 0.037 0.106 (0.098) (0.092) (0.094) Length/Height-for-age z-score -1.453 -0.0328 -0.034 0.0371 (0.106) (0.091) (0.080) Weight-for-length/height z-score 0.203 0.119 0.12 0.140* (0.079) (0.078) (0.077) Anemia (Hb < 110 g/L) 0.283 -0.0321 -0.0374 -0.046 (0.036) (0.032) (0.033) Controls NO YES YES District dummies NO NO YES Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 60 Community Treatment Control mean (1) (2) (3) Exposure+ High exposure to the treatments 0.147 0.0750* 0.0491 0.0551 (HW message through three channels) (0.042) (0.037) (0.037) Medium exposure to the treatments 0.432 0.176*** 0.138** 0.145*** (HW message through two channels) (0.059) (0.054) (0.051) [40%] [32%] [34%] Low exposure to the treatments 0.674 0.127** 0.103** 0.106** (HW message through one channel) (0.050) (0.047) (0.045) [19%] [15%] [16%] Controls NO YES YES District dummies NO NO YES + Channels: mass media (radio and/or printed materials), promotional events and personal training/educational sessions Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 Control mean (1) (2) (3) Knowledge Events that require handwashing (summary index) 0.225 0.0152 0.007 0.009 (0.020) (0.020) (0.020) Best method to wash hands is using water and soap (Percentage) 0.879 0.0402 0.0539* 0.0512* (0.028) (0.025) (0.021) [6.1%] [5.8%] Main cause of diarrhea is not washing hands with water and soap 0.94 0.0272** 0.0320** 0.0313** (Percentage) (0.012) (0.013) (0.012) [2.8%] [3.4%] [3.2%] Controls NO YES YES District dummies NO NO YES Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 61 Control mean (1) (2) (3) Handwashing facilities HW Facilities stocked with water and soap (=1) 0.653 0.0238 0.0406 0.0326 (0.049) (0.048) (0.048) Water and soap anywhere in the house (=1) 0.787 0.0126 0.0279 0.0267 (0.039) (0.039) (0.038) Hand cleanliness Hands Cleanliness Index 7.527 0.245 0.278* 0.246 (0.184) (0.161) (0.154) [3.7%] Self-Reported handwashing behaviour HW with soap and water previous to eat 0.415 0.011 0.025 0.036 (0.046) (0.046) (0.044) HW with soap and water before food prep. 0.653 0.002 0.017 0.027 (0.043) (0.038) (0.037) HW with soap and water after fecal contact 0.654 -0.004 0.015 0.0254 (0.044) (0.042) (0.038) HW with soap and water before feeding child 0.206 0.000 0.002 0.004 (0.030) (0.030) (0.030) Controls NO YES YES District dummies NO NO YES Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 62 Control mean (1) (2) (3) Child health (children < 5 years old) Diarrhea prevalence 48 hours 0.040 -0.006 -0.003 -0.002 (0.011) (0.011) (0.011) Diarrhea prevalence 7 days 0.060 -0.001 0.001 0.002 (0.016) (0.016) (0.016) ALRI prevalence 48 hours 0.041 -0.008 -0.014 -0.011 (0.019) (0.018) (0.019) ALRI prevalence 7 days 0.051 -0.009 -0.019 -0.017 (0.024) (0.022) (0.022) Weight-for-age z-score -0.69 0.006 0.041 0.000 (0.099) (0.089) (0.085) Length/Height-for-age z-score -1.453 -0.013 -0.019 -0.056 (0.099) (0.086) (0.080) Weight-for-length/height z-score 0.203 0.074 0.111 0.076 (0.086) (0.082) (0.079) Anemia (Hb < 110 g/L) 0.283 -0.049 -0.039 -0.026 (0.037) (0.035) (0.033) Controls NO YES YES District dummies NO NO YES Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 63 School Component Sample of Community Treatment Control mean (1) (2) (3) Exposure+ High exposure to the treatments 0.2 0.0113 -0.0207 -0.0275 (HW message through three channels) (0.047) (0.043) (0.043) Medium exposure to the treatments 0.499 0.100* 0.0659 0.065 (HW message through two channels) (0.058) (0.051) (0.050) Low exposure to the treatments 0.717 0.0851* 0.0591 0.0637 (HW message through one channel) (0.046) (0.041) (0.039) [8.9%] Controls NO YES YES District dummies NO NO YES + Channels: mass media (radio and/or printed materials), promotional events and personal training/educational sessions Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 Control mean (1) (2) (3) Knowledge Events that require handwashing (summary index) 0.222 0.0188 0.00894 0.0159 (0.021) (0.020) (0.020) Best method to wash hands is using water and soap (Percentage) 0.882 0.0513** 0.049** 0.044** (0.023) (0.021) (0.022) [5.7%] [5.5%] [4.9%] Main cause of diarrhea is not washing hands with water and soap 0.94 0.0131 0.0163 0.0118 (0.016) (0.016) (0.015) Controls NO YES YES District dummies NO NO YES Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 64 Control mean (1) (2) (3) Handwshing facilities HW facilities stocked with water and soap (=1) 0.67 0.057 0.071 0.062 (0.045) (0.044) (0.043) Water and soap anywhere in the house (=1) 0.768 0.056 0.065** 0.067** (0.034) (0.033) (0.033) [8.4%] [8.7%] Hand cleanliness Hands Cleanliness Index 7.403 0.205 0.226 0.21 (0.172) (0.158) (0.151) Self-Reported handwashing behaviour HW with soap and water previous to eat 0.402 0.056 0.069 0.095** (0.050) (0.049) (0.044) HW with soap and water before food prep. 0.694 -0.052 -0.047 -0.042 (0.041) (0.040) (0.039) HW with soap and water after fecal contact 0.656 -0.0121 -0.0000743 0.0199 (0.050) (0.049) (0.043) HW with soap and water before feeding child 0.194 0.0387 0.0526* 0.0546* (0.034) (0.031) (0.030) [27%] [28%] Structured observations Observed HW with soap previous to feeding baby 0.076 -0.001 0.001 0.04 (0.048) (0.054) (0.060) Observed HW with soap after fecal contact 0.342 -0.028 -0.021 -0.022 (0.059) (0.054) (0.056) Observed HW with soap previous to preparing food 0.099 0.067 0.077** 0.068* (0.041) (0.038) (0.038) [67.6%] [77.7%] [68.6%] Observed HW with soap previous to eat 0.189 0.096* 0.101** 0.115** (0.050) (0.047) (0.049) [50.7%] [53.4%] [60.8%] Controls NO YES YES District dummies NO NO YES Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 65 Control mean (1) (2) (3) Child health (children < 5 years old) Diarrhea prevalence 48 hours 0.033 0.000 -0.002 0.001 (0.010) (0.010) (0.010) Diarrhea prevalence 7 days 0.061 -0.005 -0.005 -0.005 (0.015) (0.014) (0.014) ALRI prevalence 48 hours 0.049 -0.011 -0.017 -0.018 (0.022) (0.020) (0.020) ALRI prevalence 7 days 0.056 -0.012 -0.020 -0.021 (0.025) (0.023) (0.023) Weight-for-age z-score -0.82 0.0994 0.148 0.107 (0.094) (0.091) (0.087) Length/Height-for-age z-score -1.619 0.12 0.119 0.0833 (0.094) (0.091) (0.088) Weight-for-length/height z-score 0.165 0.0743 0.135 0.106 (0.090) (0.083) (0.083) Anemia (Hb < 110 g/L) 0.278 -0.040 -0.027 -0.022 (0.033) (0.029) (0.029) Parasites in child stools Prevalence of parasites 0.227 -0.016 -0.051 -0.045 (0.055) (0.046) (0.041) Parasite count 0.303 -0.033 -0.085 -0.068 (0.078) (0.062) (0.053) Controls NO YES YES District dummies NO NO YES Clustered standard errors in parentheses; *** p<0.01 ** p<0.05 * p<0.1 66 Appendix 4: Variable Description Variable Unit of Observation Description Total Number of Observations (endline) Individual Characteristics Age: Head of HH Head of HH Average age in years 3,518 Age: Other HH members Other HH members Average age in years 14,981 Gender (male): Head of HH Head of HH Male gender indicator 3,518 Gender (male): Other HH members Other HH members Male gender indicator 14,981 Average number of children under 5 HH Children under 5 per HH 3,563 Average household size HH Members per HH 3,547 Percent of Heads of HH that ever attended school Head of HH - 3,495 Highest educational level achieved: Head of HH Head of HH - 3,308 Percent of other HH members that ever attended Other HH members - 10,345 school Highest educational level achieved: other HH Other HH members - 9,840 members Percent of teenagers that spent time on school and Attending school individuals Positive hours indicator 19903 other activities Percent of employed heads of HH Head of HH Employment status indicator 3,514 Last week activity: head of HH Unemployed head of HH - 225 Employment status indicator, +15 Percent of employed other HH members Other HH members 5,224 years Last week activity: other HH members Unemployed other HH members - 3,328 Primary work: position Employed individuals - 6,797 Primary work: monthly salary Employed individuals - 5,850 Primary work: hours per week Employed individuals - 6797 Primary work: months worked in last 12 months Employed individuals - 6,797 Households Assets HH’s assets Household Possession of assets indicator 3,547 Percent of HHs having other piece of land Household Land possession indicator 3,547 Farm equipment possession Percent of HHs having farm equipment Household 3,547 indicator Percent of HHs having animals Household Animal possession indicator 3,547 Average number of total livestock Average number of livestock owned Household 3.547 owned Dwelling ownership Household - 3,547 67 Dwelling Characteristics Type of dwelling Household - 3,544 Dwelling light source Household - 3,547 Dwelling cooking fuel Household - 3,5447 Dwelling heat fuel Household - 3,547 Material of dwelling's walls Household - 3,544 Material of dwelling's roof Household - 3,544 Material of dwelling's floor Household - 3,543 Percent of HHs with food not covered Household - 3,055 Percent of clean HHs Household - 3,527 Percent of HHs with garbage in kitchen or house Household - 3,255 Main toilet facility Household - 3,547 Toilet Facilities Percent of toilet facilities that are public Household - 3,033 Location of main toilet facility Household - 3,547 Percent of shared toilet facility Household - 3,547 Percent of safe toilet facilities during the night Household - 3,547 Disposal of child defecation Household - 3,547 Improved water source Household - 3,492 Improved sanitation Household - 3,547 Soap and water at handwashing station Household - 3,547 Water Source Rainy station only, Smaller Source of drinking water Household 3,547 categories grouped in Other Source location Household Rainy station only 263 Covered source Household Rainy station only 263 Who mainly collects water from this source Household Rainy station only 806 Satisfied with the quantity Household Rainy station only 3,526 Does the household pay for the water Household Rainy station only 3,529 Fixed, limited quantity obtained for the payment Household Rainy station only 2,374 How water was prepared (last seven days) Household - 3,067 Wash hands after going to toilet Household Yes/No: Any HH member 3,526 Handwashing Facilities Place where usually wash hands after going to toilet Household - 3,435 68 Handwashing device, toilet Household - 3,252 Water available at handwashing station Household - 1107 Soap available Household Allows multiple answers 3,547 Wash hands before/after cooking, feeding a child Household Any HH member 3,529 Place where usually wash hands before/after cooking, Household - 3,49 feeding a child Percent of HH that have a place to wash hands inside the house or in the yard (in this case, ONLY if it less Handwashing facilities stocked with water and soap Household than 3 meters away from a 3,547 bathroom); the place must have water available and soap of any kind. Percent of HH that have a place to Water and soap anywhere in the house Household wash hands (anywhere) with water 3,547 available and soap of any kind. Exposure Measures Percent of caregivers exposed to handwashing promotion through High exposure to the treatments Caregivers three channels (mass media, 3,545 promotional events and training/educational sessions) Percent of caregivers exposed to handwashing promotion through Medium exposure to the treatments Caregivers two channels (mass media and either 3,545 promotional events or training/educational sessions) Percent of caregivers exposed to handwashing promotion through Low exposure to the treatments Caregivers one channel only (mass media or 3,545 promotional events or training/educational sessions) Handwashing Knowledge Proportion of “yes� answers to the following questions: “When do you think it is necessary Events that require handwashing (summary index) to wash your hands with soap? A) Caregivers 3,591 After cleaning baby bottom; B) After using the toilet, C) Before Eating; D) Before feeding the baby; E) Before preparing food�. Best method to wash hands is using water and soap ( Caregivers Percent of caregivers that answer 3,553 69 percent) “washing hands with water and soap� to the following question: “What is the best way to clean hands� Percent of caregivers that mention “Not washing hands with water and Main cause of diarrhea is not washing hands with water Caregivers soap� to the following question: 3,582 and soup ( percent) “What causes diarrhea?� Family-School Relationship Percent of caregivers that recall any campaign on health Caregiver of child attending - and hygiene promoted by the school 2,376 primary school Handwashing Behavior Percent of caregivers that washed their hands with soap Caregiver of child under the - 3,579 since yesterday age of 5 Caregiver of child under the Last moment of hand wash since yesterday Allows multiple answers 3,196 age of 5 Caregiver of child under the Best way to clean hands Allows multiple answers 3,553 age of 5 Caregiver of child under the Caregiver's fingernails are: - 3,556 age of 5 Caregiver of child under the Caregiver's palms are: - 3,556 age of 5 Caregiver of child under the Caregiver's finger pads are: - 3,556 age of 5 Caregiver of child under the Recalls of any handwashing campaign - 3,543 age of 5 Caregiver of child attending Participation in schools activities Allows multiple answers 3,543 primary school Index composed by three elements: • Nails: +1 if visibly dirty, +2 if apparently dirty, +3 if clean. • Palms: +1 if visibly dirty, Caregiver of child attending +2 if apparently dirty, +3 if Hands Cleanliness Index 3,549 primary school clean. • Fingerpads: +1 if visibly dirty, +2 if apparently dirty, +3 if clean. The range of the index is from 3 to 9, and it goes up when hands are 70 cleaner. Handwashing w/soap and water previous to eat Self-reported handwashing with Caregiver 3,560 soap previous to eat Handwashing w/soap and water before food preparation Self-reported handwashing with Caregiver 3,560 soap before preparing food Self-reported handwashing with Handwashing w/ soap and water after fecal contact Caregiver soap after fecal contact (after using 3,560 the bathroom/after changing diaper) Handwashing w/ soap and water before feeding child Self-reported handwashing with Caregiver 3,560 soap before feeding child Structured Observations Observed handwashing w/soap and water previous to eat Observed handwashing with soap Caregiver 1,255 prior to eating Observed handwashing w/soap and water before food Observed handwashing with soap preparation Caregiver 1,256 before preparing food Observed handwashing w/soap and water after fecal Observed handwashing with soap contact Caregiver after fecal contact (after using the 901 bathroom/after changing diaper) Observed handwashing w/soap and water before feeding Observed handwashing with soap child Caregiver 250 before feeding child Childcare Situation Percent of children with clean aspect Children under the age of 5 - 4,563 Percent of children with dirty hands Children under the age of 5 - 4,563 Percent of children with dirty finger nails Children under the age of 5 - 4,563 Percent of children with dirty face Children under the age of 5 - 4,563 Percent of children wearing clothes Children under the age of 5 - 4,563 Percent of children with dirty clothes Children under the age of 5 - 4,563 Percent of children with pot-belly Children under the age of 5 - 4,563 Percent of children wearing shoes (or shoes available) Children under the age of 5 - 4,563 Percent of children that play with household objects Children under the age of 5 - 688 Percent of children that play with toys Children under the age of 5 - 688 Average number of children's books or pictures Children under the age of 5 - 688 Percent of children that attended early education programs Children under the age of 5 - 688 Percent of adults that read books with child Children under the age of 5 - 688 Percent of adults that tell stories to the child Children under the age of 5 - 688 Percent of adults that take the child outside the home Children under the age of 5 - 688 Percent of adults that play with the child Children under the age of 5 - 688 71 Caregiver of child under the Average daily caring time - 4,404 age of 5 Child Development Z-score computed child ASQ communication skills-for-age z-score Children under the age of 5 development index based on 3,628 caregivers responses to certain child’s reactions to specific stimuli and achieved milestones. The ASQ gross-motor skills-for-age z-score Children under the age of 5 3,624 questions were based on the Ages and Stages Questionnaire and the skills measured included three domains: communications, gross- ASQ social-personal skills-for-age z-score Children under the age of 5 3,627 motor and social/personal skills. age. Average number of times child was left at the charge of Children under the age of 2 - 688 another child Average number of times was child left alone Children under the age of 2 - 688 Child Health ALRI symptoms: constant cough or ALRI prevalence (48 hours) Children under the age of 5 4,563 difficulty breathing, and raised respiratory rate (according to WHO) ALRI prevalence (7 days) Children under the age of 5 4,563 Diarrhea symptoms: 3 or more stools Diarrhea prevalence (48 hours) Children under the age of 5 4,563 per day and stools were loose or watery, or blood and/or mucus Diarrhea prevalence (7 days) Children under the age of 5 were present in stool. 4,563 Percent of children under 5 with hemoglobin levels < 100 g/L. Anemia (Hb < 110 g/L) Children under the age of 5 4,083 Hemoglobin counts were altitude- adjusted Anthropometric Measures BMI-for-age z-score Children under the age of 5 Z-scores were adjusted taking into 4,347 Head circumference-for-age z-score Children under the age of 5 account child’s age; values outside 4,375 reasonable bounds were not Length/height-for-age z-score Children under the age of 5 4,354 considered (following Arm circumference-for-age z-score Children under the age of 5 4,342 72 Weight-for-length/height z-score Children under the age of 5 recommendations by WHO) 4,343 BMI stands for body mass index. Weight-for-age z-score Children under the age of 5 4,369 Percent of children under 5 whose Prevalence of stunting ( percent) Children under the age of 5 length/height-for-age indicator is 4,354 less than 2 SD from its Z-score mean. Percent of children under 5 whose Prevalence of underweight ( percent) Children under the age of 5 weight-for-age indicator is less than 4,369 2 SD from its Z-score mean. Percent of children under 5 whose Prevalence of wasting ( percent) Children under the age of 5 weight-for-length indicator is less 4,343 than 2 SD from its Z-score mean. Water Contamination =1 if E.Coli> 0.013 NMPx mL, 0 Prevalence of E.Coli Children under 5 otherwise 575 Water collected only in T3 =1 if Total Coliforms > 0.013 NMPx Prevalence of total coliforms Children under 5 mL, 0 otherwise 575 Water collected only in T3 Parasite Prevalence Percent of children with any parasites (Trichura, Ascaris, Prevalence of any parasites Children under 5 1,552 Uncinaria or others) detected in stool samples Parasite count present in child’s stool sample. It sums 1 for each Parasite count Children under 5 different type of parasite found on 1,552 the sample (Trichura, Ascaris, Uncinaria, Others) 73