The World Bank Economic Review, 37(3), 2023, 389–408 https://doi.org10.1093/wber/lhad006 Article The Effects of Community Health Worker Visits and Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 Primary Care Subsidies on Health Behavior and Health Outcomes for Children in Urban Mali Mark Dean and Anja Sautmann Abstract Subsidized primary care and community health worker (CHW) visits are important demand-side policies in the effort to achieve universal health care for children aged under 5. Causal evidence on the interaction between these policies is still sparse. This paper reports the effects on diarrhea prevention, curative care, and incidence as well as anthropometrics for 1,649 children from a randomized controlled trial in Bamako that cross-randomized CHW visits and access to free health care. CHW visits improve prevention and subsidies increase the use of curative care for acute illness, with some indication of positive interaction effects. There is no evidence of moral hazard, such as reduced preventive care among families receiving the subsidy. Although there are no significant improvements in malnutrition, diarrhea incidence is reduced by over 70 percent in the group that receives both subsidies and CHWs. Positive effects are concentrated among children under age 2. JEL classification: I11, I12, O15 Keywords: children’s health, primary care, subsidies, community health workers, diarrhea, Mali 1. Introduction Despite impressive improvements in child and maternal health over the last decade, child mortality con- tinues to be unacceptably high in many parts of the world. In Mali, and West Africa as a whole, under-five Mark Dean is an Associate Professor at Department of Economics, Columbia University, New York, United States; his email address is mark.dean@columbia.edu. Anja Sautmann (corresponding author) is a Senior Economist in the Development Re- search Group at the World Bank, Washington, DC, United States; her email address is asautmann@worldbank.org. The re- search for this article was financed by the Population Studies and Training Center and the Rhodes Center at Brown University, and the Economic and Social Research Council/Department for International Development Frontiers Award ES/K01207X/1. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessar- ily represent the views of the World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. All errors are the authors’. The authors thank Kris Ansin, Joseph Camardo, Mariam Fofana Diallo, Pierre Pratley, Diakaridia Traoré, and the team at Mali Health; Kathleen Beegle, Christine Blandhol, Samuel Brown, Seydou Doumbia, Andrew Foster, Emily Fuller, Ira Wilson, and Xinyi Zhang; and Hamadoun Bocoum, Judith Kom, Nicoló Tomaselli, and the Mali office of Innovations for Poverty Action. A supplementary online appendix is available with this article at The World Bank Economic Review website. © 2023 International Bank for Reconstruction and Development / The World Bank. Published by Oxford University Press. 390 Dean and Sautmann mortality remained above 90 per 1,000 in 2020, with even higher rates for the poorest households.1 Many of these deaths are preventable. For example, 12.2 percent of under-five deaths in Mali in 2019 were due to diarrhea. A majority of these infections are caused by unsafe drinking water and poor sanitation, and only a small share receive the recommended treatment (UNICEF 2021). In the effort to improve health outcomes for children and mothers and achieve universal health care (UHC), the global health community has focused on two key strategies: free or heavily subsidized basic Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 health care, and home visits by community health workers (CHW) who help bridge access barriers. While user fees for basic health services were considered a tool to maintain sustainability and quality of care in the 1990s (e.g. Akin, Birdsall, and de Ferranti 1987), there is now broad consensus that African countries should work towards eliminating user fees, at a minimum for mothers and young children (e.g. McPake et al. 2008; UK Secretary of State for International Development 2009; Jamison et al. 2013). At the same time, international aid organizations and advocacy groups, like UNAIDS and the “One Million Community Health Workers” campaign, are pushing for various measures to strengthen the role of CHWs in the last-mile delivery of health care (Singh and Sachs 2013). These policies are central to ongoing or proposed health reforms both in Mali and across Africa. In 2019, Mali’s Ministry of Health announced the intent to provide free children’s and mothers’ health care within three years, although there were still gaps in covering the estimated cost of $120 million (Adepoju 2019).2 By 2016, 33 out of 47 African countries had committed to a financing strategy for universal health coverage (Barroy et al. 2016; Cotlear and Rosemberg 2018). Meanwhile, the One Million Community Health Workers Campaign estimates over 300,000 CHWs in Sub-Saharan Africa alone. Given that CHW visits to households combined with subsidized primary care are cornerstones of current global health policy for children under 5, it is crucial to learn how the individual effects of these interventions compare, and how they interact with each other. Here, we study the individual and combined effects of CHW visits and subsidies on measures of preventive and curative care for diarrhea and related health outcomes in peri-urban areas of Bamako, Mali.3 We analyze data from a randomized control trial of the Action for Health program by the NGO Mali Health. Action for Health combines biweekly CHW visits with a subsidy that covers consultation fees and standard treatment for the most common illnesses among children under five at two local clinics in Sikoro. The two program components were cross-randomized at the compound level and provided to eligible households starting in winter 2012–13. The main follow-up survey at the end of 2013 collected a detailed nine-week panel of complete health diaries to analyze demand for acute health care (Sautmann, Brown, and Dean 2020). Here, we use data from a second follow-up in 2014 that collected data on diarrhea- related behaviors and outcomes: preventive measures and knowledge indicators, incidence of diarrhea symptoms and acute care from an abridged one-week health calendar, and measures of malnutrition. We are interested in the effects on both preventive and acute care and their individual and joint impact on health outcomes. Diarrhea is one of the most important causes of death for children in the region; moreover, it provides a good case study because it is affected by the full range of behaviors that the two interventions aim to change. Preventing the intake of contaminated water or food is one of the most effective ways of avoiding diarrhea, and mild diarrhea cases can be treated at home, but serious cases (e.g., dysentery) require clinical care. In other words, there is an important role for all three: prevention, home care, and formal care.4 1 Under-five mortality in 2020 in West and Central Africa was 91.4 and in Mali 91.0; for the lowest wealth quintile in Mali it was 117.1 (2019 data); Sub-Saharan Africa as a whole had an under-five mortality of 73.3 out of 1,000 live births. Source: IGME (2021). 2 The reforms have been delayed due to the Covid-19 pandemic and political disruptions. 3 While CHW programs were originally intended for rural areas, many CHWs work in urban or peri-urban areas. For example, 6 percent of Malian CHWs are estimated to work in Bamako (Saint-Firmin et al. 2021). 4 By contrast, prevention is usually the only protection from viral disease, e.g. through vaccination. Other health issues are primarily addressed through curative care, such as bacterial infections. The World Bank Economic Review 391 There are multiple reasons to think that CHWs and subsidies (free care) interact with each other in determining health outcomes. First, CHW activities focus more on preventive care, while subsidies improve access to curative care. The two types of care could be substitutes in the health production function, that is, access to one reduces the need for the other; or complements, that is, adequate prevention and acute care combined lead to greater health improvements than the sum of the individual effects.5 In addition, there is the possibility of “spillovers” on behavior: since both preventive and curative care of Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 the child depend on their caretaker’s actions, a prevention-focused intervention might affect choices about curative care and the reverse. An important concern in this respect is moral hazard; in particular, access to free curative care may lead parents to reduce prevention efforts. Conversely, the interventions could reinforce each other’s effect on parents’ behavior. For example, Sautmann, Brown, and Dean (2020) show that CHW visits could improve utilization of formal care in the presence of a subsidy: essentially, CHWs advise parents when to seek care, and the subsidy makes it possible for parents to actually follow this recommendation. On the prevention side, receiving CHW visits as part of a larger program that also provides free access to care may increase the credibility of the CHW—or it may even act as motivation for the CHW themselves. Our empirical analysis estimates the treatment effects in the three intervention groups—CHW visits only, subsidy only, and full Action for Health program—compared to the control group. We also test whether the effect of the two interventions combined is different from the CHW-only and subsidy-only effects. We find promising effects of the CHWs on measures of knowledge and prevention. Both groups that receive CHW visits have better knowledge of the recipe for oral rehydration solution and higher self-reported use of water disinfection, although they are only weakly more likely to know the correct ages for exclusive breastfeeding, and an objective measure of water chlorine contents shows no effects. However, another objective measure, whether a mosquito bed net was hung correctly, is 14 percentage points higher with CHW visits. Most knowledge and prevention measures essentially did not change in the subsidy-only group, as one might expect, and most also showed no significant difference between the CHW-only and the full-program groups. At the same time, we find positive effects of the subsidy on curative care for children with diarrhea symptoms. The groups receiving the subsidy only and the full program receive oral rehydration treatment (ORT) significantly more often, by 17 and 27 percentage points, respectively. The group receiving only CHW visits saw a 13 percentage point increase in ORT usage but the effect is not significant. There are also large (though statistically not significant) increases in formal care in the full-program group. No such increases are seen in the CHW-only or the subsidy-only group. When examining the incidence of these effects by age, we see that the increase in formal care (by 60 percentage points) accrues primarily among children under age 2 and the increase in ORT use among aged 2–4 children (both in the full-program group). Overall, the full-program group receives better care for acute bouts of diarrhea than the other groups. We see mixed effects on anthropometric danger signs of malnutrition, with most effect sizes close to zero and only marginally significant. However, a striking finding is that the full-program group experi- ences a large and significant reduction in the number of days with diarrhea symptoms (out of the previous 7 days), equivalent to over 70 percent of the control-group average. While the effects on diarrhea inci- dence in the other two treatment groups are not significant, they are negative as well.6 We do not reject 5 We use substitutes and complements here in the game-theoretic sense: x and y are substitutes in f if the marginal return to x (y) is decreasing in y (x), i.e., the cross-derivative is negative. 6 This pattern is true when looking at either any diarrhea symptoms, or for cases in which there are more than three loose stools per day. The baseline incidence of the danger sign “blood in the stool” is low at 0.045 days, and while all three treatment groups show negative point estimates that are large in relative terms (−0.022 to −0.027), none are statistically significant. 392 Dean and Sautmann that the full-program effect equals the sum of the CHW-only and subsidy-only effects. Effects on the mother/caretaker’s concern about the child’s health are noisily estimated and not significant. Our results are overall encouraging for the combination of CHW and subsidy interventions. There is no evidence for moral hazard, and if anything, preventive behavior is slightly higher in the full-program group. Children in the full-program group also get diarrhea significantly less often and receive better care when they do fall sick. Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 The findings in this paper complement the analysis of the 2013 data in Sautmann, Brown, and Dean (2020). The 2013 paper does not analyze health outcomes or prevention, but instead quantifies the impact of subsidies and CHW visits on underuse and overuse of acute care, benchmarking demand against care- seeking recommendations defined by World Health Organization Integrated Management of Childhood Illness (WHO-IMCI) protocols. Subsidies increased care seeking by 250 percent, with most of the increase in care classified as necessary by WHO guidelines. There were no CHW effects on average, but exploratory analysis indicated (large, but statistically insignificant) a complementarity between CHWs and subsidies for the youngest children, similar to the effects on care for diarrhea in the 2014 follow-up reported here. Again, these findings are encouraging for layering CHW interventions with the expansion of UHC. In both surveys, we found the strongest effects of the CHWs for the very youngest age groups, suggesting that CHW interventions could potentially focus on children under 2 rather than under 5. The existing empirical literature provides only a partial picture of the relative health effects of subsidies and CHW visits, particularly in combination. Rigorous evidence on the health outcome effects of removing user fees for primary care is fairly scarce (Lagarde and Palmer 2011; Ridde and Morestin 2012; Dzakpasu, Powell-Jackson, and Campbell 2013). Two exceptions are a randomized control trial in Ghana that found positive health effects on children who were anemic at baseline, and a difference-in-difference study using variation in access due to apartheid policies in South Africa that found positive impacts on children’s nutritional status (Powell-Jackson et al. 2014; Tanaka 2014). There is also a literature on the effects of price and non-price screening on the take-up and targeting of specific (often preventive) health products (e.g. Cohen, Dupas, and Schaner 2015; Dupas et al. 2016; Lopez, Sautmann, and Schaner 2022). By comparison, there is a large literature on the effectiveness of CHW programs (see Lewin et al. 2010; Gilmore and McAuliffe 2013; Scott et al. 2018, for systematic reviews). However, to our knowledge, no paper has studied the interaction between CHWs and the removal of user fees—a key contribution of our work.7 Indeed, some reviews note the lack of evidence on referral practices of health workers, and the barriers that households face in following the referral recommendations of CHWs (Kamal-Yanni, Potet, and Saunders 2012; Paintain et al. 2014). One related set of papers demonstrates the important role of pro-social motivations and non-monetary incentives for CHW effort (Ashraf, Bandiera, and Jack 2014; Deserranno 2019; Wagner, Asiimwe, and Levine 2020). Wagner et al. (2020) show that CHWs who provide oral rehydration solution for free, carry out more household visits than CHWs who are allowed to charge the household and keep the revenue, and argue that this is based on CHW preferences for delivering free goods over conducting “sales visits.” This suggests there could be interaction effects of CHW and subsidy policies through CHW motivation. As we argue in Section 3.2, in our setting it is a priori ambiguous whether CHWs are more or less motivated to visit households that also receive the subsidy. None of these papers test the interaction of price changes for households (for acute care) with CHW visits. Most closely related to our work is perhaps Shapira et al. (2018), who cross-randomize demand- side incentives (in-kind rewards to households) with a supply-side expansion of CHW services (pay-for- performance incentives at the community level for volunteer CHWs) and study the effect on mothers’ 7 For example, Scott et al. (2018) classifies studies into various determinants of the success of CHWs: the health system function of the CHW, the type of health issue dealt with, training, supervision, characteristics of the CHW, logistical support and supplies, remuneration and incentives, deployment, community embeddedness, and integration with the health system. None of these categories address the cost to the household for acute-care seeking. The World Bank Economic Review 393 utilization of pre- and post-natal care. The paper does not find an effect of the pay-for-performance intervention on care use. It does not study the effects of providing free access to acute care and does not measure child health outcomes. The next section describes health care and home environments in poor areas of urban Mali, as well as the Action for Health program. Experiment and Data Collection section describes the experimental design and lays out a framework for thinking about the effects of CHWs and subsidies, individually and Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 in combination. Results section describes the empirical results and the final section provides a discussion and concludes. 2. Health Care in Mali and the Action for Health Program Public health care in Mali is built around a network of community health clinics or centres de santé communautaires (CSComs). A typical clinic in Bamako is staffed with on average about 1.5 physicians, 4 medical trainees, 5 nurses and midwives, and a lab technician, as well as technical and administrative staff (Lopez, Sautmann, and Schaner 2022). Most clinics have an attached pharmacy. At the time of the study, CSComs operated under the community-funding model of public health care, advocated by the Bamako Initiative from 1987 and endorsed by governments across West Africa, meaning that the revenues from sales of medications and other user fees fund the operation of the clinic. The public health-care system is flanked by a private formal sector with higher prices, and informal sources such as market stalls that only rarely sell prescription medications. The study was conducted in a peri-urban area of the capital Bamako, in the catchment area of two local clinics partnering with the NGO Mali Health. The compounds in our sample typically lie along unpaved roads without access to sanitation (compounds are multi-family dwellings sharing a courtyard and common facilities). In the months during and after the wet season, August–November, the incidence of diarrhea and malaria is highest. Mali has high rates of maternal and child mortality, especially in rural areas, and while poor urban populations have better health-facility access and lower rates of mortality, families still often lack basic health care. Mali’s rapidly growing urban areas resemble those elsewhere in West and Sub-Saharan Africa, although literacy rates in Mali tend to be lower. In addition, fertility and child mortality rates are higher, which makes child health interventions particularly important (Sautmann, Brown, and Dean 2020). Mali Health started their Action for Health (AfH) program in 2010. Action for Health combines subsidized health care and community health worker visits. Children are enrolled at birth (or at roll-out) and receive the program until age 5. The subsidy is administered via a personalized card that entitles the child to unlimited free consultations at a partner clinic, and free treatment and medication for any illness due to diarrhea/malnutrition, malaria, vaccine-preventable diseases, and respiratory infection (together causing the vast majority of child deaths outside of neonatal conditions, e.g. WHO (2020)). Families have to cover the remaining expenses, for example for services that are not part of the standard treatment course for a given diagnosis, and any visits to non-participating providers. The subsidy reduced the average cost to the family of visiting a formal provider by 71 percent (CFA 933 vs. CFA 2,850, approximately USD 1.89 vs. USD 5.76 at 2014 exchange rates) and 70–77 percent of households who received the subsidy reported their visit as “free,” compared with 12–14 percent for households not receiving the subsidy (Sautmann, Brown, and Dean 2020). The value of care received, using prescription records and medications taken, was similar in both groups. The activities of the CHWs build on the 13 Essential Family Practices as defined by the Malian gov- ernment. They track simple health indicators including various symptoms and danger signs, height and weight, and mid-upper arm circumference (MUAC), and advise families when to visit a doctor. They teach families how to prepare and use oral rehydration solution (ORS) in mild cases of diarrhea. They also monitor and teach preventive behaviors, such as vaccinations and bed-net use, good breastfeeding 394 Dean and Sautmann practices to protect young children from ingesting contaminated water, and hand washing. They deliver water disinfection tablets to households with unsafe water access and teach water purification. CHWs are recruited locally and their training builds on the Community-Based Integrated Management of Childhood Illness (C-IMCI) (Rosales and Weinhauer 2003), a set of guidelines for community health workers that incorporates the WHO’s and UNICEF’s “Integrated Management of Childhood Illness” recommendations (WHO 2005). For the expansion covered in this study, Mali Health hired and trained Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 two additional teams of 10 CHWs each, one covering the CHW only, one covering the full-program group. All subsidized clinical care was provided by medical professionals and staff trained in country. The clinics in this study were financially supported by Mali Health. For cost control purposes, Mali Health conducted spot checks on diagnoses and prescriptions using bills submitted to Mali Health and the clinic’s treatment records and accounting. As a result, per-visit value of care was unaffected by the interventions (see above). These quality control measures mean there may be differences in the level of care received relative to an unmonitored CSCom. However, the qualifications of care providers and the facilities and materials they can access are typical for peri-urban Mali. 3. Experiment and Data Collection The experiment took advantage of a planned roll-out wave for the Action for Health program in early 2013. Mali Health conducted a household census in 2012 within their targeted expansion area in the catchment areas of their two partner clinics at the time and selected twice as many households as could have been enrolled in the absence of the experiment, using a proxy-means test for identifying the poorest third of families. The study households are located on average approximately 700 meters from the closest clinic. All compounds with one or more eligible households were randomly assigned to either a subsidy-only, CHW-only, full-program, or control group.8 The randomization was stratified by average household assets in the compound, number of eligible children at baseline, and compound location. The original teams of 10 CHWs each in the CHW-only and the full-program groups were formed by creating pairs of CHWs of comparable quality and experience and assigning one of each pair to each treatment arm. The CHWs were assigned to location strata in an overlapping pattern designed to balance the quality of CHW services between the CHW-only and full-program groups within stratum (see also Sautmann, Brown, and Dean (2020) for more detail). The teams were trained and managed separately to reduce any potential influence of one treatment arm on CHW behavior or motivation in the other. Randomization at the compound level helps account for potential spillover effects of the CHW treat- ment due to information sharing among households who also share their living space. However, given the proximity of study compounds we cannot rule out more diffuse information spillovers through contacts outside the compound. To the extent that CHW visits improve health behaviors of control or subsidy-only households, the effect estimates we report represent therefore a lower bound of the impacts of CHWs. Since the program only included a small fraction of the population,9 we deem it unlikely that the program had significant “epidemiological” spillover effects e.g. through reduced infection rates. The baseline data was collected in fall 2012, and the first follow-up took place in 2013. The 2013 data contain detailed health calendars covering nine weeks, including all health-care visits, treatments received, and symptoms observed. These data were used in Sautmann, Brown, and Dean (2020) to study 8 The per-household cost of the CHW and subsidy components is approximately the same, so Mali Health was able to provide just one arm of the program to twice as many households. 9 Approximately 30 percent of local households in the Action for Health expansion areas, a subsection of Sikoro, were eligible for the study; about 75 percent of these households received one of the treatments, and children under 5 make up about 20 percent of people living in the study households. The World Bank Economic Review 395 the effect of the interventions on the targeting of primary care for children (utilization conditional on a classification of the child’s symptoms as either “care required” or “care not (yet) required” according to IMCI symptom charts). Our analysis here uses the baseline data along with data collected in the second follow-up survey conducted October 30–November 4, 2014. These data contain an abridged health calendar that focuses on diarrhea symptoms, along with care received over the seven days preceding the survey visit. In addition, Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 we collected anthropometrics, preventive knowledge and behavior, self-reported receipt of treatment, and other indicators, with a focus on information relevant to malnutrition and diarrhea. Enumerators were unaware of the experiment or treatment arms. An English translation of the survey instrument is available in the Online Supplemental Appendix. The original treatment groups had been kept intact for a second year, and children in enrolled families in one of the treatment arms continued to receive the subsidy, CHW visits, or both, regardless of age. However, it is worth noting that Mali Health was aware of the findings from the 2013 data collection that the CHW program had relatively low impacts on acute-care seeking (Sautmann, Brown, and Dean 2020). While the interventions were ongoing, Mali Health continued to provide additional training to CHWs. This might affect comparability of the CHW effects in 2013 and 2014. Between the two follow-up surveys, Mali Health had received a gift that allowed them to supply all beneficiary households in need with treated mosquito nets (both in the three treatment groups and in other Action for Health households not part of the sample). For ethical reasons, it was decided not to withhold this benefit from any households in the treatment groups that did not currently own treated nets. This may affect the incidence of health concerns in the three treatment groups. 3.1. Balance and Attrition The primary unit of analysis is the child or the caretaker (typically the mother). We describe the evolution of the sample here in terms of the number of children in the different waves. At baseline, we enrolled 1,732 children in the data collection. In both follow-up survey rounds, some children could not be found. In both years, five children died; additionally, families had moved or were traveling and in one case a family refused to further take part in the study. Between the baseline and 2013 survey, newborn children in the treatment households were enrolled in Action for Health either by the CHWs or by a program officer who visited study households at least every three months. This follows the standard protocol for Action for Health, and since the caretakers’ treatment choices and resulting health improvements are our outcome of interest, we consider these children part of the sample.10 It is possible that the treatment allocation affected fertility, but we consider the risk to representativeness greater if we were to exclude these children. Figure S1.1 in the supplementary online appendix includes a CONSORT diagram detailing both attri- tion and addition of new children throughout the study. After the first follow-up in 2013, which included 1,768 children, 217 children left the survey for various reasons, 12 percent of the sample. An ANOVA test fails to reject the hypothesis that attrition is equal across treatment groups. At the same time, 98 children were added to the sample.11 Again, ANOVA fails to reject the hypothesis that additions were equal across treatments. Table 1 tests for balance for a set of covariates at the child, household, and compound levels. Each column is a regression on three treatment arm dummies with stratum fixed effects and clustered standard errors, using the same specification as in the regressions below. Only one coefficient in the table is significantly different from zero. We control for these covariates in the main regressions. 10 Children born after 2013 were not added, in order to reduce the complexity of the 2014 survey and to capture health outcome effects of at least several months of program exposure. 11 These were children who were part of the baseline sample, but whose caretaker could not be interviewed in the 2013 survey round. 396 Dean and Sautmann Table 1. Balance: Covariates by Treatment Group Child Household head Household Compound Is Age Head is Majority Head Head has Head is No. of Own Log Distance male (years) literate ethnicity is male salaried job over 50 members home assets to clinic (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 CHW visits −0.015 0.075 −0.024 −0.065 −0.041 0.010 0.001 −0.133 0.024 −0.227 −0.015 (0.037) (0.106) (0.048) (0.053) (0.032) (0.032) (0.046) (0.306) (0.046) (0.268) (0.060) Subsidy 0.022 0.031 0.031 −0.049 −0.014 −0.001 −0.019 0.152 0.024 0.233 −0.034 (0.034) (0.101) (0.047) (0.052) (0.031) (0.032) (0.046) (0.299) (0.046) (0.229) (0.061) Full program 0.014 −0.001 0.107** −0.080 −0.003 −0.022 0.052 0.359 0.045 0.265 −0.055 (0.035) (0.102) (0.047) (0.052) (0.032) (0.031) (0.047) (0.285) (0.046) (0.224) (0.060) Mean 0.515 4.148 0.464 0.648 0.852 0.117 0.424 6.050 0.421 6.205 6.024 N 1,649 1,649 900 900 900 900 900 900 900 900 583 Source: Data for this study come from a 2014 survey of families enrolled in the “Action for Health” randomized controlled trial in Bamako, Mali. The survey was conducted by Brown University and Innovations for Poverty Action. Note: Each column is a regression of the variable in the top row onto the three treatment group dummies. All regressions include stratum fixed effects and cluster the standard errors at the compound level. There are between 1 and 6 households and up to 12 children in a compound. The number of observations reflects the total number of children (columns (1)–(2)), households (columns (3)–(10)), and compounds (column (11)), not including missing observations. Significance levels: *0.1, **0.05, ***0.01. Figure 1. Schematic Pathways for Treatment Effects Source: The authors. Note: The vertical arrows denote the direct effects of the subsidy treatment on acute care, and the community health workers (CHWs) on preventive care, both provided by the child’s caretaker. The diagonal solid and dotted arrows show potential “cross-effects” as well as interaction effects in the full-program arm through the behavior of caretakers and CHWs, respectively. Acute and preventive care combine in an (unobserved) health production function to determine overall health outcomes. 3.2. Framework and Hypotheses Figure 1 shows in schematic form what effects we might expect from the two treatment components, sub- sidies and CHW visits. Subsidies that reduce the cost of care at the CSCom are likely to primarily increase care seeking with a formal provider in the case of acute illness. Positive effects arise from better access to curative care, provided that the child indeed receives this care when it is needed.12 The CHW visits are more likely to improve preventive care, through teaching caretakers appropriate health behaviors, such 12 We are able to analyze in detail how the subsidy changes the utilization of (formal) curative care conditional on the child’s health status, and in particular how it affects the risk of health-care overuse (vertical arrow on left in fig. 1). This The World Bank Economic Review 397 as correct breastfeeding, hand washing, and bed-net use, and through providing supplies such as water disinfection. These direct effects are represented by the vertical arrows in fig. 1. Acute care and preventive care combine to generate better health outcomes by reducing the incidence and severity of illness. Our experimental design allows us to analyze the effects of each program component separately as well as in combination. The full program may have different impacts on health outcomes from the sum of each component alone. The first pathway for such an interaction is through the health production function Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 (bottom of fig. 1). In principle, acute and preventive care may be substitutes in the health production function: acute care may affect overall health less when prevention is strong, or conversely, prevention may be less needed when the child always receives prompt acute care in the case of illness. Alternatively, both inputs affect health in different ways and work together: prevention reduces the incidence of illness and acute care reduces the severity of any remaining illness spells. This means that their health effects are additive and may even be complements, i.e., each type of care becomes more effective in the presence of the other.13 In addition, there may be interaction effects of the two interventions on the levels of each type of care the child receives. These are represented by the diagonal arrows in fig. 1. For prevention, for exam- ple, the CHWs provide advice and instruction on sanitation and deliver water disinfectant, but it is up to the parents to actually implement the measures promoted by the CHWs, including using the disinfection tablets regularly. Caregivers may or may not increase their preventive behavior, and moreover, their re- sponse may differ depending on whether they also receive the subsidy treatment. An immediate concern is that parents take less care to use prevention knowing their child can see a doctor for free: a form of moral hazard where the beneficiaries reduce effort in response to program benefits. On the other hand, parents may be more willing to follow the advice of the CHW if they also receive the subsidy, for example because the link with the CSCom strengthens the CHW’s credibility. On the acute-care side, there is a similar possibility of an interaction effect with the CHWs through the behavior of parents. First, in mild cases of diarrhea, the caretaker should monitor hydration and may provide ORT as necessary. Rehydration solution can be prepared at home using salt, sugar, and water, and the recipe is taught by the CHWs. Positive effects on curative care may therefore arise because the CHWs can nudge the parents to be proactive and provide knowledge to deal with mild cases of illness. One of the main activities of the CHWs in the realm of acute care, however, is that they directly monitor the child’s health during the biweekly visits and in the process teach parents to spot symptoms. The IMCI algorithms are designed to discern signs of serious illness and ensure that a child in need of evaluation sees a formal care provider. We may think of this as the CHWs providing information about the child’s health status that is aimed at improving the targeting of formal care. For a child with diarrhea, a visit to a formal health-care provider is recommended in more serious cases, such as symptoms for five days or more, dehydration, or blood in the stool. As argued in Sautmann, Brown, and Dean (2020), the effect of information provision can be ambigu- ous, because the child’s caretakers will in general use new information to align care seeking with their own preferences, rather than those of the policy maker (here represented by the CHW care-seeking recommen- dations). Information on its own can improve targeting—e.g., by alerting parents who are not aware how ill their child is—but it may also have no or even a negative effect: simply put, the parent might learn to interpret symptoms but use their new knowledge to postpone a doctor visit as long as possible if they feel the illness is not serious enough. This friction is akin to moral hazard and may arise in particular risk arises more frequently in curative care than in preventive care, because formal care only benefits a child with an illness that actually requires care and responds to treatment, whereas non-indicated treatment may have no, or even negative, effects on health outcomes, not to mention that it wastes resources. The 2013 data collection and analysis in Sautmann, Brown, and Dean (2020) pursue these questions in detail. 13 We use complements and substitutes to mean that one input increases or decreases the marginal productivity of the other; see footnote 5. 398 Dean and Sautmann due to high (private) costs of formal care to the parents, vs. the (social) benefits to the child or to others. This provides a strong case for CHWs and subsidies acting as complements: the subsidy helps align the preferences of parents and policy maker by reducing private cost, while the CHWs provide the necessary information to act on these preferences.14 Lastly, while we focused here on the behavior of parents as a potential factor that introduces in- teraction effects, note that another channel is the behavior of the CHWs themselves (dotted arrows in Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 fig. 1). Specifically, it is possible that the CHWs exert different levels of effort in the CHW-only vs. the full-program group. The direction of the effect could go either way. The CHW may feel more motivated when families are able to act upon advice to seek care due to the subsidy, in particular when it comes to tracking the child’s symptoms week to week. Conversely, the CHW may perceive their work as more important when the family does not have access to the subsidy, and this may in particular affect their preventive work. 4. Results All our regressions estimate intent-to-treat effects and include three treatment group dummies—CHW visits, subsidy, and full program—along with the covariates from table 1 (dummying out missing variables) and stratum fixed effects. Each column is a regression with the variable in the top row as the outcome. In different specifications we also report p-values for various tests: (a) “CHW = full” and “subsidy = full” test whether the full program has the same effect as one of the program arms separately, and (b) “CHW + subsidy = full” tests whether the arms are complements or substitutes, that is, whether the full-program effect is larger or smaller than the sum of effects of the individual program components. All regressions cluster the standard errors at the compound level. The number of observations varies depending on whether the unit of analysis is the child, their caregiver (the respondent), the household, or the compound. For age-standardized danger signs of malnutrition, we report results only for children un- der 5. We first report the main results on prevention, acute care, and illness incidence and anthropometric outcomes, and then discuss the findings further in the next section. The analysis focuses on health care and health outcomes related to diarrhea and malnutrition.15 Diar- rhea is most frequent—and particularly dangerous—among the youngest children, and Sautmann, Brown, and Dean (2020) showed that there may be heterogeneity in the utilization of curative care by age. We therefore also analyze age-specific effects of subsidies and CHWs for acute care and illness incidence. 4.1. Prevention Behaviors and the Effects of CHWs Table 2 shows effects on indicators that measure knowledge and prevention behavior of the caregiver and should be primarily affected by the activities of the CHWs. In areas with poor sanitation, exclusive breast- feeding helps prevent gastrointestinal (GI) disease before month 6. Afterwards, safe water is an important factor in prevention. If diarrhea does occur, a simple and effective home remedy is oral rehydration treat- ment. Columns (1) and (2) in table 2 show indicators for whether the respondent could report the recipe for homemade ORS and whether they knew the age until which children are ideally exclusively breastfed (6 months). Column (3) is an indicator for whether the caretaker (mother) reports that she received wa- ter tablets as part of a health NGO program last year. Columns (4)–(5) show self-reported use of water disinfectant, and (6) shows water chlorine content measured with detection strips. Column (7) shows a 14 A direct negative effect of the CHW visits on acute care is also possible; for example, parents may not visit a doctor even when the child appears ill, perhaps in the belief that the illness could not be serious given all the preventive measures taken. This is again a form of moral hazard. We consider this less plausible as the available evidence shows that parents are very aware when their child is not well. 15 The program’s effect on malaria care and incidence would have been of interest as well, but the distribution of insecticide- treated malaria nets to all households in the three treatment groups invalidates comparisons between them. The World Bank Economic Review 399 Table 2. Prevention: Caretaker’s Knowledge and Behavior Caretaker’s knowledge Water disinfection Malaria Self-reported Measured protection Knows Knows age Given Used Used Water Bed net ORT of excl. last last today chlorine hung Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 recipe breastfeeding year 7 days content (ppm) correctly (1) (2) (3) (4) (5) (6) (7) CHW visits 0.330*** 0.096* 0.230*** 0.125*** 0.095*** 0.056 0.146*** (0.044) (0.053) (0.030) (0.038) (0.030) (0.040) (0.046) Subsidy 0.067* −0.057 0.023 −0.013 0.006 0.020 0.082* (0.037) (0.054) (0.022) (0.035) (0.027) (0.043) (0.045) Full program 0.498*** 0.046 0.357*** 0.169*** 0.083*** 0.027 0.141*** (0.040) (0.052) (0.037) (0.038) (0.028) (0.048) (0.044) Control group mean 0.167 0.554 0.008 0.120 0.058 0.170 0.662 N 1,016 1,016 1,015 1,015 1,015 1,016 1,649 p-value: CHW = full 0.00 0.33 0.00 0.26 0.69 0.51 0.91 p-value: C + S = full 0.10 0.93 0.03 0.27 0.65 0.43 0.17 Source: Data for this study come from a 2014 survey of families enrolled in the “Action for Health” randomized controlled trial in Bamako, Mali. The survey was conducted by Brown University and Innovations for Poverty Action. Note: The unit of observation is the mother/caretaker in columns (1)–(6) and the child in column (7). All regressions include stratum fixed effects, control for covariates, and cluster standard errors at the compound level. “Knows ORT recipe”: mother can explain how to make oral rehydration treatment for diarrhea. “Knows age of exclusive breastfeeding”: mother answers “6 months” when asked how long a baby should be exclusively breastfed. “Given last year” reports receiving water disinfectant tabs in the last year. “Used last 7 days”: self-reported use of water disinfectant. “Water chlorine content” measured chlorine in parts per million. “Any chlorine detected”: dummy for nonzero chlorine content. “Bed net hung correctly”: the enumerator asked to see the bed net under which the child sleeps (if any) and noted whether it was hung correctly or could be hung easily, vs. stored or not available. Significance levels: *0.1, **0.05, ***0.01. dummy for whether the mother could show the enumerator a correctly hung mosquito bed net under which the child sleeps. The CHWs have significant positive impacts on health knowledge (columns (1)–(2)) and self-reported water disinfectant use (columns (4)–(5)) in both the CHW-only and the full-program groups. In order to overcome experimenter demand effects in self-reported use, we also measured water chlorine content using test strips. There is no evidence for any objective impact of the free chlorine tablet distribution on chlorine in the water (column (6)).16 Households either overreport disinfectant use or use the tablets incorrectly. It is worth noting that nearly 50 percent of the control group show some chlorine detected in the drinking water, indicating that many households do have access to chlorinated water sources such as community taps or undertake disinfection on their own accord. In contrast, we find strong effects of both the CHW-only and the full-program intervention on the correct use of the mosquito net (column (7)). We report this measure here because it is another objective check of an important day-to-day health behavior (over and above the free distribution of the bed nets that benefited all three treatment groups, likely the cause of the weakly significant positive effect in the subsidy group). Note that even in the control group, 66 percent of children had a correctly hung net. However, the CHWs raise the share of children who sleep with mosquito protection by over 14 percentage points.17 16 The point estimates of the program impact are in some cases relatively large—one-third or the control group mean in the case of the CHW only group—but also estimated with large standard errors. 17 In table S2.1, we show self-reported receipt of a free mosquito net in the last year. The effect sizes in all three treatment groups are almost exactly identical, reflecting the distribution to all of Mali Health’s beneficiary households that did not yet own a net; see Section Experiment and Data Collection. Note also that we see these strong CHW effects despite potential information spillovers to households without CHWs that received a net, who may have been particularly motivated to seek out information on how to use it. 400 Dean and Sautmann For most measures of knowledge and prevention, there is no strong evidence of a subsidy effect, as we would expect: the point estimates are typically close to zero and in some cases negative. The two exceptions are for ORT knowledge and bed-net usage, where we find small, but significant positive effects. In most measures we also do not see a strong interaction effect between the subsidy and CHWs. There is no evidence of moral hazard, but there is also no strong positive interaction. One exception is column (1): mothers are significantly more likely to know the recipe for ORT in the full-program group Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 than in the CHW group. Despite the quantitatively large, positive effect of 6.7 percentage points in the subsidy-only group, the test that the effect of CHWs and subsidy combined equals the full-program effect is still marginally rejected at the 10 percent level, due to the strong full-program effect of a nearly 50 percent increase. We will discuss ORT more below, because it may also be prescribed or recommended by a provider and therefore constitutes a special case. The second exception is column (3), which is a self-reported measure of whether the household has ever received water disinfectant tablets from an NGO. Here, the effect of the full project is significantly larger than that of the CHWs only or of the sum of the CHW and subsidy effects. Interestingly, this holds only for this one water disinfection indicator. Self-reported use of the tablets over the last seven days is somewhat higher in the full-program group, but the difference is not significant, and self-reported use on the same day is almost identical in magnitude. We note that the CHW distribution of water tablets is partially determined by the household’s prior access to safe water: some compounds have access to communal taps or protected wells and do not receive the tablets, and this partially explains the over- all low rates of receipt. It does not explain the difference between CHW and full-program effects in column (3). As discussed earlier, it is possible that the subsidy program increases the effectiveness or motivation of CHWs, and this could be a potential source for the interaction effect in columns (1) and (3). Ideally, we would have administrative data or other objective measures of CHW effort, but these records were not available to us. We therefore analyzed other measures of self-reported intervention receipt, shown in detail in table S2.1. There is no clear evidence for a complementarity driven by higher CHW effort or credibility in the full-program group. The overall low number of reported visits suggests that many parents do not see the CHW visits as significant, do not recall them, or do not receive them, for whatever reason (this could include the respondent being absent during the day due to work). While the share of reported visits is 11–13 percent higher in the full-program arm than in the CHW-only arm, a non-negligible share of families also report CHW visits in the subsidy-only arm, significantly more than in the control. Mirroring this pattern, many families report receiving free care in the CHW-only arms, and the share that reports the subsidy in the full-program group is higher than in the subsidy-only group. Overall, the most likely explanation for these patterns seems to be that parents underreport the individual program components more often when they do not receive both program parts, perhaps because the full program is more valuable and salient, rather than differences in CHW behavior. 4.2. Acute Illness: Care Received Table 3 combines information on health-care utilization conditional on the child being ill (columns (1)– (3)) with data on the incidence of symptoms (columns (4)–(7)), obtained from a seven-day health calendar over the week preceding the survey. For diarrhea, we first asked respondents to report all days on which the child had diarrhea in the last week. If diarrhea was reported, we asked about days with more than three loose stools, and days with blood in the stool (a danger sign that may indicate dysentery). We also asked whether the respondent was concerned about the child’s health on any day. For care received, we recorded any visits to CSComs or associated reference hospitals (CSRef), any other formal provider visits (private or non-profit clinics and hospitals) and whether ORT was given. We begin by analyzing curative care, and specifically acute care received for diarrhea. Accordingly, the sample in columns (1)–(3) includes only children with diarrhea symptoms. As discussed in detail The World Bank Economic Review 401 Table 3. Acute Care for Diarrhea and Incidence of Diarrhea (Last Seven Days) Acute care received Incidence (0/1) (no. of days out of 7) CSCom/ Any ORT Any Any ≥3 loose Blood in Mother CSRef formal care given diarrhea diarrhea stools/day the stool concerned Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 (1) (2) (3) (4) (5) (6) (7) (8) CHW visits 0.041 −0.014 0.127 −0.020 −0.047 −0.046 −0.026 0.114 (0.089) (0.139) (0.108) (0.019) (0.057) (0.049) (0.027) (0.112) Subsidy −0.060 −0.088 0.174** −0.024 −0.061 −0.051 −0.027 −0.037 (0.076) (0.119) (0.079) (0.019) (0.057) (0.050) (0.023) (0.101) Full program 0.170 0.088 0.266** −0.049*** −0.149*** −0.131*** −0.022 −0.118 (0.105) (0.184) (0.121) (0.018) (0.051) (0.043) (0.026) (0.100) Control group mean 0.083 0.222 0.000 0.085 0.210 0.161 0.045 0.492 N 102 102 102 1,649 1,649 1,649 1,649 1,649 p-value: CHW = full 0.28 0.59 0.39 0.06 0.02 0.01 0.87 0.03 p-value: subsidy = full 0.04 0.31 0.45 0.14 0.04 0.02 0.79 0.40 p-value: C + S = full 0.21 0.38 0.85 0.84 0.56 0.57 0.33 0.18 Source: Data for this study come from a 2014 survey of families enrolled in the “Action for Health” randomized controlled trial in Bamako, Mali. The survey was conducted by Brown University and Innovations for Poverty Action. Note: The unit of observation is the child. “CHW visits” refers to visits from community health workers. In columns (1)–(3), the sample is restricted to children for whom the caretaker reported any diarrhea in the observation period. All regressions include stratum fixed effects, control for covariates, and cluster standard errors at the compound level. Columns (1)–(3) report measures of acute care received, including oral rehydration therapy (ORT), and include only children with any diarrhea symptoms in the last seven days. Column (1) Includes Centres de Santé communautaires (CSCom) and associated reference hospitals (CSRef). Column (2) includes private, faith-based, and other providers in addition to public clinics. Columns (4)–(8) report diarrhea symptom incidence and caretaker’s concern for the child’s health. Column (4) shows the proportion of children with any diarrhea. Columns (5)–(8) report incidence as the number of days with occurrence, out of seven. Significance levels: *0.1, **0.05, ***0.01. in Sautmann, Brown, and Dean (2020), this reflects our interest in whether care is received conditional on the child experiencing a diarrhea episode. We discuss incidence in more detail below, but note that column (4) shows the share of children with any diarrhea symptoms in the last week and therefore gives an idea of how the sample for acute-care use is selected. The share of children with any diarrhea lies between approximately 3 and 9 percent across treatment groups. Unconditional use of care is not informative about changes in caregivers’ health-care behavior, precisely because of the variation in illness incidence between treatment groups (see also below)—a parent is unlikely to give ORT to a child who does not have diarrhea.18 A downside of reporting condi- tional use of care is the differential selection of the sample of sick children. This means that the results in table 3 provide unbiased estimates of the health-care utilization rate among the subset of children who are expected to fall sick under the different program variants, but they cannot be used to directly compare care-seeking probabilities for all children (or all households). Columns (1) and (2) show no significant effect of CHW visits on care received. More surprisingly, there is no measurable effect of the subsidy on the use of formal care, either on its own (where the point estimates are negative, making a large positive effect unlikely) or as part of the full program (where the point estimate is large and positive, but not significantly different from zero). While this may seem surprising in the light of the results of Sautmann, Brown, and Dean (2020), the same paper also provides a potential explanation for this finding. Using the more detailed data from the 2013 survey, Sautmann, Brown, and Dean (2020) find that the increase in health-care utilization following the subsidy largely occurs on days 18 The overall use of care in the full sample is very low and the differences between treatment arms are not significant, but relative average formal care use echoes the estimates of unconditional treatment effects with the larger 2013 sample, where formal care use in the subsidy groups more than doubled; see Sautmann, Brown, and Dean (2020). 402 Dean and Sautmann in which formal care is required according to the WHO guidelines. Diarrhea on its own does not fall into this category, unless it has lasted for more than five days or is accompanied by other symptoms. Thus the number of diarrhea spells in the sample for which formal care is the appropriate response is likely to be even smaller than our already small set of observations, making our point estimates very noisy.19 Indeed, mild diarrhea cases can be treated with ORT, which can be prepared at home. Alternatively, pre-mixed ORS packets can be purchased, and providers may also recommend ORT in addition to or in Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 place of prescription medication. Column (3) of table 3 shows that ORT use is higher than in the control by 12.7 to 26.6 percentage points in all three treatment groups, and significantly so in the groups that receive the subsidy. The difference between the full-program group and the other groups is not significant but of large magnitude. Recall that we saw that the CHWs significantly increase knowledge of ORT among caretakers. The effect is stronger in the full-program group, and there is an effect in the subsidy- only group as well. These patterns are consistent with parents learning about ORT from CHWs as well as from CSCom staff, and possibly using the subsidy to purchase more ORS packets. The results of table 3 mask important differences across age groups. Panels 1 and 2 of fig. 2 additionally show the effects on use of any formal care (corresponding to column (2)) and ORT (column (3)) broken out by age group.20 Panel 1 shows that the increase in formal care use in the full-program group is concentrated among young children. Children under 2 who receive the full program have an over 60 percent higher chance than the control group of seeing a formal care provider, and the effect is statistically significant at the 1 percent level. Children who receive only the subsidy do not show a corresponding effect, suggesting a positive effect of the CHWs on care-seeking propensity. Panel 2 of fig. 2 shows that the age 2–4 group who receive the full program see a significant increase in ORT use (whereas the subsidy group shows moderate increases in all age groups that are not individually significant). Although we need to interpret these findings with caution due to the small sample, the evidence overall suggests that the full-program group benefits from significantly better care for acute diarrhea. Parents in the full-program group more often seek formal care for the youngest, most vulnerable children, and use ORT more often for the older children. This is consistent with the CHWs helping steer parents’ use of different curative-care options. Both the CHWs and the subsidy contribute in different ways to better awareness and use of ORT, a low-cost, effective treatment for mild diarrhea cases. The age-specific effects on formal care use complement those found in Sautmann, Brown, and Dean (2020) from the first year of the intervention: the CHWs in the full-program group appeared to increase medically recommended curative care for the youngest children, over and above the subsidy-only group. A possible explanation is that information about the child’s health status provided by CHWs improves acute-care seeking for young children, even when access is free. An alternative, program-specific explanation may be that the CHWs educate parents of newborns on eligibility or support them in claiming the subsidy benefit. 4.3. Acute Illness: Incidence Columns (4) to (8) of table 3 report on the incidence of diarrhea symptoms, as well as the subjective concern of the mother about the child’s health. Children who are being weaned are the most vulnerable to gastrointestinal problems, since they are exposed to food and water for the first time, and diarrhea incidence for the under 2 age group is typically highest. We therefore show age-specific effects on diarrhea and mother concern in panels 3 and 4 in fig. 2 as well. Strikingly, table 3, columns (4)–(6) show large and significant reductions in diarrhea incidence in the full-program group. The effect size in column (5) is equivalent to over 70 percent of the control group mean of 0.21 days. The effects in the subsidy-only and CHW-only groups are also negative but not sig- nificant. Indeed, we can reject that the effect is the same in the full-program group as it is in either of 19 The 2014 data reported here constitute a small sample of spells and do not cover a long enough period to identify spells in which care was required per WHO guidelines in an unbiased manner. 20 Note that children over 5 would have aged out of the sample under normal conditions. The World Bank Economic Review 403 Figure 2. Treatment Effects in Two-Year Age Groups, Relative to Control Group, with 95 Percent CI 1. Any formal care (conditional on diarrhea symptoms) CHW Subsidy Full program 1.5 1.5 1.5 1 1 1 Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 .5 .5 .5 0 0 0 −.5 −.5 −.5 −1 −1 −1 Age 0 Age 2 Age 4 Age 0 Age 2 Age 4 Age 0 Age 2 Age 4 2. ORT given (conditional on diarrhea symptoms) CHW Subsidy Full program 1 1 1 .5 .5 .5 0 0 0 −.5 −.5 −.5 Age 0 Age 2 Age 4 Age 0 Age 2 Age 4 Age 0 Age 2 Age 4 3. Incidence: any diarrhea (number of days out of 7) CHW Subsidy Full program .5 .5 .5 0 0 0 −.5 −.5 −.5 −1 −1 −1 Age 0 Age 2 Age 4 Age 0 Age 2 Age 4 Age 0 Age 2 Age 4 4. Incidence: concerned about child’s health (number of days out of 7) CHW Subsidy Full program 1 1 1 .5 .5 .5 0 0 0 −.5 −.5 −.5 −1 −1 −1 Age 0 Age 2 Age 4 Age 0 Age 2 Age 4 Age 0 Age 2 Age 4 Source: Data for this study come from a 2014 survey of families enrolled in the “Action for Health” randomized controlled trial in Bamako, Mali. The survey was conducted by Brown University and Innovations for Poverty Action. Note: Panels 1 and 2 report care received conditional on exhibiting diarrhea incidence (sample of 102 children); panel 3, number of days with diarrhea; panel 4, number of days caretaker was concerned. 404 Dean and Sautmann Table 4. Anthropometrics and Malnutrition Indicators Weight (kg) WfA less than −2 MUAC (mm) MUAC below 125 (1) (2) (3) (4) CHW visits 0.023 −0.060* −0.392 −0.004 (0.204) (0.036) (1.027) (0.012) Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 Subsidy −0.448** 0.012 −0.616 −0.000 (0.185) (0.036) (0.971) (0.012) Full program −0.015 −0.035 −1.206 −0.017* (0.172) (0.033) (1.020) (0.010) Control group mean 14.653 0.196 152.990 0.022 N 1,635 1,069 1,631 1,060 p-value: CHW = full 0.85 0.42 0.45 0.20 p-value: subsidy = full 0.01 0.13 0.55 0.07 p-value: C + S = full 0.13 0.78 0.89 0.39 Source: Data for this study come from a 2014 survey of families enrolled in the “Action for Health” randomized controlled trial in Bamako, Mali. The survey was conducted by Brown University and Innovations for Poverty Action. Note: The unit of observation is the child. “CHW visits” refers to visits from community health workers. Inconsistent repeat measurements for the same child were dropped. All regressions include stratum fixed effects, control for covariates, and cluster standard errors at the compound level (except regression (2) which does not include covariates). “MUAC” is mid upper arm circumference and “WfA” is weight-for-age (z-score). Values below 125 mm and −2, respectively, indicate moderate acute malnutrition for children under 5. Correspondingly, regressions (2) and (4) only include children under 5. Significance levels: *0.1, **0.05, ***0.01. the groups that received a single program arm, but do not reject that it is the sum of the two individual effects. From fig. 2, incidence reductions are driven by children under age 2, and the effect is significant at the 1 percent level in the full-program group. The magnitude is remarkable: the average number of days with diarrhea symptoms in the under 2 age group is 0.75, and the estimated effect is −0.67 days, or a reduction by 90 percent. We see negative but not significant effects on days with “blood in the stool,” which is a fairly rare danger sign. The effects in column (8) of table 3 for mother’s concern are not statistically significant, although the large negative coefficient in the full-program group indicates a similar pattern to the other measures. The effect is again strongest among under 2-year-olds but still not significant at the 5 percent level (fig. 2, panel 4). The subsidy-only group shows a smaller and also insignificant negative effect. There is an unex- plained positive effect in the CHW group, driven by children 4 years and older. If this is a robust effect, it could indicate that the CHWs make caregivers aware of potential health problems, but do not provide them with the resources to seek care. It is worth noting that mothers report on average 0.5 days out of 7 (7 percent) with a concern about their child’s health, highlighting the pressure that the caregivers of the children in our sample experience. 4.4. Malnutrition Indicators: Weight and MUAC In table 4, we report results for weight and MUAC, two indicators of malnutrition that may respond to lower incidence and better treatment of diarrhea. Columns (1) and (3) report raw treatment effects on weight in kilograms and MUAC in millimeters. Columns (2) and (4) report indicators for two danger sign indicators for children under 5 years: weight for age below −2 standard deviations, and MUAC below 125 mm. Note that many children at the time of the 2014 follow up are older than 5 and therefore excluded from these indicators. Other reasons for missing data were outliers flagged during the age normalization procedure.21 21 We used the standards defined by the UNICEF IGROWUP macro for weight for age; see the latest version at UNICEF (2019). The World Bank Economic Review 405 The effects of the full program, while generally of the right sign, are largely insignificant. The exception is a significant negative effect on the number of children exhibiting malnutrition according to the MUAC measure. The CHW-only group also has effects which are largely of the expected sign, with a borderline significant effect on the fraction of children showing weight for age below −2 standard deviations. The negative effect of the subsidy on raw weight is puzzling and may be a random outlier. A lot of precautions were taken to minimize measurement error, partly informed by the challenging measurement conditions Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 (e.g. no flat, hard surfaces to place a scale). Each child was weighed twice, and the two measurements compared. If there was a difference of more than 0.1 kg, a third measurement was taken.22 Weakly, columns (2) and (4) suggest that the CHWs might have some effect on the incidence of malnutrition danger signs. The age-specific effects in fig. S2.2 for weight for age (significant reductions in dangerously low weight for age in the under 2 age group in the CHW-only and the subsidy-only group) suggest again that program effects occur primarily at the lowest ages, likely due to the incidence profile of diarrhea, but we do not see a corresponding effect in the full-program group. One possible reason for the limited effect of the program on these measures is that the population has relatively small numbers of children exhibiting signs of malnutrition. Only 2 percent had a MUAC measurement below 125 mm in the control group, while 20 percent had a weight for age below −2 standard deviations. 5. Discussion and Conclusion Overall, our results are encouraging for combined interventions of both removal of user fees and employ- ment of community health workers who visit households in an urban context. The CHWs have significant effects on indicators of preventive knowledge and behavior, such as breast- feeding practice and knowledge of ORT. Even though we do not find significant impacts on objectively measured water disinfection, self-reported disinfectant use is higher in the CHW treatment arms. There is a possibility that self-reports are inflated due to experimenter demand effects. However, the significant effects of the CHWs on (objective) mosquito net hanging are not consistent with mere overreporting and suggest that the problem could also be incorrect use of the disinfectant by the household, or of the water tests that our enumerators used. It will be important to understand better why self-reported and measured disinfection diverge, especially in the context of evidence from Kenya that chlorine solution dispensers at water sources increase use and greatly reduce child mortality (Haushofer et al. 2021). On the curative-care side, the subsidy significantly increases the use of oral rehydration treatment for children acutely ill with diarrhea. The effects on preventive and curative-care behaviors by parents are equal or greater in the full-program group in almost every aspect of care seeking: In families receiving both interventions, caretakers are more knowledgeable about ORT and self-report higher water-disinfectant use. Moreover, formal care is in- creased significantly in the under 2 age group, while ORT use is significantly higher among 2–4-year-olds. Overall, the findings suggest better targeting and use of care in the group receiving both interventions, although with the caveat that the sample size of 102 children with diarrhea is small. The two interventions seem to work particularly well together for the use of ORT, likely due to the fact that ORT use is promoted by both CHWs and formal health-care providers. We saw that the subsidy-only group significantly more often receive ORT than the control. They are also 9.7 percentage points more likely than the control to know the recipe for ORT (table 3, column (4)). For both receipt of ORT and ORT knowledge, effects are largest for households that received both CHW visits and the subsidy. Overall, the interventions are successful at changing parents’ care-giving behavior, and we do not find any evidence of moral hazard, in the sense that households for example reduce prevention in response 22 Children were weighed on their own if over 2, and with their mother if below 2 years of age. 406 Dean and Sautmann to subsidized care. If anything, both preventive knowledge and behaviors and curative care are most improved in the full-program group. Likely as a result, we find that the full-program group has significantly lower diarrhea incidence than the other groups, driven by the youngest children, who are also the most vulnerable group, with the highest incidence in the control. The number of days with diarrhea is reduced by over 70 percent of the control group incidence for the sample as a whole, and by over 90 percent in the under 2 age group. Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 While the reduced incidence implies gains in children’s well-being, we do not find clear effects of the two interventions on indicators of malnutrition. A robust interpretation of the evidence is that the CHWs improve prevention, the subsidy improves curative care, and, when households receive both interventions, the reduction in diarrhea incidence repre- sents the sum of these two effects. There is suggestive evidence of stronger CHW effects on some preven- tion activities when parents have access to free care as well. We argued earlier for two possible channels by which the subsidy treatment might strengthen CHW effectiveness: one is through motivating CHWs, the other is through increasing their credibility with parents. The former is of interest for any CHW pro- grams, while the latter is particularly relevant for non-profit health program providers designing their own interventions, such as Mali Health. The indirect question format of the data on treatment receipt that we used introduces measurement error (see discussion in appendix S2.2), and therefore does not provide a conclusive test of these channels. However, the fact that respondents report the Action for Health benefits more often in the group receiving both interventions suggests that each may make the other more salient, possibly contributing to better health-care behaviors on the parents’ part. These possibilities may be a fruitful area for further research. Any policy recommendations of course depend on the objectives and cost constraints of the implement- ing body. However, our results make a case for the application of subsidies and CHWs in tandem. The two policies seem to address different constraints that households face, and for at least one key outcome measure—diarrhea incidence—we find that the two are at worst additive in their effects. Moreover, the impact of the individual arms on health outcomes are similar and, in the case of Action for Health, they also cost about the same, suggesting comparable marginal benefits per dollar spent. Our results also show possible areas in which programs of this type can be improved. For example, we find strong suggestive evidence that program benefits are concentrated in children under 2 years old, suggesting that resources could be most usefully directed towards the youngest age group. Data Availability Statement The data used in this paper are available in the World Bank Microdata Library at https://microdata. worldbank.org/index.php/catalog/5683. References Adepoju, P. 2019. “Mali Announces Far-Reaching Health Reform.” Lancet 393(10177): 1192. Akin, J., N. Birdsall, and D. de Ferranti. 1987. Financing Health Services in Developing Countries: An Agenda for Reform. Washington, DC: World Bank Group. http://documents.worldbank.org/curated/en/ 585551468345859470/Financing-health-services-in-developing-countries-an-agenda-for-reform. Ashraf, N., O. Bandiera, and B. K. Jack. 2014. “No Margin, No Mission? A Field Experiment on Incentives for Public Service Delivery.” Journal of Public Economics 120: 1–17. Barroy, H., L. Musango, J. Hsu, and N. Van de Maele. 2016. Public Financing for Health in Africa: From Abuja to the SDGs. Geneva: World Health Organization. https://www.afro.who.int/publications/public-financing-health-africa- abuja-sdgs. Cohen, J., P. Dupas, and S. Schaner. 2015. “Price Subsidies, Diagnostic Tests, and Targeting of Malaria Treatment: Evidence from a Randomized Controlled Trial” American Economic Review 105 (2): 609–45. The World Bank Economic Review 407 Cotlear, D., and N. Rosemberg. 2018. Going Universal in Africa: How 46 African Countries Reformed User Fees and Implemented Health Care Priorities. Universal Health Coverage Studies Series; No. 26. © World Bank, Washington, DC. http://hdl.handle.net/10986/29177. License: CC BY 3.0 IGO. Deserranno, E. 2019. “Financial Incentives as Signals: Experimental Evidence from the Recruitment of Village Pro- moters in Uganda.” American Economic Journal: Applied Economics 11(1): 277–317. Dupas, P., V. Hoffmann, M. Kremer, and A. P. Zwane. 2016. “Targeting Health Subsidies through a Nonprice Mech- Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 anism: A Randomized Controlled Trial in Kenya.” Science 353(6302): 889–95. Dzakpasu, S., T. Powell-Jackson, and O. M. Campbell. 2013, 01. “Impact of User Fees on Maternal Health Service Utilization and Related Health Outcomes: A Systematic Review.” Health Policy and Planning 29(2): 137–50. Gilmore, B., and E. McAuliffe. 2013. “Effectiveness of Community Health Workers Delivering Preventive Interventions for Maternal and Child Health in Low- and Middle-Income Countries: A Systematic Review.” BMC Public Health 13(1): 847. Haushofer, J., M. Kremer, R. Maertens, and B. J. Tan. 2021. “Water Treatment and Child Mortality: Evidence from Kenya.” NBER Working Paper 29447. IGME U. 2021. “UN Inter-Agency Group for Child Mortality Estimation: Under Five Mortality.” Accessed January 28, 2022. https://childmortality.org/data. Jamison, D. T., L. H. Summers, G. Alleyne, K. J. Arrow, S. Berkley, A. Binagwaho, and F. Bustreo et al. 2013. “Global Health 2035: A World Converging within a Generation.” Lancet 382(9908): 1898–955. Kamal-Yanni, M. M., J. Potet, and P. M. Saunders. 2012. “Scaling-Up Malaria Treatment: A Review of the Performance of Different Providers.” Malaria Journal 11(1): 1–10. Lagarde, M., and N. Palmer. 2011. “The Impact of User Fees on Access to Health Services in Low- and Middle-Income Countries.” Cochrane Database of Systematic Reviews 2011(4): CD009094. Lewin, S., S. Munabi-Babigumira, C. Glenton, K. Daniels, X. Bosch-Capblanch, B. E. Van Wyk, and J. Odgaard-Jensen et al. 2010. “Lay Health Workers in Primary and Community Health Care for Maternal and Child Health and the Management of Infectious Diseases.” Cochrane Database of Systematic Reviews 2010 (3): CD004015. Lopez, C., A. Sautmann, and S. Schaner. 2022. “Does Patient Demand Contribute to the Overuse of Prescription Drugs?” American Economic Journal: Applied Economics 14(1): 225–60. McPake, B., A. Schmidt, E. Araujo, and C. Kirunga-Tashobya. 2008. Freeing Up Healthcare: A Guide to Removing User Fees. London: Save the Children. https://resourcecentre.savethechildren.net/pdf/1975.pdf/. One Million Community Health Workers Campaign. 2021. “Operations Room.” Accessed February 1, 2022. http://1millionhealthworkers.org/operations-room-map/. Paintain, L. S., B. Willey, S. Kedenge, A. Sharkey, J. Kim, V. Buj, J. Webster, D. Schellenberg, and N. Ngongo. 2014. “Community Health Workers and Stand-Alone or Integrated Case Management of Malaria: A Systematic Literature Review.” American Journal of Tropical Medicine and Hygiene 91(3): 461. Powell-Jackson, T., K. Hanson, C. J. Whitty, and E. K. Ansah. 2014. “Who Benefits from Free Healthcare? Evidence from a Randomized Experiment in Ghana.” Journal of Development Economics 107: 305–19. Ridde, V., and F. Morestin. 2012. “A Scoping Review of the Literature on the Abolition of User Fees in Health Care Services in Africa.” Health Policy and Planning 27(3): 1–11. Rosales, A., and K. Weinhauer. 2003. C-IMCI: Community Integrated Management of Childhood Illness. Catholic Relief Services. https://www.crs.org/our-work-overseas/research-publications/community-integrated-management- childhood-illness. Saint-Firmin, P.P., B. Diakite, K. Ward, M. Benard, S. Stratton, C., Ortiz, ..., and S Traore. 2021. “Community Health Worker Program Sustainability in Africa: Evidence From Costing, Financing, and Geospatial Analyses in Mali.” Global Health: Science and Practice 9(Supplement 1): S79–97. Sautmann, A., S. Brown, and M. Dean. 2020. “Subsidies, Information, and the Timing of Children’s Health Care in Mali.” World Bank Policy Research Working Paper (WPS9486). Scott, K., S. W. Beckham, M. Gross, G. Pariyo, K. D. Rao, G. Cometto, and H. B. Perry. 2018. “What Do We Know about Community-Based Health Worker Programs? A Systematic Review of Existing Reviews on Community Health Workers.” Human Resources for Health 16(1): 39. Shapira, G., I. Kalisa, J. Condo, J. Humuza, C. Mugeni, D. Nkunda, and J. Walldorf. 2018. “Going beyond Incentivizing Formal Health Providers: Evidence from the Rwanda Community Performance-Based Financing Program.” Health Economics 27(12): 2087–106. 408 Dean and Sautmann Singh, P., and J. D. Sachs. 2013. “1 Million Community Health Workers in Sub-Saharan Africa by 2015.” Lancet 382(9889): 363–5. Tanaka, S. 2014. “Does Abolishing User Fees Lead to Improved Health Status? Evidence from Post-Apartheid South Africa.” American Economic Journal: Economic Policy 6(3): 282–312. UK Secretary of State for International Development. 2009. Eliminating World Poverty: Building Our Common Fu- ture. London: Department for International Development White Paper. Downloaded from https://academic.oup.com/wber/article/37/3/389/7167207 by World Bank and IMF user on 14 September 2023 UNICEF. 2019. “Update of IGROWUP Stata Macro Prevalence Module.” Accessed January 28, 2022. https:// github.com/unicef-drp/igrowup_update. ———. 2021. “Diarrhoea.” Accessed: January 28, 2022. https://data.unicef.org/topic/child-health/diarrhoeal-disease/. Wagner, Z., J. B. Asiimwe, and D. I. Levine. 2020. “When Financial Incentives Backfire: Evidence from a Community Health Worker Experiment in Uganda.” Journal of Development Economics 144: 102437. World Health Organization. 2005. Handbook: IMCI Integrated Management of Childhood Illness. World Health Organization. https://apps.who.int/iris/handle/10665/42939. ———. 2020. “Children: Improving Survival and Well-Being.” Accessed September 8, 2020. https://www.who.int/ news-room/fact-sheets/detail/children-reducing-mortality.