Community Health Educators and Maternal Health: Experimental Evidence from Northern Nigeria

Abstract The slow pace of improvement in service delivery and health outcomes for pregnant women and newborns in developing countries has been a major concern for policy makers in recent decades. This paper presents the results from a randomized controlled trial of a community health worker program designed to enhance uptake of child and maternal health services in Northern Nigeria. Three interventions were evaluated: the deployment of community health educators, health educators with the provision of safe birth kits, and health educators with community dramas. The results suggest that the interventions increased utilization of antenatal, postnatal, and infant care. Maternal and newborn health practices improved as well as health knowledge. In addition, the community health worker program was more effective when supplemented with additional program components.


Introduction
Enhancing health outcomes for pregnant women and newborns in developing countries has been a key policymaking goal over the last 20 years, but progress has been stubbornly slow. The World Health Organization (WHO) estimates that more than 300,000 women and more than two million newborns die each year during pregnancy and childbirth (WHO, 2016). The majority of maternal and newborn deaths can be prevented through the appropriate provision of antenatal care in pregnancy, skilled care during childbirth, and postnatal care after delivery (Goodburn & Campbell, 2001). However, availability and utilization of these services remains low in much of the developing world (WHO, 2016). Barriers to utilization of maternal health care include an inability to access services due to transportation, logistical or financial enhanced program intensity. Finally, this evaluation provides experimental evidence around the efficacy and feasibility of the distribution of birth kits outside of the formal health system, and evidence on whether community dramas increase the take-up of formal health care services.

Institutional setting
This program evaluation was conducted in 96 rural communities in Jigawa state in Northern Nigeria, an environment that ranks among the most challenging in the world for maternal and child health. Though Nigeria accounts for only 2 per cent of the world's population, it accounts for 10 per cent of worldwide maternal deaths (Hogan et al., 2010).
Jigawa is a poor, rural state with an overwhelmingly Muslim and Hausa-speaking population (EIU Canback, 2016), characterized by extremely poor health outcomes. The maternal mortality rate in Jigawa state in particular is estimated to be around 1012 per 100,000 live births, compared to 576 per 100,000 nationwide (Sharma, Brown, Kainuwa, Leight, & Nyqvist, 2017). In the 2013 Demographic and Health Survey, Jigawa reported the third lowest rate of facility births in Nigeria; 7 per cent of women in the state reported their most recent birth was in a health facility, compared to a national average of 36 per cent (NPC and International, 2014). In addition, this region of Nigeria has been particularly affected by the ongoing violence linked to the Boko Haram rebellion. Between 2012 and 2016, the Armed Conflict Location and Events Dataset (ACLED) reported 23 violent events in Jigawa, of which at least 6 are linked to Boko Haram, and the close neighboring states of Yobe and Bauchi reported 224 and 116 violent events respectively.
A survey of 24 primary health centers serving the evaluation sample was conducted to provide additional information about the availability of formal health care services in the area, and summary statistics are reported in Table S1 in the supplementary material. Primary health centers provide basic preventive and curative care. While a majority of facilities report having access to electricity and water over the last six months, only 4 per cent report access to a telephone, and 21 per cent report access to an ambulance. Eight of the 24 primary health centers reported having a general doctor on staff, but no health center was staffed by an obstetrician or other specialist.

The program
This project evaluates the impact of a basic community-based health worker intervention implemented in conjunction with two additional health programs in Northern Nigeria. These interventions were implemented in areas that had benefited from the Midwives Service Scheme (MSS), a program launched by the Nigerian government in 2009 to fund the deployment of midwives to rural primary health centers .
In Northern Nigeria, our partner organization, the Planned Parenthood Federation of Nigeria (PPFN), rolled out these three interventions to stimulate utilization of maternal health services that were newly available under the MSS. 2 The core intervention entailed the deployment of community health educators, designated community resource persons (CORPs). CORPs were defined as 'a bridge/link crucial to foster trust, confidence and acceptance between the midwives and their clients and … increased access to maternal health services' (PPFN, 2012b). Women between the ages of 20 and 45 who were married, widowed, or divorced and possessed a minimum of primary school education were eligible to serve as CORPs. Recruitment was managed by PPFN.
CORPs were supposed to provide health information to pregnant women in a series of up to six home visits. The objective was to conduct visits at different points during pregnancy in order to monitor women's health status, encourage them to seek formal health care if needed, and promote the benefits of delivery at the primary health center. Importantly, the CORPs are not health workers and do not provide any health services (that is, they cannot provide antenatal, delivery, or postnatal services). Moreover, qualitative data collection implemented by the research team as well as ongoing monitoring by the intervention team suggests there is no evidence that CORPs were simultaneously serving as traditional birth attendants.
At the launch of the intervention in 2012, the CORPs received a one-week training, and a small monthly stipend of 2000 naira (about $5) thereafter; the absence of any substantial financial incentive is consistent with the evidence that community health workers are not highly responsive to financial incentives (Ashraf, Bandiera, & Jack, 2014). Each CORP was responsible for roughly 150 households, and given the estimated birth rate, would conduct approximately 15 household visits to pregnant women each month.
In the first treatment arm, the basic CORPs intervention was implemented in isolation, but in the other two treatment arms, it was supplemented by additional interventions. In the second treatment arm, CORPs were provided with birth kits (also known as safe delivery kits or clean delivery kits) to distribute at no cost to all pregnant women in their third trimester. The objective was for women to utilize the sterile birth kits during delivery either at home or in the health facility. The kits included a plastic sheet for the woman to lie on during delivery, surgical gloves for the birth attendant, a sterilized razor and cord clamps to cut and tie the umbilical cord, methylated spirit, clean gauze, swabs and perineal pads to be used by the mother after birth, a gallipot, a mechanical suction tube to clear secretions from the baby's airways, and a wrapper and diapers (PPFN, 2012a). All materials are packaged in a single sterile unit and had an estimated cost per kit of $8. 3 The kits distributed in this intervention were comparable to those promoted by the World Health Organization as a promising strategy to strengthen hygienic standards for deliveries conducted at home and in low-capacity facilities (Hundley, Avan, Ahmed, & Graham, 2012).
The goal of the birth kits intervention is twofold: first, utilization of the sterile materials by a birth attendant will reduce infection risk and thus reduce morbidity and mortality for mothers and infants. Second, provision of materials may increase utilization of health services by ensuring that women are confident that appropriate supplies will be available for their delivery (if they bring the kit to the facility), and by reducing the perceived risk that they will be requested to pay for materials. This second channel of increased utilization may also reduce maternal and neonatal morbidity and mortality risk.
In the third treatment arm, the basic CORPs program was implemented in conjunction with a series of community dramas conducted in order to promote the importance of safe motherhood to a broader audience. The dramas were conducted quarterly by a local drama organization called YARAC (Youth, Adolescent, Reflection and Action Center), and included a performance followed by a community forum. The objective of this intervention was to shift norms around the utilization of maternal health care, and target community members who are primary decision makers but might not be well-informed about maternal health challenges, especially men but also elders and traditional leaders. The implementing organization reported that the dramas were widely attended by both men and women in the villages, and that was also confirmed in video recordings from some sample villages.
All three interventions were targeted to address the challenges around maternal and child health observed at baseline. Given low utilization rates of health services and perceived low levels of trust between communities and facility staff, the CORPs intervention was designed to provide information about the newly available midwife services, increase knowledge about the benefits of utilizing these public health services, and increase confidence in service quality. The CORPs also sought to enhance maternal and neonatal health by providing information about recommended health practices including nutrition during pregnancy, danger signs during pregnancy, delivery and the postpartum period, and breastfeeding and infant care. The birth kits intervention was designed to provide a safer, more sterile environment for women during delivery, regardless of whether they delivered at home or in the health facility. Finally, the drama intervention aimed to change the social norms around safe motherhood among men and other community stakeholders, given the evidence that health care decision-making in this context is dominated by men.
The basic CORPs intervention and the birth kit intervention were launched in January 2013, and the drama intervention was launched in the third quarter of 2013. Data collection was conducted between 2012 and 2016. A timeline is provided in Figure 1.

Related literature
There is increasing evidence around the effects of community health worker programs, and systematic reviews of randomized controlled trials of these interventions have found mixed evidence, reflecting the wide heterogeneity in supervision and incentives provided to the health workers themselves (Lewin et al., 2010;Okwundu, Nagpal, Musekiwa, & Sinclair, 2013). Bj€ orkman Nyqvist, Guariso, Svensson, and Yanagizawa-Drott (2019) analyze a microentrepreneurship program in Uganda in which community health workers were incentivized to provide education and sell health products during home visits to households with children younger than five years old, and report a significant reduction in child mortality over a threeyear period. Comfort et al. (2016) and Comfort et al. (2019) analyze the effect of programs that engage community health workers in distributing pregnancy tests on utilization of both antenatal care and contraceptive services. Our paper contributes to this broad literature by analyzing the effects of community health workers in a region where utilization of formal maternal and child care is particularly low, and in a context of ongoing civil unrest, making utilization of health services even more challenging.
There is also a limited evidence base around the use of safe delivery kits, an intervention widely promoted by the international health community (including the WHO) in recent years (WHO, 1996). The majority of the evaluations are small-scale and non-experimental, and focus on safe birth kits distributed through the formal health system (Darmstadt et al., 2009;Winani et al., 2007). This study adds to this literature by evaluating take-up and usage of safe birth kits distributed through community channels to women at their home, and in a context where delivery outside the facility is the norm.
Finally, our paper also joins a small but growing debate around the returns to utilizing formal health care for delivery. A recent global analysis estimates mortality due to low-quality care and finds that poor quality of health care is a larger contributor to excess mortality globally than low utilization of care (Kruk et al., 2018). Okeke et al. (2016) found that the expansion of the Midwives Service Scheme in Nigeria itself had no significant effect on maternal or Community health educators and maternal health 77 neonatal outcomes in a quasi-experimental evaluation. 4 In this paper, we are exploring a community-based strategy designed to enhance maternal and child health: training volunteers with no formal education as community health workers in order to encourage pregnant women to utilize the existing formal health system.

Evaluation design and sample
The study was a parallel-group, stratified cluster randomized controlled trial, where randomization was conducted at the community level. The evaluation design called for the inclusion of 96 clusters (communities), with 24 clusters in each of four arms and where each community included around 500 households. The surveyed sample comprises 15 per cent of households reporting a woman of reproductive age at baseline. Births among the sampled women are monitored continuously for a two-year period of intervention implementation, and all women are surveyed again at endline. Given this design, the evaluation was powered to detect a 25 per cent decrease in the maternal morbidity and neonatal mortality rates with 90 per cent power to detect an effect in a one-sided test at a 10 per cent significance level. 5 Ethical approval for the trial was provided by the Massachusetts Institute of Technology and the Jigawa State Operations Research Advisory Committee (ORAC).
In order to identify the sample communities, we first identified local government areas (a subdivision analogous to a district). Jigawa state has 27 local government areas (LGAs) and out of those, 24 LGAs containing a primary health center receiving services under the Midwife Services Scheme were included in the evaluation. Within each of the 24 LGAs, we identified all villages of the target size that were located in the catchment area of the MSS facility. Thereafter, a random subset of 96 communities in total were selected to be included in the evaluation sample. Sampling procedures are described in more detail in Section S1, but in brief, we selected any community in the target facilities' catchment areas that had a reported population between 500 and 850 households for inclusion in the sample. If there were more than four communities of this size, we randomly selected a subset of these communities; if there were too few communities of the target size, we identified clusters of adjacent villages and jointly treated them as a cluster.
In sum, each of the three treatment arms and the control arm include 24 clusters and approximately 1600 women in fertile age were surveyed in each arm. Figure 2 provides the trial profile. Finally, we conducted a partial census in all communities that was employed as the basis for baseline sampling. 6

Data
The evaluation included three phases of data collection: baseline, ongoing data collection (surveys conducted 3 days and 28 days after birth in addition to a sample-wide audit of recent births), and an endline survey, as summarized in the timeline in Figure 1. 7 The baseline survey was conducted between March and June 2012 and included 7069 households with a female household member of reproductive age (corresponding to roughly 15% of the listed households). More details are provided in Panel A of Table S2 in the supplementary material. The respondent of the baseline survey was a woman in the household of reproductive age (between 15 and 49). If more than one eligible woman was present, the respondent was randomly selected utilizing an on-the-field randomization protocol.
Ongoing data collection entailed continuous monitoring of births among the baseline sample during the two-year study period. Female monitors in each community were recruited and trained to provide a cardboard chip to all baseline households and send a simple text message by SMS to the survey team following a birth or infant death in the baseline households. The SMS messages were redirected to one of our survey enumerators, who had the responsibility of following up on the SMS messages by identifying the household and conducting surveys 3 days and 28 days after birth. 8 Continuous monitoring and data collection was initiated in November 2013 and continued until November 2015. Unfortunately, this process was incomplete given that female monitors in the villages did not fully comply and failed to identify some births in the baseline sample households. More specifically, in total 1791 3-day and 28-day surveys were conducted, corresponding to 41 per cent of all births reported for the sample during this period. The survey team also conducted an audit between January and March 2015; approximately 70 per cent of baseline households were reached in the audit survey (4674 households), and enumerators conducted detailed surveys on 802 additional births in 802 distinct households; the majority of these surveys were for births that had not been included in the ongoing data collection.
Finally, the endline survey was conducted between February and July 2016. 9 We successfully re-surveyed 90 per cent of the baseline sample or 6350 households. Some households that could not be reached during the endline survey (households that had migrated, divorced, etc.) had already been surveyed previously in the post-birth or audit surveys. Therefore, the total number of women observed in any follow-up data collection was slightly larger than the endline sample (6494 women, or 92% of the original sample). Anthropometric data collection was conducted in all surveys and was implemented using SECA portable stadiometers, digital baby scales, and measuring mats and MUAC measuring tapes following standard procedures (Cogill, 2003).
Given that the entire state was significantly affected by ongoing violence and instability linked to Boko Haram attacks during the evaluation period, we regard the observed pattern of attrition as low. In the analysis, we find no evidence that the correlation between baseline characteristics and a dummy for attrition differs significantly across treatment and control arms. More details on attrition are found in Section S6.

Outcomes and statistical framework
We analyze the impact of the interventions on five primary categories of outcomes: intervention exposure, utilization of maternal health care, maternal and newborn health practices, child anthropometrics, and maternal and child morbidity and mortality, as well as secondary outcomes related to knowledge and attitudes and fertility. These variables were all included in the endline survey as well as in the post-birth survey (3-day or 28-day).
We begin by analyzing patterns of household exposure to the community health educator program, and identify whether the intensity of the CORPs program was different when additional interventionssafe birth kits and community dramaswere implemented in conjunction with the basic CORPs program. In particular, we report whether women who experienced Community health educators and maternal health 79 a pregnancy during the follow-up period had received any visit from a CORP member, and how many visits. For the birth kits intervention, we report whether the woman had received a kit, if she knows how to use the kit, whether she can name at least one item in the kit, and whether she reports utilizing the kit in her most recent delivery. Finally, we report whether the woman and her spouse attended the community drama.
Thereafter, we analyze the impacts of the intervention on health service utilization for maternal health care. Variables of interest capturing the utilization of maternal health care include the respondent's reported antenatal care visits, the number of visits, an index of antenatal care quality, and whether the woman received a postpartum check-up within two months. For delivery care, variables of interest include whether the program impacted facility-based delivery or skilled attendance at birth. We also analyze intervention effects for maternal and newborn health practices: whether the respondent developed a birth plan, breastfeeding practices, and immunizations and infant check-ups. We further analyze an index of maternal morbidity (during pregnancy, delivery, and postpartum), an index of neonatal morbidity and infant and neonatal mortality. Finally, we analyze child health outcomes as captured by height-for-age, weight-for-age, and MUAC-for-length, measured at endline for all children under five in households reporting a birth during the intervention period. In addition to these five primary outcome categories, we also analyze a set of additional secondary outcomes capturing knowledge and attitudes and fertility.
3.5.1. Sample for primary analysis. The sample for the analysis of the primary outcomes linked to maternal and newborn health is restricted to the subset of women who reported a birth (live birth or stillbirth) during the intervention period, or women who were pregnant at endline. It is important to note that women who had not been pregnant during the intervention period do not report any pregnancy or delivery-related outcomes, and are thus not included in the analysis. 10 Specifically, 4420 births corresponding to 4290 women are observed in the intervention period; thus 61 per cent of the original sample reported at least one birth.
Selection into this subsample is potentially a source of bias if the intervention affected choices around fertility or birth timing. Given the relatively short evaluation timeline, the vast majority of women delivered only once, and thus there is very limited scope to observe any effect on timing following the first birth. Only 130 women, or 2 per cent, reported two births during the period. Moreover, since CORPs did not interact with women until they were pregnant, there is similarly very little scope for an effect on birth timing for the first birth during the evaluation period. 11 3.5.2. Econometric specification. The primary specification used to estimate the effect of the interventions on an outcome variable X bicg for birth b observed for respondent i in cluster c in LGA g can be written as follows: CORPS cg , Kits cg , and Drama cg are indicator variables denoting treatment assignment, and v icg is a vector of baseline control variables. 12 All regressions include LGA fixed effects l g , and standard errors clustered at the cluster level.
For some outcomes, we have a group of related outcome measures. To assess the impact of the intervention on a set of K related outcomes, we follow Kling, Liebman, and Katz (2007) and estimate a seemingly unrelated regression system. We then derive average standardized treatment effects,b ¼ 1 K P K k¼1b k r k , whereb k is the point estimate on the treatment indicator in the kth outcome regression andr k is the standard deviation of the control group for outcome k (Duflo, Glennerster, & Kremer, 2007).
In addition, we report at the bottom of each table tests of equality across b 1 , b 2 , and b 3 . In addition, we estimate a separate regression including a dummy variable for assignment to the basic CORPs-only arm and a dummy variable for assignment to any 'enhanced intervention' arm (CORPs and birth kits, or CORPs and dramas) and report a test b 1 ¼ b enhanced , where b enhanced denotes the effects of the enhanced interventions. This test allows us to identify whether the effect of the arms including enhanced interventions is significantly larger. Most important, we report a joint test of the impact of any intervention b 1 ¼ b 2 ¼ b 3 ¼ 0 to evaluate whether any intervention including the CORPs program had a statistically significant effect.

Balance at baseline
Tables S3 and S4 in the Supplementary Material present mean pre-treatment characteristics for the treatment (CORPs only, CORPs and birth kits, and CORPs and community dramas) arms as well as the control arm. We report in the tables the p value on the joint test The average respondent is age 28 at baseline, with a reported age at marriage of 15; around a third live in polygamous households, and only about 17 per cent of the respondents report ever attending school. The baseline values of the main outcome variables of interest are reported for the subset of 4007 women who reported having had a birth during the two years preceding the baseline survey. At baseline, around 65 per cent of women report having utilized antenatal care in their most recent pregnancy, but only 9 per cent of women reported their most recent delivery was in a health facility.
In the last row of Table S4, we report the p value corresponding to the joint F-test of the hypothesis b 1 ¼ b 2 ¼ b 3 , testing across all variables examined in both tables of baseline characteristics. From this joint test we can conclude that there are no significant differences in observable characteristics in aggregate when comparing across arms.

Intervention exposure, care utilization, and health practices
We first analyze the evidence around pregnant women's exposure to the interventions, patterns of utilization of maternal health services, and health practices. Table 1 reports the results of estimating Equation (1) for variables capturing intervention exposure. Pregnant women in all three treatment arms were significantly more likely to report a visit by a CORP during their pregnancies compared to pregnant women in the control group, despite the fact that we observe some spillovers in the control arm. 13 Intervention coverage was 37 per cent higher in the CORPs-only communities compared to the control group, when coverage is defined as the percentage of women reporting a birth who had a visit from a CORPs.
We find even higher coverage in the treatment arms that had the CORPs program implemented in conjunction with an additional health program: a 12.6 percentage increase in birth kits communities and a 10 percentage point increase in drama communities. The difference in program coverage between the basic CORPS program and the enhanced CORPs program is statistically significant at the 1 per cent level, suggesting that CORPs seem to have been more motivated to identify and visit pregnant women when other interventions entailing provision of health inputs or community education were also implemented. On average in the three treatment arms, 22 per cent of women reporting a birth interacted with a CORP during the program period, and the difference between the treatment and the control arms is statistically significant. Conditional on reporting a CORPs visit, the average respondent reported 1.6 visits in total. Within the sample of women who reported a CORP visit, the most common topics reported discussed were the importance of antenatal care (70%) and healthy pregnancy practices (69%), followed by the importance of delivering at a clinic (63%).

Notes:
All regressions include LGA fixed effects and standard errors clustered at the cluster level. The sample includes all respondents reporting a pregnancy during the intervention period who were interviewed in at least one follow-up survey. We report tests of equality across the estimated coefficients; a test of the hypothesis that the joint effect of treatment is zero (b and a test of equality of impact across the basic arms and the arms including additional interventions (drama and birth kits) Asterisks denote significance at the 10, 5, and 1 per cent level. 82 M. Bj€ orkman Nyqvist et al.
Next, Column (3) in Table 1 reports results for the birth kit arm. Here, we observe that among the 28 per cent of women who reported they received a visit from a CORP, only 8.8 per cent reported they received a birth kit. When asked specific knowledge questions about the birth kits, 7 per cent of respondents stated they knew how to use the kit and 8 per cent could name at least one item included in the kit. Clearly, the birth kit intervention was not as inclusive as intended, as only a small fraction of pregnant women received and used the kits, and the reason for this pattern is unclear.
By contrast, the drama intervention was more successful in achieving coverage, as reported in Columns (7) and (8) in Table 1. We find that 36 per cent of respondents in the drama arm state that events were conducted in their community, and 27 per cent state that they attended. Among respondents who attended a drama, 32 per cent report their husbands also attended, and 51 per cent report that they discussed the content with a relative or friend. Conditional on attending at least one drama, the average number attended was 1.6, and the effect is statistically significant at the 1 per cent level. We also observe some spillovers of the drama intervention in the other arms, as 8-9 per cent of the respondents in the control arm and 19 per cent of women in the birth kits arm report exposure to a drama in their community.
Summing up, it is clear that the CORPs program did provide services to beneficiaries, but did not reach the majority of pregnant women. However, the intervention was more effective when it was accompanied by additional interventions (birth kits or dramas). Supplementary data from a qualitative analysis suggest that CORPs encountered hostility from women and particularly their husbands if they made household visits without the provision of birth kits or other hoped-for incentives or material benefits (Sharma, Giroux, Leight, Abdulaziz, & Nyqvist, 2020). Accordingly, one hypothesis that would be consistent with the observed pattern is that CORPs are more active in the birth kits and drama arms because the additional services provided rendered beneficiaries more receptive to engagement. This is also consistent with existing literature suggesting non-financial incentives are highly effective in motivating health workers (Ashraf et al., 2014;Deserranno, 2019). Table 2 reports results analyzing women's utilization of maternal health care services during pregnancy. For antenatal care, we observe in Columns (1) and (2) an increase in the probability of using antenatal care in the birth kit and drama treatment arms of 6.3 to 8.6 percentage points that is statistically significant. Relative to the control arm, these effects correspond to a 11 per cent increase. In the same treatment groups, we also observe an increase in the number of antenatal care visits of around 10 per cent that is also statistically significant. In order to capture the quality of antenatal care, we construct an index, and the estimated coefficients reported in Column (3) suggest there was also an increase in the quality of antenatal care in the birth kits and drama arms. 14 In addition, Column (4) reports an increase in the probability of utilizing postnatal care of around two to three percentage points in all treatment arms, though this coefficient is significant only for the basic CORPs and drama arms. Given the control mean of 7 per cent, this is a proportionally large effect of 45 per cent. Column (5) reports the average standardized treatment effects for these four outcome variables. Here, we observe a 0.12 standard deviation increase in utilization of antenatal and postnatal care, statistically significant for the birth kits and drama treatment arms. The enhanced interventions (where the CORPs program is implemented in conjunction with another health program) had larger impacts on antenatal and postnatal care compared to the treatment arm with the basic CORPs program, and the difference is statistically significant at the 1 per cent level.

Utilization of maternal health care.
In Columns (6) through (8), we report treatment effects for the probability of a facility delivery, skilled attendance at birth, or the probability of delivering with another person present. Here, we find a uniformly null effect. 15 In Column (9), we report the average standardized effect for the outcomes capturing utilization of maternal health services (that is, Columns 1-4 and Community health educators and maternal health 83 Notes: All regressions include LGA fixed effects and standard errors clustered at the cluster level. The sample includes all respondents reporting a pregnancy during the intervention period who were interviewed in at least one follow-up survey and who report the indicator of interest. We report tests of equality across the estimated coefficients; a test of the hypothesis that the joint effect of treatment is zero (b and a test of equality of impact across the basic and enhanced arms b 1 ¼ b enhanced : Asterisks denote significance at the 10, 5, and 1 per cent level. 6-8); the estimated coefficients suggest there was an increase in care utilization of around 0.07 standard deviations that is statistically significant in the birth kits arm and drama treatment arm. Again, the health educator program was more effective when implemented in tandem with other health interventions. The joint tests of the effects of any intervention reported at the base of the table reinforce this evidence. The effect of any CORPs intervention is significant and positive for the three antenatal care variables and narrowly insignificant (p ¼.110) for postnatal care. The hypothesis that the average standardized effect across any intervention is equal to zero can be rejected at the 1 per cent level for non-delivery care utilization measures, and at the 5 per cent level for the full set of utilization measures including facility births.

Maternal and newborn practices.
A second objective of the CORPs program was to enhance health practices, and Table 3 reports the estimated effects of the intervention on these outcomes. Two of the CORPs' specific learning objectives included encouraging pregnant women to develop a birth plan and providing information about the importance of breastfeeding. In Column (1), we find evidence of a significant and positive effect on the probability the respondent reported a birth plan in all treatment arms (an increase of between 21 and 40% relative to the control mean of 11%). We do not find that the intervention had any impact on breastfeeding behavior.
We also find a significant increase in the number of immunizations administered to newborns in the first month of life in the birth kit and drama arms (an additional 0.1 immunization administered on average, relative to a control mean of one), and a significant increase in the probability of an newborn check-up in the first month (an increase of around 12 percentage points, relative to the control mean of 29%). These are proportionally large effects, suggesting an increase of roughly 40 per cent in infant check-ups. Again, we report the average standardized treatment effect for these variables in Column (7), and observe that there is an enhancement in maternal and newborn health practices of roughly 0.06 standard deviations compared to the control group, and this impact is significant in all three treatment arms. 16 The joint tests of the effects of any intervention reported at the bottom of the table suggest that any CORPs intervention had a significant and positive effect on the utilization of birth kits and postnatal infant check-ups, and the average standard treatment effect for any intervention is similarly positive and significant at the 5 per cent level.

Maternal and child morbidity, mortality, and anthropometrics
The interventions did not seem to have any effect on self-reported measures of maternal morbidity during pregnancy, delivery, and the postpartum period or neonatal morbidity, as reported in Table S5 in the Supplementary Material. We do observe a decline in maternal morbidity in the birth kits arm, and this pattern would be consistent with some limited use of the birth kit yielding benefits in terms of maternal health. 17 However, the average standardized effect for the morbidity indices reported in Column (3) shows no significant effects. We also fail to identify any significant impact on the probability of stillbirths and infant and neonatal mortality; the mortality rates are calculated at the cluster level, yielding a sample of 96 observations.
We also evaluate the effects of the interventions on the anthropometric status of all children under five for respondents reporting a birth in the intervention period. 18 The final sample thus includes 5332 children with data on anthropometric measures reported at endline. In Table 4, we report the results of estimating Equation (1) for height-for-age, weight-for-age, and midupper-arm-circumference-for-age. Here, we find evidence of a somewhat heterogenous pattern across indicators, but the standardized effects reported in Column (5) are generally Notes: All regressions include LGA fixed effects and standard errors clustered at the cluster level. The sample includes all respondents reporting a pregnancy during the intervention period who were interviewed in at least one follow-up survey and who report the indicator of interest. We report tests of equality across the estimated coefficients; a test of the hypothesis that the joint effect of treatment is zero (b  Table S6, and show a similar pattern. 19

Knowledge and attitudes
In addition to the primary outcomes enumerated above, we analyze a set of additional secondary outcomes: knowledge and attitudes around facility delivery care, and knowledge about infant care and fertility. The results of estimating Equation (1) for these variables are reported in Table 5. 20 We find that the interventions increased knowledge of relative risk and pregnancy complications and enhanced attitudes toward health facility use. The average standardized effect for the attitudinal variables, reported in Column (6), shows increases of around 0.05 standard deviations that are statistically significant in the birth kit and drama treatment arms. Results for the subsample of respondents reporting a birth during the intervention period are reported in the Supplementary Material, Table S7, and are broadly similar. We also analyze whether the intervention generated any effects on fertility. The results are reported in Table S8 in the Supplementary Material and the coefficients of interest are small in magnitude and insignificant, consistent with the hypothesis that the intervention did not generate any differential patterns of selection into the subsample of women reporting a birth.

Treatment on the treated
To evaluate the effects of the interventions on women who did in fact receive the interventions, we estimate treatment on the treated specifications. We focus on the CORPs intervention given that this intervention was implemented across all three treatment arms. The specification of interest is as follows: Notes: All regressions include LGA fixed effects and standard errors clustered at the cluster level. The sample includes children of respondents born in the intervention period who have anthropometric measurements collected at endline. We report tests of equality across the estimated coefficients; a test of the hypothesis that the joint effect of treatment is zero (b 1 ¼ b 2 ¼ b 3 ¼ 0; and a test of equality of impact across the basic and enhanced arms b 1 ¼ b enhanced : Asterisks denote significance at the 10, 5, and 1 per cent level. Notes: All regressions include LGA fixed effects and standard errors clustered at the cluster level. We report tests of equality across the estimated coefficients; a test of the hypothesis that the joint effect of treatment is zero (b where v icg denotes the same vector of control variables utilized in the intent-to-treat estimates, and the dummy variable for exposure to the CORPS intervention is instrumented by a dummy variable for assignment to any treatment arm. 21 The results of two-stage least squares estimation for care utilization, health practices, and attitudes and knowledgethe primary outcomes for which significant ITT estimates are observedare reported in Table S9 in the Supplementary Material. 22 The estimated treatment on the treated coefficients are large in magnitude; the probability of accessing antenatal care services increases by more than 50 percentage points, and the probability of a postnatal care visit increases by 32 percentage points. These results provides suggestive evidence that a community health worker program that was more aggressively implemented could have very large effects on care utilization even in an extremely low-resource setting.

Conclusion
Over the last 20 years, progress in reducing maternal and neonatal deaths in developing countries has largely stagnated, particularly in sub-Saharan Africa. Accordingly, identifying interventions that can increase utilization of maternal health services and enhance health outcomes, particularly in challenging regions characterized by extremely low baseline human capital outcomes and persistent violence, is a key priority for policymakers and researchers.
This evaluation adds to the body of evidence around community-level interventions designed to improve maternal and child health in low-resource settings. The results suggest that the interventions reached approximately 22 per cent of women reporting a birth during this period and generated increased utilization of antenatal care, postnatal care, immunizations, and newborn check-ups in communities exposed to the intervention. We also find positive shifts in attitudes around facility-based delivery care. The community health program was most effective when implemented in conjunction with additional health programsthe birth kits and the drama interventions.
However, the increases in service utilization and health practices did not yield any detectable improvements in anthropometric outcomes for infants or young children or any decreases in maternal or neonatal morbidity or mortality. The persistently low quality of health services available through the MSS may have limited both the intervention's effectiveness in increasing utilization of services and its effect on health outcomes. Previous literature about the MSS did not identify any positive effect of the deployment of these services on health outcomes, and found significant quality challenges . Additional qualitative research conducted as part of this project similarly identified the absence of health providers (particularly female health providers), the absence of drugs and equipment, and unexpected facility closures as significant barriers to the use of facilities (Sharma, Leight, Abdulaziz, Giroux, & Nyqvist, 2017;Sharma, Leight, Giroux, Abdulaziz, & Nyqvist, 2019).
Our results add to an existing evidence base suggesting that while community health worker programs can be effective in enhancing health outcomes in developing countries, there is huge heterogeneity in their effects (Lewin et al., 2010;Okwundu et al., 2013;Scott et al., 2018). It is important to note that the evaluation has limitations, and questions linked to external validity of the intervention for other contexts must be carefully considered. This trial was conducted in one state in Northern Nigeria characterized by very low utilization of care and poor human capital outcomes ex ante; in addition, ongoing patterns of conflict, described in more detail above, contributed to very limited penetration of the CORPs program and associated interventions in the study communities. In other, more stable contexts, interventions such as the CORPs program may be able to achieve higher penetration rates in communities and more effectively reach target women. On the other hand, receptiveness to information provided by actors such as the CORPs may vary unpredictably in different communities, and if supply-side Community health educators and maternal health 89 barriers are salient, demand-side interventions may not shift utilization of maternal health care significantly.
Further research, both qualitative and quantitative, may productively explore the conditions under which community health worker programs can effectively shift maternal and neonatal health outcomes. Ultimately, the observed pattern here is consistent with the hypothesis that demand-side interventions such as community health educator programs can effectively stimulate increased utilization and change health behaviors in a challenging context such as Jigawa state. Notes 1. In practice, MSS-served facilities were in fact characterized by frequent stockouts . 2. Founded more than 25 years ago, PPFN is now one of the oldest indigenous organizations in Nigeria offering sexual and reproductive health services; however, the interventions proposed as part of this evaluation were new to the organization. 3. The kits were identical to those available to midwives working in the MSS primary health care centers. 4. Chari and Okeke (2014) conclude that a policy-induced shock to the supply of institutional deliveries did not have a significant effect in reducing newborn mortality in Rwanda. Evidence from Malawi suggests a government ban on the use of traditional birth attendants did result in a significant increase in utilization of formal sector care, but no overall decline in newborn deaths (Godlonton & Okeke, 2016). 5. Power calculations are conducted following Hayes and Bennett (1999). This is assuming a baseline maternal morbidity rate of 35 per cent, and a baseline infant mortality rate of 47 deaths per 1000 births. In addition, the design assumed a birth rate of 46 per 1000 population and accordingly 21 sampled births observed per year per community of 3000 individuals given that 15 per cent of households are sampled. The coefficient of variation between clusters k was assumed to be.2. 6. The trial was registered with the American Economic Association registry (Leight, Nyqvist, & Sharma, 2016), and registered at clinicaltrials.gov. The protocol number is NCT01487707. 7. All respondents provided informed written consent, and all data were collected by electronically by trained same-sex enumerators using ODK software. 8. In the event an enumerator became aware of a birth more than three days after the birth, she was still instructed to conduct the three-day survey as soon as possible, and then return for the 28-day survey. In the event she became aware of a birth more than 28 days after birth, she was instructed to conduct the survey up to three months after birth. 9. Some additional intensive data collection targeted to minimize attrition continued until October 2016. 10. No systematic data was collected on miscarriages, other than one question posed at the endline as to whether the respondent experienced at least one miscarriage in the preceding two years. 11. More evidence around the hypothesis that there is no selection into the subsample of women reporting births can be found in Section S6 in the Supplementary Material. 12. The control variables employed include all those reported in the balance tests in Panel A of Table S3: a dummy variable for whether the respondent is married, the number of co-wives, age at marriage, the number of marriages reported, age, a dummy for whether the respondent has ever attended school, a dummy variable for whether the respondent reads Hausa, a dummy variable for Muslim, current birth parity, and a wealth index. We also include dummy variables equal to one if the respondent is observed in the 3-day and 28-day surveys, and in the audit survey. 13. There is no evidence that spillovers are higher in control communities that are geographically more proximate to treatment communities. 14. The antenatal care index is equal to the mean of indicator variables for receiving important components of antenatal care: utilizing care in the first trimester, receiving more than half of available ANC services, receiving iron folic pills and the tetanus vaccine, and receiving advice on danger signs during pregnancy. The observed intervention effect is largely driven by increases in the receipt of iron pills and tetanus vaccines, as well as reported counseling about pregnancy danger signs. 15. In practice, facility delivery and skilled attendance at birth are almost equivalent, given that health personnel generally do not attend home births in this region. 16. Given that the effect is observed in all three treatment arms, it is plausible to hypothesize that it primarily reflects the effects of the CORPs, rather than the ancillary interventions. 17. It should be noted that the sample for neonatal morbidity is restricted, given that this information was reported only in the 28-day survey. We also collected limited data on miscarriages at endline, and find no significant effect of the interventions on the rate of miscarriage, defined as the loss of a pregnancy in the first two trimesters. Data on induced abortions were not collected, but this could be another potential channel for selection in the sample of households reporting a birth. 18. Endline anthropometric data is missing for children corresponding to births observed only in ongoing surveys (133 observations), as well as for an additional 650 observations; for the latter subsample, the adult respondent was surveyed in the endline, but the enumerator assigned to follow-up with a separate anthropometric survey did not locate the household, the respondent declined to provide consent for measurement, or the child was not available. 19. It is important to emphasize that the results should be interpreted cautiously given that the children observed in the anthropometric data are drawn from a subsample of respondents. However, we present evidence in Section S6 that respondents observed only in the endline and those observed in ongoing surveys are not characterized by significant differences in observable characteristics, and thus we cannot reject the hypothesis that the respondents observed in the anthropometric data constitute a random subsample. 20. All the indices are coded such that a higher value indicates more knowledge or more positive attitudes. 21. In the first stage, the coefficient on the treated dummy is.101, significant at the 1 per cent level. 22. Estimating the treatment on the treated specifications for the outcomes reported in Table S5 shows null effects.
Limited power is available to estimate treatment on the treated for the anthropometric variables given the reduced sample.