Baray estriktirèl ak konpòtman yo ki anpeche amelyore rezilta devlopman yo: Ka swen matènèl yo nan Ayiti

Public Disclosure Authorized Policy Research Working Paper 10421 Public Disclosure Authorized Structural and Behavioral Barriers to Improving Development Outcomes The Case of Maternal Care in Haiti Public Disclosure Authorized Emilie Perge Jimena Llopis Abella Anna Fruttero Public Disclosure Authorized Poverty and Equity Global Practice April 2023 Policy Research Working Paper 10421 Abstract This paper investigates the interplay between structural and behavioral barriers that discourage pregnant women from accessing institutional care in Haiti, where despite some improvements in the past decades, maternal mortality remains a significant challenge. The analysis complements household survey data with data on service provision and qualitative data on beliefs, perceptions, and attitudes toward maternal health care. Using a mixed-methods approach, the paper confirms that transportation and pov erty are important barriers that decrease the likelihood of attending maternal health care services. At the same time, the findings show that women suffer from optimism bias and are uncomfortable with the current model of received care. These barriers discourage women from seeking, reach ing, and receiving maternal health care services at health institutions. Tackling structural barriers while finding ways to encourage women to shift their beliefs, perceptions, and attitudes are key recommendations to improve maternal health in Haiti. This paper is a product of the Poverty and Equity Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at afruttero@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Structural and Behavioral Barriers to Improving Development Outcomes: The Case of Maternal CVre in Haiti∗ Emilie Perge, Jimena Llopis Abella, and Anna Fruttero† JEL: D91, I12, I15, I18. Keywords: Maternal health, behavioral biases, multilevel model, mixed methods, Haiti. 1 Introduction With 480 deaths per 100,000 live births in 2017, Haiti, was far from achieving the Sustainable Development Goal of having fewer than 70 maternal deaths per 100,000 live births by 2030 (UNICEF, 2022). In the Latin America and Caribbean region the average maternal mortality rate was 74 deaths per 100,000 live births. Pregnant women die as a result of complications such as preeclampsia, eclampsia, severe bleeding, and infections. While some intrapartum complications cannot be reliably predicted or prevented, most of them can be successfully detected and treated with prompt diagnosis and care (Say et al., 2014). Adequate professional care before, during, and after childbirth has been proven to reduce death rates (UNICEF, 2014). Thus, all women should access antenatal care (ANC) ∗Acknowledgments: The authors thank the two peer reviewers, Jorge Luis Casta˜neda and Nicolas Collin, for their detailed comments; Ondine Berland for excellent research assistance; Donald Antoine, Jamesson Vamblain, Mayerline Antoine, Manouchka Justin, Louise Estavien, Tania Mathurin and Fleurimonde Charles Joseph for their help collecting qualitative data in Haiti; Ingrid Dallmann for her help with the SPA data; and Lauren Manning and Chiara Broccolini for support with editing. All findings, interpretations, and errors belong to the authors. †Perge is affiliated with UN Sustainable Development Solutions Network, Llopis Abella is affiliated with Save the Children, and Fruttero is affiliated with the World Bank Group. Correspondence: Emilie Perge (emilie.perge@unsdsn.org), Jimena Llopis Abella (jimena.llopis@savethechildren.org and Anna Fruttero (afrut tero@worldbank.org)). during their pregnancy, skilled care during childbirth, and postnatal care (PNC) and support in the weeks after childbirth.1 Haiti has the lowest rates of ANC and PNC in the region, with 67 and 31 percent of Haitian women receiving these services in 2017, respectively, compared to 91 and 88 percent for the whole region (UNICEF, 2022). Only 42.1 percent of women deliver with a skilled health professional, while 48 percent of women deliver with a matron or traditional birth attendant (IHE and ICF, 2018), who have little formal training, and often receive knowledge only from their elders. The remaining 10 percent delivers with family members, community health workers or on their own. Low utilization rates of health care services can be explained by structural barriers, which do not depend on the individual such as costs, distance to facilities (either because of physical distance or poor state of roads), and poor quality of health centers infrastructure, or by individual behaviors and beliefs (Datta and Mullainathan, 2014), such as optimism bias 2, uncertainty aversion 3, status quo bias 4, or discomfort with the quality of care. Pregnant women may underestimate the likelihood of pregnancy complications, or of needing complex care beyond the capabilities of the matrons. Likewise, if they cannot immediately recall a family member or friend who might have required more care, they are less likely to pursue care themselves. Matrons may also fall victim to these heuristics and underestimate the need for care or probability of pregnancy complications, referring women to hospital care too late in a delivery scenario to save lives. Therefore, in order to improve outcomes it is essential to address both structural issues and individual behaviors. As Charter and Loewenstein (2022) highlight, policies focusing on the individual (i-frame) should be seen as complementing policies addressing the system in which individuals operate (s-frame). This paper documents the structural and behavioral barriers that discourage pregnant women from attending institutional care during their pregnancy and delivery in Haiti. It builds on earlier research by Gage and Calixte (2006) and Wang et al. (2017), which examined the impact of physical access to health services on the use of ANC and delivery care services. Using data from Haiti, Gage and Calixte (2006) found that limited access to obstetric services and limited use of existing facilities discourage delivery at a hospital. This paper uses data from the 2017 Haiti Demographic and Health Survey (DHS), the 2017 service provider assessment (SPA), and qualitative data collected during fieldwork in May 2018 to shed light on other factors that influence women’s decisions and emphasizes the importance of the quality of health services in shaping these decisions. The quantitative analysis uses a multilevel model, which accounts for the fact that women are nested within geographic clusters with similarly available health services, to identify determinants of women’s decision to seek and reach care and to receive adequate care. This is then complemented by the analysis of perceptions and attitudes through the qualitative data. We find that structural factors, including difficult access to healthcare centers, can be significant for women seeking care. Many women have rational concerns about the impact of these barriers on their health. For example, traveling on rough roads by motorcycle during pregnancy and labor can be frightening and dangerous. Additionally, uncertainty about the 1During the pregnancy, the World Health Organization (WHO) recommends four visits providing essential evidence-based interventions such as identification and management of obstetric complications (preeclampsia), and of infections (HIV, syphilis...) as well as promoting the use of skilled attendance at birth and healthy behaviours. After delivery, the WHO recommends that all mothers and babies have at least four postnatal checkups in the first 6 weeks. 2A cognitive bias that causes someone to believe that they themselves are less likely to experience a negative event. 3Preference for known risks over unknown risks 4Preference for the maintenance of one’s current state of affairs, or a preference to not undertake any action to change this current state. 2 cost of hospital stays, medication, and other expenses may prevent women in poverty from seeking clinical care. It is also common for women to be unaware of when they should seek additional medical attention. Lack of transportation is also a significant barrier to accessing maternal healthcare services. Women without transportation often have to walk long distances or rely on scarce public transport options like motorbikes or pick-up trucks. This is especially difficult in rural areas where such transportation is scarce and the roads are in poor condition. These challenges can lead to fears about the safety of both the woman and her baby during travel. Additionally, the high cost of transportation may prevent women from seeking care, even if they intend to. Structural barriers and concerns about their impact are valid and warranted, and women may be unaware of when they should seek additional care. The poor condition of roads and the risk of injury or delivery during travel can also be frightening and dangerous for pregnant women and those in labor. Behavioral factors also play a role. Biases, such as availability and optimism bias, can prevent women from seeking necessary healthcare. Perceptions of the quality of care and the way hospitals and medical staff treat women are important factors that can affect women’s decisions to seek care. Even if women are able to access hospitals, they may be deterred by negative experiences or expectations of poor treatment. Some women report feeling inferior, receiving condescending or rough treatment, or being made to deliver in uncomfortable sit uations. In interviews, women have expressed fears about hospital settings, including being left alone after operations or seeing infants receiving negligent care. The burden of needing family members to bring food to the hospital can also discourage women from seeking care. Lastly, husbands may play a key role in encouraging their wives to attend institutional care. Not only are they often convinced of the benefits, but husbands also seem to care for the social status attached to being able to afford institutional care. The paper is organized as follows. Section 2 describes the background. Section 3 focuses on the methodological framework. Section 4 presents some descriptive statistics of pregnant women in Haiti and section 5 presents the findings. Section 6 discusses the findings and concludes. 2 Background Haiti, which spends less than 5 percent of its national budget on health (World Bank, 2017), has insufficient health infrastructure and limited healthcare workforce and medical resources.5 A substantial part of the population has difficulty accessing health centers due to distance, poor road conditions and/or limited access to transportation. There are 1,048 health institu tions in Haiti that serve 11 million people. These health institutions are organized into three levels: the primary level, which includes community health centers or dispensaries, health centers with or without beds, and community reference hospitals; the secondary level, which includes departmental hospitals; and the tertiary level, which includes university hospitals and specialized hospitals (MSPP, 2014, 2015). Women can receive basic obstetric services at all three levels, but they are encouraged to seek antenatal care at the primary level. Women with complications are referred to higher levels. In 2017, there were a total of 59 health facilities for moderate risk pregnancies and 41 for high-risk births (SONU-B and SONU-C,6respectively). More than half of these institutions lack the trained staff to provide antenatal, postnatal, and childbirth delivery services. Haiti suffers from a severe shortage of skilled healthcare providers and has only one midwifery 5Haiti had less than 7 hospital beds and 2.34 medical doctors per 10,000 people compared to 20 hospital beds and 30 for 10,000 people in the LAC region. 6Soins Obst´etricaux et N´eonataux d’Urgence de Base or Complet 3 education program. According to the WHO, there were a total of 2,606 physicians in Haiti in 2018 while the combined number of nurses and midwives was 4,424 for a population of over 11 million 7. Health centers, on average, do not have two out of the six recommended medications needed for childbirth delivery (IHE and ICF, 2018). The Government of Haiti and its partners have recently focused on improving physical access to healthcare. For example, the 2008-2013 maternal health services program provided free services to low-income women in selected health centers, with funding from the United Nations Population Fund (UNFPA) and the Canadian International Development Agency (CIDA). UNFPA and non-governmental organizations (such as Midwives for Haiti) also at tempted to bring mobile prenatal clinics to populations in remote areas with limited access. Additionally, Midwives for Haiti has provided training to nurses since 2006 to increase the number of skilled midwives, while other NGOs focus on promoting continuing education for official midwives after they have received their diploma. However, to the extent of our knowl edge, rigorous evaluations of these approaches are missing. 2.1 Literature on barriers to institutional care Timely access to institutional care is a key factor to reducing maternal mortality rates, as most of the pregnancy and labor complications that can lead to death can be detected and treated by timely medical attention (Pfeiffer and Mwaipopo, 2013). Thaddeus and Maine (1994) organized barriers to accessing healthcare in a conceptual framework known as the three-delay model, which groups barriers around three different moments: (i) decision to seek care, (ii) identifying and reaching care facility, and (iii) receiving adequate and appropriate care. For each of these moments, several types of barriers have been documented. Structural barriers - Financial and opportunity costs. Through interviews in Kenyan slums, Essendi et al. (2010) observe that most residents want their children to be born at the hospital. How ever, the high cost of a hospital-based delivery and additional indirect costs (e.g for transportation) prevent them from doing so. Kyei-Nimakoh et al. (2017) conclude that the uncertainty of the delivery cost at a health institution due to the complexity of the billing system also discourages women from seeking care. Whether or not a woman has a social support network can play an important role in the decision to seek care (Thaddeus and Maine, 1994), as this network can help reduce women’s opportunity cost to seek care by, for example, taking care of other children or helping with household chores. - Distance to the health facility. Besides the physical distance other factors come into play. Thaddeus and Maine (1994) posit that the effect is even stronger when combined with lack of transportation and poor roads. For example, pregnant women that would have to walk for hours over rugged terrain will be disincentivized from even seeking care. - Physical accessibility to health facilities. Shortage of health facilities is so substantial that even if women intend to deliver in a health facility, they may be simply unable to do so (Gage and Calixte, 2006), especially at night when transportation options are scarcer. - Lack and/or high cost of available means of transportation results in women having to rely on walking or motorcycles to reach the health facility (Llopis Abella et al., 2020; Kyei-Nimakoh et al., 2017). 7Global Health Observatory data repository, WHO 4 - Poor road conditions and safety risks discourage women’s willingness to reach obstetric care. The risk of miscarriage increases if the roads are in poor condition (Gage and Calixte, 2006). Moreover, women cannot always travel alone safely, discouraging them further. This is especially the case in very poor urban areas such as slums (Essendi et al., 2010). - Shortages of supplies or equipment. These cause delays in receiving adequate and ap propriate care at the facility. Numerous health facilities are poorly equipped, delaying care and, in some instances, forcing patients to buy the supplies, including essential drugs, which when available, many cannot afford (Essendi et al., 2010). - Proficiency of health staff. The scarce, unqualified, and unprepared health staff cause long delays in receiving care (Thaddeus and Maine, 1994). In a qualitative study in Haiti, Mirkovic et al. (2017) observe that 26 percent of the women in their sample waited over four hours to be seen for antenatal care, while 28 percent spent less than five minutes with their provider. Behavioral barriers - Previous experience with the healthcare system, or perceived quality of care (Bohren et al., 2015) influence women’s perceptions and thus their future decision to seek care (Kyei-Nimakoh et al., 2017). - Mental models: the recognition of the need to seek care can be considered unnecessary, even more so if pain does not materialize. Pregnancy and delivery are considered nat ural events in almost all societies, where even death during labor can be considered as something inevitable. - Agency: women’s agency plays an important role in some areas of the world where the decision to seek care depends on the women seeking permission from their spouse or other family members to leave the house (Kyei-Nimakoh et al., 2017). - Fear of mistreatment: In a systematic review, Bohren et al. (2015) find that women are sometimes mistreated during childbirth in health facilities through physical, verbal, or sexual abuse perpetrated by health providers. - Discomfort with the model of care: It can discourage women from receiving institutional care as preferences around birth practices might not match practices at modern medical facilities. For instance, not being given the freedom to choose birthing positions, retain the placenta for burial, or have relatives nearby were reported in several studies as bar riers to seeking institutional care (Bohren et al., 2015; Kyei-Nimakoh et al., 2017). Fear of surgery, episiotomy, and blood transfusions also discourage women. The discomfort influences the time women want to spend at health institutions and their willingness to come back. Because these barriers are usually interconnected, efforts to increase the use of institutional care by building more health facilities or reducing the costs of care are often not enough. While earlier studies in Haiti have shed light on structural barriers such as distance to health facility and lack of transportation (Wang et al., 2017; Gage and Calixte, 2006), more needs to be uncovered with respect to the beliefs, perceptions, and attitudes of pregnant women towards institutional care and the role that quality of care and women’s perception of this quality have on encouraging the use of institutional care. 5 3 Methodological framework Extending earlier studies (Wang et al., 2017; Gage and Calixte, 2006), we use a mixed method design that combines findings from both quantitative and qualitative methods. The analy sis of the quantitative data provides a general overview of factors that influence pregnant women’s decisions regarding institutional care while the qualitative data analysis explores in depth beliefs, perceptions, and attitudes. During the interpretation stage, findings from the econometric model and the behavioral sciences literature are brought together, with the same weight, for a deeper understanding of the interplay of structural and behavioral barriers that prevent women from accessing institutional care (Greene et al., 2001; Cullen et al., 2011). In addition, this design provides an opportunity for triangulation of the findings across methods. 3.1 Quantitative research data and analysis The quantitative data come from two datasets: the 2017 Demographic and Health Survey (DHS), called EMMUS (Enquete Mortalite, Morbidite et Utilisation des Services - Survey of Mortality, Morbidity and Service Utilization) in Haiti, and the 2017 Service Provision Assess ment (SPA). The DHS dataset consists of a nationally representative sample of households interviewed between November 2016 and April 2017 by the Institut Ha¨ıtien de l’Enfance with support from Institut Ha¨ıtien de Statistiques et d’Informatique (IHSI). Besides information on household characteristics and an asset measure of wealth, typically collected in DHS-type surveys, it contains data from an in-depth survey conducted on a sample of 15- to 49-year-old women who answered questions about access to ANC, source, place, and type of assistance received during delivery for each child born within the five years prior to the survey. Surveyed women also answered questions related to decision-making in the household and agency. In 2017, 14,371 15-49 year old women from 13,405 households were surveyed.8In ad dition to household- and individual-level characteristics, community characteristics can be retrieved using GPS coordinates of the cluster centroid. The DHS website provides access to topographical characteristics for all clusters.9 The SPA data were collected between December 2017 and May 2018 on all 1,033 health centers across the country (IHE and ICF, 2018). This assessment made an inventory of all the facilities and equipment, surveyed the staff, observed visits for ANC and family planning services, as well as services for sick under-5-year-old children. Women who were observed attending ANC or family planning consultations and the families of sick children were also surveyed. All these survey tools allow the SPA to assess how much the services work and how satisfied service users are. In this paper we use data from facility inventory provided by the facility manager or most knowledgeable person on the infrastructure, supplies, staffing, and routine practices.10 The quantitative data are analyzed using a multilevel, hierarchical model to investigate the neighborhood and individual effects, while controlling for the clustered nature of the data (Bafumi and Gelman, 2007; Gage and Calixte, 2006). Multilevel models allow one to look at 8The DHS uses a two-stage cluster sample design, where clusters are the enumerating areas provided by the IHSI based on a 2011 update of the 2003 population census. The sampling methods are described in IHE and ICF (2018) but in this study, weights are computed to ensure the representativeness of the evidence at the national, urban-rural, and department-levels. 9The cluster data are anonymized by displacing the cluster centroid by, on average, 0.8 km for urban clusters (maximum displacement of 2 km), and 2.1 km for rural clusters (maximum displacement of 5 km and of 10 km for 1 percent of rural clusters) from its original location (Wang et al., 2017). This issue was not considered when linking the cluster to all available health centers within a 10 km radius from the cluster centroid from the SPA data (Burgert and Prosnitz, 2014). 10Because the SPA data are collected as a census and the level of non-response is small, differences between weighted and unweighted results are minor; the results presented are unweighted. 6 within-cluster and between-cluster determinants that affect the outcomes of interest, allowing for the relationships between the outcomes and the factors to vary depending on the context (Jones, 1993). Within clusters, individual characteristics are highly correlated as communities are quite homogeneous (especially in rural areas). Women are nested within clusters (level 1), and clusters are the higher level of analysis (level 2).11 Multilevel modeling treats the outcome of interest as a function of individual-level charac teristics while controlling for interactions between demographics and cluster characteristics. A cluster random intercept term represents the extent of the differences in the outcome between clusters. Let Yis be an indicator variable for woman i in cluster s defined as follows: Yis ={1 if woman i attends institutional care, 0 if she does not} where i = 1, . . . , n, and s = 1, . . . , S. Defining institutional care as a binary variable, we model πi = P(Yis = 1) the probability that woman i in cluster s attends institutional care when pregnant using a multilevel logistic regression model with varying intercepts and varying slopes to control for the selection bias since the outcome is only observed for pregnant women. The model is as follows: πi = logit−1(αs + βsxis | {z } space variant + θZis |{z} space invariant + i) for i = 1, . . . , n (1) where s represents the cluster s where the woman i resides. logit−1is the inverse logistic function. x are individual-level and time-variant predictors while z are individual-level and space-invariant predictors. αs and βs are space-varying intercepts and slopes, respectively taking the following form: αs ∼ N(αs + αUs, σ2αs) for s = 1, . . . , S βs ∼ N(βs + βUs, σ2βs) for s = 1, . . . , S Us are contextual predictors at the cluster level; αs and βs can be further modelled as a function of cluster (Gs) predictor αs ∼ N(γ0 + γ1Gs, σ2αs) for s = 1, . . . , S The following analysis models two outcomes: • attending at least the recommended four ANC visits, and • delivering in a health institution. Using the three-step delivery framework described above, the predictors are grouped into the three steps: 1. Women’s socioeconomic factors (having a job, secondary education, access to informa tion, permission to attend health centers, and belonging to the bottom 40 percent of asset index distribution) which may explain whether she seeks care; 11Since most households only have one pregnant woman, household level characteristics are treated as individual-level characteristics. Thus, there are only two levels in the present analysis: women level (level 1) and cluster level (level 2). 7 2. Women’s place of residence (rural or urban, mountainous cluster, cluster with access to ANC available at least 18 days a month, cluster with access to normal delivery services) and access to transportation means, which account for women’s ability to reach care; and 3. Women’s access to ANC with medications, to ANC with equipment, or access to delivery with equipment, which account for access to adequate care. The cluster-level variables are built once a service area of 10 km around the cluster centroid has been identified using Euclidean distance12 from within the cluster centroid to all health institutions. Health institutions can serve multiple clusters. Using the characteristics of all these health institutions, dummy variables are defined to characterize whether a certain share of health centers offer the services or have the equipment or medication for ANC or delivery. 3.2 Qualitative research design Qualitative data come from fieldwork undertaken by two of the authors with support from Haitian researchers, in May 2018. The fieldwork explored: pre- and post-natal care behav iors, attitudes and opinions around institutional delivery, perceptions, social structures, and relationships, among other contributing factors. The instruments chosen for this study were focus group discussions (FGDs), semi-structured interviews (SSIs), and field observations of local health facilities. Site selection was done through a two-stage process, as is common with qualitative research methods (Tracy, 2010). The first stage consisted of selecting the d´epartement, subnational administrative level, with the highest presence of hospitals with obstetrician care per women, and a high rate of institutional births and of births attended by a skilled provider (IHE and ICF, 2018) to ensure access. The second stage controls for availability of a SONU-B or SONU C and identifies communal sections with a low and high percentage of births at an institution. Following these selection criteria, two communal sections, 2`eme Fonds-des-N`egres and 1`ere Chalon in the Nippes d´epartment (southwest of Port-au-Prince) were selected, as they have the lowest (25.3) and highest (67.9) percentage of institutional births respectively. In addition to pregnant women, respondents were recruited based on their role in the decision-making process of pregnant women: traditional birth attendants, health workers, family members, community health workers, and community leaders. In total, 20 SSIs and 9 FGDs were conducted with a pre-mobilized sample of respondents in a public space in each communal section.13 Field observations were conducted in two health facilities: Ste-Th´er`ese Hospital, a public community reference hospital in 1ere ` Chalon, and B´ethel de L’Arm´ee du Salut, a private health center in 2eme Fonds-des-N`egres. After data collection, interviewers transcribed and translated recordings from Haitian Creole to French. Once the transcripts were ready, an examination of the raw data was conducted to identify key categories and patterns supported by the data. From these key categories, a code scheme was developed and discussed before being analyzed in NVivo software. The first-level codes were “Facts about maternal health system”, “Beliefs and opinions about ANC and PNC”, “Beliefs and opinions about delivery at health institutions”, “Beliefs and opinions about home delivery”, and “Profiles of actors”. Sub-codes for “Facts about maternal health” explored themes such as types, costs, equipment, staffing, and procedures for all types of health facilities and more 12Euclidean distance is defined as the length between two points drawn with a straight line. 13SSIs were conducted with community health workers (5), community leaders (5), traditional birth atten dants (4), health workers (4), a pregnant woman (1), and a family member (1).FGDs were conducted with pregnant women (3), health workers (2), family members (2), and traditional birth attendants (2). A total of 64 people participated in the FGDs. 8 specifically for ANC, PNC, and childbirth deliveries. When coding about “Beliefs and opin ions”, we explored women’s experiences of going to the health facilities for maternal healthcare and their reasons for going (safety, incentives) or not going (transportation, costs, experience with healthcare workers). The code “Profiles of actors” helped describe who are the main decision-making actors. 4 Descriptive statistics In Haiti, in 2017, 15-49 year old women who had given birth in the previous 5 years had an average of 1.3 number of births and were predominantly from rural areas (62 percent), married (89 percent), had a job outside the home (55.8 percent), access to information (64.3 percent), did not have a secondary education (65.9 percent); nor any means of transportation (85 percent). Women with four ANC visits and who delivered in an institutions were less likely to be in the bottom 40 percent of the distribution (Table 1). There were small but significant differences between women who did all four ANC visits or institutional deliveries and those who did not. Those who did were less likely to live in rural areas, more likely to have secondary education, access to information, and to be in the top 60 percent of the distribution. At the cluster level, they were also less likely to live in a mountainous cluster or a cluster with a high percentage of poor people, but more likely to have a greater share of adults with secondary education or with a greater share of women who have done at least four recommended ANC visits. Women who accessed institutional care also lived in areas with better access to health facilities that provide ANC and normal or C-section delivery services. Women who delivered at a health institution had on average 18 health institutions providing normal delivery services in their vicinity compared to 13 health institutions providing these services overall. However, there are no differences with respect to quality of care, such as living in an area with health services providing ANC more than 18 days per month14 or being well equipped.15 14This was chosen as it corresponds to nearly every day of a working week in a month. 15We consider that a health facility is well equipped when it has at least 70 percent of the recommended equipment for ANC or delivery. 9 Table 1: Characteristics of 15-49 year old women who gave birth in the 5 years before the survey Variable name - Definition All 15-49 year old women Women characteristics Women with 4 ANC Women doing institutional delivery Lives in rural area (%) 62.3 (0.014) 57.1*** (0.017) 46.1*** (0.021) Age (average in years) 29.8 (0.123) 30.0** (0.140) 29.5** (0.205) Married (%) 85.6 (0.007) 86.3 (0.008) 83.5*** (0.011) Has secondary education (%) 44.1 (0.014) 52.9*** (0.014) 65.6*** (0.014) Has a job outside home (%) 55.8 (0.011) 59.6*** (0.011) 59.0*** (0.016) Has access to information (%) 64.3 (0.013) 70.0*** (0.012) 77.1*** (0.014) Number of births (average) 1.29 (0.012) 1.2*** (0.011) 1.2*** (0.011) Household characteristics Average size 5.89 (0.056) 5.8*** (0.059) 5.8* (0.067) In bottom 40 of wealth distribution (%) 42.0 (0.018) 33.3*** (0.011) 19.8*** (0.013) Doesn’t have any means of transportation (%) 85.1 (0.009) 81.7*** (0.017) 77.7*** (0.013) Doesn’t have media (radio, TV) to access info (%) 48.6 (0.013) 41.6*** (0.014) 33.7*** (0.017) Cluster characteristics Mountainous (slope greater than 10%) (%) 24.7 (0.025) 21.0*** (0.024) 19.4*** (0.027) Hhs in the cluster that are in Q1 (%) 21.4 (0.013) 16.7*** (0.010) 9.5*** (0.007) Women with at least 4 ANC visits (%) 66.6 (0.013) 74.1*** (0.010) 60.1*** (0.011) Adults (15 to 65) in the cluster that have at least secondary education (%) Health facility characteristics in cluster vicinity 5.8 (0.004) 6.7*** (0.004) 8.9*** (0.006) Total (average) 46.75 (2.29) 48.58*** (2.33) 61.47*** (3.17) Providing ANC services (average) 40.81 (2.01) 42.34*** (2.05) 53.53*** (2.79) Providing normal delivery services (average) 13.49 (0.62) 13.92*** (0.64) 17.41*** (0.88) Providing c-section services (average) 8.19 (0.46) 8.54*** (0.47) 11.09*** (0.65) Providing ANC services at least 18 days a month (%) 0.86 (0.01) 86.1 (0.006) 86.8*** (0.006) Facilities w/ ANC services w/ medication for ANC (%) Facilities w/ ANC services w/ equipment for ANC (%) Facilities w/ delivery services w/ equipment for de livery (%) 25.0 (0.001) 25.2 (0.011) 23.5** (0.011) 49.1 (0.010) 49.7** (0.011) 49.5 (0.010) 53.5 (0.015) 54.5* (0.015) 55.1* (0.014) Note: Linearized standard errors into brackets. Testing against pregnant women not doing the mentioned institutional healthcare. Adjusted Wald test *** significant at 1% level; ** significant at 5% level; * significant at 10% level. New population weights for each year to reflect total women population in Haiti. Source: Authors’ estimates with EMMUS 2017 5 Findings 5.1 Bivariate analysis First, we investigate how individual, household, cluster, and health facility characteristics affect women’s prevalence of institutional care depending on the barriers identified as salient in the literature (Table 2). We use a Wald test when the differences tested are binary and a t-student test by coefficient from an OLS regression when the variables are categorical. In terms of the decision to seek care, women with better socioeconomic conditions (such as outside employment, secondary education, information, being in the top 60 percent of the 10 Table 2: Prevalence of institutional care by barriers. Three-delay model. Women Characteristics. At least four ANC Inst’l delivery Decide to seek care based on socioeconomic and cultural factors Has job No 61.0 (0.016) 39.1 (0.017) Yes 71.2*** (0.013) 44.4** (0.016) Has secondary education No 56.2 (0.017) 25.9 (0.014) Yes 80.0*** (0.013) 62.5*** (0.014) Has access to information No 56.0 (0.019) 27.0 (0.017) Yes 72.6*** (0.012) 50.4*** (0.016) Is independent - does not need permission No 54.1 (0.031) 29.0 (0.027) Yes 67.9*** (0.013) 43.3*** (0.014) Lives in household from bottom 40 No 76.8 (0.012) 58.1 (0.017) Yes 52.8*** (0.019) 19.9*** (0.013) Identify and reach care Lives in rural households No 76.0 (0.016) 60.2 (0.022) Yes 61.0*** (0.017) 31.1*** (0.017) Has no transport means No 82.1 (0.017) 63.1 (0.024) Yes 64.0*** (0.014) 38.4*** (0.014) Lives in mountainous cluster No 70.0 (0.012) 45.0 (0.015) Yes 56.6*** (0.032) 33.0** (0.032) Lives in service area where 50% of facilities with ANC 18 days/month No 61.3 (0.043) 23.1 (0.064) Yes 66.7 (0.013) 42.3** (0.014) Lives in service area where 50% of facilities with inst’l delivery available No 65.7 (0.014) 39.8 (0.015) Yes 72.3* (0.027) 54.4** (0.042) Receive adequate care Lives in service area where share of facilities with ANC with medications is Low (< 25%) 66.4 (0.017) 46.6 (0.019) Medium (25%<HC<40%) 69.8 (0.026) 39.1* (0.029) High (>40%) 64.7 (0.031) 32.3*** (0.033) Lives in service area where share of facilities with ANC with equipment is Low (<30%) 56.8 (0.041) 25.1 (0.031) Medium (30%<HC<50%) 68.7** (0.017) 52.0*** (0.024) High (>50%) 67.7* (0.020) 39.0*** (0.020) Lives in service area where share of facilities w/ deliveries with equipment is Low (<35%) 62.5 (0.028) 33.2 (0.026) Medium (35%<HC<70%) 69.1* (0.015) 48.6*** (0.020) High (>70%) 69.8 (0.026) 39.1 (0.033) Note: Linearized standard errors into brackets. Adjusted Wald test except for receive adequate care barriers where we use linear regression coefficients. New population weights for each year to reflect the total women population in Haiti. ***p < 0.001; **p < 0.01; *p < 0.05. Source: Authors’ estimates with EMMUS 2017 distribution of assets) are more likely to receive institutional care. Thus, 80 percent of women who delivered in the previous five years and had a secondary education had at least four ANC visits, compared to 56.2 percent of women with lower or no education. Better educated women 11 were more likely to have at least four ANC visits and to deliver in a medical institution than less educated women. Women’s empowerment seems to matter as well; women who needed permission from their husbands or fathers to go to a health center in any circumstances were less likely to get institutional care than more independent women. In terms of the decision to reach care, physical barriers are also a deterrent. Women living in rural households, without access to transportation or in a mountainous cluster are less likely to get this type of care than urban women, women with a car or motorbike in the household, or women living in a non-mountainous cluster. These differences are the largest when looking at rural women’s likelihood to deliver in a medical institution: rural women are half as likely as urban women to deliver in a medical institution. In addition, women living near health institutions that provide ANC consultations at least 18 days per month are more likely to deliver at a medical institution than women living in a cluster where less than 50 percent of health institutions in a 10km radius provide ANC consultations 18 days a month.16 Interestingly, having access to ANC services more than 18 days per month does not affect women’s likelihood to go to all four prenatal visits. Living in a cluster where at least 50 percent of the health institutions offer normal or C-section delivery services appears to positively influence women to receive ANC consultations. In terms of the decision to receive care, quality of care seems to matter but the relationship is not always linear. For instance, more than 65 percent of women living in clusters with fewer than 25 percent of health institutions having three types of medication attend at least four recommended ANC consultations. Similarly, women living in clusters where more than 40 percent of health institutions have three types of medication are also more likely to make at least four recommended ANC visits. At the same time, women living in a cluster where health facilities are well equipped in terms of ANC equipment encourages women to attend ANC visits but this is not the case when the health facilities are well-equipped with equipment used for delivery. Living in a cluster where 25 to 40 percent of health facilities have at least three types of ANC medication increases the likelihood of delivering in a medical institution. Similarly, women living in an area where 30 to 50 percent of health institutions are well equipped for ANC care are also more likely to deliver in a medical institution. Living in a cluster where 35 to 70 percent of institutions have 40 percent of delivery equipment appears to encourage women to deliver in a medical institution. While insignificant, fewer women would be encouraged to deliver in a medical institution if more than 70 percent of the latter were well-equipped for delivery. 5.2 Econometric results with the multilevel model To perform the multilevel regression, we start by estimating grand mean centered variables to be able to make inferences on the absolute effect of women and household-level characteristics and on cluster- and service area-level variables (Sommet and Morselli, 2017). After confirming that there is enough variability across clusters to justify the use of a multilevel model, we check the effect of lower-level variables across clusters. Given that including residual terms associated with each individual level (mother is head, mother has secondary education) does not significantly improve the fit of the regression, we use a constrained form. The best model for each outcome of interest is reported in table 3.17 For all outcomes, mother-level variables seem to matter more than cluster or service level variables when explaining access to institutional care (Table 3). When controlling for other 16All clusters have at least one health facility providing ANC services in a 10km radius which is the service area. 17The table reports the Odds Ratios, that is the odds of the outcome for women with a certain characteristic relative to those without it. Thus is the case of having at least four ANC visits, pregnant women in the bottom 40 are 0.755 times less likely to have at least four ANC visits compared to pregnant women in the top 60 12 Table 3: Determinants of the use of maternal healthcare Variables At least four ANC Institutional delivery OR CI OR CI Women’ and hhs’ characteristics Head 0.673 (0.209 2.170) 2.614 (0.620 11.02) Spouse 1.217* (1.018 1.455) 0.914 (0.762 1.096) Age 1.056*** (1.041 1.072) 1.048*** (1.033 1.064) Number children 0.741*** (0.661 0.831) 0.552*** (0.488 0.623) Squared number children 1.011* (1.001 1.021) 1.037*** (1.026 1.048) Job outside 1.286*** (1.117 1.481) 1.125 (0.970 1.305) Secondary edu 1.664*** (1.404 1.972) 1.881*** (1.600 2.213) Access information 1.260** (1.071 1.483) 1.130 (0.935 1.366) Permission 0.839 (0.654 1.076) 0.718* (0.530 0.973) Bottom 40 (B40) 0.755* (0.605 0.943) 0.609*** (0.486 0.762) No transportation 0.856 (0.678 1.082) 0.755** (0.616 0.926) Cluster characteristics Mountainous 0.868 (0.733 1.028) 0.910 (0.722 1.146) Share adults w/ secondary edu is Medium 0.719 (0.501 1.032) 1.084 (0.723 1.625) High 0.734 (0.467 1.153) 1.027 (0.647 1.630) Share hhs in bottom quintile is Medium 1.163 (0.926 1.461) 0.630*** (0.481 0.824) High 1.266 (0.985 1.626) 0.449*** (0.329 0.612) Share women w/ 4 ANC visits is Medium 3.120*** (2.652 3.671) 1.601*** (1.257 2.040) High 11.10*** (8.874 13.88) 2.368*** (1.816 3.089) Health centers characteristics Dum ANC18dys 0.803 (0.482 1.338) 1.995 (0.886 4.490) Dum instDEL 0.941 (0.710 1.246) 1.131 (0.834 1.532) Share facilities w/ ANC with medications available is Medium 1.005 (0.830 1.217) 0.721** (0.562 0.924) High 1.061 (0.898 1.255) 0.837 (0.668 1.047) Share facilities w/ ANC with equipment available is Medium 0.924 (0.727 1.173) 0.956 (0.684 1.337) High 0.940 (0.759 1.166) 1.019 (0.750 1.385) Share facilities w/ delivery with equipment available is Medium 1.048 (0.883 1.243) 0.968 (0.773 1.212) High 1.126 (0.927 1.366) 1.109 (0.857 1.435) Interaction factors Head + dum ANC18dys 1.547 (0.477 5.017) 0.355 (0.0837 1.504) Perm + dum instDEL 1.090 (0.516 2.300) 1.509 (0.700 3.252) B40 + dum instDEL 0.991 (0.655 1.500) 1.788* (1.108 2.885) Access + C secedu=MEDIUM 1.517 (0.984 2.338) 1.146 (0.750 1.749) Access info + C secedu=HIGH 1.032 (0.634 1.680) 1.503 (0.952 2.375) Constant 0.803 (0.470 1.373) 0.319** (0.143 0.713) Random intercept 2.97e-08 (0 .) 0.552*** (0.451 0.675) N 4896 4899 Note= ***p<0.001; **p<0.01; *p<0.05. Reference categories are indicated in parentheses after the names of the characteristics being considered. OR: odds ratios; CI: confidence interval. Dum ANC18dys: dummy if ANC available at least 18 days per month. Dum instDEL: dummy if institutional delivery available to women in the cluster. C secedu: share of adults with secondary education. Source: Authors’ estimates with EMMUS 2017 13 variables across clusters, older mothers, mothers with a job outside their house, mothers with at least secondary education, and mothers with access to information are more likely to make at least four ANC visits than others. However, neither having a job outside the house nor having access to information increases a mother’s odds of delivering in a health institution. These four characteristics either directly or indirectly relate to experiences that would encourage women to seek care (ANC or institutional delivery). They are likely to indicate that mothers are more knowledgeable about the risks associated with their pregnancy as women are more educated and, as a result, more likely to make informed decisions about their healthcare. In addition, older women might be more aware that risks increase with age, helping explain why they are more likely to consult with or deliver in a health institution. Finally, women’s networks are likely to expand with employment in activities outside their house; women could have met more women throughout their lives who had faced such risks or had read about such cases. If a woman is the spouse of the head of the household, she is more likely to attend at least four ANC visits. However, if a woman is the head of the household, her probability of attending at least four ANC visits does not increase. This may be because husbands encourage their wives to seek care (as confirmed in the qualitative data analysis) and may provide financial or other types of support to help them access and receive care. However, the effect disappears when considering the decision to deliver in a health institution. We find that having children discourages women from attending four ANC visits or de livering in a health institution, and this effect becomes even stronger as women have more children. There are two potential reasons for this. First, women who have gone through pregnancy and delivery before might feel more knowledgeable about what happens during pregnancy and childbirth and have less of a need for consultation. Second, women need to make arrangements for someone to take care of their children while they attend these health services. Poverty, as measured by being in the bottom 40 of the wealth index distribution, decreases the odds of doing at least four ANC visits and, even more importantly, to deliver in a health institution supporting the hypothesis that poorer women are less likely to seek care. Even when ANC visits are free, additional costs (blood tests, medicines) and also hidden costs (care of other children, transportation, medications) can be a hindrance for poorer women. These costs might be even more in the case of an institutional delivery. The lack of means of transportation decreases women’s odds of delivering in a health institution, even when controlling for accessibility with the existence of ANC services within 10km of the woman’s cluster. Women without transportation means are less capable to reach care than those with transportation means. Finally, women who need permission to attend normal health visits are less likely to deliver in a health institution. This suggests that empowering women to make other health decisions could significantly impact their decisions to seek prenatal and delivery care at a health institution. Turning to cluster-level effects, living in a cluster where a large share of women that attend at lea