UNDERSTANDING STUNTING IN PERUVIAN CHILDREN: A MULTILEVEL ANALYSIS By Claudia F. Benavides Vizcarra B.A, Pontificia Universidad Catolica del Peru (PUCP), 2003 THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS IN DEVELOPMENT ECONOMICS UNIVERSITY OF NORTHERN BRITISH COLUMBIA May, 2015 Claudia Benavides © UMI Number: 1526534 All rights reserved INFORMATION TO ALL USERS The quality of this reproduction is dependent upon the quality of the copy submitted. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed, a note will indicate the deletion. Di!ss0?t&iori Publishing UMI 1526534 Published by ProQuest LLC 2015. Copyright in the Dissertation held by the Author. Microform Edition © ProQuest LLC. All rights reserved. This work is protected against unauthorized copying under Title 17, United States Code. ProQuest LLC 789 East Eisenhower Parkway P.O. Box 1346 Ann Arbor, Ml 48106-1346 Abstract This thesis applies the Chronic Undemutrition Framework by Tufts University (2001) in order to analyze the determinants o f chronic malnutrition, or more specifically stunting, in Peru. Using cross-sectional data for children aged 6-59 months old taken from the Peruvian Demographic and Health Survey for 2009-2012,1 apply a multilevel analysis approach to test the effect of individual, household and community factors for stunting in two groups o f children from ages 06 to 24 months and 25 to 59 months old. The findings of this study are consistent with the literature, and support the multisectorial approach to decrease stunting. The results indicate considerable effects at each level (individual, household and community). Significant determinants at the individual level are age (for children 6-24 months old), gender, low birth weight and diarrheal disease. At the household level, wealth and mother’s education are clearly associated with child growth. Access to healthcare services and antenatal care and living in mountains areas are significant factors at the community or environmental level. Also, the Articulated Nutrition Program implemented by the Peruvian government since 2008 appears to have favourable impact on children between the ages of 25 and 59 months old. 2 Table of contents A bstract...................................................................................................................................................................2 Table of contents.................................................................................................................................................... 3 List of Tables.......................................................................................................................................................... 5 List of Figures........................................................................................................................................................ 6 Abbreviations......................................................................................................................................................... 8 Acknowledgment and Dedication.......................................................................................................................10 Chapter 1: Introduction...................................................................................................................................... 11 1.1 Background................................................................................................................................................ 11 1.2 Research Problem, Objective, Questions and Thesis Statem ent........................................................ 24 1.3 C ontributions............................................................................................................................................. 25 1.4 Summary of main findings....................................................................................................................... 26 Chapter 2: Literature Review............................................................................................................................28 2.1 Socioeconomic and Demographic D eterm inants.................................................................................. 29 2.2 Child Determinants - Morbidity S tatus.................................................................................................32 2.3 Environmental Conditions....................................................................................................................... 34 2.4 Child and M aternal Caring Practices.....................................................................................................35 2.5 The Hierarchical Models of Chronic Undernutrition and S tunting.................................................. 38 2.5.1 Studies on Developing Countries outside Latin America............................................................. 38 2.5.2 Studies in Latin American Countries (excluding P e ru )............................................................... 42 2.5.3 Studies on P e ru ...................................................................................................................................45 Chapter 3: Data and M ethods........................................................................................................................... 49 3.1 Data............................................................................................................................................................. 49 3.1.1 Summary of statistics........................................................................................................................ 50 3.2 Conceptual Fram eworks.......................................................................................................................... 57 3.3 M ethods...................................................................................................................................................... 61 3.3.1 The Empirical Models....................................................................................................................... 61 3.3.1.1 Dependent Variable: Chronic Under nutrition S tatus................................................................61 3.3.1.2 Independent V ariables................................................................................................................... 64 3.3.1.3 Modeling A pproach........................................................................................................................ 66 3.3.1.4 Multilevel Analysis......................................................................................................................... 70 Chapter 4: Results and Discussion.................................................................................................................... 74 3 4.1 Estimation Results for the age group between six to twenty four months o ld ................................. 74 4.2 Estimation Results for the age group between 25 to 59 months old ................................................... 78 4.3 Logistic Regression Results for the age group between 25 to 59 months o ld ................................... 82 Chapter 5: Conclusions.......................................................................................................................................86 5.1 Summary of Main Findings..................................................................................................................... 86 5.2 Policy Implications, Limitations and Further Studies..........................................................................90 5.2.1 Policy Im plications............................................................................................................................ 90 5.2.2 Lim itations..........................................................................................................................................91 5.2.3 Future Studies.....................................................................................................................................92 Bibliography........................................................................................................................................................ 93 Appendix 1: List of Variables of Interest....................................................................................................... 101 4 List of Tables Table 3.1 Summary of statistics 50-57 Table 4.1 Determinants of growth (stunting) for children between six and twenty-four months old 77-78 Table 4.2 Determinants of Stunted for children between twenty five months old and before fifty nine months old 80-81 Table 4.3 Determinants of Stunted for children between twenty five months old and before fifty nine months old: Logistic multilevel effects regression models 82-84 5 List of Figures Figure 1.1 Estimated Global Prevalence (%) of children under five moderately or severely stunted and of children under five moderately or severely underweight for the period 1990-2012 18 Figure 1.2 Estimated prevalence (%) and number o f children under-five years of age by stunting (moderate or severe) by MDG region: 1990 and 2011* 18 Figure 1.3 Estimated Prevalence (%) o f children under-five years o f age affected by stunting (moderate or severe) for the period 20062012 by Income Group, Peru, Americas Region and Global 21 Figure 1.4 Estimated Prevalence (%) o f children under-five years o f age affected by stunting (moderate or severe) for the period 20062012 by Latin America & Caribbean and Upper Middle Income Countries (%) 21 Figure 1.5 Gross Domestic Product (GDP) for the period 2000-2012 by Latin America & Caribbean and Upper Middle Income Countries (average annual % growth) 22 Figure 1.6 Estimated Prevalence (%) of children under-five years of age affected by stunting (moderate or severe) in Peru for the period 1990-2012* 23 Figure 3.1 UNICEF Conceptual Framework of Undemutrition 58 e Figure 3.2 Conceptual Framework o f Stunting 60 Figure 3.3 Histogram, kernel density of height by age Z-score 64 Figure 3.4 Trend and line o f mean height for age z-score by children’s age 67 Figure 3.5 Trend and line o f mean height for age z-score by children’s before -6 months old 68 Figure 3.6 Trend and line of mean height for age z-score by children’s after 5 and before -59 months old 68 Figure 3.7 Trend and line of mean height for age z-score by children’s after 24 and before 60 months old 69 Figure 3.8 Scatter of cluster mean o f height for age z-score against clusters 71 7 Abbreviations ANP Articulated Nutrition Program CIP Comprehensive Implementation Plan CM Chronic Malnutrition CU Chronic Undemutrition CDC Centers for Disease Control and Prevention DHS Demographic and Health Survey HAZ Height for age z-score INEI National Institute of Statistics and Informatics of Peru MDG Millennium Development Goal MEF Ministry of Economy andFinance o f Peru MIDIS Ministry of Development and Social Inclusion of Peru MMPV Ministry of Woman and Vulnerable Population NCHS National Center for Health Statistics NPTAI National Plan to Take Action for the Infancy and Adolescence OLS Ordinary Least Square PSU Primary Sample Unit BfR Budget for Results SD Standard deviation SPSS Statistical Package for the Social Science STATA Data Analysis and Statistical Software 8 UNICEF The United Nations Children's Fund UNSSCN United Nations System Standing Committee Nutrition WHO World Health Organization Acknowledgment and Dedication I would like to thank my supervisor Dr. Jalil Safaei, for his support throughout this project. This thesis would not have been written without his guidance and encouragement. I also would like to thank my committee members, Dr. Fiona MacPhail and Dr. Mamdouh Shubair for taking an interest in my investigation and for their useful and invaluable comments. I owe a great thank you to my friend Dr. Luis Cordero, who, when I thought I was not able to understand how to apply a multilevel analysis, told me I was. His knowledge, experience, guidance and encouragement throughout my research brought me to this point. I want to thank my parents, Luis and Carmen, and my sisters, Paola and Carmen, for their unconditional support, love and encouragement. Thank you for always believing in me. Gracias a mis padres, Luis y Carmen, a mis hermanas, Paola y Carmen, por su apoyo incondicional, amor y por creer en mi siempre. 10 Chapter 1: Introduction 1.1 Background The importance of child nutrition on development outcomes has gained increased international awareness over the years. Strong evidence shows that inadequate nutritional status during pregnancy and in the first two years of life leads to lower human capital endowments, negatively affecting physical strength and cognitive ability in adults. This contributes directly to a reduction of individuals earnings potential and damages national economic growth (World Bank, 2007; Save the Children, 2013). Save the Children (2013) based on a study with a panel data of 3 000 children in four countries (Ethiopia, India, Peru and Vietnam) bom between the years 1994-1995 and 2000-2001, estimated that the global economic impact of child malnutrition at 125 billion dollars a year by the time today’s children reach working age in 2030. Specifically, chronically malnourished children are significantly less able to read, write a simple sentence, or perform basic arithmetic, and their average adult earnings will be 20% less than healthy children. Therefore, child undemutrition1 is a global problem that has been addressed by different developing countries as well as different aid organizations. It is a major Millennium 1 The literature uses the terms: undemutrition and malnutrition interchangeably. H ow ever, it must be taken into consideration that they have different meanings. A ccording to Prakash (2006) “Malnutrition refers to all deviations from adequate and optimal and nutritional status, including energy undemutrition and over-nutrition (obesity is a form o f malnutrition). The term ‘undemutrition’ is used to “refer to generally poor nutritional status” (Shetty, 2006, p. 526). In this sense malnutrition im plies undemutrition but also over-nutrition. For the purpose o f this thesis I will use the term “undemutrition” . 11 Development Goal (MDG)2 target which is a key indicator for measuring progress towards MDG1: Reduce poverty, hunger and undemutrition, and a critical input factor for achieving MDG4: reduce child mortality, and MDG5: Improve maternal health. Undemutrition is defined by UNICEF (2003) as the outcome of insufficient food intake and repeated infectious diseases. However, their manifestations on children health status will vary depending on severity, duration and age (Wright & Garcia, 2012). In this sense, undemutrition can be separated in two groups acute and chronic, most often demonstrated by wasting and stunting, respectively (Reinhardt & Fanzo, 2014), or by underweight if it is a combination of both types (UNICEF, 2013)3. Acute undemutrition (AU) is an effect of an immediate problem such as sudden catastrophes, seasonal depressions/shortages, highly infection-disease environments. Whereas, chronic undemutrition (CU) is more closely associated with latent poverty, chronic food insecurity, poor feeding practices, and prolonged health problem (Bergeron & Castleman, 2012; Reinhardt & Fanzo, 2014). 2 At the M illennium Summit in September 2000 the largest gathering o f world leaders in history adopted the U N M illennium Declaration, comm itting their nations to a new global partnership to reduce extreme poverty and setting out a series o f tim e-bound targets, with a deadline o f 2015 that have becom e known as the M illennium D evelopm ent G oals. There are 8 M illennium D evelopm ent G oals (M D G s); M DG1: Reduce poverty, hunger and undemutrition; M DG2: A ch ieve universal primary education; M DG3: Promote gender equality and em pow er women; M D G 4: Reduce child mortality; M DG5: Improve maternal health; M DG6: Combat diseases (HTV/AIDS, malaria); M DG7: Ensure environmental sustainability and M DG8: D evelop a global partnership for developm ent. (United N ations, 2014) 3 Stunting, underweight, wasting and severe acute undemutrition are defined as follow s: Stunting “reflects chronic undemutrition during the most critical periods o f growth and developm ent in early life” and it is measured on the height for age o f the children aged to 0 to 59 months; underweight “is a com posite form o f undemutrition that includes elem ents o f stunting and w asting” here the indicator takes into account the w eight for age o f children aged 0 to 59 months old, wasting “reflects acute undemutrition, and both wasting and severe acute undemutrition consider the w eight for height”, but for children aged 0 to 59 months and 6 to 59 months, respectively (U NICEF, 2 013, p. 7) 12 Chronic Undemutrition “is defined as a form of growth failure that causes both physical and cognitive delays in growth and development”. Stunting is the indicator used as a proxy o f CU, and it is also known as linear growth failure, defined “as the inability to attain potential height for a particular age” (Reinhardt & Fanzo, 2014, p. 1). The causes o f CU are multidimensional, as are the consequences, which can be both short-term and long-term Chronic Undemutrition is a priority for the Peruvian Government. It is a main topic in all the policies oriented to early infancy. It is one of the main indicators of the National Plan to Take Action for the Infancy and Adolescence for the year 2021 (MMPV, 2012). It is also one of the main outcomes o f the Budget for Results (BfR) policy BFR’s policy was implemented since 2007 and it is ruled by the General Law o f Budgeting (MEF, 2011). Regardless of the many efforts and strategies of the Peruvian Government, CU is still a major problem in Pern. According to the Demographic and Family Health Survey (DFHS) for the years 2000-2012, stunting in children under five years old began to decline in 2009 after a period of stagnation from 2000 to 2007, where it stayed at a level close to 30%. In 2012, the rate was 18.1%, a significant decline from a rate of 28.0% in 2005. In rural areas the drop was even more significant as it went from 47.1% to 31.9%, a drop of 15.2 percentage points. Despite these improvements, the rates of stunting still remain high in the rural areas and in some geographic regions o f Peru (INEI, 2014). The high rates o f stunting in Peru shows a need for the Peruvian government to take on the major challenge of devising interventions that can address stunting both in rural and urban areas. Successful social interventions in developing countries and in their rural regions, demands not only a good strategy but also a comprehensive strategy that applies different kinds of resources such as human capital, government funding, as well as supportive infrastructure. The key challenge of developing a comprehensive strategy to reduce the deprivation that contributes to stunting is the main motivation for my research, which I hope will contribute to the dialogue and development in improving health outcomes of children and infants in Peru. In this thesis I analyze the determinants of CU by using a multilevel econometric model based on popular conceptual frameworks of undemutrition4. I argue that by exploring the relationship between CU and its determinants, we can contribute to the design, monitoring and evaluation of public policy that could more successfully reduce CU in Peru. Chronic Undemutrition is a major problem o f public health and its consequences are manifested throughout the human life cycle. The poor and the extreme poor are mainly affected. The most recent evidence reinforces the importance o f facing CU as it has immediate negative effects. These effects include a greater likelihood of illness or premature death in children under five years of age, a greater likelihood of later entry to school, poor cognitive development, poor school performance, reduced productivity and low work capacity leading to increased costs and economic losses to the family and society (Black et al., 2008; Bhutta, et al., 2013; Crookston, et al., 2011). There is evidence that children with CU have lower cognitive development related to anxiety and low self-esteem, with the worst emotional outcomes and behavior during adolescence (Walker, et 4 In 1990, UNICEF developed the “Strategy for improving nutrition o f children and w om en in developing countries” document which formed the basis o f analysis in most o f the papers where malnutrition or child mortality were analyzed. In this docum ent, a conceptual framework o f child undemutrition and its determinants w as proposed. The document was updated in 2013 with the title “Improving child nutrition the achievable imperative for global progress”. 14 al., 2007). Even if children experience an excellent nutritional rehabilitation process, the child malnutrition observed before 60 months of life is associated with a persistent attention deficit in adulthood regardless o f socioeconomic conditions (Galler et al., 2012). In undemutrition children more emotional problems have been identified (Liu et al, 2004; Pollit et a l., 2000; Gardner et al., 1999), as well as, decreased activity, less play and exploration (Meeks Gardner et al, 1999), less vocalization (Femald & Grantham-McGregor, 2002), a tendency to stay closer to the mother, be more apathetic, and more inhibited (Aburto et al., 2009; Baker-Henningham et al., 2009). Therefore, undemutrition is one of the major targets in Public Policy around the World due to its negative impacts on child health leading to increased rates of child mortality under five years of age (WHO, 2014). According to Black et al (2013), undemutrition caused 45% of the total child deaths in 20115, this percentage implied a staggering 3.1 million child deaths that year. The United Nations (UN) indicates the decrease o f undemutrition in any of its types is a requirement in order to achieve the Millennium Development Goals (MDGs) to reduce poverty, hunger and undemutrition (MDG1) and reduce child mortality (MDG4)6 (UNSSC, 2010). This decision is based on the evidence that links malnutrition and child mortality (Pelletier et al., 1993) and also because there are effects o f malnutrition on delayed mental development, poor school performance and reduced intellectual capacity on children (Mendez & Adair, 1999; WHO, 1999; de Onis, 2006). 5 The estimate includes fetal growth restriction, stunting, wasting, and deficiencies o f vitamin A and zinc along with suboptimum breastfeeding. 6 The target for the M DG1 w as to reduce the proportion o f people who suffer from hunger by half, between 1990 and 2015, and for the M D G 4 w as to reduce by two-thirds over the sam e period. (W H O, 20 1 4 ). 15 There are two main indicators used to analyze and to track the progress towards the achievement o f the MDGs in undemutrition: i) the high prevalence of stunting and ii) high prevalence of underweight children, through the new WHO growth standards developed in 2006 (UNSSC, 2010). The first main indicator (stunting) measures CU and “it is defined as the percentage of children aged 0 to 59 months whose height for age is below minus two standard deviations (moderate and severe stunting) and minus three standard deviations (severe stunting) from the median of WHO child growth standards” (UNICEF, 2013, p. 7). The second main indicator (underweight) reflects a combination o f chronic and acute undemutrition and “it is defined as the percentage o f children aged 0 to 59 months whose weight for age is below minus two standard deviations (moderate and severe underweight) from the median of the WHO child growth standard” (UNICEF, 2013, p. 7). The monitoring of stunting is a recent addition in the reports about MDGs and this incorporation is supported on the irreversible damage done to a child’s body and brain that it is different from the effects produced by children that are underweight. In fact, it is true that both, indicators of stunting and being underweight, provide understanding about the relationships and effects of dietary intake and illness. However, stunting (height for age) is taken as a better indicator to measure the effects of undemutrition at the population level since this indicator reflects a cumulative growth deficit that has long-term consequences for adult health and human capital that cannot be reversed (Luter, Chaparro, & Munoz, 2011; UNICEF, 2013; UNICEF, 1990; UNSSC, 2010). The Millennium Development Goals set a 50% reduction, for both child underweight and for population undernourishment by 2015 when compared to 1990 (Gentilini & Webb, 2008). 16 According to the data for 2012 (The World Bank, 2014) there is evidence of improvements in children’s nutritional status. The percentage o f underweight children is estimated to have declined from 25% in 1990 to 15% in 2012 (Figure 1.1). Stunting in children under five years of age has decreased globally from 40% to 25% over the same period (Figure 1.1). These are encouraging results, but there is still work to be done. For example, above all other continents, Asia and Africa have the worst results and it is a concern, that there is inequality on these achievements between different regions and within countries, especially for the CU and its consequences. In terms o f absolute numbers, the number o f stunted children in Asia is estimated to have reduced by more than half between 1990 and 2012. However, in Africa the number of stunted children is estimated to have increased from 46 million in 1990 to 59 million in 2012. The inequality in the achievements highlights the low probability to reach the MDG1 and their related health indicators related by 2015 (United Nations, 2014). As a result, the global agenda still considers CU as a priority even after 2015. At the 65th World Health Assembly in May 2012, WHO’s Member States endorsed a Comprehensive Implementation Plan (CIP) on Maternal, Infant and Child Nutrition that included a 40% reduction in the number of stunted children by 2025 against the 2010 baseline as a part o f six global targets to be achieved (WHO, 2012). 17 Figure 1.1- Estimated Global Prevalence (%) of children under five moderately or severely stunted and o f children under five moderately or severely underweight for the period 1990-2012 45 40 35 30 25 20 15 10 5 0 1990 1995 2000 2005 ■ S tu n te d 2010 2012 ■ U n d e rw e ig h t N ote: This figure is constructed with information from integrated data o f U NICEF, W HO and World Bank 1990- 2 0 127 Figure 1.2- Estimated prevalence (%) and number o f children under-five years of age by stunting (moderate or severe) by MDG region: 1990 and 2011 60 60 47 50 i 47 40 37 40 38 1 37 28 30 30 40 37 38 33 26 23 ; 18 20 i 12 II il 10 0 NA SSA I.A C LA SA SLA \V A [18 17 OCLN I . IIII CCA IX-v G lo b a l Cl K* a 1990 ■ 2011 Note: This figure is constructed with information from Peruvian D H S and from integrated data o f U NICEF, W HO and W orld Bank 1990-2012. The prevalence o f stunted children in the case o f Peru is for the year 2012 not for the year 2011. The Abbreviations are: N A (Northern Africa); SSA(Sub-Saharan Africa); LAC (Latin A m erica & Caribbean); EA (Eastern A sia); SA (Southern Asia); SEA (South-Eastern Asia); W A (W estern A sia); OCEN (Oceania); CCA(Caucasus & Central Asia); D ev (D eveloped Countries); PER(Peru) 7 In September 2013 UNICEF, W HO, and the W orld Bank updated their joint database on child malnutrition and released new global and regional estim ates for 2012 (W H O, 2014). 18 As I pointed out, the prevalence of stunted children in the world presents a downward trend from 1990 to 2012 (Figure 1.1), which provides evidence that the efforts o f undemutrition policies around the world have had some effects. However, there were no major differences in the reduction in the rates for before 2000 and after 2000, when the MDGs were set. The reduction was seven percentage points for the 1990-2000 period and eight percentage points for the 20002012 period, respectively. Estimates from 2011 suggest stunting prevalence reductions o f more than 40% in Asia and Latin America and the Caribbean since 1990 (Figure 1.2). On the contrary, the decreases in Africa and Oceania have been more modest (10-15%). These estimates show the inequality in the progress by regions, meaning that not only stunting is a challenge around the world, but also that we experience differences between regions. It is notable that developed countries have the lowest rate of stunting (Figure 1.2). This implies that economic development or level of income is a condition for a low stunting rate (Figure 1.3). The association between economic growth and the rate of stunting is demonstrated by Heltberg (2009).However, he concludes that the association is quite small, suggesting that it is neccesarry to implement direct interventions to reduce CU. For example, by taking into account a period of 25 years and with the assumption o f only economic growth alone as an intervention to reduce CU, a country with a rate of stunting at 30% would require a constant real growth of 3.7% per capita per year in order to halve the stunting rate and if a country has a stunting rate at 50%, they would require a constant real growth of 5.9% per year to achieve this (p. s85). This shows that CU is not an economic matter only. UNICEF (1990) points out that the causes of CU are numerous ranging from immediate causes (inadequate dietary intake and access to food), underlying (household food insecurity, inadequate care, unhealthy household environment and lack of health services) and basic causes (income poverty, lack o f capital and socioeconomic conditions). The experts suggest that in order to “comprehensively address malnutrition, programs thus need to blend available knowledge and combine multiple approaches” (Bergeron & Castleman, 2012, p. 242)8. The main objective of my thesis is to contribute to the literature by analysing how multiple causes affect the nutrition status of the children in Peru. Peru is an interesting case because its economy has been doing well. The Peruvian economy has maintained steady growth since 2002. During the period 2002-2012 the growth of the Peruvian economy was 6.4% in real terms (MEF, 2013). However, its stunting indicator has not reached the expected results showing higher rates in comparison to other countries in Latin America with similar or even lower economic growth rates (Figure 1.4). Aa key question that arises is what kind o f factors can be affecting the nutrition status of Peruvian children? The purpose of my research is to address this question. 8 There are different sectors who are pointed out to have responsibility in the policy towards C U , such as education, agriculture, health, social protection, w om en empowerment, developm ent and poverty reduction. 20 Figure 1.3- Estimated Prevalence (%) o f children under-five years of age affected by stunting (moderate or severe) for the period 2006-2012 by Income Group, Peru, Americas Region and Global Americas Region* Global High Income Group UMI 3 Upper middle income Group l.ower middle income Group Low Income Group N ote: This figure is constructed with information from Peruvian D H S and from integrated data o f U NICEF, W HO and W orld Bank 1990-2012. The High, Upper, Lower and L ow Incom e Group are classifications from the W orld Bank List o f E conom ics. At regional level there are different classifications made by UNIC EF, the United N ations and the W orld Health Organization. Figure 1.4- Estimated Prevalence (%) of children under-five years of age affected by stunting (moderate or severe) for the period 2006-2012 by Latin America & Caribbean and Upper Middle Income Countries (%) 25 20 15 13 12 10 CO I CR DR BR LAC M EX PAN PER Note: This figure is constructed with information from Peruvian D H S and from integrated data o f UNICEF, W HO and W orld Bank 1990-2012. The countries are labeled as: CO (Colom bia); CR(Costa Rica); D R (D om inican Republic); LAC (Latin A m erica & Caribbean); BR (Brazil); M E X (M exico); PA N (Panama), PER (Peru). 21 Peru is one o f the most economically dynamic countries in Latin America - second only to Panama (Figure 1.5, MEF, 2013). However its prevalence o f stunted growth in children under five years old is less promising. Going further with numbers, the Peruvian economic growth measured was twice the Global economic growth for the period 2000-2012. Economic growth in Peru is more than two points higher than Latin America Region (LAC) and at a similar level to that of Upper Middle Income (UMI) countries (Figure 1.5). Still, the prevalence of stunted children under five years old is higher than countries with worse economic indicators such as Mexico and Brazil, and it is six points more than LAC (Figure 1.4). Figure 1.5- Gross Domestic Product (GDP) for the period 2000-2012 by Latin America & Caribbean and Upper Middle Income Countries (average annual % growth) 6.1 3.6 = — PER 2.7 - LAC G lobal N ote: This figure is constructed with information from W orld Bank (The W orld Bank, 2014). The countries are labels as: CO (Colom bia); C R(Costa Rica); D R (D om inican Republic); U M I (Upper M iddle Income); LAC (Latin A m erica & Caribbean); BR (Brazil); M EX (M exico); PA N (Panama), PER (Peru). As mentioned before, there is a significant reduction in the prevalence of stunting in Peru from 38% in 1990 to 18% in 2012 (Figure 1.6). However, stunting remains a major problem for Peru in comparison with other countries with similar characteristics (Figures 1.4). Evidence also shows that there is inequality in stunting rates between urban and rural areas within Peru (Figure 1.6). As figure 1.6 shows, stunting had a significant decline in 2000 and then in 2009 after a phase of stagnation from 2000 to 2007, when the rate stayed close to 30%. However, during the period 2011 to 2012 there were no major changes. In2012the indicator was 18.1%, only 1.4 percentage points decrease from 2011. Regardless o f these improvements, the rates o f stunting remain high in the rural areas. Even in urban areas the rate increased in 2012 compared to 2011 (Figure 1.6). Moreover, in some specific regions o f Peru such as Huancavelica, Cajamarca and Loreto the stunting prevalence is as high as 50, 35 and 30%, respectively, showing vast regional variation. (MIDIS, 2012). Figure 1.6- Estimated Prevalence (%) of children under-five years of age affected by stunting (moderate or severe) in Peru for the period 1990-2012* 60 30 20 10 11 2011 2012 0 1990 2000 2005 2007 " • ■ “ N a tio n a l 2008 —•" U rb a n 2009 2010 — • — R ural Note: The information is only available at national level for the w hole period 1990-2012, but for urban and rural areas only since the year o f 2 0 0 7 (except for the year 2 0 0 8 ) when the sam ple o f Dem ographic Health Survey in Peru was increased in order to obtain significant estim ations for urban and rural areas Institution (INEI, 2014). 23 1.2 Research Problem, Objective, Questions and Thesis Statement The inequalities showed above demonstrate the need for studies that can help elucidate the factors behind the rates of nutritional deprivation faced by Peruvian children. This also identifies a need for us to develop ways to better help public programs or policies to reduce the problem. As I pointed out previously, this is a crucial aspect of Peru’s development when considering the- vast international evidence on the negative effect that nutritional deprivation during pregnancy and the first two or three years of life has on a child’s future performance in school and in the labor market. These early periods are important because of the high level of sensitivity o f child growth impacted by deprivation. These impacts are proportional to the velocity of growth under normal conditions, which is extremely high in the early years (Valdivia, 2004; Hoffman & Klein, 2012). Many studies refer to the fact that there is a critical window o f opportunity to prevent undemutrition, this is both during the pregnancy of the mother and during the first two years o f life o f a child9. This is when interventions can be implemented to reach optimal growth and development o f the children (Bhutta et al., 2008; Bhutta et al., 2013; Crookston, et al., 2011; UNICEF, 2013). In the 1990s, UNICEF developed a conceptual framework that emphasized several levels of causality for undemutrition (UNICEF, 1990). Pelletier (2002) pointed out that in the UNICEF conceptual framework, “malnutrition and child death are viewed as two of the manifestations of a multisectoral development problem that can be analyzed in terms of immediate, underlying and basis causes. The immediate causes are inadequate dietary intake and infectious disease; the underlying causes are household food insecurity, inadequate maternal and child care and inadequate health services and health environment; the basic causes include formal and non-formal institutions, political and ideological superstructure, economic structure and potential resources’’ (p. 2). The 9 Stunting and C ognitive impairment due to under-nutrition is largely irreversible after the first tw o years o f life; and that preference is being placed on preventing rather than incur huge econom ic costs towards reversing either effects. 24 Peruvian Government used this framework to design the Articulated Nutrition Program (ANP), the main goal o f which was to decrease the prevalence o f CU in children under five years old. This program was part of the implementation o f BfR Policy in 2007 (MEF, 2011). The design includes different level of outcomes related to the causes identified by UNICEF conceptual framework (final, intermediate and immediate) with a particular set o f interventions. However, this program is based on international data and investigations, not based on national data or the Peruvian experience. This thesis is a result of an initiative to explore what factors predict and produce differences in CU status as measured by stunting in Peruvian children. Thus, the main objective of this study was to assess the relationship between stunting o f children aged six to fifty nine months and its determinants factors in Peru. The objective was explored through the following questions: 1. What factors can explain stunting in children aged six to fifty nine months in Peru? 2. Do the same factors that affect Peruvian children during the window o f opportunity o f intervention (before two years old) also affect the children above two years old? 3. Is ANP affecting stunting in Peruvian children? 1.3 Contributions I have applied a multilevel model to the data of the Demographic and Health Survey (DHS) for Peru from 2009-2012 to identify various factors affecting stunting in Peruvian children. Multilevel modelling is a “statistical technique that extends ordinary regression analysis to the situation where the data are hierarchical” (Leyland & Groenewegen, 2003, p. 267). This is the case o f the DHS data as it is organized in hierarchical form which is currently the commonly used evidence base for informing public health policy. Therefore, it is important that policy makers should be aware of the methodology used in this study. Multilevel modelling is a recent 25 technique for the social and economic field, but is already well known and established methodology in the health and education sectors (Goldstein, 2003). Still, it is not so well widespread among Peruvian investigations. It is my understanding that my research will be the first to do so with the most recent data available. Marini & Gragnolaty (2006) and Arocena (2010) used this technique with information from 2000 DHS and 2009 DHS, respectively. This will further the pool of knowledge created by the various investigations which address CU in Peru as developed by Marini & Gragnolaty (2006), Shin (2007), Ravina & Chavez (2007), Arocena (2010), Cruzado (2012), and Gutierrez, et al (2014), among others. My thesis contributes in several aspects. First, I am applying both, linear and binary multilevel regressions to find the interplay of factors at the children, household, and community levels affecting CU in Peru for age groups 6 to 24 months old and 25 to 59 months old for the first time for Peru. Second, I am using most recent data from DHS in Peru, covering 4 years of the survey (2009-2012) with information o f children under five bom during the period 2004-2012. Third, I am introducing new variables in the analysis such as a child birth’s cohort. Also, I am using the new WHO reference growth curve developed in 2006 to estimate the prevalence o f stunting in Peruvian children in an oficial way since 2011 (WHO, 2006; MINSA, 2011). Finally, I am investigating the effect of the ANP on CU since its implementation in 2008. 1.4 Summary of main findings The purpose of this thesis was to empirically examine how stunting in Peru is driven by socioeconomic, demographic, and health condition factors at the individual, household, and community levels. The results generally support the conceptual framework proposed by UNICEF (1990) and adapted by Tufts University (2001). The findings identify several factors at the 26 individual, household and community levels that are associated with stunting in Peruvian children. Age is significant only for the children between 6 to 24 months old, but not for the children in the second group of age. This finding is related to the investigations that point out that there is a window o f opportunity to implement interventions in the two first years of life, suggesting a policy of prevention (Crookston, et al., 2011; UNICEF, 2013). Gender, mother’s education level, household wealth, place of residence, low weight at birth, diarrhea, lack of access to antenatal care and delivery services, and lack of access to piped water are among factors contributing to stunting in Peru. The results also indicate that ANP may have reduced stunting among the 25 to 59 months old children. The present thesis is organized in five chapters. In the subsequent chapter I review the literature on this topic. The first part of that chapter focuses on the literature that explore the different relationships between CU and its determinants considering the following categories: i) socioeconomic and demographic determinants, ii) child determinants - morbidity status, iii) environmental condition, and iv) child and maternal caring practices. The second part provides an overview of the literature that explores the multi-causal nature of CU and its determinants highlighting the existing hierarchical models. In the third chapter, I provide preliminary analysis o f the data, the conceptual framework that guides the construction of the variables, and an explanation about some of the interactions between them and the dependent variables as used for the modelling. Also, I explain the multilevel methodology and my empirical models. The estimation results are provided in chapter four. The final chapter of this thesis, chapter five, summarizes the main findings of my thesis and discusses the limitations o f the study, and proposes future investigations. 27 Chapter 2: Literature Review Chronic undemutrition (CU) defined as prolonged undernutrition after a child is exposed to different factors such as inadequate food intake and illness that leads to reduced growth, that is, lower than normal height for the same age. Chronic undemutrition is the outcome of multiple risk factors, causes and determinants. Their interplay is complex, ranging from biological and social to environmental and economic factors. Therefore, the analysis must be done carefully as variations in undemutrition are known to occur between countries of similar economic status, between regions of the same country (Wagstaff & Watanabe, 2000), and even between individuals in the same household (Bemt & Tadesse, 1997). The goal of 50% reduction of children undernourishment as a part o f the MDGs by 2015, and the goal of 40% reduction o f stunted children by 2025 as a part o f the implementation of Comprenhensive Implementation Plan (CIP) on Maternal, Infant and Child Nutrition by the WHO's Member States, have demanded in recent years not only the commiment of governments but also investigations that provide information to help to achieve the goals. CU, its causes and effects, and possible interventions have been extensively studied in developing countries. In this chapter, I first review the investigations related to the determinants o f CU in developing countries, Latin America and Pern by taking into account first four group of determinants: i) Socioeconomic and demographic determinants, ii) Child determinants - Morbidity Status, iii) Environmental condition, and iv) Child and maternal caring practices. Then, I review the literature that considered CU and the complex relationship between the determinants. 2.1 Socioeconomic and Demographic Determinants Van de Poel et al. (2008) uses component analysis to estimate socioeconomic status using a set of household assets and living conditions. Socieconomic inequeality was measured using the concentration index that avoids problems with dependence on the mean level of malnutrition. The study points out that “socieconomic inequality in malnutrition refers to the degree to which childhood malnutrition rates differ between more and less socially and economically advantaged groups” (p. 282). The study estimates both stunting and wasting rates in children aged up to five according to their economic status. The rates were measured using the WHO growth standars released in 2006. The investigators analyse DHS’s data from 47 countries from 4 regions: 26 in sub-Saharan Africa, seven in the eastern Mediterranean, five in south and south-east Asia, nine in Latin America and the Caribbean, including Peru. Their conclusions have shown that economic inequality in malnutrition is present througouth the 47 developing countries, and in almost all the countries stunting affected the poor, however there was no clear association between average stunting and socioeconomic inequality. Moreover, reducing the overall rate of malnutrition did not necessarily lead to a reduction o f inequality. Also, Barros et al. (2010) found that stunting and overweight were usually twice as prevalent for the poor than the rich in low and middle income countries. The goal o f the study was to describe the effects of social inequities on the health and nutrition o f children in low and middle income countries by revieweing existing data from nearly 100 countries (DHS data and UNICEF Multiple Indicators Cluster Surveys) on socioeconomic disparities within countries as related to the use o f services, nutritional status, morbilidity, and mortality. Particulary, in the case of stunting and underweigth, the analysis and estimations were based on concentration index for both prevalence of stunting and underweight. The main conclusion of this study is that the poor 29 children are more likely of being stunted. Barros et al notes that “these inequities in health outcomes result from the fact that poor children, relative to those from wealthy families, are more likely to be exposed to disease-causing agents; once they are exposed, they are more vulnerable due to lower resistance and low coverage with preventive interventions; once they acquire a disease that requires medical treatment, they are less likely to have access to services, the quality of these services is likely to be lower, and life-saving treatments are less readily available” (p. 1). Similar conclusions were obtained by Wagstaff & Watanabe (2000) in a study of 20 developing countries from 11 Living Standars Measurement Study (LSMS) surveys. The study concludes that the inequalities and socieconomics disadvantages in malnutrition always disfavor the poor, and it is not only that the poor have the highest rates in malnutrition, but also the rates declined with rising living standard.Particulary, the authors referred that Peru was one o f the countries with highly unequal distributions, and it had both, a high average level o f stunting (higher than Egypt for example) and higher poor-non poor inequality (Wagstaff & Watanabe, 2000, p. 15). Wamani et al. (2004) examine the association of four socioeconomic indicators namely: mothers' education, fathers’ education, household asset index, and land ownership with growth stunting among infants in the rural district o f Hoima, Uganda. The authors implemented a cross sectional survey using two-stage sampling design to obtain 720 child/mother pairs. The results indicated a higher prevalence of stunting among children o f non-educated mothers compared to mothers educated above primary school, children of non-educated fathers compared to those with fathers who were educated above secondary school, and children from households belonging to the “poorest” quintile of the asses index compared to those from the “least poor” quintile. However, adjusting all socioeconomic indicators in conditional regression analysis left 30 mothers’ education as the only independent predictor of stunting. Finally, more boys than girls were significantly stunted in poorer than wealthier socio-economic status (Wamani, Tylleskar, Nordrehaug, Tumwine, & Peterson, 2004, p. 5). In the same vein, Smith et al. (2003) concluded that the role of the women’s social status in determining their children’s nutritional health is important. The study included 36 developing countries. The authors used an Ordinary Least Square (OLS) regression to analyse the relationship between stunting and underweight, with mother’s status, mother’s education, income, access to sanitation services, among others things. The results o f the study indicate that the mother’s education positive contribute to the indicator of height for age (Z-score). Mothers with high school education have a twice positive effect on stunting than mothers with only primary education. Similary, Baldarrago (2010) investigates the relevance of the mother’s education on CU. He evaluated the effect of the mother’s education on child’s health using height for age z-score as a proxy o f the child health status. The study is based on DHS data o f the year 2008. The sample included 3 958 children under five years old and their mothers. It was found that the mother’s education has a positive impact on her child’s growth. Part of this effect is explained for their reading and writing skills, and for the access to information (radio, tv or newspaper). Shin (2007) studied the effect of the education of the mother on the children's on stunting (measured as a height for age z-score). Using the 2000 DHS of Peru and with a multilevel analysis the study found that in urban areas, maternal education is less important for child health than in poor rural areas, and a higher level o f education has a greater effect in rural areas. Multilevel analysis shows that a significant part of the observed correlation between maternal 31 education and child health is moderated by regional differences and community characteristics (p. 430). Urke et al. (2011) also found that stunting is associted with geographical areas, education of the mother and socieconomic factors. They investigated the association of parents socieconomic status with child stunting in the Peruvian Andes and in Peru nationally. The data were from DHS 2004 to 2006. A comparison of two samples o f children from 3 to 60 months old: a national sample (1426 childlren) and Andes sample (543 children) were selected. The socieconomic status was significantly related to stunting for both, the national and the Andes sample. The odds o f stunting in the poorest wealth index quintile were significantly higher than the richest quintile. The same pattern was observed in children o f mothers having incomplete primary education compared with children o f mothers having complete secondary or higher education. The associations o f wealth and maternal education with stunting were significantly stronger in the Andes compared with the national sample. 2.2 Child Determinants - Morbidity Status There are several investigations that examine the relationship between related child health condition and undemutrition. The health conditions are usually diarrhea, respiratory infections and low weigth at birth. Medhin et al. (2010) conclude that Low Birth Weight (LBW) is a factor associated with a 1.69 increased risk of CU (Table 4, p. 4). The study recruited 1 065 women in the third trimester of pregnancy from the demographic surveillance site in Butajira, south-central Ethiopia, and followed up until the infants were at one year of age. The study analyzed two indicators of undemutrition: underweight and stunting using the 2006 WHO child growth standard. By applying logistic and linear multiple regression models it was found that male gender, LBW, 32 poor maternal nutritional status, poor household sanitary facilities and living in a rural residence are significant and strong explanatory variables. Similar findings were found by Ngoc & Kam (2008). They estimated that LBW increases by 5.69 times the risk of CU in children under five in Nghean, Vietnam (table 5, p.236). In this study, 650 child/mother pairs were selected using a two-stage cluster sampler methodology. There were estimated underweight, wasting and stunting using the reference data from the National Center for Health Statistics (NCHS)AVHO. The study explored the hierachical relationships between potential risk factors and malnutrition by using a logistic regression analysis. Adair & Guilkey (1997) also find LBW as a strong predictor o f CU for children under one year old in Philipines. The study focuses on the relationship between age-specific factors and stunting in children from birth to 24 months of age. This community-based study was conducted from 1983 to 1995 with data from nearly 3 000 children in the Cebu Longitudinal Health and Nutrition Survey. Length, morbidity, feeding and health-related data were collected every two month during home visits. Stunting was estimated using the WHO reference growth curve released in 2006. They estimated the likelihood of becoming stunted by using a multivariate discrete time hazard model. In addition to LBW, diarrhea, febrile respiratory infections, and early supplemental feeding were found to be statistically significant determinants. Checkley et al. (2008) completed a systematic review. The review finds that high level of retarded growth and risk of CU are consequences o f the number and duration o f diarrhea episodes for children under 24 months old. The data covered a 20 years period (1978-1998) and five countries: Peru, Bangladesh, Guinea-Bissau, Brazil and Ghana. They found that five or more episodes o f diarrhea before the age of 24 months old explain 25% of CU observed in children at the age of 24 months (p. 821). 33 Similary, Lee at al. (2012) estimate the impact on stunting o f diarrhea episodes in children from the Peruvian Amazon (jungle). Diarrhea incidence predicted decreases in linear velocity over four and six months intervales of 0.029 and 0.028 cm/episode, respectively (p. 534). The study also analysed the adverse impact of vivax malaria. The ponderal and linear growth velocities over two, four and six month periods were examined using longitudinal models with data o f 442 children under 43 months. Concerning literature related to respiratory infections, Coles et al. (2012) investigates the effects of Stretococcuspneumoniae (Spn) infection on stunting and underweight in children at ages two, four and six months among 389 infants living in rural South India. The data was from “an Spn carriage study nested within a randomized, double-blind, placebo-controlled community trial designed to evaluate the impact of newborn vitamin A supplementation on Spn carriage in the first 6 months of life” (p. 1088). In this experiment children with Spn at two months old have 3.07 times greater risk of CU (Table 2, p. 1091) and the infection was associated with the decrease of 1.31 in height for age z-score (Table 3, p. 1091). For the case o f children at six months old the decrease was in 0.59 o f height for age z-score (Table 3, p. 1091), and for children at four months old there was no effect. 2.3 Environmental Conditions Smith & Haddad (2014) estimated the effects of underlying and basic determinant o f stunting in children under five years old for a 40 year period. They considered as a part of underlying determinants the access to safe drinking water and improved sanitation. The study is based on data for 116 out o f total o f 132 developing countries between 1970 and 2012. It uses fixed effect regression. In addition to the educacion of the mother, the access to safe drinking water and improved sanitation were strongly linked to stunting reduction. Access to safe water (25%) and 34 improved sanitation (14%) explained 39% of the variation in stunting rates across countries and time periods (Figure 3b, p. 24). Similar conclusions were found by Gunther et al. (2011), they evaluated the relation between household access to water and sanitation and child health measured as a mortality, stunting and diarrhea. This study used a cross sectional data from 171 DHS of 70 developing countries that includes information on 1.1 million children under the age o f five years over the period 19862007. By logistic modelling they estimated the effect of water and sanitation access on infant and child mortality, diarrhoea and stunting. Children with sanitation access have lower risk (0.87) of suferring diarrhea episodes and lower risk o f mild or severe stunting (0.74). Also, children with access to improved water was associated with lower risk o f diarrhoea (0.91) and lower risk of mild or severe stunting (0.92) (Table 3, p. 1202). 2.4 Child and Maternal Caring Practices The appropriate feeding practices positively affect the growth of children. Marriott et al. (2012) analysed the relationship between eight WHO food indicators and children health status (stunting and undemutrition). The eight WHO food indicators were: i) early initiation of breastfeeding, ii) exclusive breastfeeding under six months, iii) continued breastfeeding at one year, iv) introduction of solid, semi-solid foods or soft foods, v) minimum dietary diversity vi) minimum meal frequency, vii) minimum acceptable diet, and viii) consumption of iron-rich or iron fortified food (Table 1, p. 357). They used individual-level logistic regression models for different food indicators along with age, gender, mother’s education and urban/rural status. The study was based on 14 countries DHS databases from 2000-2005, there were no significant association between breastfeeding initiation or exclusive breastfeeding and stunting, but there was for underweight. For children between six to eight months old, the consumption of solid 35 foods, minimum acceptable diet, the consumption of iron-rich and dietary diversity were associated with significantly lower risk of both stunting and underweight. Also, the risks for stunting and underweight are even lower when it is also associated with maternal education (Table 6, p. 365) Similar conclusions were found by Lamberti et al. (2011). They conducted a systematic review of literature about breastfeeding as a risk factor for selected diarrhea morbidity and mortality outcome for children under 24 months old. They conducted a random effects meta analyses in order to generate pooled relative risk by diarrhea incidence, diarrhea prevalence, diarrhea mortality, all cause o f mortality and diarrhea hospitalization by group o f age. The results o f 18 studies show that children younger than 1 year old who did not receive exclusive breastfeeding compared to those with exclusive breastfeeding have 2.65 times the risk of diarrhea, and they are 19.48 times more likely to be hospitalized for diarrhea (Table 2, p.5). About the access to adequate health services and CU, Penny et al. (2005) found some effects on CU. They evaluated the effects on nutrition of child oriented national programmes such as immunization, monitoring growth and development, and management of acute respiratory infections and diarrhoea in a periurban area in Trujillo, Peru. They concluded that improvement o f nutritional education delivered through health services can decrease the prevalence of CU in children at age 18 months old in areas where access to food is not a limiting factor. Six pairs of health facilities were randomly selected. The types of health centers considered were: community hospitals offering maternal and perinatal specialist services; health centers with medical staff always in attendance; and health centers with more limited services. Two outcomes were considered: i) the nutritional status measured by weight, length, and Z scores for weight for age and height for age at age 18 months, ii) feeding practices, children receiving recommended 36 feeding practices and the 24-h dietary intake of energy, iron, and zinc from complementary food at ages 6, 9, 12, and 18 months. Results were compared by use o f ANOVA and non-parametric statistical tests, and random effects models were applied. The study included 187 infants from the health centers of intervention and 190 from control areas. Related to the feeding practices the adequate access to health services had a positive effect, more children at six months in the intervention areas were fed the recommend complementary nutrient at lunch than were controls. Fewer children in intervention areas failed to meet dietary requirements for energy, for iron than did controls (Table 5, p. 365). Children in control areas were more likely to fail to grow at 18 months than children in intervention groups, and had 3.04 times as much the risk of being stunted (Table 4, p.365). Moreover, the use of health services could be linked to the level of the mother’s education. Abuya et al. (2011) concluded that mother’s education has an impact on the use of health services and stunting. They conducted a study with 2003 DHS data in Kenya. They estimated odd ratios in order to show the effect of maternal education on children completed immunization and stunting. Children with mothers with only a primary education were 2.17 times more likely to be fully immunized compared to those whose mothers lacked any formal education. So far, I have reviewed the literature that focuses on a simple or several factors affecting CU and stunting, without taking into account the hierarchical nature o f the relationships between the determinants. The following section reviews the literature that has considered the hierarchical nature o f relationships among the determinants of CU and stunting. 37 2.S The Hierarchical Models of Chronic Undernutrition and Stunting 2.5.1 Studies on Developing Countries outside Latin America There are several investigations that analyze the multilevel causality o f the determinants of CU. Such studies recognize that the determinants are intertwined with each other and are hierarchically related. In order to address these complex relationships. The studies (such as those in the case of Peru) use conceptual frameworks to understand the determinants of nutrition outcomes. The main focus is to develop multilevel relationship between the determinants and CU. The conceptual frameworks set the determinants at different levels while recognizing the relationships between them. In this case, the authors are not concerned with the detailed characteristics of each determinant, unlike the studies in the previous section. However they focus more on the econometric and statistical methods that capture the multilevel relationships among the determinants and how they affect children’s CU or stunting. For example, Chopra (2002) takes into account the different pathways o f the determinants o f CU in South Africa. It considers a conceptual framework for the rural area o f South Africa where they consider socioeconomic, environmental and birth weight factors (Figure 1. p. 647). The socioeconomic variables measured were: migrant father, mother’s education, proportion of income, spent on food, material o f housing, among others. In the case of environmental factors the main focus was on access to water, sanitation and electricity. And for the birth weight factors the variables included current breastfeeding, age o f introduction of solids, weaned before 4 months, introduction of feeding and other feeding practices. The nutritional conditions studied were underweight (weight for age z-score) and stunting (height for age z-score). The effects were measured through a multivariate analysis by taking into account first control variables as age and gender. Next all the socioeconomic variables, then environment factors and finally the birth weight variables. The results confirm the multicausality of undemutrition and that the risk factors vary according to the type o f undemutrition. For stunting, low birth weight, distance form health clinic and breastfed last child were the most important risk factors. Nevertheless, mother literacy and duration o f breastfeeding were not significant (Table 3, p. 650). This paper discussed the differences between the risk factors according to the type of undemutrition in order to emphasize the different policies approach that could be considered and implemented. However, it did not included other factors related to child illness or sanitation. Ngoc & Kam (2008) constructed a conceptual framework o f CU for Vietnamese children under five years old based on previous studies. In this study, the model considers three groups of variables as a determinants: i) distal factors (socioeconomic variables) ii) intermediate factors (family variables), and iii) proximal factors (individual variables) (Figure 1, p. 264). They used 2007 data from cross-sectional survey in the province of Nghean, Vietnam. A multiple logistic regression analysis was used to capture the hierarchical effects between potential factors and malnutrition. Malnutrition was estimated in its three forms, underweight (weight for age), wasting (weight for height) and stunting (height for age) based on reference data from the National Center for Health Statistics (NCHS) / WHO. The results show that malnutrition, independent o f its type, was found to be a result of maternal, socio-economic and environmental factors. In the case of stunting, children in rural and mountainous areas were 1.9 times more likely to be stunted (Table 4, p. 235). Children from families that had > 3 children were found to 39 be 5.4 times more likely to be stunted than children from families with <3 children (Table 5, p. 235). The odds ratio o f being stunting was lower among female children than among male children. And children between 0-11 months old were found to have a lower odds ratio, and infants with low weight at birth were 5.6 times more likely than normal birth weight infants to be stunted. Children who did not receive exclusive breastfeeding were 3.7 times more likely than children who did (Table 6, p.236). This study contributes to the literature on malnutrition in rural areas. However, it must be noted that the three types of malnutrition considered in that study have different causes that may not be explained by the same set of factors (UNICEF, 1990). Kabubo-Mariara et al. (2008) investigated the impact of child, household and community characteristics on children’s height in Kenya. These characteristics were taken from the conceptual framework proposed by UNICEF (1990). The nutritional status of the children were measured as height for age z-score, and the results were presented in four different models, two models with controls for cluster-fixed effects for both dependent variables, and the other two without such controls. The determinants were grouped by child characteristics, household characteristics, health care variables and enviromental factors. The authors conclude that for all the regressions, there were no significant unobserved community-level characteristics affecting child nutritional status.. At least one variable from each group was significant. More specifically boys were more likely to suffer CU than girls, also children of multiple births had a higher risk of being stunted than singletones. Maternal education is more important than parental education and household assets and the use of public services had a significant association with CU (Table 3, p. 374-375). Kandala et al. (2011) analyze CU in children under five in Democratic Republic of Congo (DRC). The study takes into account the multicausal nature of CU by controlling the regressions 40 for distal, proximate and intermediate factors following the framework proposed by UNICEF (1990). It is important to note, that both linear and nonlinear effects between the determinants and nutritional status were incorporated. They evaluated the significance o f the determinants on z-score height for age estimating fixed, non-linear effects and spatial effects. CU was high in all the provinces of the DRC. However, it was significantly higher in rural areas compared to urban centers. Moreover, age, access to delivery services, education o f the mother and assets index were found to affect CU status (Tables 1 and 2, p. 7-8). Mazumdar (2012) also addressed the multicausality of CU through an adaptation of the conceptual frameworks proposed by FAO/FIVIMS and UNICEF (1990) that consider determinants at child, households and community levels (Figure 2, p.3). The author analyzed CU in children under five years old in Egypt. The data used is from 2008 DHS and child nutrition status was determined by height for age z-score for each by using the WHO reference population data in 2006 (de Onis, 2006). The total o f children considered in the sample was 8 505 from 6 487 households distributed in 1 214 clusters. The research used a multilevel analysis (linear and logistic) in order to capture fixed and random effects at the three levels (child, household and community). Results indicate considerable community effects influencing a household's nutritional choices. However, education of the mother or women's decision-making power could not be linked to CU, but there was a positive association between a better nutritional status and the health service utilization as well as child care, economic status and living standards and feeding practices. The authors conclude that such results, support the hierarchical and multicausal nature of CU, addressing the effects at individual, household and community level (Table 2, p.8). 41 Fenske et al. (2013) studied the determinants of CU in India for children under 24 months old by using the conceptual model elaborated by UNICEF in 1990, identifying possible causes o f child stunting structured by the layers (immediate, intermediate and underlying determinants). They used cross-sectional data from the Indian National Family Health Survey for 2005/2006. The methods of analysis included an additive quantile regression for four quantiles in order to capture the linear, non-linear, spatial and age-varying effects. Also, they modelled a logistic regression for stunting and severe stunting. The nutritional status of the children was analyzed by z-score for height for age and stunted (z-score < -2), the explanatory variables were created considering the following groups: child demographics, maternal characteristics, household characteristics, regional characteristics, environmental characteristics (such as water, sanitation and hygiene, indoor air pollution, along with curative and preventive health care, breastfeeding factors, complementary feeding practices, and micronutrient deficiency) (Table 1, p. 4-6). Their results confirmed the multilevel nature of child stunting. For the 35% Z-score quantile regression at least one variable within the groups o f determinants was significantly associated with the dependent variable (z-score height for age) (Table 2, p. 7-8). The variables that showed the largest effects were: child age and sex, household wealth, maternal education and Body Mass Index (BMI). They found a non-linear relationship between child stunting and the variables: maternal age, maternal BMI, birth order and number of antenatal visits (Table 2, p. 7-8). One of the strengths of this research is that through the statistical modelling they could capture linear and nonlinear effects. 2.5.2 Studies in Latin American Countries (excluding Peru) In the case of Latin American countries, there several studies that approach CU by taking into account the multicausal characteristic of the problem. For Guatemala, Gragnolati (1999) 42 investigates the determinants o f poor child growth in rural areas (measured as height for age zscore). This study is based on the data o f the Guatemalan Survey of Family Health conducted in 1995. The estimates were guided by the economic model o f the family and the proximate determinants framework. The study applied multilevel models to hierarchically clustered data to control for family and community heterogeneity. Therefore, the complexity o f the relationships between the determinants were taking into account by estimating fixed and random effects by individual, household and community covariates in shaping differentials in children’s height. The education o f the parents is found to have a positive effects on child malnutrition status. Children had a better height where community infrastructure (such as piped water and garbage disposal) and health care facilities were better (Table 5, p.39-40). In Ecuador, the World Bank (2007) used an adaptation of the conceptual framework proposed by Tufts's University (Figure 13, p. 22). In this case, a multivariate analysis was applied with data from the 2004 Demographic and Maternal and Child Health Survey (ENDEMAIN). CU status was determined by z-score height for age and the determinants were specified at individual, household and community levels. The sample only included children under five years old. However, not all the variables considered in the conceptual framework could be included. The most important at child level were: inadequate food intake, low weight at birth and child health status (diarrhea disease, inadequate immunizations and respiratory infections). It is pointed out that the exclusion of some o f the variables was needed to avoid problems of endogeneity. Although the study recognizes that hierarchical factors, it does not take into account the hierarchical form o f the cross sectional data during the estimations. However, some effects were captured such as the height of the mother with a positive impact. Altitude, rural status, and household resources and composition had negative relationships with CU, whereas 43 adequate access to sanitary conditions had a positive impact on the nutritional status of the children (Table 7, p. 24-25). Monteiro et al. (2010) analyzed stunting in Brazilian children under five years old. The main goal was to evaluate the effects of income and basic service redistribution policies implemented by Brazilian Government through the trends in the prevalence and social distribution of child stunting over a period of 33 years (1974-2007). Although, different pathways o f causes o f CU were recognized by identifying underlying determinants, intermediate determinants and proximate determinants, only the underlying and intermediate determinants such as household assets, maternal education, maternal antenatal health care, water supply and sanitation services, and reproductive health indicators were included. The effects of the variables at each level were captured for the period 1996-2007 in order to explain part of the drop of CU from 37.1% to 7.1%. Prevalence of stunting for quintile was the outcome to be evaluated10. The data was collected from four national surveys. The study used the Slope Index o f Inequality (SII) to quantify absolute socioeconomic disparities in child stunting. The SII, which is based on a weighted linear regression o f the observed prevalence of stunting in the quintiles, expresses the absolute difference in outcome between the lowest and the highest quintile. Also, the changes in the rest of determinants were tested for statistical significance by fitting an interaction term between the year o f the survey and the socioeconomic quintiles. The main finding is that in the last 10 years the decline in the rate o f stunting was particularly steep. Also the gaps between poor and wealthy families with children under five were also reduced in terms o f purchasing power; 10 The W HO growth reference curve released in 2006 was used to estimate the prevalence. 44 access to education, health care and water and sanitation services; and reproductive health indicators (p. 305). 2.5.3 Studies on Peru A few studies have analyzed CU and its determinants in Peru. Lescano (2002) use the Peruvian 1996 DHS data to explain CU trends in Peruvian children between two and four years old. This study uses variables from four categories: i) access to quality dietary, ii) access to water and sanitation, iii) access to health services, and iv) education and practices o f hygiene, care and feeding (p.23) to predict the rates of prevalence o f CU in seven areas of Peru: Lima Metropolitana, urban coast, urban mountain, urban jungle, rural coast, rural mountain and rural jungle. The dependent variable was a dummy (stunted or not sunted) measured according to the growth reference data of NCHS/WHO proposed in 1977. Using logistic regression, the study finds that at least one variable from each category is found significant (Table 10, p.40). Similary, Marini & Gragnolaty (2006) analysed the nonlinear effects of altitude on child growth in Peru. This relationship was explored within a multivariate framework. They also recognize the role of the factors at community, household and individual level. They pointed out that due to the cluster nature o f the data they applied a multilevel model in order to control for correlation of variables in the same cluster. The data used was from DHS collected in 2000, once again z-score height for age was the dependent variable. After controlling for characteristics of the children, families and communities, the data shows a significant nonlinear relationship between altitude and stunting, Peruvian children living at altitudes between 2500 and 3500 meters appeared to be worse off than children living at altitudes above 3500 metres. Other variables that were found to be linked to CU were education of the mother at secondary level, age of the children, and access to sanitation (Table 3, p .16-18). The multilevel analysis provided a better fit o f the coefficients 45 and allowed to captur both random and fixed effects. However, important variables at child and community level such as low birth weight, and access to health services such as growth control or vaccination were not included. Ravina & Chavez (2009) analyzed the determinants of CU in Peruvian children under five years old by using the conceptual model of UNICEF developed in 1990 and updated in 1998 (UNICEF, 1998). The determinants were identified by immediate causes (at individual level), underlying causes (at household level) and basic causes (at society level) that were intertwined (p.21). The data was collected from pooled DHS for the years 1992, 1996, 2000 and 2004-2005. The dependant variable was the z-score height for age according to the growth reference database o f WHO. The explanatory variables were: breastfeeding, diarrhea episodes, respiratory infections, birth order, sex o f the child, age, education of the mother, fixed assets of the household, access to sanitation and water services, among others. The study included three linear regressions: OLS, fixed effects at cluster level and random effects at individual level, suggesting that the fixed effect model is the most appropiate. The variables that were found associated with CU were breastfeeding, diahrrea, age o f the children and access to water and sanitation. The study concludes that endogeneity o f the variables was a limitation. Arocena (2010) estimated effects o f location o f the Regional Health Directions and child and household variables on stuning for children under five years old. It is implemented a logistic multilevel regression using 2008 DHS data. The location of Regional Health Directions, the education of the mother, mother’s height, economic status of the household and the type of floor of the house were associated with the stunted status of Peruvian children (p. 50). However, the study does not indicate how many levels there were considered. 46 Gutierrez et al. (2014) estimated statistical probabilities for being stunted (height for age < 2 SD below the mean) for Peruvian children under five years of age using 2011 DHS data. They explored the association o f factors such as sex, age, geographic location, education o f the mother, wealth index, natural region, access to water, access to sanitation, public health insurance, altitude, child illness, among others. The calculation of the adjusted odds ratios showed significant differenes between CU in mother’s education, Andes region, altitudes greater than 2500 m, presence of two or more children in household, and being the third or succesive child (p. 108). However, hierachical relationships are not explained in this study. The reviewed literature suggest that it is important to consider the multicausal relationship between CU or stunting and its determinants. The determinants are hierarchically related. The conceptual frameworks cited in the investigations considered at least three layers of group of determinants. My thesis also takes into account the complex relationship between stunting and its determinants. However, I consider additional determinants related to the Peruvian context, such as APN and its possible effect on stunting. Some o f the researchers cited above mention that due to the hierachical nature o f the data, it is necessary to use multilevel analysis that captures the effects of the determinants in the different levels (child, household and community level). In the case of Peru, Marini & Gragnolaty (2006) and Arocena (2010) used that method to analyse stunting in Peruvian children under five years old. However, the former study is on Peruvian children bom before 2000. It was before the political reforms in Peru and the creation o f the Peruvian regions. In the second study, the analysis is based on Peruvian children bom from 2003 to 2008, when the rates o f stunting were stagnent. My thesis considers children bom from 2006 to 2012, when the national rate of 47 stunting presents a downward trend at least from 2008 to 2012 that may allow to capture different effects from previous studies in Peru. 48 Chapter 3: Data and Methods 3.1 Data The database used for the analysis in this thesis is taken from the Peruvian Demographic and Health Survey (DHS) n . The data has been pooled from the years 2009, 2010, 2011 and 2012. The data presents a hierarchical structure and has comparable information on community and household characteristics as well as on nutrition and health o f women aged 15-49 and their children under five years old at the time o f the survey. The samples cover all Peruvian regions, and both urban and rural areas. All women registered as living in the households were interviewed. The database contains basic information of 38, 766 Peruvian children under five years old (from 0 to 59 months), in 30, 446 households and in 2, 259 clusters or Primary Sampling Unit (PSU). The number of children from 6 to 24 months old is 12, 149 in 11, 616 households and 2, 186 clusters (PSU), and the number of children from 25 to 59 months is 23, 277 in 20, 860 households and 2, 245 clusters (PSU). The total number of children between 6-59 months old is 35 326. However, there is a high proportion o f missing data (up to 63% in some cases such as diversity of food). Therefore, the number o f observations vary between models depending on the variables included in the model. The information is about children (provided by the mother), their household and their area of residence. Due to the nature of the data, a hierarchical model similar to other studies such as in " The databases are available online at http:/ dhsprogram.com/W here-W e-W ork/CountrvMain.cfm'/ctrv id 33&c Peru&C ountr\ Peru&en &r 6 or at httn: ;/iinei.inei.eob.pe'm icrodaios; 49 Marini & Gragnolaty (2006) and Mazumbar (2012) is used. Child and mother are considered to be nested within the same household, which in turn is nested into the community (PSU) variable. The DHS data is cross-sectional which can cause some difficulties during the inteipretation o f the results, as cross-sectional studies may not provide definitive information about cause-andeffect relationships (Kuhn et al, 1997). DHS is the main source to estimate health indicators around the world and in Peru the database is used to estimate the National health indicators and it is managed by the National Institute o f Statistics (INEI, 2014). The key aim of DHS “is to provide quality data for policy development and programme planning, monitoring and evaluation. To build the policy and programmatic evidence base, DHS data must first be transformed into information, which must then be made accessible to decision­ makers” (Short et al., 2012, p. 604). 3.1.1 Summary of statistics The summary of data, statistics, and the variables in the sample used in the present study are given in the following tables: Table 3.1- Summary of statistics Variables Age in Number o f months Observations Mean Dummy 0 Height for age z- 0-5 3 340 -.8452246 6-24 12 149 -1.243199 25-59 23 277 -1.292984 1 score< continuous variable) 50 Variables Age in Number of months Observations Mean Dummy 0 Stunted (Dummy 1 Total 38 766 -1.238804 0-5 3 221 .1477802 2 745 476 6-24 11 861 .2468595 8933 2 928 25-59 22 765 .2513508 17 043 5 722 Total 37 847 .2411288 9 126 28 721 0-5 3 340 2.785629 6-24 12 149 15.0214 25-59 23 277 42.17378 Total 38 766 30.2708 0-5 3 340 0.4868263 1 714 1626 6-24 12 149 0.4934563 6 154 5 995 25-59 23 277 0.4955106 11743 11 534 Total 38 766 0.4941186 19611 19 155 0-5 3 340 0 3 340 0 6-24 12 149 0 12 149 0 25-59 23 277 0.1697384 19 326 3 951 Total 38 766 0.1019192 34815 3 951 0-5 3 340 0 3 340 0 6-24 12 149 0 12 149 0 variable) Age (months) (hcl), continuous variable Gender, male (0), female (1) Birth’s Cohort -Cohorts 2004 and 2005; “coh200405" (1), otherwise (0) 51 Variables Age in Number of months Observations Mean Dummy 0 -Cohort 2006; 1 25-59 23 277 0.2039782 18 529 4 748 Total 38 766 0.1224785 34 018 4 748 0-5 3 340 0 3 340 0 6-24 12 149 .0753972 11233 916 25-59 23 277 .2287236 17 953 5 324 Total 38 766 0.1609658 32 526 6240 0-5 3 340 0.0344311 3 225 115 6-24 12 149 0.2406782 9 225 2 924 25-59 23 277 0.2213773 18 124 5 153 Total 38 766 0.2113192 30 574 8 192 0-5 3 340 0.2649701 2 455 885 6-24 12 149 0.2471808 9 146 3 003 25-59 23 277 0.1312884 20 221 3 056 Total 38 766 0.179126 31 822 6944 0-5 3 340 0.2329341 2 562 778 6-24 12 149 0.2436415 9 189 2 960 25-59 23 277 0.0448941 22 232 1 045 Total 38 766 0.1233813 33 983 4 783 0-5 3 340 0.2389222 2 542 798 6-24 12 149 0.1672566 10 117 2 032 “coh2006” (1); otherwise (0) -Cohort 2007; “coh2007" (1), otherwise (0) -Cohort 2008; "coh2008’ (I), otherwise (0) -Cohort 2009 ; “coh2009” (1), otherwise (0) -Cohort 2010; "coh2010" (1), otherwise (0) 52 Variables Age in Number o f months Observations Mean Dummy 0 -Cohort 2011; 1 25-59 23 277 0 23 277 0 Total 38 766 0.0730021 35 936 2 830 0-5 3 340 0.2287425 2 576 764 6-24 12 149 0.0258457 11 835 314 25-59 23 277 0 23 277 0 Total 38 766 0.0278079 37 688 1 078 0-5 519 0.9749518 13 506 6-24 8085 0.2512059 6054 2031 25-59 4 321 0.1603795 3 628 693 Total 12 925 0.2499033 9 695 3 230 0-5 2 921 0.0568299 2 755 166 6-24 10 624 0.0701242 9 879 745 25-59 18 969 0.0779166 17 491 1 478 Total 32 514 0.073476 30 125 2 389 0-5 3 189 0.1059893 2 851 338 6-24 11 582 0.2311345 8905 2 677 25-59 13 720 0.1197522 12 077 1 643 Total 28 491 0.1634902 23 833 4 658 0-5 3 189 0.1326435 2 851 338 6-24 11 582 0.1955621 8905 2 677 “coh2011" (1), otherwise (0) -Cohort 2012; “co h 2 0 1 2 "(l); otherwise (0) Access to diversity o f food; Access to less than 4 types o f food (1); otherwise (0) Low Birth Weight; birth weight <2.5 kg (1); otherwise (0) Diarrheal disease; if child had diarrhea in the last 2 weeks (1); otherwise (0) Acute respiratory infections;; if child had 53 Variables Age in Number o f months Observations Mean Dummy 0 cough in the last 2 weeks 1 25-59 21 319 0.1456447 12 077 1 643 Total 36 090 0.1605154 23 833 4 658 0-5 3 330 0.6543544 1 151 2 179 6-24 11 949 0.6331074 4 384 7 565 25-59 22 049 0.5861037 9 126 12 923 Total 37 328 0.6072385 14 661 22 667 0-5 3 174 0.1792691 2 605 569 6-24 11 363 0.1880665 9 226 2 137 25-59 17 120 0.2098131 13 528 3 592 Total 3 1 657 0.1989449 25 359 6 0-5 3 165 0.7740916 715 2 450 6-24 11 283 0.8003191 2 253 9 030 25-59 15614 0.8229794 2 764 12 850 Total 30 062 0.8093274 5 732 24 330 0-5 3 340 0.3098802 2 305 1 035 6-24 12 149 0.3068565 8 421 3 728 25-59 23 277 0.3182541 15 869 7 408 Total 38 766 0.3139607 26 595 12 171 0-5 3 340 0.2790419 2 408 932 ( 1 ); otherwise (0 ) M other's education, if the mother completed secondary school or higher level o f education ( 1 ); otherwise (0 ) Access to delivery services; i f the mother has not received delivery services ( 1 ), otherwise (0 ) Access to antenatal care; 298 if the mother has received at least 6 control during pregnancy ( 1 ); otherwise (0 ) Wealth Index Score -q uintilnl, household in the poorest quintile ( 1 ); otherwise ( 0 ) Variables Age in Number o f months Observations Mean Dummy 0 -quintiln 2 , household in 1 6-24 12 149 0.2643839 8 25-59 23 277 0.264553 17 119 6 Total 38 766 0.2657483 28 464 10 302 0-5 3 340 0.2125749 2 630 710 6-24 12 149 0.2158202 9 527 2 25-59 23 277 0.2066417 18 467 4 810 Total 38 766 0.2100294 30 624 8 0-5 3 340 0.1254491 2 921 419 6-24 12 149 0.138283 10 469 1 680 25-59 23 277 0.1351978 20 130 3 147 Total 38 766 0.1353248 33 520 5 246 0-5 3 340 0.0730539 3 0% 244 6-24 12 149 0.0746564 11 242 907 25-59 23 277 0.0753534 21 523 1 754 Total 38 766 0.0749368 35 861 2 905 0-5 3 186 0.1848713 2 384 800 6-24 11 565 0.0999568 5 805 5 741 25-59 21 230 0.0581253 16410 4 742 Total 35 981 0.0827937 24 599 11 283 0-5 3 189 0.9708373 93 2 276 937 3212 the poorer quintile ( 1 ); 158 otherwise (0 ) -quintiln3, household in the middle quintile ( 1 ); 622 otherwise ( 0 ) -quintiln4, household in 142 the richer quintile ( 1 ); otherwise (0 ) -quintilnS, household in the richest quintile ( 1 ); otherwise (0 ) Access to growth control, child has received growth control services according to the norm ( 1 ); otherwise (0 ) 55 Variables Age in Number o f months Observations Mean Dummy 0 Access to immunization 1 6-24 11 582 0,6040408 4 586 6 25-59 6 683 0.3405656 4 407 2 276 Total 21 454 0.5764892 9 086 12 368 0-5 3 340 0.2763473 2417 923 6-24 12 149 0.2693226 8 887 3 272 25-59 23 277 0.2729304 16 924 6 Total 38 766 0.2720941 28 218 10 548 0-5 3 340 0.3281437 2 244 1 0 % 6-24 12 149 0.3243065 8 209 3 940 25-59 23 277 0.320531 15 186 7 461 Total 38 766 0.3223701 26 269 12 497 0-5 3 340 0.5835329 1 391 1 949 6-24 12 149 0.5880319 5 005 7 144 25-59 23 277 0.5781243 9 820 13 457 Total 38 766 0.5816953 16216 22 550 0-5 3 340 0.4508982 1 834 1 506 6-24 12 149 0.4583924 6 785 5 364 25-59 23 277 0.4583924 12 607 10 670 Total 38 766 0.4524583 21 226 17 540 9% services child has received vaccines according to the norm ( 1 ); otherwise (0 ) Access to treated water, household has not access to treated water ( 1 ), otherwise (0 ) Access to water, 353 household has not access to piped water ( 1 ), otherwise (0 ) Access to improve sanitation; household has not access to improve sanitation ( 1 ), otherwise (0 ) Area o f residence; if the cluster is in rural area ( 1 ), urban area (0 ) 56 Variables Age in Number o f months Observations Mean Dummy 0 1 Place o f Residence -Lima Metropolitana; if 0-5 3 340 0.0625749 3 131 209 6-24 12 149 0.0704585 11 293 856 25-59 23 277 0.0663316 21 733 1 544 Total 38 766 0.0673012 36 157 2 609 0-5 3 340 0.2416168 2 533 807 6-24 12 149 0.240349 9 229 2 920 25-59 23 277 0.2366714 17 768 5 509 Total 38 766 0.23825 29 530 9 236 0-5 3 340 0.395509 2 019 1 321 6-24 12 149 0.392131 7 385 4 764 25-59 23 277 0.4157323 13 600 9 677 Total 38 766 0,4065934 23 004 15 762 0-5 3 340 0.3002994 2 337 1 003 6-24 12 149 0.2970615 8 540 3 609 25-59 23 277 0.2812648 16 730 6 Total 38 766 0.2878553 27 607 11 159 the cluster is in Lima Metropolitana ( 1 ); otherwise ( 0 ) -Rest Coast Area; if the cluster is in rest o f the coast area ( 1 ); otherwise (0 ) -Mountain or Andes Area; if the cluster is in the Mountain o f Andes area ( 1 ); otherwise (0 ) -Jungle Area; if the cluster is in the Jungle area ( 1 ); otherwise ( 0 547 3.2 Conceptual Frameworks In order to guide the analysis o f the relationships between the variables, this thesis uses two conceptual models. First, is the conceptual framework designed by UNICEF (1990) as part o f its 57 Nutrition Strategy. It was updated in UNICEF (1998) and in Black, et al (2008). This framework is shown in Figure 3.1. It captures the multifactorial causality of undemutrition. The strength of this framework is that it recognizes and integrates the biomedical as well as the underlying socieconomic determinants o f undemutrition. The framework considers three layers o f causality: immediate, underlying and basic. The inmediate causes are at the individual level, the underlying causes influence households and communities, and the basic causes are at the structural level of the society, including the environmental, economic, and socio-political contextual factors, with poverty having a central role at this level. Figure 3.1- UNICEF Conceptual Framework of Undemutrition .M a te rn a l an c chikL (Source: Black, et al. Maternal and child undemutrition: global and regional exposures and healthconsequences. 2 0 0 8 -F ig u r e l , p . 244) However, in order to model the interactions of the variables, it is required specific information. For example, one of the immediate causes is diseases. What kind of disease must be taken into consideration for the Peruvian context? According to Moradi (2009) “the list can include diarrhoea, pneumonia and other respiratory illnesses, diphtheria, measles, malaria, yellow fever, 58 sleeping scikness, dengue, plague, smallpox, schistosomiasis and other parasites like hookworms but their relevance depends on the epidemiological environment” (p. 16) The adaptation of the UNICEF framework for stunting developed by Tufts University (2001) is useful because its focus is on stunting, and also specifies the general causes proposed in the UNICEF framework, such as diarrhea and acute respiratory infections instead of diseases. This adaptation takes into acount the particular characteristis o f Peruvian context. Also it considers most of the factors and the causal relationships that the Peruvian government assumed during the desing and implementation of the Articulated Nutrition Program (ANP) with the reduction o f the prevalence of stunting in children under five years old as the final outcome. ANP is part of the Budget for Results policy (BfR) that has been implemented in Peru since 2007 (Cruzado, 2012; MIDIS, 2012; MEF, 2011). The determinants at the immediate and intermediate level have been laid out, isolating factors operating at the community, household and child level. This framework gives a schematic overview o f the causes of CU as shown in Figure 3.2. At the individual or child level there are the immediate causes: poor health status, inadequate food consumption and low birth weight, all of which, have possible explanations or causes. For example, in poor health three possibilities are pointed out i) diarrheal disease, ii) acute respiratory infections and iii) inadequate immunizations; Low birth weight is considered to be caused by the poor maternal and nutrition health. And inadequate food intake include i) poor quality, ii) inadequate quantity, iii) short exclusive breastfeeding duration and iv) inadequate complementary feeding. The above factors are linked to factors that are part of the underlying causes at the household level. For example, households with low incomes lead to low food intake in children. Large family sizes and short birth spacing lead to low birth weight, as do poor feeding practices during 59 pregnancy. Poor feeding practices can also lead to inadequate food intake in children, even in households that do not face economic constraints. Inadequate water and sanitation measures lead to increase in diseases. Some of the factors identified at this level of the model are linked to the behaviour of individuals, mainly mothers and families. Feeding practices, health care and hygiene practised at home are all critical determinants of the adequate growth of the children. Similarly, the household level causes are affected by the factors at the community level: the local economic and social infrastructure, the educational system, the existing health services and the water and sanitation system all provide critical effects on the determinants at the household level. Figure 3.2- Conceptual Framework o f Stunting in Peru STUNTING Child Low birth weight Inadequate food Intake ■Poorqutfty ■ Poor maternal nuMton andhaetei ■ Inadequate quantity ■ Short «xdmhe_ taMN&Mdng dmallon a madaqum comptemenlaiyterthg Household jpwfcqlate looci Inadequate practices koSehoW resources Community cMdlMdng ■MteK* .Economic social !infrastructure % ■ ttoadasoaat BlnocrteBonaraton opporturaM Inadequate healthcare Inadequate aandcaa ■QOMtl ■Chan water te&MMon e Mart an) ■Brtiepadag t ■ Acute m pM uy infecUona a Inadquate immunization* r pfwUOVfr f Poor health status ■DtentimlMMM IflOMOrtRQ aSaikeanln Educational services encewaaaachoeN aOuMtyeducation aiMetyaftnda iPsmmv X Health services X Water and sanitation services ■ QuaRy nsatei cm • KMMaiailee Source: (R eduction de la Desnutricion Cronica en el Peru: Propuesta para una Estrategia N acional.T uftsU niversity, 2001, p. 13) Both conceptual frameworks provide a better understanding about the different pathways related to undemutrition and stunting. However, they do not assign relative weights to the different 60 pathways, which makes it difficult for decision makers to choose the most cost-effective interventions. To have more information, it is necessary to apply them in a specific context, with the best available methodology and data. It must be noticed that the hierarchical form of the framework and the variables are somewhat related between levels. Therefore their association to stunting is not necessarily directly causal. For example, optimum water access may not be statistically significant in a regression model for stunting but it can be a key variable for diarrhea, which is linked to stunting. In this sense, if a model points out that access to water is not significant for stunting, it does not mean that interventions to improve water resources should not be considered as part o f a strategy to reduce CU .Thus, taking into account these links will provide adequate conclusions and appropriate recommendations. 3.3 Methods 3.3.1 The Empirical Models 3.3.1.1 Dependent Variable: Chronic Undernutrition Status Undemutrition status in children is commonly determined by using anthropometric information12. In the case of CU status the basic information and measurement used is the heightfor-age (WFP, 2005; de Onis, 2008; Mora, 1989; de Onis et al, 2006). That involves the use o f growth standards and/or growth references for assessing children growth status and well-being (Wang et al, 2006). A growth standard reflects optimal growth, suggesting that all children have potential to achieve that level, while a growth reference is simply the distribution used for comparison. The anthropometric indices can be expressed in relationship to the reference 12 The basic information and measurements and measurement that constitute anthropometric m easurements in children are: age, sex, length, height, w eight, and oedema. “The measurements are the key building blocks o f anthropometries and are essential for measuring and classifying nutritional status in children under 5 years" (W FP, 2005, p. 13) 61 population by three different statistical terms: i) standard deviations from the median (Z-score), ii) percentage o f the median or, iii) percentiles (WHO, 2006; Wang & Chen, 2012; de Onis, 2008). For population-based assessment, including surveys and nutritional surveillance, the preferred and most common way o f expressing anthropometric indices is in the form o f Z-scores. (WFP, 2005; de Onis M, 2008; WHO, 2006). According to WHO (2006) and de Onis (2008) the main advantage o f z-scores is that they can be described in terms of means and standard deviations of population or subpopulations13. The z-score is measured by the standard deviation o f the height of the child relative to the median size o f healthy children, as determined by the reference population. The International Growth Reference Curves are normalized, that means that in the case of height-for-age: at each age, there is a normal distribution of heights. This is really helpful from a statistic point o f view because that implies that the observed Z-score for age from a population are likely to be normally distributed so it can be applied regressions and t-tests where the normality is assumed (Gorstein, et al., 1994, p. 276). The formula for calculating the Z-score is: height fo r age z —score measured value —median o f reference population = — ---— —— :— :------- — ------------------------------Standard deviation o f the reference population Also, the use of z-score for ageallows to easily determine the proportion o f children from a group o f population who fail or achieve the reference growth by falling below or above a cutpoint, so the prevalence of stunting can be calculated (Gorstein, et al., 1994). In general, this cut­ 13 A ccording to the same authors this cannot be m eaningfully done with percentiles. M oreover, at the extreme o f the distribution, large changes in height or weight are not necessarily reflected in changes in percentile values. The percent o f median is deficient relative to the z-score in that it expresses deviation from the reference median without standardizing for the variability in the reference population. 62 off is -2 standard deviations (SD) from the median. This criterion is supported by the statistical property where 95% o f observations fall within 2 standard deviations from the mean (or median). According to the WHO (2006), the values o f -3.0 to less than -2.0 indicate moderate stunting, but when this value reaches -3 SD, it is considered to be severe stunting (WHO, 2006; Wang & Chen, 2012). It is important to explain about all the types o f growth curves o f the reference population, because the value o f the z-score depends on the adoption o f a particular set o f growth curve. In 2006, the WHO published the height o f reference children obtained as a result of a longitudinal study from six countries representing different regions of the world: Brazil, Ghana, India, Norway, Oman and the United States (Onyango, et al., 2007). This set of growth references is recommended by WHO since year 2006 (WHO Multicentre Group Reference Study, 2006). The recommendation about the use of WHO/2006 for international reference started because some concerns were raised that the use of National Center of Health Statistics (NCHS)-WHO/1977 growth reference's curves “may lead to the early introduction of complementary foods in exclusively breast-fed infants, which often has adverse consequences for the health and nutritional well-being o f infants” (de Onis, 2008, p. 119). Its adoption was evaluated for different countries since then, for example the United States (Grummer-Strawn et al. ,2010) and Australia (Australian Health Ministers Conference, 2009) decided to take them into account in 2010 for children under two years old after a period o f three years of evaluation. For Peru, the adoption was in 2011 and it was for children from 0 to 59 months old, making it official through the Ministerial Resolution of the Health Ministry (MINSA, 2011). Prior to the evaluation, most countries used the growth charts developed by the 63 US NCHS in 1977: the NCHS-WHO/1977, or the CDC/2000 released by the US Center for Disease Control (CDC) in 2000. I use the standardized height for age Z-score as the dependent variable for Peruvian children under five years o f age according to the WHO/2006 growth curve reference. The Figure 3.3 shows a histogram and kernel density of the distribution o f the Z-scores o f Peruvian children under five years old in my database. Figure 3.3- Histogram, kernel density o f height for age z-score -6 -4 -2 0 2 4 Source: Peruvian DHS data 2 0 0 9-2012 3.3.1.2 Independent Variables The selection of the explanatory or predictor variables is guided by the conceptual framework outlined in section 3.2 regarding determinants at individual, household and community level. Unfortunately, not all of the determinants are available in the survey, especially those at the “community” level14. The measurement of each variable is obtained from the DHS database and 14 It is important to recognize that given the data source used for this analysis there are limitations that are worth mentioning. The D H S data set uses a standard questionnaire approach with a predetermined set o f variables. The 64 questionnaires using the definitions for the indicators used by the National Institute of Statistics of Peru and based on UNICEF and WHO definitions (INEI, 2014; UNICEF, 2013; WHO, 2014). Also, I follow the guide elaborated by Rutstein & Rojas (2006) and provided by the Measure DHS/ICF International team of the DHS in order to create the explanatory variables for use in the Statistical Packages15. I am interested in the analysis of associations between CU and the explanatory variables in the context o f the Budget for Results (BfR) approach that has been implemented since 2008 by the Ministry o f Economy. BfR seeks to organize budgetary management around those outcomes that citizens require and value (MEF, 2011). Five Strategic Programs were initially defined as part of BfR, which responded to 11 priority actions for early childhood established by the Government of Peru as related to the Millennium Development Goals. These actions were translated into five outcomes to be achieved through implementation o f RBB. The five outcomes were: (i) reducing chronic malnutrition in children younger than five, (ii) reducing neonatal and maternal mortality, (iii) improving learning attainments in children between six and seven years old children; (iv) improving access to identity (birth’s registration); and, (v) improving access to basic social services and market opportunities. Behind the formulation of the first strategic programs there was a clear focus on reducing poverty and social exclusion, as well as on improving the quality of life of Peruvians (Cruzado, 2012, p. 2). One of the program was the Articulated Nutrition Program (ANP) designed by the BfR methodology, linking results with outputs within an evidence-based logical model. The ultimate focus is primarily on gathering demographic, health, and nutrition information. The surveys also collect basic data on a number o f other aspects such as maternal em ploym ent, use o f alternative childcare arrangements, housing quality, and ownership o f assets. H ow ever, information on these constructs is limited in scope. 151 use tw o programs SPSS 21 and Stata 12 in order to create the variables. 65 goal of the ANP is to reduce the prevalence of CU in children under five years old. To achieve this objective, the program focuses on a set o f intermediate outcomes that the literature identified as important determinants of child malnutrition: i) Reducing incidence o f low birth weight through improvement in the nutrition of pregnant women and control o f infections during the first term o f the pregnancy, ii) Reducing incidence o f acute diarrhea, respiratory infections and other prevalent diseases, and iii) Improving adequate dietary intake for children under 36 months (MEF, 2011). The current DHS database does not allow for a direct examination of the effects o f ANP program on CU. However, I use a “cohort’' variable that could capture the potential impact of the program on CU. A detailed overview of variables, both dependent and independent, and their creation is provided in Appendix 1. 3.3.1.3 Modeling Approach Some researchers have directly dealt with non-linear relationships between the dependent variable and the explanatory variables in their analysis of CU in children under five years old. For example, Marini & Gragnolaty (2006) show the non linear effects o f altitude on child growth in Peru. In a similar way Kandala, et al (2011) explore the effects o f geographic locations on malnutrition among children in the Democratic Republic of Congo (DCR). In both investigations the main purporse is to determine those nonlinear effects (altitude or geographic impact). Also, Grab (2009) analyzes and points out that the nonlinearity relationship between some o f the predictor variables with the response variable is one o f the challenging situations that researchers in malnutrition in children must be face. So, the assumption o f a strictly linear effect on the response variable (height for age z-score) may not be be appropriate for all the cases. In the case 66 of the present research I find similarities to Grab (2009) and Marini & Gragnolaty (2006), where there are nonlinear trends between the height for age z-score for Peruvian children under five years old, and the variables breastfeeding and geographic altitude. As shown in Figure 3.4, it is clear that the growth trajectory by age is nonlinear. The decline in children’s growth as measured by z-score for age is sharp between the ages o f 6 and 24 months and then slows down, changing the form of the trajectory. With this in mind, following the recommendations of Rabe-Hesketh & Skrondal (2012), I split the sample into 0-5, 6-24 and 2559 months old age groups (See Figures 3.5 - 3.7). The line-segments were fitted to show the agegroup specific trajectories. Figure 3.4- Trend and line o f mean height for age z-score by children's age .00 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 -.20 -.40 .80 £ - 1.00 1.20 -1.40 1.60 age in m onths Source: Peruvian DHS data 2 0 09-2012 It must be noted that there are fewer observations for children under six months old. The main purpose of my research is to analyse the determinants of CU older than 6 months. As Arocena 67 (2010) points out we should focus on children older than six or 12 months old because their stunting status is not yet evident. Figure 3.5-Trend and line of mean for height for age z-score by children’s before 6 months old -.80 -.82 4 -.84 chi2 0.0000 Standard errors are given in the parentheses Note: Labels ***, **, and * indicate significance at 1%, 5% and 10%, respectively. 10 352 0.2138 -14741.37 10 352 10 112 2 161 -14676.661 128.32 0.0000 4.2 Estimation Results for the age group between 25 to 59 months old The analysis of the determinants of CU in children 25 to 59 months old consists o f four linear models and two logistic regression models that are dreported and discussed in the next section. Models 3A and 3B are OLS regressions (with and without dietary intake, respectively), and the Models 4A and 4B are the multilevel linear regressions (with and without dietary intake, respectively). I discuss the regression results for the multilevel models as they are preferred over the OLS results bases on likelihood ratio tests. As I previously pointed out, the Models B do not take into account the variable dietary intake, and the number of observations is increased. However, for this group o f age dietary intake is significant at 10% o f significance, but there are no major differences in the effects for the other explanatory variables. Similar to the results for the age group 06-24 months old, the research findings for the age group 25-59 months old are in line with those found in the literature. First of all, there is a significant association between child growth and gender. Female children show better growth than males. This finding confirms the resuly by Ravina & Chavez (2007). Secondly, child growth is negatively associated with LBW. That is, children bom with a weight less than 2.5 kg have lower growth . Medhin et al. (2010) concluded for developing countries that LBW is a factor associated with a risk of being stunted. The age for Peruvian children above twenty four months old is not significantly related to child growth, a conclusion in line with the findings in in Fenske et al (2013) and Bryce et al (2008), that the status of stunted can be determined after twenty four months old and after this point the status is irreversible. Furthermore, in the present case, the dietary intake variable shows a certain association at 10% level o f significance, so Peruvian children with access to less than four kind of foods present worse growth outcomes, or are more prone to suffer from CU. Also, in this case not only living in a mountain range (Andes region) shows negative effect but also living in a rural zone and jungle region have a significant negative effect similar to the conclusions in Arocena (2014). Children who live in rural areas and in mountain range have lower growth. Moreover, similar to the case o f children between 6 to 24 months old, the higher wealth and educational attainment of the mother, the better the height-for-age z-scores for the child. 79 About health variables, diarrhea shows a negative effect at 1% level o f significance, respiratory infeccions still has no effect. Among all community variable services lack o f access to delivery services and antenatal care show a significant negative effect on height for age. In contrast to the results of Models 2A and 2B, the cohorts 2009 and 2010 are associated with child growth in Models 4A and 4B at significance level o f 1% and 10% , respectively.. This implies a positive effect for the ANP program, which is consistent with the findings of Cruzado (2012) for 2009. Table 4.2- Determinants o f Stunted for children between twenty five months old and before fifty nine months old M odel 3 A (OLS) M odel 4A (M ultilevel) M odel 3B (O LS) M odel 4B (M ultilevel) 0.00357 (0.00489) 0.125*** (0.0292) 0 .00334 (0.0 0 4 8 8 ) 0.124*** (0.0291) 0.0000153 (0 .0 0 4 3 6 ) 0.100*** (0.0 2 6 2 ) -0.000393 (0.00434) 0.1000*** (0 .0 2 6 2 ) 0.0243 (0.0498) 0.0988*** (0.0355) 0 .0 7 2 8 (0.0425) -0.455*** (0.0543) 0.0222 (0 .0 5 0 1 ) 0.0987*** (0 .0 3 5 6 ) 0 .0 7 2 5 (0 .0 4 2 8 ) -0.442*** (0 .0 5 4 1 ) -0.0256 (0 .0 4 8 4 ) 0.0263 (0 .0 3 8 4 ) 0.0936** (0 .0 3 6 5 ) 0 .0 7 2 1 (0 .0 4 3 4 ) -0.454*** (0.0 48 5) -0.0339 (0 .0 4 8 9 ) 0.0212 (0.0390) 0.0953** (0.0365) 0 .0 6 8 9 (0.0438) -0.441*** (0 .0 4 8 2 ) -0.00485 (0.0403) -0.0910** (0.0411) -0.0837* (0.0438) -0.0 07 10 (0.0403) -0.0916** (0.0411) -0.0802* (0 .0 4 3 8 ) -0.00541 (0 .0 3 60 ) -0.101*** (0 .0 3 7 0 ) -0.00867 (0.0360) -0.0996*** (0.0370) 0.250*** (0.0473) 0.489*** (0.0596) 0.244*** (0 .0 4 7 7 ) 0.483*** (0 .0 6 0 2 ) 0.258*** (0.0 41 9) 0.564*** (0 .0 5 2 4 ) 0.247*** (0 .0 4 2 4 ) 0.556*** (0.0530) Fixed Effects Individual Variables A ge o f children (in months) Fem ale Birth's Cohort (base cohort 2008) Cohort 2006 Cohort 2007 Cohort 2009 Cohort 2010 W eight at birth; b elow 2.5 kg Poor Health Status A cute Respiratory Infection A cute Diarrheal D isease Dietary intake Household Variables Wealth index: quintile) 2nd quintile M iddle quintile (base poorest 80 4th quintile R ichest quintile M other's education (base: no education or primary education) 0.751*** (0.0709) 1.104*** (0.0820) 0.286*** 0.745*** (0 .0 7 1 4 ) 1.091*** (0 .0 8 2 6 ) 0.283*** 0.817*** (0.0 62 9) 1.167*** (0 .0 7 3 3 ) 0.317*** 0.805*** (0.0635) 1.145*** (0.0 7 4 0 ) 0.311*** (0.0372) (0 .0 3 7 2 ) (0 .0 3 3 2 ) (0.0332) -0.0266 (0.0359) -0.00462 (0.0 3 6 6 ) -0.00158 (0.0 3 6 9 ) -0.0258 (0.0363) -0.00480 (0 .0 3 68 ) 0.000758 (0.0 37 2) -0.00164 (0 .0 3 2 4 ) 0.0161 (0 .0 3 2 9 ) -0.0203 (0 .0 3 2 8 ) -0.00144 (0.0 3 2 9) 0.0170 (0.0 3 3 1) -0.0149 (0.0330) -0.240*** (0.0 4 6 3 ) 0.0646 (0.0 4 8 6 ) -0.000336 (0.0 3 3 1 ) 0.0770* (0.0419) -0.237*** (0 .0 4 64 ) 0.0650 (0 .0 4 8 9 ) -0.000662 (0 .0 3 3 1 ) 0.0789* (0 .0 4 1 7 ) -0.215*** (0 .0 4 08 ) 0.0498 (0 .0 4 87 ) 0.00695 (0 .0 2 8 9 ) 0.0695* (0 .0 3 6 6 ) -0.208*** (0.0 4 0 9) 0.0552 (0.0490) 0.00286 (0.0 2 8 9) 0.0713* (0.0 3 6 4) 0.0921** (0.0 4 2 3 ) 0.0815* (0 .0 4 3 3 ) 0.0831** (0 .0 3 7 7 ) 0.0672* (0.0 3 9 2) -0.0336 -0.0311 0.0158 0.0163 (0.0596) -0.465*** (0.0625) -0.170*** (0 .0 6 4 5 ) -1.778*** (0.181) (0 .0 6 0 9 ) -0.465*** (0 .0 6 3 8 ) -0.173*** (0 .0 6 5 8 ) -1.761*** (0 .181) (0 .0 5 3 9 ) -0.433*** (0 .0 5 6 3 ) -0.142** (0 .0 5 8 1 ) -1.754*** (0 .163) (0.0 5 5 6) -0.437*** (0 .0 5 8 2) -0.150** (0 .0 6 0 0) -1.724*** (0.163) .9120393 .1878704 0.0000 0 0 0 9 7 7 .8925512 Community Characteristics Water and sanitation services N o access to piped water inside N o access to improved sanitation N o access to treated water Health services N o access to delivery service A ccess to grow ing control A ccess to vaccination A ccess to antenatal care Place o f residence (base: urban); rural Natural R egion (base: M etropolitana) Rest o f the Coast (no M etropolitana) Mountain Range Jungle Constant Lima Lima Random Effects Cluster random effect H ousehold Random effect Individual Random effect .2355158 0.000000286 .9100064 Observations Children 3 947 3 947 5 168 Households 3 911 Clusters 1 705 Adj- R2 0.2753 0.2773 Integrated log Likelihood -5237.1415 -5233.6841 -7 0 12.4087 LR-Test chi 1(2) 6.91 Prob > chi2 0.0315 Standard errors are given in the parentheses Note: Labels ***, **, and * indicate significance at 1%, 5% and 10%, respectively. 5 168 5 118 1 901 -7000.7399 23.34 0.0000 81 4.3 Logistic Regression Results for the age group between 25 to 59 months old I also attempted two logistic multilevel effects regression models (Model 5A and Model 5B, with and without dietary intake, respectively). Here the dependent variable is the stunting status status represented by a dummy variable (0,1). The results are presented below in the table 4.3. Table 4.3- Determinants of Stunted for children between twenty five months old and before fifty nine months old: Logistic multilevel effects regression models Odds Ratio SE M odel 5A 95% Conf. Interval Odds Ratio SE M odel 5B 95% Conf. Interval Individual Variables A ge o f children (in months) 0.9796743 (.0 1 4 7 0 9 6 ) [0.9512642 1.008933] 0.988712 (.0 1 2 6 7 4 8 ) Female 0 .6755059*** (.0607588) [0.5663271 0.8057326] 0 .7309028*** (0 .0 5 6 6 9 1 ) [0 .6278239 0.8509055] 1.352473** (0 .1 7 8 1 92 3 ) [1.044673 1.750963] [0.7948756 1.420876] 1.265629** (0 .1 2 4 3 86 3 ) [1.043879 1.534485] Birth’s Cohort (base cohort 2008) Cohort 2006 Cohort 2007 0.0607588 [0 .9641794 1.013869] (0 .1 5 7 4 7 3 8 ) Cohort 2009 0 .7584419** (0 .0 8 3 0 5 2 2 ) [0.6119453 0 .9400089] 0 .7 584419 (0 .1 0 9 9 6 5 9 ) [0.7245781 1.208286] Cohort 2010 0 .8 0 6 6 5 8 2 (0 .1 0 4 4 1 5 6 ) [0.7281375 1.214094] 0.93902 (0 .1 1 0 2 8 9 8 ) [0.7459325 1.182089] 2 .264197*** (0.3 2 3 3 2 3 7 ) [1.711449 2.995466] 2 .222979*** (0 .2 7 6 4 9 3 2 ) [1.742061 2.836661] 0 .9296485 (0.1 1 3 7 5 0 3 ) [0.7314205 1.1816] 1.015322 (0 .1 0 4 5 1 2 3 ) [0.8298225 1.242288] A cute Diarrheal D isease 1.248368 (0 .1 4 7 9 4 7 ) [0.9896117 1.574782] 1.240343 (0 .1 2 8 4 3 6 4 ) [0.012513 1.519438] Dietary intake 1.197225 [0.9511991 1.506886] W eight at birth; below 2.5 kg Poor Health Status A cute Respiratory Infection (0 .1 4 0 5 1 6 8 ) Household Variables Wealth index: (base poorest 82 quintile) 2nd quintile 0.818893 (0 .1 0 0 9 2 8 ) [0.6431572 1.042647] 0.7482 2 4 5 * * (0 .0 7 8 6 3 9 9 ) [0.6089322 0.9193798] M iddle quintile 0 .4 5 1 3 268***(0.0788 121) [0.3205173 0.6355222] 0 .3858476*** (0 .0 5 7 1 9 5 9 ) [0.2885616 0.5159327] 4th quintile 0 .2 5 3 6 1 2 2 * * * (0 .0 6 15 153) [0.1576542 0.4079759] 0 .2249381*** (0 .0465418) [0.1499483 0.3374306] Richest quintile 0.1204 8 2 4 * * * (0 .0 4 6 8 7 2 5 ) [0.0562046 0.2582709] 0 .1069198*** (0 .0 3 5 9 9 6 2 ) [0.0552698 M other’s education (base: no education or primary education) 0.5412 4 9 8 * * * (0 .0 5 5 4 9 9 ) [0.4427074 0.6617268] 0.4960094*** (0 .0 4 3 4 8 7 8 ) [0.4176961 0.5890055] 1.225522** (0.1 26 6 2 1 7 ) [1.000863 1.50061] 1.167338 (0 .1 0 5 0 6 9 5 ) [0.9785472 1.392552] N o access to improved sanitation 1.053976 (0.1 14 5 9 1 1 ) [0.8516982 1.304294] 0.9692237 (0.0 9 1 6 7 2 1 ) [0.8052205 1.16663] N o access to treated water 1.052142 (0 .1 1 4 4 2 8 9 ) [0.8501574 1.302115] 0 .9 912957 (0.0 9 2 6 3 2 1 ) [0 .8253959 1.190541] 1.799275*** (0 .2 1 3 5 6 5 9 ) [1.425817 2.270552] 1.704981*** (0 .1 7 2 53 57 ) [1.398241 2.079012] A ccess to grow ing control 0.9402612 (0 .1 2 6 3 3 5 7 ) [0.7225685 1.223539] 0.9820016 (0.1 2 7 9 9 5 4 ) [0.7606161 1.267824] A ccess to vaccination 0.8314125* (0 .0 8 5 0 5 6 5 ) [0.6803549 1.016009] 0.8591958* (0 .0 7 3 4 2 2 6 ) [0.7266958 1.015855] A ccess to antenatal care 0.8455382 (0 .0 9 8 4 4 6 5 ) [0.6730191 1.06228] 0 .8 677474 (0 .0 8 6 1 1 1 9 ) [0.7143702 1.054055] Place o f residence (base: urban); rural 0.8491955 (0 .1 0 3 9 7 3 ) [0.6680194 1.079509] 0.8922567 (0 .0943785) [0.7251935 1.097806] Natural R egion (base: Lima Metropolitana) Rest o f the Coast (no Lima Metropolitana) 0.8991498 (0 .2479523) [0.5237234 1.543697] 0.9044206 (0 .2 1 5 6 7 5 1 ) [0.5667432 1.443293] Mountain Range 2.337663*** (0 .6 3 1 2 5 5 5 ) [1.376975 3.968603] 2.269244*** (0 .5 2 9 5 8 4 7 ) [1.436256 3.585339] Jungle 1.363991 (0 .3 7 4 8 7 7 2 ) [0.7959188 2.337515] 1.394607 (0.331687 ) [0.8749954 2.222786] 0.206837] Community Characteristics Water and sanitation services N o access to piped water inside Health services N o access to delivery service 83 Constant 0.8670675 (0 .4 9 6 9 2 3 6 ) [0.2819826 2.6661441 0 .6 411708 (0 .3 1 4 5 41 5 ) [0.2451301 1.677069] Observations Children H ouseholds Clusters 3 940 3 904 1 704 -1663.8472 Integrated log Likelihood 1.45 Integrated log Likelihood 0.4836 LR-Test chi 1(2) Standard errors are given in the parentheses N ote: Labels ***, **, and * indicate significance at 1%, 5% and 10%, respectively. 5 154 5 104 1 899 -2225.3496 3.40 0.1829 Generally, the results are largely similar to the finding from the Models 3A and 3B, thus, male Peruvian children are more vulnerable to stunting compared to females, children with LBW have more probability of stunting, as well as children from poor households, with no educated mother, with no access a health services delivery services, or living in the mountain range. The Odd Ratio (OR) of stunting in girls was 0.73 relative to boys. Also, children who settled in mountain range were significantly more likely to be stunted as compared to those of Metropolitan area of Lima (OR= 2.22). Higher wealth status of the family was determined as the protective factor of stunting. Children from families in the richest quintile were significantly less likely to be stunted compared to those with a poorer wealth status (OR=0.11 to 0.12). UNICEF (2013) points out that poor families have limited access to healthy and varied diets, health care, clean water or basic sanitation. Moreover, lack of piped water was another predictor of stunting (OR= 1.22). According to UNICEF (2009), the access a safe water have an impact on CU but that impact will be higher if there is a mixed policy about sanitation, concealing to families and improved sanitation services. It is important to note that there are higher probability of being stunted for children who were bom in 2006 (cohort 2006 - OR=l .35) and 2007 (cohort 2007 -O R =l .26) suggesting an improvement over time, which may partly be related to the effect of the APN program. 84 Overall, the results are consistent with the findings from the literature and with the conceptual framework of CU (Figure 3.2). Although not all the variables have been found to be significant, several variables at individual, household and community levels are identified as important determinants or factors associated with the child growth. 85 Chapter 5: Conclusions This thesis applies the UNICEF’s (1990) malnutrition framework adapted by Tufts University (2001) for stunting in the Peruvian context. By using multilevel modelling, I analyze the determinants of CU for two group of children, those between 6 and 24 months old and those who are older than 24 months but not younger than 59 months old. The purpose of this chapter is to summarize the main findings o f this thesis and highlight the contributions of this investigation. Thus, in the first section, I summarize the main findings from the models as relate them to the literature. In the second section, I discuss the thesis policy implications. I also recognize some of the study’s limitations and briefly suggest areas o f further research that could contribute to a better understanding of stunting in Peru as basis for effective interventions to reduce stunting. 5.1 Summary of Main Findings In this study, the CU status was analyzed based on the modified UNICEF (1990) malnutrition framework by Tufts University (2001). The framework recognizes that CU has many possible causes and that solving malnutrition problems requires an understanding of the importance of various determinants at different levels. Therefore, I use a multilevel modeling techniques that allow for the estimation of different level effects for two group o f ages, which enables us to separate effects working at the individual, household and community level. The patterns of association with the outcome of interest (height for age z-score - HAZ) was analyzed for the two groups of age: children from 6 to 24 months old 86 and children from 25 to 59 months old, I also considered a binary logistic multilevel model for the older group. The main motivation for using a multilevel analysis stemmed from the hierarchically nature of the DHS data, where children are nested into households, households nested into PSU’s, and PSU’s into the communities. The findings for the two group o f age provide enough information to answer the three questions I posed at the beginning: 1. What factors can explain CU in children aged six to fifty nine months in Peru? The findings for the two groups o f age (As reported in tables 4.1, 4.2 and 4.3) demonstrate that there are different factors at individual, household and community level that explain CU in Peruvian children from 6 to 59 months old. At the individual level for both age groups demonstrated child’s gender, the occurrence o f diarrhea and LBW are risk factors for stunting. The results show lower probability o f stunting in female children compared to male children. A result consistent with that of Ravina & Chavez (2007), but not with those of Marini & Gragnolaty (2006) and Gutierrez et al (2014) for Peruvian children. The occurrence o f diarrhea was found to be significantly associated with height for age z-score. In contrast, the respiratory infections did not show any effect on the growth z-score. This negative effect o f diarrhea but not o f respiratory infections, is consistent with other studies (Adair & Guilkey, 1997; Checkley, et al, 2002). In addition, LBW is find to have a significant association with stunting, this has also been documented in previous investigations (Adair & Guilkey, 1997; Ngoc & Kam, 2008; Medhin, et al., 2010). No previous research has analyzed the effect of LBW on children in Peru. Such 87 findings supports the importance of prenatal caring and the access to services that it is also significant for children between 6 to 24 months. At the household level, the mother education is positively associated with child's growth. Such findings are in line with those of Marini & Gragnolaty (2006) for Peruvian children and Smith & Haddad (2014) for children in developing countries. As educated mothers have enough knowledge about their children’s health and nutrition, the nutritional status of children can be developed through improvement of child care, health services usage, hygiene and sanitation, etc. (Baldarrago, 2010). Also, the wealth index is a strong predictor o f child growth. Children from richer households have better results on their CU status. For the group between 25 to 59 months old, the odd ratios (Table 4.3) indicate that as a more household cumulative living standard (better wealth index position) a lower risk to be stunted. This significant association may be due to high quality of life, giving essential health services and also adequate dietary intake. Moreover, low literacy and low purchasing capacity, and also poor sanitation and hygiene are common results o f poverty inequality (Javenic et al, 2010). At community level, the children from the mountain range have worse height for age z-score results. This findings echoes those of Arocena (2010) and Marini & Gragnolaty (2006) for the impact o f geographic altitude of Peruvian regions on stunting. I do not find a significant association between rural areas and stunting. The high prevalence o f stunting in rural areas in Peru may be result of low educational level, low wealth status, poor water supply, and high incidence o f infectionus disease. It is possible that the rural effects, are captured by those factors. This possibility is also demostrated by Ngoc & Kam (2008) for Vietnam. 88 Also there is a significant association o f access to health services, such as the delivery and antenatal services with CU status. However, access to water and sanitation services are not associated with stunting, this is also summarized in the findings of UNICEF (2009) and in Esrey (1996). As for the Peruvian investigations only Ravina & Chavez (2009) and Gutierrez et al. (2014) found a correlation between stunting in children and access to sanitation. Those services are important to prevent diarrhea cases and depend on a combination o f services like improved sanitation, treated water at the point of use and handwashing practices (UNICEF, 2009). This mixed results in this case indicate that further investigation is needed. 2. Do the same factors that affect Peruvian children during the window o f opportunity of intervention (before two years old) also act in children above two years old? This question was inspired by the UNICEF (2009) and Bhutta et al (2013) recommendations for an early intervention policy to prevent undemutrition during the critical window o f opportunity that is during the pregnancy of the mother and during the two years of life of a child. As expected, age is associated with stunting for the younger group o f children (6-24 months), but not for the older group (25-59 months). Such finding confirm the critical role o f age for stunting during the first two years of life. The investigations in Peru did not consider the critical role o f younger age into their analysis. For example, Ravina & Chavez (2009), Marini & Gragnolaty (2006) and Gutierrez et al. (2014) did not differentiate between age groups before and after 24 months old. They all find that age is significant for stunting. This leaves the question of intervention open to the first five years old as opposed to first two years which may be too risky. Other difference between the two groups of age is the level of significance of dietary intake, not having more than at least four kinds of food per day shows a negative association only for the 89 case o f children above twenty four months old. Bhutta et al., (2013) conclude that dietary diversification strategies have not been proven to affect nutritional status, however the access is still an important factor. I think this is an aspect that needs more research. I cannot find any investigation that linked that factor exclusive to stunting. Some of them analyze the effects of the early introduction of dietary intake, breastfeeding and stunting (Chung, et al., 2007). Therefore, I think it is important to complement the analysis with other variables, such as breastfeeding and health status o f children. 3. Is ANP affecting CU in Peruvian children? I do not find a strong association between the birth cohorts and stunting for the age group between 6 to 24 months. However, I do find a signifficant association for the group older than 24 months. This implies that the ANP program may have affected the cohorts of children after 2008. Interestingly, the binary multilevel regression reveals that children who belong to the birth’s cohorts before 2008 have higher risk to be stunted. Cruzado (2012), whose research was the only one that took into account the ANP program,found a positive effect for the year 2009. ANP program as a public policy needs further investigation in order to find out which of its many interventions is the most effective one. 5.2 Policy Implications, Limitations and Further Studies 5.2.1 Policy Implications The findings of my thesis have important implications for the development o f policy interventions by the Peruvian Government for controlling and reducing CU. My findings can inform policy implementation such as BfR nd the ANP program, which is a multi-sector effort 90 with a preventive approach to reduce CU. Specifically, my results for children between 25 to 59 months old show that ANP program can be effective in reducing the prevalence of stunting. Given that age is a signifiant variable linked to height for age z-score only for children between 6 to 24 months old, it is very important to intervene in a timely fashion. Morre importantly, since lack of access to nutrition, clean water and sanitation, health care servies and adequate income are routed in poverty and inequality, a general policy targeted at poverty eradication and inequality reduction could go along way in improving the prospect o f healthly child growth. For example, antenatal care and access to delivery services have a possitive association with CU. Also, low birth weight is a strong predictor of stunting and stunted. Investing in interventions that prevent low birth weight through the antenatal care, access access to delivery and adequate health services is an important aspect to be considered in APN program. The causes of stunting are complex and difficult to measure in any cross-sectional survey. Although the dataset was not ideally suitable for the use of conceptual framework to analyze the causes o f stunting, some interesting associations between CU and determinants at the individual, household and communities levels are identified. To repeat, the maternal level o f education, gender, place of residence, wealth status, child’s health status, and access to some services, are found to be important in determining the generation o f CU the two age groups. 5.2.2 Limitations One o f the limitations of the study is that the analysis is based on cross-sectional data, thus I cannot discuss causalities in my conclusions. It could have been preferred if the data were panel so that the trajectory of stunting over time, and the causal dynamics o f the determinants could be better examined. Following a child during his or her growing process would make it easier to 91 account for the effect of unobservable health determinants, and also to capture the time dynamics of the relationships between growth determinants. Also, the characteristic of the survey might have affected the findings. As the survey is questionnaire-based, questions that required a good memory were vulnerable to recall bias. Finally, there was no reliable variable to use for capturing the potential effects of the ANP policy. Using cohorts to measure such effects is less than satisfactory. 5.2.3 Future Studies My study can be extended in several ways. For instance, considering cultural barriers such as those addressed by Grab (2009) may shed light on the use of health services as related to child growth. Considering the quality of the services in addition to their quantity or availability is important. Moreover, examining the issues o f poverty and inequality - as social determinants o f health including healthy child growth - that underlie poor hygiene, food insecurity and lack of access to health care and education will improve our understanding o f stunting in Peru. 92 Bibliography Aburto, N., Ramirez-Zea, M., Neufeld, L. M., & Flores-Ayala, R. (2009). Som e indicators o f nutritional status are associated with activity and exploration in infants at risk for vitamin and mineral deficiencies. The Journal o f nutrition, 139{9), 1751-1757. Abuya, Onsomu, Kimani, & M oore. (2011). Influence o f maternal education on child im munization and stunting in Kenya. M atern Child Health J, 1 5 , 1389-1399. Adair, L., & Guilkey, D. (1997). A ge-specific determ inants o f stunting in Filipo children. The Journal o f Nutrition, 127(2), 314-320. Arocena, V. (2010). Factores asociados a la desnutricion cronica infantil en Peru: una aplicacidn de m odelos multinivel. Revista Latinoamericana de Poblacidn, 3(6), 41-56. Australian Health M inisters C onference. (2009). The Australian National Breastfeeding Strategy 2010-2015. Sidney: Australian Governm ent D epartm ent o f Health and Ageing. Baker-Henningham, H., Hamadani, J. D., Huda, S. N., 8i Grantham-McGregor, S. M. (2009). U ndernourished children have different tem p eram en ts than better-nourished children in rural Bangladesh. The Journal o f nutrition, 139(9), 1765-1771. BaldSrrago, E. (2010). La education de la madre com o determ inante de la salud del nino: Evidencia para el Peru en base a la ENDES. CIES, 1-58. Barros, F., Victora, C., Scherpbier, R., & Gwatkin, D. (2010). S ocioeconom ic inequities in th e health and nutrition of children in low / middle incom e countries. Rev. Saude Publica, 44(1), 1-16. Bergeron, G., & Castleman, T. (2012). Program R esponses to Acute and Chronic Malnutrition: D ivergences and C overgences. Am erican Society fo r Nutrition, 34, 242-243. Bernt, L., & T adesse, A. (1997). Intra-household correlations of nutritional status in rural Ethiopia. International Journal o f Epudemiology, 26(1), 160-165. Bhutta, Z. A., Das, J., Rizvi, A., Gaffey, M., Walker, N., Horton, S., Black, R. (2013). Evidence based interventions for im provem ent o f maternal and child nutrition: w hat can be d on e and at w hat cost? Lancet, 382, 452-477. Bhutta, Z., Ahmed, T., Cousens, S., D ewey, K., Giugliani, E., Haider, B., Shekar, M. (2008). W hat Works? Interventions for maternal and child undernutrition and survival. Lancet, 3 7 1 , 417-440. Black, R., Allen, L., Bhutta, Z., Caulfield, L., de Onis, M., Ezzati, M., Rivera, J. (2008). Maternal and child undernutrition: global and regional exposures and health con seq u en ces. Lancet, 371(9608), 243-260. Bryce, J., Coitinho, D., Darnton-Hill, I., Pelletier, D., Pinstrup-Endersen, P., & Group, f. t. (2008). Maternal and child undernutrition: effective action at national level. The Lancet, 371(9611), 5 10-526. d oi:1 0 .1 0 1 6 /S 0140- 6736(07)616 9 4 -8 Checkley, W., Buckley, G., Gilman, R., Assis, A., Guerrant, R., Morris, S., Black, R. (2008). M ulty-country analysis o f th e effects o f diarrhoea on childhood stunting. International Journal o f Epidemiology, 37(4), 816-830. 93 Checkley, W., Epstein, L., Gilman, R., Cabrera, L., & Black, R. (2002). Effects o f Acute Diarrhea on Linear Growth in Peruvian Children . American Journal o f Epidemiology, 157(2), 166-175. Chopra, M. (2002). Risk factors for undernutrition of young children in a rural area of South Africa. Public Health Nutrition, 6(7), 645-652. Chung, M., Raman, G., Chew, P., Magula, N., DeVine, D., Trikalinos, T., & Lau, J. (2007). Breastfeeding and m aternal and infant health outcom es in developed countries. Agency for Healthcare Research and Quality. Rockville: AHRQ. Clasen, T., Bostoen, K., Schmidt, W., Boisson, S., Fung, I., Jenkins, M., Cairncross, S. (2010). Interventions to improve disposal of human excreta for preventing diarrhoea (Review). The Chocrane Library. Coles, C., Rahmathullah, L., Kanungo, R., Katz, J., Sandiford, D., Devi, S., Tielsch, J. (2012). Pneum ococcal Carriage at Age 2 M onths is A ssociated with Growth Deficits at Age 6 m onths am ong infants in South India. The Journal Crookston, B., Dearden, K., Alder, S., Porucznik, C., Stanford, J., Merrill, R Penny, M. (O ctober de 2011). Impact of Early and Concurrent Stunting on Cognition. M aternal & Child Nutrition, 7(4), 397-409. Cruzado, V. (2012). Analisis del im pacto del program a presupuestal Articulado Nutritional sobre la desnutricidn crdnica en ninos m enores de 5 ahos. Lima: Ministerio de Economia y Finanzas. de Onis, M. (2008). Chid grow th and d evelop m en t. En R. Semba, & M. Bloem, Nutrition and Health in Developing Countries (peigs. 71-91). Totowa, NJ: Humana Press. de Onis, M., M onteiro, C., & Cluaston, G. (1993). The W orlwide M agnitude o f Protein-Energy Malnutrition: an overview from th e WHO Global Database on Child Growth. World Health Organization, 71(6), 703-712. O btenido de http://w w w .w h o.in t/n u trition /p u b lication s/en /ch ild grow th _d atab ase_overview .p d f. de Onis, M., Onyango, A., Borghi, E., Garza, C., & Yang, H. (2006). Comparison of th e World Health Organization (WHO) Child Growth Standars and th e National Center for Health Statistics/W HO international grow th reference: im plications for child health programm es. Public Health Nutrition, 9(7), 942-947. DHS Program. (15 de 02 d e 2015). W ealth Index. Retrieved from th e Demographic and Health Surveys: http://w w w .d h sp rogram .com /top ics/w ealth -in d ex/in d ex.cfm Esrey, S. A. (1996). W ater, W aste, and Well-Being: A Multicountry Study. American Journal o f Epidemiology, 143(6), 608-623. Fenske, N., Burns, J., Hothorn, T., & Rehfuess, E. A. (2013). Understanding Child Stunting in India: A Com prehensive Analysis o f Socio-Econom ic, Nutritional and Environmental Determ inants Using Additive Quantile Regression. PLoS ONE, 8(11), e7 8 6 9 2 . Fernald, L. C., &. Grantham-McGregor, S. M. (2002). Growth retardation is associated with changes in th e stress response system and behavior in school-aged Jamaican children. The Journal o f nutrition, 132(2), 3674-3679. Galler, J. R., Bryce, C. P., Zichlin, M. L., Fitzmaurice, G., Eaglesfield, G., & W aber, D. P. (2012). Infant Malnutrition is A ssociated with Persisting A ttention Deficit in Middle Adulthood. American Society fo r Nutrition, 142(4), 788-794. doi:10.3945/jn . 111. 145441 94 Gardner, J., Grantham-MC Gregor, S., Himes, J., & Chang, S. (1999). Behavior and D evelopm ent of stunted and n o n ­ stunted Jamaican children. Joural o f Child Psychology and Psychiatry, 40(B), 19-27. Gelman, A., & Hill, J. (2007). Data analysis using regression and m ultilevel/hierachical models. New York: Cambridge University Press. Gentilini, U., & W ebb, P. (2008). How are w e doing on poverty and hunger reduction? A new m easure o f country perform ance. Food Policy, 1-12. Goldstein, H. (2003). Multilevel Statistical Models. London: Edward Arnold. Gorstein, J., Sullivan, K., Yip, R., de Onis, M., Trowbridge, F., Fajans, P., & Clugston, G. (1994). Issues in th e a ssesm en t of nutritional status using anthropom etry. Bull Wolrd Health Organ, 72, 273-283. Grab, J. (2009). Econometric Analysis in Poverty Research: with Case Studies fro m Developing Countries (Vol. 26). Frankfurt am Main, Berlin, Bern, Bruxelles, New York, Oxford, Wien: Gottingen Studies in D evelopm ent Economics. Gragnolati, M. (1999). Children's Growth and Poverty in Rural Guatemala. The World Bank, 1-50. Griffiths, M., & Del Rosso, J. (2007). Growth monitoring and th e prom otion o f healthy young child grow th. 1-31. Grummer-Strawn, L, Reinold, C., & Krebs, N. F. (2010). Use of th e World Health Organization and CDC Growth and Charts for Children Aged 0-59 m onths in th e United States. Centers o f Disease Control and Prevention, 59(RR09), 115. Gunther, F., Gunther, I., & Hill, K. (2011). The e ffect of W ater and Sanitation on Child Health: Evidence from The Dem ographic and Health Surveys 1986-2007. Int. J. Epidemiol, 40(5), 1196-1204. Gutierrez, C., Cunha, A., Deivila, M., Alarcbn, J., & Sobino, M. (2014). Desnutricibn infantil en m enores d e cinco ahos en Peru: tendencias y factores determ inantes. Rev Panam Salud Publica, 35(2), 104-112. Habicht, J.-P., & Pellettier, D. L. (1990). The im portance o f con text in choosing nutritional indicators. The Journal o f Nutrition, 120(11), 1519-1529. Heltberg, R. (2009). Malnutrition, Poverty, and Economic Growth. Health Econ, 18, S77-S88. Hoffman, D., & Klein, D. (2012). Growth in transitional countries: The long-term im pact of undernutrition on health. Annals o f Human Biology, 39(5), 395-401. INEI. (2014). Instituto Nacional de Estadlstica del Peru. Retrieved 01 04, 2013, from h ttp ://w w w .in e i.g o b .p e/: h ttp ://w w w .in ei.g o b .p e/esta d istica s/in d ice-tem a tico /so cia les/ INEI. (2014). Peru: Encuesta Demografica y de Salud Familiar 2013. Lima: INEI. Javenic, T., Petrovic, O., Bjelic, I., & Kubera, A. (2010). Risk factors for childhood malnutrition in Roma settlem en ts in Serbia. BMC Public Health, 1 0 , 1-8. Kabubo-Mariara, J., N denge, G., & M wabu, D. (2008). D eterm inants o f Children's Nutritional Status in Kenya: Evidence from Demographic and Health Surveys. Journal o f African Economies, 18(3), 363-387. 95 Kandala, N.-B., M andungu, T., Emina, J. B., Nzita, K. P., & Cappucio, F. P. (25 de April de 2011). Malnutrition am ong children under th e age o f five in th e Democratic Republic o f Congo (DRC): d oes geographic location m atter? BMC Public Helath, 11(261), 1-15. Kuhn, J., Lou, M., Greenfield, H., & W ojtys, E. (1997). A Statistics Primer: Types o f Studies in th e M edical Literature. The American Journal o f Sports Medicine, 25(2), 1-3. Lamberti, L. M., Fischer, C. W., Noiman, A., Victora, C., & Black, R. (2011). Breastfeeding and th e risk for diarrhea morbidity. BMC Public Health, 11(3), 1-12. Lee, G., Yori, P., Paredes, M. 0 ., Pan, W., Caulfield, L., Gilman, R., Kosek, M. (2012). Comparative effects o f vivax malaria, fever and diarrhoea on child growth. International Journal o f Epidemiology, 41(2), 531-539. Lescano, A. (2002). M odelo para elAnalisis d e Politicos de Intervencion en Desnutricion. Lima: INEI. Leyland, A. H., & G roenew egen, P. P. (2003). M ultilevel m odelling and public health policy. Scandinavian Journal o f Public Health, 31, 267-274. Liu, J., Raine, A., V enables, P. H., & Mednick, S. A. (2004). Malnutrition at age 3 years and externalizing behaviour at ages 8 ,1 1 , and 17 years. The American journal o f psychiatry, 161(11), 2005. Luter, C. K., Chaparro, C. M., & Munoz, S. (2011). Progress tow ards Millennium D evelopm ent Goal 1 in Latin America and th e Caribbean: The Im portance o f th e Choice o f Indicator for Undernutrition. Bulletin o f the World Health Organization, 89(1), 22-30. Marini, A., & Gragnolaty, M. (2006). Nonlinear Effects o f Altitude on Child Growth In Peru - A multilevel Analysis. World Bank, 3823, 3823. Marriott, B. P., W hite, A., Hadden, L., Davies, J. C., & Wallingford, J. C. (2012). World Health Organization (WHO) infant and young child feeding indicators: associations with grow th m easures in 14 low -incom e countries. M aternal & Child Nutrition, 8(3), 354-370. Mazumdar, S. (2012). A ssessing Vulnerability to Chronic Undernutrition am ong Under-Five Children in Egypt: Contextual D eterm inants o f and Individual C onsequence. International Journal o f Population Research, 1-12. M edhin, G., Hanlon, C., Dewey, M., Alem, A., Testaye, F., Worku, B., Prince, M. (2010). Prevalence and predictors of undernutrition am ong infants aged six and tw elve m onths in Butajira, Ethiopia: th e P-MAMIE Birth Cohort. BMC Public Health, 10(27), 1-15. M eeks Gardner, J. M., Grantham-McGregor, S., Himes, J., & Chang, S. (1999). Behaviour and d evelop m en t o f stu nted and nonstunted Jamaican children. Journal o f child psychology and psychiatry, 40(5), 819-827. MEF. (2011). Programa Articulado Nutricional. Lima: Cooperacidn Alemana (GIZ). O btenido de http://w w w .m ef.gob.pe/contenidos/presu_publ/ppr/P P R _articulado_nutricional.pdf MEF. (2013). Marco M acroeconomico M ultianual 2014-2016. Lima: MEF. M endez, M., 8i Adair, L. (1999). Severity and timing o f stunting in th e first tw o years o f life perform ance on cognitive te st in late childhood. J Nutr, 129(8), 1555-1562. 96 MIDIS. (2012). Desnutricidn Crdnica Infantil: M etas al 2016. Lima: MIDIS. O btenido de h ttp ://w w w .m id is.gob .p e/im ages/d ireccion es/d gp ye/rep orte_p eru .p d f MINSA. (2011). Norma Tecnica de Salud para el Control de Crecimiento y Desarrollo d e la Nina y el Nino. Lima: Ministerio de Salud. MMPV. (2012). Plan Nacional de Accidn por la Infancia y la Adolescencia 2012-2021. Lima: MIMP. M onteiro, C., D'Aquino, M., Lisboa, W., Konno, S., Lovadino, A., Barros, A., & G om es, C. (2010). Narrowing Socioeconom ic Inequeality in Child Stunting: The Brazilian Experience, 1974-2007. World Health Organization, 88, 305-311. Mora, J. (1989). A n ew m ethod for estim ating a standarized prevalence of child malnutrition from anthropom etic indicators. The World Health Organization, 62(2), 133-142. Moradi, A. (2009). Toward and Objective Account of Nutrition and Health in Colonial Kenya: A Study o f Stature in African Army Recruits and Civilians. The Journal o f Economy Histoty, 69, 719-754. National Center for Health Statistics. (1977). Growth curves for children birth to 18 years. United S ta tes D epartm ent o f Health Education and Welfare, Vital and Healht Statistics, 11(165), 1-74. Ngoc, N., & Kam, S. (2008). Nutritional Status and th e Characteristics Related to Malnutrition in Children Under Five Years of Age in Nghean, Vietnam. Journal o f Preventive M edicine and Public Health, 41(4), 232-240. Onyango, A. W., de Onis, M., Caroli, M., Shah, U., Sguassero, Y., Redondo, N., 8t Carroli, B. (2007). Field-Testing the WHO Child Growth Standars in Four Countries. J.Nutri, 1 3 7 , 149-152. Pelletier, D. L. (2002). Toward a com m on understanding o f Malnutrition: Assessing the Contributions o f the UNICEF Framework. W ashington: World Bank/Unicef. Pelletier, D., Frongillo, E., & Habicht, J. (1993). Epidemiologic evidence for a potentiating effect o f malnutrition on child mortality. A m J Public Health, 83(8), 1130-1133. Penny, M., Creed-Kanashiro, H., Robert, R., Narro, R., Caulfield, L., & Black, R. (2005). E ffectiveness o f an educational intervention delivered through th e health services to improve nutrition in young children: a cluster-random ised controlled trial. Lancet, 3 6 5 , 1863-1872. Pirez, V. F. (2013). Los m od elos multinivel en el an^lisis de factores d e riesgo en lactantes. Enfoque frecuentista y bayesiano. Universidad de Murcia, 1-145. Pollit, E., Saco-Pollitt, C., Jahari, A., Huasaini, M., & Huang , J. (2000). Effects of an energy and micronutrient supp lem ent on m ental d evelop m en t and behavior under natural conditions in undernourished children in Indonesia. European Journal o f Clinica Nutrition, 54(2), 80-90. Rabe-Hesketh, S., & Skrondal, A. (2012). M ultilevel and Longitudinal M odeling Using S ta ta (Vol. 1: Continuos R esponses). Texas: Stata Press. Rajkumar, A., Gaukler, C., 8i Tilahun, J. (2012). Combating Malnutrition in Ethiopia: An Evidence-Based Approach for Sustained Results. World Bank, 1-226. 97 Ravina, R., & ChSvez, J. C. (2007). Andlisis de lo sfa cto res Subyacentes Asociados a la Desnutricidn Crdnica Infantil. Lima: Instituto Nacional d e Estadfstica e Informcitica. Reinhardt, K., & Fanzo, J. (2014). Addresing chronic malnutrition through m ulti-sectoral, sustainable approaches: a review o f th e cau ses and co n seq u en ces. Frontiers in Nutrition, 1 , 1-11. Rutsein, S., & Johnson, K. (2004). The DHS w ealth index. DHS com parative reports no 6 . Calverton: ORC Macro. Rutstein, S. O., & Rojas, G. (2006). Guide to DHS Statistics. Calverton, Maryland: USAID. Obtenido de http://w w w .dhsprogram .com /pubs/pdf/D H SG l/G uide_to_D H S_Statistics_29O ct2012_D H SG l.pdf Save th e Children. (2013). Food fo r Thought: Tackling Child M alnutrition to Unlock Potential and Boost Prosperity. London: The Save th e Children Fund 2013. Scientific Advisory C om m ittee on Nutrition (SACN). (2011). The influence o f m aternal, fe ta l and child nutrition on the d evelopm ent o f chronic disease in later lafe. London: SACN. Shetty, P. (2006). Malnutrition and undernutrition. Medicine, 34(12), 524-529. Shin, H. (2007). Child Health in Peru: Im portance o f Regional Variation and Community Effects on Children's Height and W eight. Journal o f Health and Social Behavior, 48(A), 418-433. S h o rt, M. F., Joung, Y. C., & Bird, S. (2012). A system atic review o f Demographic and Health Surveys: data availability and utilization for research. World Health Organization, 604-612. Smith, L., & Haddad, L. (2000). Explaining child malnutrition in developing countries. W ashington D.C.: International Food Policy Research Institute. Smith, L., & Haddad, L. (2014). Reducing Child Undernutrition: Past Drivers and Priorities for th e Post-MDG Era. Institute o f D evelopm ent Studies, 2014(441), 1-47. Smith, L., Ruel, M„ 8t Ndiaye, A. (2005). Why Is Child Malnutrition Lower in Urban Than in Rural Areas? Evidence from 36 D eveloping Countries. World Development, 33(8), 1285-1305. The World Bank. (2007). Nutritional Failure in Ecuador: Causes, Consequences a n d Solutions. W ashigton, DC: World Bank. The World Bank. (2014). Data: M alnutrition prevalence, height fo r age (% o f children under 5). Retrieved from de The World Bank. Working for a World free of poverty: http://data.worldbank.org/indicator/SH.STA.STNT.ZS/countries/XJ7displaysgraph The World Bank. (2014). The World Bank - 2014 Economy. O btenido de 4.1 World D evelopm ent Indicators: Growth o f output: http://w di.w orldbank.O rg/table/4.l Tufts University. (2001). Reduccidn de la Desnutricidn Crdnica en el Peru: Propuesta para una Estrategia Nacional. M edford, MA: Tufts University and United States A gency for International D evelopm ent. Recuperado el 27 d e 02 de 2013, de h ttp ://w w w .m ef.gob .p e/con ten id os/p ol_econ /d ocu m en tos/R ed u ccion _d esn u tricion _T U F T S .p d f UNICEF. (1990). Strategy fo r im proved nutrition o f children and w om en in developing. N ew York, N.Y., USA: UNICEF. UNICEF. (1998). The sta te o f the world's chilren 1998: Focus on nutrition. New York: UNICEF. 98 UNICEF. (2003). Strategy to reduce m aternal and child undernutrition. Bangkok: UNICEF - East Asia and Pacific Regional Office. UNICEF. (2009). Evidence Base: W ater, Sanitation and Hygiene Interventions. New York: UNICEF. UNICEF. (2013). Improving child nutrition. The achievable im perative fo r global progress. N ew York: United Nations. United Nations. (2014). Millenium Project. Retrieved from UN Millenium Project: http://w w w .unm illennium project.org/ UNSSC. (2010). 6th Report on the World Nutrition Situation: Progress in Nutrition. G eneva, Switzerland: United N ations Standing C om m ittee on Nutrition. Urke, H., Bull, T., & Mittelmark, M. (2011). Socioeconom ic status and chronic child malnutrition: W ealth and maternal education m atter m ore in th e Peruvian Andes than nationally. Nutr Res, 31(10), 741-747. USAID. (2014). The DHS Program. Retrieved from Demographic and Health Surveys: http://dhsprogram .eom /topics/N utrition.cfm #sthash.lafiJZhC.dpuf Valdivia, M. (2004). Poverty, health infraestructure and th e nutrition of Peruvian children. Economics and Human Biology, 2(3), 489-510. Van de Poel, E., H osseinpoor, A., Speybroeck, N., Van Ourti, T., & Vega, J. (2008). Sociecononom ic inequality in malnutrition in developing countries. Bulletin o f th e World Health Organization, 86(A), 282-291. Victoria, C. G., Fuchs, S. C., Huttly, S. R., 8t Olinto, M. A. (1997). The Role of Conceptual Frameworks in Epidemiological Analysis: A Hierarchical Approach. International Journal o f Epidemiology, 26(1), 224-227. W agstaff, A., & W atanabe, N. (2000). Socioeconomic Inequalities in Child M alnutrition in th e Developing World. W ashington, DC: World Bank. Walker, S., Chang, S., Powell, C., Sim onoff, E., & Grantham-M c Gregor, S. (2007). Early Chilhood Stunting is Associated with Poor Psychological Functioring in Late A dolescence and Effects are Reduced by Psychosocial Stimulation. The Journal o f Nutritions, 137, 2462-2468. W amani, H., Tylleskar, T., Nordrehaug, A., Tumwine, J., & Peterson, S. (2004). M others’ education but n ot fathers' education, household a ssets or land ow nership is th e b est predictor o f child health inequalities in rural Uganda. Int J Equity Health, 3(9), 1-8. W ang, Y., & Chen, H.-J. (2012). Use of Percentiles and Z-Scores in Anthropometry. En S. Nandy, P. Sverderg, S. Nandy, & P. Sverberg (Edits.), Handbook o f Anthropom etry: Physical M easures o f Human Form in Health and Disease (pags. 29-48). New York: Springer. Wang, Y., M oreno, L., Caballero, B., & Cole, T. (2006). Limitations o f th e current world health organization growth references for children and ad olescen ts. Food N utr Bull, 27, S175-88. W aterlow, J. (1972). Classification and definition o f protein-calorie malnutrition. British m edical journal, 3, 566-569. WFP. (2005). A Manual: M easuring and Interpreting M alnutrition and Mortality. Rome: WFP. 99 WHO. (1999). A Critical Link Interventions fo r Physical Growth and psychological d evelopm ent a review. Geneva: WHO. WHO. (2006). WHO child grow th standars : length/height-for-age, weight-for-age, weight-for-length, weight-for- height and body m ass index-for-age: m ethod s and developm ent. Geneva: WHO. WHO. (2012). Com prehensive im plem entation plan of maternal infant and young child nutrition. Sixty-fifth World Health Assembly, WHA resolution 65.6 and A nnex 2, (pags, 12-13). Geneva. WHO. (2014). Global Database on Child grow th and M alnutrition. Obtenido de h ttp ://w w w .w h o .in t/n u tg ro w th d b /estim a tes2 0 1 2 /en / WHO. (2014). Millennium D evelopm ent Goals (MDGs). Retrieved from World Health Organization: h ttp ://w w w .w h o .in t/m ed ia c en tr e/fa c tsh e e ts/fs2 9 0 /e n / WHO M ulticentre Group Reference Study. (2006). WHO Child Growth Standards based on length/height, w eight and age. A cta Paedriatica, 450, 76-85. WHO Working group on infant grow th. (1994). An evaluation o f infant growth. Geneva: World Health Organization. WHO Working group on infant growth. (1995). An evaluation o f infant growth: th e use and interpretation of anthropom etry in infants. Bulletin o f th e World Health Organization, 7 3 , 165-174. World Bank. (2006). Reposition Nutrition as a central to developm ent: a stra teg y fo r a large scale action. . W ashington. World Bank. (2007). Nutritional Failure in Ecuador: Causes, Consequences and Solutions. W ashigton, DC: World Bank. Wright, C. M., & Garcia, A. L. (2012). Child undernutrition in affluent societies: w hat are w e talking about? Proceedings o f the Nutrition Society, 71, 545-555. 100 Appendix 1: List of Variables of Interest Variables Characteristics Number of Name and type in the database Observations Stunting (Z -score for age) height H eight-for-age Z-score based on the new W HO child 38 766 growth “Z o m s t e ” is provided in database; Continue Variable reference standards (W HO, 2006) H eight-for-age Z-score: >-2 Stunted “C M O M S 1 ”; Dum m y variable. standard deviations (sd) (0), <-2 sd 0= 28 721; 1=9 126 It is created with the information below the median height for age (1) o f the binary variable “dc06” (Chronic Undemutrition according to the W HO l:yes, 2:no); A t the sam e time, that variable is created based on the variable “h v070” : “height standard deviations according to W HO” Individual variables Gender Whether the child is male (0 ) or 0 = 1 9 6 1 1 ; 1=19 155 fem ale (1) “Gender2” is created with the information “hc27” of the variable (1: male, 2: fem ale); Dum m y variable A ge o f the children from 0 to 59 A ge months old 38 766 “h cl" , continues variable. Based on the variable “agegrp”, information of it created w as it is also this categorical variable, with values 1: children o f 0-5 months old, 2: children o f 6-2 4 months old and children o f 101 25-59 months old. N: 1 - 3 340, 2 - 12 1 4 9 , 3 - 2 3 277 Birth’s Cohort To identify in which cohort o f birth Cohorts 2004 and 2005= the child was bom. 3 951 “co h 200405”, “coh 2006”, “coh 2007” Cohort 20 06= 4 748 Cohort 20 07= 6 240 Cohort 2009=6 944 , “coh 2008, “coh 2009”, “coh 2010”, “co h 2 0 1 1”, " coh2012”; dum m ies variables. T hey were created based on the variable “cohort” by making a group o f Cohort 20 10= 4 783 children according to the sem ester and year o f birth. Cohort 2011=2 830 Cohort 2012=1 078 Whether the child after six months Inadequate food intake -A ccess to variety o f food 0= 9 665 o f age accesses more than 4 types o f food in the last 7 days (0), less than 4 types o f food in the last 7 days (1) “d iv fo o d l”; dumm y variable. To 1= 3 230 get this variable it was generated first the variable dietary diversity: “divfeed” 1: L ess than 4 types o f food; 2: M ore than 4 types o f food. This categorical variable was created with the information o f the follow ing variables in the database that collect information of the follow ing questions: do you feed your child in the last 7 days with? S496a (flour, bread, cookies), noodles, s4 9 6c (juice, pumpkin), s4 9 6 e (beans, lentils, peas, soy), s4 96g (tubercle, root), s496i (vegetables), s496k (papaya, mango, am ong others), s496m (others fruits), s496o (meat: beef, chicken, fish), s496aa (cheese yogurt), s496ak (porridge from a program). 102 social's “breastfl”, This 0= 17 648 -Exclusive Breastfeeding Whether the child has had exclu sive breastfeeding during the first six months o f its life (1 ) if not (0) dummy variable generated based on “lac ex ” that have the categorical 1=2 499 variable. was values 1: breastfeeding and aged between 0-6 months old, 2: no breastfeeding or aged more than 6 months old. As w ell this variable w as created with the combination o f the variable age (h cl was grouped for children betw een 0 and 6, and older), “m 4”, duration o f breastfeeding - selecting for ex clu sive for 6 or more months of duration, “v 4 0 4 '\ currently breastfeeding - selecting as an exclu sive i f the child is more than 06 months old. A lso the information from “v 4 0 4 ” and “m 4” was contrasted to the questions from v409 to v414 where the mothers were asked if they were feeding his/her child with other alim ents like water, liquid, am ong others. 103 formula, “birthw eightl”, dummy variable. It w as created based on Whether the weight o f the child at 0 = 3 0 125 birth is more than 2.5 kilograms (0) L ow birth weight 1=2 389 or is less than 2.5 kilograms (1 ) the variable “p esom ” 1: less than 2.5 kilograms, 2: more than or equal to 2.5 kilograms. This variable w as based on “m l 9” original variable in the database with the w eight of information the child o f the at the moment o f birth i f the mother can remember. Poor Health Status “addl", dummy variable. Is w as Whether Diarrheal D isease the child has hadn’t 0=23 833 created based on the variable 1 = 4 658 w as diarrhea in the last tw o w eeks (0) or has had diarrhea in the last two “eda” 1: yes 2: no. This variable w eeks (1) created information based of on the “h 11”: had diarrhea recently? A cute respiratory infections W hether the child has hadn’t a cough in the last tw o w eeks (0 ) or 0= 30 297 has had a cough in the last two 1=5 793 w eeks (1) “a r il”, dummy variable. It was created based on the information o f “ira059” 1: yes, 2: no. This variable w as created based on tw o variables “h 3 1 had a cough in last tw o w eeks, and “h31b” : short, rapid breaths? For children only under 59 months old. H ousehold variab les W om en’s education Whether the mother education or she has not 0 = 1 4 661 completed “educ_m om 2”, dummy variable. This variable was created based 1=22 667 104 primary school (0), and whether the on the variable “le v m o m e d u ” 1: mother no com pleted secondary education or primary school or higher level o f education education, 2: (1) higher education. secondary or The information was provided by the variable “m h v l0 6 ”: level o f the education o f the mother. “deliv2’\ dummy variable. This A ccess to delivery services W hether the mother has received 0=25 359 variable w as created with the delivery services 0; otherwise 1 information o f the follow ing 1=6 298 variables: “m3n” : no assistance during delivery, “m3a': assistance by a doctor, “m3b: assistance by a nurse/m idw ife”, “m 3c” : assistance by an obstetrician “cfantenatal", dumm y variable. This variable w as created with Whether the mother received the information o f tw o variables A ccess to antenatal care antenatal care (at least 6 controls 0=5 732 during pregnancy in a hospital, health center or post center) 1; otherwise 0 “m l 4 ” : antenatal pregnancy and visits the for variables 1=24 330 “m 57i-m 57q” : antenatal kind o f health center. 105 care, Household econom ic W ealth Index S core17: Whether the Poorest: 12 533 “quintilnl - quintiln5”, dum m ies variables. 5 status household is destitute, poor, These Poorer: 10 594 middle status, or wealthy. The W ealth Index Score categorizes households into 5 wealth quintiles variables were created based on Middle: 8 437 Richer: 5 547 the wealth index provided by the survey “hv270” (between poor and wealth Richest: 3 218 Community variables A ccess to health services A ccess to growth Whether the child has not received 0=34 710 “cr e d l”, dummy variable. This control services18 checkup growth control services - variable w as created based on 1=2 136 that it is according to the Peruvian ”S 4 6 6 c” : number o f check-up Health Norm (at least 6 controls) 0; growth controls. otherwise 1 0=12 768 17 The wealth index is presented in the DHS Final Reports and survey datasets as a background characteristic. It is generated by the D H S program with principal com ponents analysis and it categorizes households into 5 wealth quintiles. It is “a com posite measure o f a household's cum ulative living standard. The wealth index is calculated using easy-to-collect data on a household’s ownership o f selected assets, such as television s and bicycles; materials used for housing construction; and types o f water access and sanitation facilities” (D H S Program, 2015; Rutsein & Johnson, 2 0 0 4 ). 18 The growth control services is provided in a health center. B asically is a prevention activity comprised o f growth monitoring (process o f follow ing the growth rate o f a child in comparison to a standard by anthropometric measurements) linked with promotion and counseling about child growth; im proves caring practices and serves as the core activity in an integrated children health and nutrition program. In the Peruvian case it is considered at least 11 controls during the first year o f age (M IN SA , 2011; Griffiths & D el R osso, 2007). 106 A ccess to Whether the child has not received 1=9 287 “vaccination 1” , dummy im munizations the basic vaccines for his/her age variable. This variable services according to the Peruvian Health created based on the variable Norm 0; otherwise 1 provided by the w as survey: “vaccom ” . This variable was constructed with the information o f the variables “s45d -s45v” w hose cover the number, the types o f vaccines and age o f the child at w hich the vaccine was administered. A ccess to water and sanitation services 0=28 218 “safew atl 1”, dumm y variable. It w as A ccess to safe (treated) water Whether the household has access created based on the 1= 10 548 variables hv237: anything done to treated water at the point o f use to water to make safe to drink, (0), otherwise (1) hv237a: water usually treated by: boil, hv237b: water usually treated by: add bleach/chlorine, hv237e: water usually treated by: solar disinfection and hv237d: water usually treated by: use water filter) 107 A ccess to water Whether the household has 0=26 269 accessed to piped water into the " w ataccl”; dummy variable. This variable was created with 1= 12 497 house (0), otherwise (1) the information o f the variable “h v201” : source water; by taking of drinking only into consideration options 11 and 12: piped into dw elling and piped outside dw elling but within building. A ccess to improve sanitation Whether the household has accessed to improved sanitation services (0), otherwise 1 "sanitation 1” , dummy variable. 0 = 16 216 T his variable was created with 1=22 550 the information o f the variables "hv225” : hare toilet with other households and "hv205”: type o f toilet facility, by taking only into consideration options 11, 12, 21 and 22: inside dw elling, outside dw elling, ventilated latrine, septic w ell. Whether i f the cluster is in urban Area o f residence 0=21 226 area (0) or in rural areas (1) "area2”, dummy variable. This variable w as created based on 1=17 540 “hv025” type o f residence urban, 2: rural 108 1: Place o f residence At what natural region o f Peru the Lima Metropolitana= 2 “n a tr e g io n 1”, “nat_region2”, cluster is. Lima M etropolitana, rest 609 “nat_region3”, “nat_region4”. o f the coast area, mountain or They are 4 dum m ies variables Rest o f coast area= A ndes area and jungle area created based on the information 9 236 Mountain o f “shregion”. or A ndes area= 15 752 Jungle area= 1 1 1 5 9 109