Family-level clustering of childhood mortality risk in Kenya

Slides:



Advertisements
Similar presentations
Multilevel Multiprocess Models for Partnership and Childbearing Event Histories Fiona Steele, Constantinos Kallis, Harvey Goldstein and Heather Joshi Institute.
Advertisements

The Economic Consequences of the Transition into Parenthood Wendy Sigle-Rushton Paper presented at the GeNet Seminar: Low Fertility in Industrialised Countries.
Multilevel Event History Analysis of the Formation and Outcomes of Cohabiting and Marital Partnerships Fiona Steele Centre for Multilevel Modelling University.
Multilevel Event History Modelling of Birth Intervals
BR and Fertility Why do some area’s of the world have higher fertility rates? BABY O MATIC How many will you have? Why do governments care about fertility?
National Institute of Statistics of Rwanda
Fertility history and health in later life: A study among older women and men in the British Household Panel Survey Sanna Read and Emily Grundy Centre.
Single Motherhood and Child Mortality in Sub-Saharan Africa: A Life Course Perspective Shelley Clark Associate Professor of Sociology Canada Research Chair.
Unpacking ‘Son preference’: the trajectory of a demographic variable Danièle Bélanger, PhD Associate Professor The University of Western Ontario.
The Forgotten Beneficiary of the Medicaid Expansions Andrea Kutinova and Karen Smith Conway Department of Economics University of New Hampshire.
The Tanzania Demographic and Health Survey (TDHS) June 2005.
Intra-urban differentials in early marriage: Prevalence and consequences Zeinab Khadr Combating Early Marriage and Young People’s Reproductive Risks in.
Risk of Low Birth Weight Associated with Family Poverty in Korea Bong Joo Lee Se Hee Lim Department of Social Welfare, Seoul National University. A Paper.
2015 TANZANIA DEMOGRAPHIC AND HEALTH SURVEY (TDHS)
The impact of job loss on family dissolution Silvia Mendolia, Denise Doiron School of Economics, University of New South Wales Introduction Objectives.
BACKGROUND RESEARCH QUESTIONS  Does the time parents spend with children differ according to parents’ occupation?  Do occupational differences remain.
Migration, methodologies and health inequality SEED Group
Mortality Rates LEARNING OBJECTIVES 1.TO DEFINE THE DIFFERENT MEASUREMENTS OF MORTALITY 2.TO IDENTIFY THE DIFFERENT GLOBAL MORTALITY TRENDS.
Chapter Objectives Define maternal, infant, and child health.
Exposure to Family Planning Messages through Mass Media and Interpersonal Communication and Current Contraceptive Use in Ghana Claire Bailey
Health and Living Conditions in Eight Indian Cities
Uses of Population Censuses and Household Sample Surveys for Vital Statistics in South Africa United Nations Expert Group Meeting on International Standards.
1. POPULATION IN TRANSITION IBDP Expectations: Population Change: Explain population trends and patterns in births (Crude Birth Rate), natural increase.
Analysis of Clustered and Longitudinal Data
SITUATION ANALYSIS AND IDENTIFICATION OF NEEDS IN THE AREA OF FAMILY POLICY IN SLOVENIA Ružica Boškić Child Observatory Social protection Institute of.
Child deaths: Causes and epidemiological dimensions Robert E. Black, M.D., M.P.H. Johns Hopkins Bloomberg School of Public Health.
DISENTANGLING MATERNAL DECISIONS CONCERNING BREASTFEEDING AND PAID EMPLOYMENT Bidisha Mandal, Washington State University Brian E. Roe, Ohio State University.
10 th INDEPTH AGM, 27 th -30 th September, 2010 ACCRA-GHANA “ Whose Child is not Immunized” Trends in Prevalence and Predictors in Child Health Care. Evidence.
School Dropout in Rural Vietnam: Does Gender Matter?
Afghanistan Mortality Survey 2010 Key Findings. What is the AMS? The AMS 2010 is the first comprehensive mortality survey in Afghanistan. It is a nationally.
Kevin Kovach, DrPH(c), MSc, CHES Johnson County Department of Health and Environment – Olathe, Kansas Does the County Poverty Rate Influence Birth Weight.
It is estimated that over 50 per cent of the African population do not have access to modern health facilities and more than 60 per cent of people in rural.
Mother’s, Household, and Community U.S. Migration Experience and Infant Mortality in Mexico Erin R. Hamilton, Andres Villarreal, and Robert A. Hummer Department.
National Institute of Population Studies Islamabad.
Impact of Maternal Education and Health Related Behaviors on Infant and Child Survival in Pakistan G. Mustafa Zahid University of Western Ontario London,
SEMINAR PRESENTATIONS
SEMINAR PRESENTATIONS Cambodia DHS and Measure DHS+ Survey Objectives and Methodology Housing and Characteristics of the Population Fertility and its Determinants.
Far and few between? The child- bearing decisions of Portuguese women Author: Márton Varga Conference on the Impact of Day-Care Services in Visegrad Countries.
Factors influencing transition to marriage among females in the Kassena-Nankana District, Ghana University of Cape Coast – Navrongo DSS Collaborative Team.
Time-invarying Covariates of Successive Births in Pakistan Ali Muhammad Ph.D. Candidate Department of Sociology University of Western Ontario London, Ontario.
ETHNIC DIFFERENCES IN FERTILITY TRANSITION IN TURKEY Political Demography: Ethnic, National and Religious Dimensions Session I : Differential Fertility.
PREAICE GEOGRAPHY POPULATION AND SETTLEMENT. POPULATION DYNAMICS 1 MILLION YEARS AGO: 125,000 PEOPLE. 10,000 YEARS AGO WHEN PEOPLE DOMESTICATED ANIMALS,
Multiple Indicator Cluster Surveys Data dissemination and further analysis workshop Further Analysis: Youth and Adolescents MICS4 Data dissemination and.
Rwanda: The impact of conflict on fertility Kati Schindler & Tilman Brück Gender and Conflict Research Workshop 10/06/2010.
Sterilisation uptake in the Dominican Republic: are women begging for it? Tiziana Leone Department of Social Policy.
Prelacteal feeding practices in Vietnam: Problems and determinant factors Poster Reference Number: PO0724 Background and Objectives: Figure 1: Conceptual.
Learning Objectives To understand the strengths, limitations and factors that affect different countries’ fertility rates.
The dynamics of poverty in Ethiopia : persistence, state dependence and transitory shocks By Abebe Shimeles, PHD.
Age at First Marriage in Palestine Niveen ME Abu-Rmeileh, MPH, PhD Institute of Community and Public Health-Birzeit University Ulla Larsen, PhD University.
1 Confidence Intervals for Two Proportions Section 6.1.
MDG 4 Target: Reduce by two- thirds, between 1990 & 2015, the mortality rate of children under five years.
2010 World Programme on Population and Housing Censuses Workshop on Civil Registration and Vital Statistics in the UNESCWA Region Cairo, Egypt, December.
Social-economic Differentials of the Dying Risk of the oldest- old Chinese Liu, Guiping Max-Planck-Institute for Demographic Research.
Measuring the population: importance of demographic indicators for gender analysis Workshop Title Location and Date.
Demographic models Lecture 2. Stages and steps of modeling. Demographic groups, processes, structures, states. Processes: fertility, mortality, marriages,
The Changing Population. What is Population? Population – a group of people living in a particular place at a specified time. The scientific study of.
Household Structure and Household Structure and Childhood Mortality in Ghana Childhood Mortality in Ghana Winfred Avogo Victor Agadjanian Department of.
Adult and Child Mortality 2010 Cambodia Demographic and Health Survey.
Ethiopia Demographic and Health Survey 2011 Mortality.
2014 Kenya Demographic and Health Survey (KDHS) Key Indicators Report.
2015 Afghanistan Demographic and Health Survey (AfDHS) Key Indicators Report.
Partner violence among young adults in the Philippines: The role of intergenerational transmission and gender Jessica A. Fehringer Michelle J. Hindin Department.
Son preference, maternal health care utilization and infant death in rural China Jiajian Chen 1, Zhenming Xie 2, Hongyan Liu 2 1 East-West Center, USA,
2014 Kenya Demographic and Health Survey (KDHS) Key Indicators.
Intimate Partner Violence in Peru: An assessment of competing models Corey S. Sparks Alelhie Valencia Department of Demography Institute for Demographic.
Correlates of HIV testing among youth in three high prevalence Caribbean Countries Beverly E. Andrews, Doctoral Candidate University.
Follow along on Twitter!
Follow along on Twitter!
DEMOGRAPHIC TRANSITION
DEMOGRAPHIC TRANSITION
Presentation transcript:

Family-level clustering of childhood mortality risk in Kenya D. Walter Rasugu Omariba Department of Sociology Population Studies Centre University of Western Ontario London, Ontario

Background Mortality decline in Kenya began in late 1940s. E.g. under-five mortality: 220 in 1958-62 period, declined to 89 in 1984-1989 period Reversals in the downward trend started in 1986 (see figure 1). Infant mortality increased by 24 % and Under-five mortality by 25 % in 1988-98 period. -Source- (NCPD and Macro International, 1989; Brass and Jolly, 1993). -Similar declines were also recorded in infant mortality rates. It declined from 103 deaths per one thousand live births in 1975, to 83 in 1985 and reached 67 in 1995 (UN, 2001).

Figure 1: Child mortality trends 1974-1998, Kenya Source: National Council for Population and Development and Macro International, 1989, 1994; 1999. -It is evident that mortality increased from 1986.

Existing research Focuses on determinants and differentials of mortality (See, for instance, Kibet, 1981; Ewbank et al., 1986; Kichamu, 1986; Omariba, 1993; Obungu et al., 1994; Ikamari, 2000). This study’s focus: Familial child death clustering: In the literature, defined in two ways: 1) Expected vs. observed- Higher observed deaths indicate death clustering 2) Control for unobserved heterogeneity through inclusion of random effects in models- correlation of risks at different levels. -Relied on census and survey data

Rationale Random-effects models used yet to be applied on Kenyan data. Child mortality remains an important public health issue. Reducing mortality important for sustaining country’s incipient fertility transition. -Recent increases in mortality are likely to affect the government’s two-pronged approach to population growth management, i.e., through birth spacing and reduction of mortality. Fertility declined from 5.4 in 1993 to 4.7 in 1998 and expected to reach about 2.5 in 2010. -Need to reexamine mortality determinants especially in changing conditions

Sources of unobserved heterogeneity Differential competence in childcare (Das Gupta, 1997). Biological factors e.g. genetically determined frailty, ‘improvident maternity’ syndrome (Guo, 1993; Das Gupta, 1997). Socioeconomic, cultural factors and environmental factors. All unmeasured and unmeasurable factors. -In relation to child care, such women are likely to be poor at making effective home diagnoses of their children’s symptoms and taking active steps to help them (Das Gupta, 1990). Guo (1993): -Gupta (1990): ‘Improvident maternity’ syndrome whereby certain families suffer brief birth intervals and/or large families in which higher-parity children receive less parental care and other resources. -Socioeconomic and socio-cultural factors: Poverty in certain households/ communities, lack of infrastructure in community, cultural practices that favour boys in childcare and feeding practices, infanticide in certain cultures -Environmental factors: ecological conditions associated with the aetiology of diseases

Death clustering? In this study: Measured by unobserved heterogeneity term indicating correlation of risks in family. Most studies only select one child, truncate data by certain date or ignore first child- Biased results especially when variables such as preceding birth interval and survival status are considered. -Clustering of mortality risks among siblings (or among children residing in the same community) is due in part to children sharing the same observed family and community characteristics. But the correlation may persist even after controlling for observed covariates- The remaining correlation is a consequence of genetic, behavioural and environmental factors that are related mortality risks and are common to groups of children but that are unmeasured or unmeasurable. -Correlated observations can be used to estimate the extent of clustering in mortality risks and can help us to determine the importance of unmeasured factors for child survival.

Implications of data structure Children in same family are more alike than children from different families. covariates’ estimates biased. Consequences of violation of independence: standard errors of parameters underestimated– spurious precision. biases baseline hazard duration pattern downward in survival analysis. -The sameness of children from same community/family violates the assumption of independence of observations required for statistical analyses. -If standard errors are underestimated we are likely to make Type I error- rejecting true null hypothesis. -biases baseline hazard duration pattern downward- This is similar to having a constant hazard (The best way of understanding this is by imagining a process of constant hazard).

Implications of data structure Random-effects models: Correct for the biases in parameter estimates, provides correct standard errors and correct confidence intervals and significance tests Separates impact of individual and social context If contextual effects significant, using a random effect (or multilevel model is reasonable). If not, then we need only adjust the error term for dependence of units. -Covariates’ estimates biased? Means magnitude may be erroneously large and direction of effect could be wrong. -Random-effects models? The random-effects have to correspond with the levels that the data is conceptualized to have.

Data and methods Data source: Demographic and Health Survey for Kenya, 1998. 7,881 women 15-49, all marital statuses from 8,380 households and 8,233 eligible women. 3,407 husbands/partners of the women Largely rural sample, 81.4% of the women’s sample Methods: Weibull hazard models and random-effect hazard models. The latter tests for family-level variance. -DHS are a replacement of the WFS for developing countries. The 1998 DHS for Kenya was the third such survey in the country. Others in 1989 and 1993. The 2003 DHS is yet to be released to the public. - Although we can include a random term for unobserved heterogeneity at the individual-level, a major advantage of the multilevel analysis is that by splitting of the variance in the multilevel model enables us to determine how much of the variation between individual survival chances is due to family or community effects.

Conceptual framework Study is guided by the Mosley and Chen (1984) ‘proximate determinants’ model (see Figure 2). Individual characteristics: Migration status, education, year of birth, ethnicity, religion, survival status of preceding child, birth interval, birth order and maternal age at birth. Household characteristics: socioeconomic status, sanitation and source of water. Strategy of analysis: first estimate effect of distant factors and then include proximate factors.

Figure 2: Conceptual framework for studying the determinants of infant and childhood mortality Proximate Determinants Distant Factors -Reproductive healthcare behaviour e.g. prenatal care, place of delivery, delivery care, tetanus injection, breastfeeding -Biodemographic factors e.g. maternal age at birth, birth interval, birth order, age at marriage, child loss experience -Household environmental conditions e.g. source of water, toilet facility. -Socio-economic factors: e.g. maternal & paternal education, place of residence, region, migration, occupation, household socioeconomic status, marital status, year of birth, period of child birth. -Socio-cultural factors: e.g. religion, ethnicity. Outcome Variable Risk of child death -The premise of this model is that all social and economic determinants of child mortality operate through a common set of biological mechanisms or proximate determinants, which include maternal factors, environmental contamination, nutrient deficiency, injury and personal illness control to affect child survival.

Data description: Of the 7881, 5716 had at least one child, while 2165 had never had a child. 23348 children born to 5716 women (family) 2325 children had died before their fifth birthday: Infancy- 1620(0-12 months) Childhood- 705 (Age 13-59 months)

Table 3: Distribution of children and child deaths per family in Kenya, DHS 1998 Children per/fam   Deaths in family Percent of 1 2 3 4 5 6 7 8 Total Children Deaths 1012 87 1099 4.7 3.7 884 99 991 8.5 4.9 632 130 16 778 10.0 7.0 523 131 30 689 11.8 9.0 366 128 36 11 542 11.6 10.2 327 115 47 15 509 13.1 11.9 193 100 42 14 9 359 10.8 11.5 129 81 35 19 275 9.4 11.0 105 62 29 18 227 8.7 10 41 40 23 139 5.9 9.5 12 53 2.5 4.3 38 2.0 4.5 13 0.7 1.6 0.2 0.5 0.1 0.4 4233 993 286 111 58 21 5716 % of children 22 .8 .5 .2 .03 ----   ----- % of deaths 43 25 0.3 -Over 80 percent of the children belong to families contributing two or more children to the sample. -Families with six or more children comprise about 28 percent of the families yet contribute over half of the children. -57 percent of the deaths occurred to 8.6 percent of the families with two or more deaths. -About 2 percent of the families contribute four or more deaths; together accounting for about 18 percent of the deaths.

Does clustering exists? Over 80 percent of the children belong to families contributing two or more children to the sample. Families with six or more children comprise about 28 percent of the families yet contribute over half of the children. 57 percent of the deaths occurred to 8.6 percent of the families with two or more deaths. About 2 percent of the families contribute four or more deaths; together accounting for about 18 percent of the deaths. Before clustering- Comments: -The magnitude of the family effect in the model is measured by the number of deaths per family, because children in families in which there are a large number of deaths face higher mortality risks. -To the extent that these risks are not associated with observed covariates, they will result in greater clustering of mortality in the multilevel model. Is there a basis for considering death clustering? (see Table 1) -The distribution of number of children born to a woman in the reference period shows that we cannot ignore the issue of dependence of sample and hence correlation of risks of death (Death clustering)- We need to control for correlation between births to the same mother.

Results There is significant unobserved heterogeneity both in infancy and childhood (Tables 3 &4): The estimated random parameters, θ, in the models with unobserved heterogeneity are 0.40 and 0.78 for infant and child mortality respectively. There is significant familial variation in the risk of infant and child death. Maternal education, period of birth, ethnicity, type of toilet facility, birth interval and maternal age at birth of child important for both infant and child survival (Tables 1&2). Migration status, religion, survival status of previous child and birth order significant only for infant mortality, while household SES significant only for child mortality. -Methodology: Still grappling with the challenges of data analysis – Revising this chapter as I understand the procedures involved by doing.

Results There are large ethnic differences in risk of death with children Luo mothers being most disadvantaged. Secondary or higher education associated with a 22 % and 42% reduction in risk of infant mortality and child mortality respectively. Risk of infant death higher for children born after 1990, while that of child death is higher for all children born after 1985. The risk of infant death is higher for children whose sibling died, were born less than 19 months after preceding sibling, and when the mother was less than 20. -Unobserved heterogeneity captures all unobserved and unobservable factors. It is statistically easy to establish the presence of unobserved heterogeneity, the challenge is establishing the source of the heterogeneity. This calls for qualitative research.

Conclusions The determinants of death have different effects on infant and childhood mortality. Biodemographic factors have greater effect in infancy, while education and ethnicity have greater effect in childhood. Suggests varied policy actions: Infancy: longer birth intervals through family planning and breastfeeding, later age at birth etc. Childhood: improvement in education, socioeconomic status and poverty eradication programs. -Evidently, child death clustering and its causes in different societies is an area that deserves further research

Conclusions Death clustering is non-ignorable – Needs further research: Healthcare factors- Information available only for children born three years before the survey. Qualitative research at community level. Death clustering, another measurement: Consider unobserved heterogeneity in the context of each woman’s sequence of births. The heterogeneity term used in this paper does not reflect this fact. Health factors: Analyze the data for children born in recent three years. More data collection through community surveys.