Menu
Menu
Menu
Menu
This paper uses longitudinal data from two informal settlements of Nairobi, Kenya to examine patterns of child growth and how these are affected by four different dimensions of poverty at the household level namely, expenditures poverty, assets poverty, food poverty, and subjective poverty. The descriptive results show a grim picture, with the prevalence of overall stunting reaching nearly 60% in the age group 15-17 months and remaining almost constant thereafter. There is a strong association between food poverty and stunting among children aged 6-11 months (p<0.01), while assets poverty and subjective poverty have stronger relationships (p<0.01) with undernutrition at older age (24 months or older for assets poverty, and 12 months or older for subjective poverty). The effect of expenditures poverty does not reach statistical significant in any age group. These findings shed light on the degree of vulnerability of urban poor infants and children and on the influences of various aspects of poverty measures. © 2011 Elsevier Ltd.
The study settings are two informal settlements of Nairobi, Kenya, namely, Viwandani and Korogocho where the African Population and Health Research Centre (APHRC) runs a longitudinal demographic surveillance system referred to as the Nairobi Urban Health and Demographic Surveillance System (NUHDSS). The NUHDSS has been in operation since 2002 and has about 60,000 registered inhabitants in nearly 20,000 households. These two densely populated communities have higher unemployment, poverty, crime, poor sanitation and generally poorer health indicators than Nairobi as a whole (APHRC, 2002). The two communities, however, exhibit structural differences: Viwandani is bordered by an industrial area and attracts relatively younger, more educated, and shorter term migrants, while the population in Korogocho is more stable and has higher levels of co-residence of spouses (Emina et al., 2011). This study uses data from the Maternal and Child Health (MCH) component of a broader project entitled “Urbanization, Poverty and Health Dynamics” being implemented in the NUHDSS. All NUHDSS female members who gave birth since September 2006 and their children were enrolled in the project, and anthropometric measurements taken. Updates were done during follow-up visits every four months, and also when new children were recruited into the study for the first time to form new cohorts. Some children could not be immediately traced until after several visits due to the high population mobility in urban poor settings. For the purpose of this study, we use data on 3693 children from six cohorts as described in Table 1. These children contribute data at different time-points (surveys) totaling 14,410 observations. The first baseline observations (Cohort 1 and survey 1) took place between February and April 2007 with follow-up visits and new recruitments made routinely thereafter. The first wave of cohort 3 was done during a prolonged period (between October 2007 and May 2008) as a result of the political and social instability that followed Kenya's 2007 elections. Sample size. Note: The total number of children enrolled across all six cohorts is 3693. The data in Table 1 show a relatively high level of attrition across successive waves. For instance, of the 568 children enrolled in the first cohort, 474 were successfully re-contacted in the first follow-up, and only 178 were reached during the eighth visit, for an average annual attrition rate of about 24%. The average annual attrition rate for the other cohorts ranged from 21 to 28%. Once a year the NUHDSS collects data on various aspects of well-being at the household level. The questionnaires cover monthly expenditures (on rent, food, energy, water, transport, electricity, health care, and school fees), assets (or possessions), dwelling characteristics (floor, wall, roof, drinking water, toilets, and garbage collection), subjective poverty on a scale from 1 (poorest) to 10 (richest), and access to food (e.g. number and quantity of meals, failure to eat, going to bed hungry). These data for 2007, 2008 and 2009 are also used in the analysis. The dependent variable is based on height-for-age Z-scores (HAZ), computed using the 2000 CDC growth reference standards using zanthro command in STATA. While child weight-for-age fluctuates with environmental influences such as acute infections and poor nutritional intake, the height-for-age indicator represents a long term measure of health or chronic undernourishment (FAO, 1997). As recommended by the World Health Organization, overall stunting is defined as HAZ below −2 standard deviations (SD) from the median of the WHO/NCHS reference, while severe stunting is defined as HAZ below −3SD from the median of the WHO/NCHS reference (WHO, 1995, 2010). HAZ score below −2SD for children in the age group below 2–3 years represents stunted growth which reflects a continuing process of ‘failing to grow’ or chronic malnutrition. In a healthy, well-nourished population of children, it is expected that approximately 2.3% of children will fall below two standard deviations of the reference population and will be classified as stunted, wasted or underweight (WHO, 1995). The World Health Organization considers the severity of malnutrition to be ‘high’ when the prevalence of stunting exceeds 30% and wasting reaches 10%. High levels of stunted growth are often associated with poor socio-economic conditions, frequent illness and inappropriate healthcare practices (WHO, 1995). In this and other similar studies, infants were measured in the recumbent position and ‘length’ was used rather than ‘height’. We operationalize alternative measures of poverty which capture not only the money-metric dimension, but also the broader aspects of human deprivation. First, we constructed a money-metric indicator of poverty using information on monthly household consumption. This indicator allows us to assess the relationship between access to cash income and child growth. Second, we derived an assets index using information on household ownership of durable assets. As indicated earlier, the assets index is considered a good measure of long-term wealth, and is expected to have an impact on stunting which represents a long term nutritional deficiency. Third, we derived a food poverty index using information of household's access to food. This index allows us to assess the effect of household food insecurity on child growth. Lastly, we included a measure of subjective poverty, derived from households' perceptions of their relative wealth status in the community, on a scale from 1 (poorest) to 10 (richest). Table 2 describes the five alternative measures of household welfare. Alternative measures of household welfare used in the study. Apart from the subjective poverty variable which was recoded in three categories using the cut-off points of three and six, the three other welfare indicators were recoded as tertiles. The categories were labeled “poorest”, “middle” and “least poor”. All four measures of poverty are time varying: the 2007 poverty indices were linked to the 2007 anthropometric data, the same for the 2008 and 2009 data. All poverty variables were measured at the NUHDSS level and tertiles derived after merging with the MCH data. There was about eight percent of missing values due to the fact that not all households had poverty information for the three time points. These missing values were imputed using the STATA add-on for imputation by chained equations (ICE) procedures (Royston, 2005). The following variables were used in the imputation equations: village where the household is located; mother's marital status, age, education and parity at the time of the first interview; household size; slum of residence; as well as the values of poverty measures for the preceding and/or the following time point. There were 100 observations with missing welfare data that were excluded from the analyses, hence a final sample of 14,310 observations from 3692 children. In the models we control for a set of characteristics and the child, mother, household and community levels which previous studies have hypothesized to have an impact on child growth. These include the sex and age of the child, and child's mother's education, length of stay in the study area, marital status, and parity. Besides mother's parity, we also control for household size since children may not necessary live with their biological parents. Using PCA, we also construct a household environment index from information on the type of dwelling's floor, wall and roof; toilet facilities, the type of drinking water source and garbage collection – factors expected to have a direct effect on risk of infections. Finally, we control for the slum of residence (Korogocho or Viwandani). The analysis is conducted in three steps: First, univariate and bivariate analyses are used to describe the patterns of stunting as the children age, and to depict the differences across the five poverty measures. Second, four multivariate models are used to test robustness of each poverty measure as a predictor of child growth and development and the statistical significance of the differences observed in the descriptive phase. Third, we stratify the analysis by age to examine how the overall effect of poverty on child nutritional status may vary by age. Given that the data are made up of repeated longitudinal observations, we use the random intercept multilevel models to control for clustering of observations at child level. The model is specified as follows: where i and j refer to the observation and child, respectively; πij is the probability that the child referenced (i, j) is stunted; xij(k) is the kth covariate; β0j represents the intercept modelled to randomly vary between children; βk is the regression coefficients of the kth explanatory variables; and u0j is the random coefficient distributed as N(0,σu2) (Rasbash et al., 2002). The equations used to fit the interaction models are derived from eq. (1). Models are fitted using the STATA “xtlogit” command. The third category (least poor) is used as the reference group for all five measures of poverty. The presentation of results will focus primarily on the coefficient of the first category (poorest).
N/A
