Menu
Menu
Menu
Menu
Background: Poverty is a multidimensional phenomenon and unidimensional measurements have proven inadequate to the challenge of assessing its dynamics. Dynamics between poverty and public health intervention is among the most difficult yet important problems faced in development. We sought to demonstrate how multidimensional poverty measures can be utilized in the evaluation of public health interventions; and to create geospatial maps of poverty deprivation to aid implementers in prioritizing program planning.
In 2009, World Vision International was awarded a United States Agency for International Development (USAID) grant called Strengthening Communities through Integrated Programming (SCIP) for a 5-year multi-sector program aimed at improving the health and livelihoods of children, women, and families in Zambézia Province, Mozambique. Known locally as the Ogumaniha project, which means “united for a common purpose” in the local language of Echuabo, SCIP is based on a consortium of five international non-governmental organizations led by World Vision. The broad goals of the 5-year project are to: 1) reduce poverty in Zambézia Province by pursuing the consolidation of an integrated, innovative, and sustainable community-based program in the province; and 2) integrate current and future United States Government (USG) investments in Zambézia Province in the areas of health, HIV/AIDS, water and sanitation, income generation, and institutional capacity building. In 2012, Mozambique ranked 185 of 187 nations on the UNDP’s HDI, and the gross national income was estimated at US $906 per capita [8], with male and female life expectancies of 47 and 51 years, respectively, in 2009 [17]. Although Mozambique’s health expenditure has risen substantially over the past 10 years, as a proportion of total GDP it was only 6.6% in 2011 (66 USD per capita). [18] Mozambique is one of the sub-Saharan African countries most affected by the HIV/AIDS epidemic, with a national adult HIV prevalence of 11.5% in 2009. [19] Nationally, 12% of children were considered orphans or vulnerable children, and only 43% of households had access to clean drinking water in 2009 [20]. The magnitude of poverty is especially evident in Zambézia Province, Mozambique’s second largest province and home to about 4 million persons (Figure 1). [21], [22] While Mozambique ranks among the poorest of the poor nations, Zambézia consistently ranks among Mozambique’s lowest performing provinces with low literacy rates, poor maternal and child health (MCH) indices, high rates of tuberculosis and malaria infections, and high levels of malnutrition. [20], [23], [24] Zambézia Province is overwhelmingly rural and depends almost entirely on subsistence farming and fishing. The province has the highest estimated number of persons living with HIV in the country (∼275,000 or nearly 20% of Mozambique’s HIV-infected population) as of 2009 [19], [25] and yet only 31 of Zambézia’s 214 health facilities provided antiretroviral therapy (ART) as of December 2012. This is partially because Zambézia Province housed much of the armed conflict in Mozambique’s 16-year civil war (1976–1992) and suffered disproportionately in destruction of its healthcare infrastructure [21], [22]. Integral to Ogumaniha’s design is a strong monitoring system and project evaluation based on performance indicators agreed upon with USAID and the provincial government. Because the project involves multi-sectoral interventions and an interdisciplinary approach to implementation, the consortium opted for a multidisciplinary evaluation design. A survey instrument used at Ogumaniha’s initiation (baseline survey in 2010) and at the project’s end (final survey implemented June 2014) was designed based on the human development theory originated by Sen (1999) and further developed by researchers from OPHI. This instrument uses multiple dimensions to measure poverty including health, education, and income; access to goods and services; and self-empowerment. The vision of this pre-post project evaluation is that the information collected can provide a more thorough and holistic measure of the impact of this large-scale, multi-sector intervention on the overall health and well-being of the households in Zambézia Province, more so than analysis of the individual sector specific measures when viewed in isolation. The Ogumaniha survey tool collects information on over 500 variables in 8 dimensions and was developed by a multi-disciplinary team of researchers including staff, faculty, and graduate students from Vanderbilt University and the Universidade Eduardo Mondlane. To design the survey, we used many questions and validated scales from previous national surveys in Mozambique, including various National Institute for Statistics (Instituto Nacional de Estatísticas [INE]) surveys focusing on poverty and economic status; and other international surveys such as the Demographic Health Survey (DHS) and the Multiple Indicator Cluster Survey (MICS). The survey was designed to collect household information from the female head of household, defined as the principal wife of the nuclear family (polygamy is common practice), because she is thought to be most familiar with the majority of topics of interest. Survey questions covered household demographics; economic status; health knowledge, attitudes and practices; access to health and HIV-related services and products; access to improved water and sanitation; nutritional intake; and others. The poverty index used to identify households or areas of poverty in the province following baseline survey data collection, was modeled after the Multidimensional Poverty Index (MPI) founded on the OPHI methodology which calculates the quantity of defined “poor” (the headcount), multiplied by their average amount of deprivation, called the Adjusted Headcount Ratio [15]. Mobile survey teams conducted interviews with 3,916 (98%) of 3,960 planned female heads of households in 259 Enumeration Areas (EAs) across 14 of Zambézia’s 17 districts. Complete data for analysis was available from 3,749 (96%) of the interviews conducted. EA selection was not stratified by district, thus 3 districts were randomly excluded from the province-wide sample. Interviews were conducted either in Portuguese or in one of the five predominant tribal languages of the province (Cisena, Elomwe, Echuabo, Cinyanja, and Emakhuwa), and data were collected using mobile cell phones. Interviewers received intensive training on the use of mobile phones for data collection prior to implementation. Satellite and census maps were used to locate the EAs. Initial plans for household selection included administering questionnaires at 15 household structures identified through a color threshold algorithm and randomly selected on the satellite map; however, the actual implementation resulted in division of the EA into four quadrants and collecting 3–4 interviews starting at the household nearest the center of each quadrant (household selection paper in preparation). In a subset of 95 randomly selected EAs, anthropometric measures of a random selection of children under 5-years residing in participating households were also included. In these EA’s, households with one or more children aged 0–12 months, one child was randomly selected for weight and height measurements. Similarly, for households with one or more children aged 13–59 months, one child was randomly selected for height and weight measurements. Baseline survey data were collected between August and September, 2010. Fourteen teams of five female surveyors were recruited, each with prior experience in survey work. The teams were assigned by language proficiency to a specific region, working under the supervision of a regional supervisor, and were trained on general aspects of survey conduct. Mozambique’s 2007 census served as the sampling frame. To appropriately capture Ogumaniha’s public health and development interventions without increasing the sample size and survey costs, data were collected in two phases; a concentrated sample of 2,878 households in 193 EAs in three selected focal districts (Namacurra, Alto Molócuè, and Morrumbala) (Figure 1) and a smaller sample of 871 households in 66 EA from the remaining districts. These three districts were selected because they represent 3 distinct geographical regions, and Ogumaniha interventions were anticipated in each, allowing future analysis of intervention impact on poverty. We provide the sample size justification online: http://globalhealth.vanderbilt.edu/manage/wpcontent/uploads/Ogumaniha_SampleSize_20100613.pdf. Methods to identify and aggregate the poor using multiple dimensions have mathematical properties that allow for decomposition of poverty indices by subgroups or by indicators. [15] The Alkire and Foster method was used to construct three key poverty indices: headcount, average poverty gap, and adjusted headcount (called the Ogumaniha MPI). Application of this method is detailed elsewhere (see http://www.ophi.org.uk/research/multidimensional-poverty/how-to-apply-alkire-foster/); briefly, the steps for identification and aggregation of households include: Proportions include 95% confidence intervals that incorporate the effects of stratification and clustering due to the sample design. [26] Descriptive analysis of continuous variables includes weighted estimates of median, 25th and 75th quantiles (interquartile range). Categorical variables are reported as weighted percentages, with each observation being weighted by the inverse of the household sampling probability. Tests of association by poverty ignore effects of clustering, and they include Wilcoxon rank sum (continuous) and chi-square (categorical) tests. Using Esri shape files provided by INE to identify EA boundaries, a basic heat-map of poverty metrics was generated for the three focus districts. To enhance readability of the maps, ordinary kriging was used to predict poverty metrics for unsampled areas, assuming only spatial correlation. The ‘krige’ function is in the gstat package of R, which uses generalized least squares prediction with spatial covariances to produce smoothed geospatial representations of poverty. [27] R-software 2.15.1 (www.r-project.org) was employed for all data analyses; analysis scripts are available online at (http://biostat.mc.vanderbilt.edu/ArchivedAnalyses). The measurement of household income is particularly problematic in high poverty areas. [28] In the current sample, 49% report no monetary income whatsoever. Increasingly in economics and development, monetary income is no longer the preferred measure. Instead, a “permanent income”, [29] or wealth measure based upon ownership of selected assets is employed. Poverty stemming from lack of resources is associated with low income, but it is perhaps more closely related to low wealth. Low wealth individuals always have low income, but not all low income individuals have low wealth. In that sense, wealth and poverty are more closely related than income and poverty. We applied a measure of permanent income developed by the World Bank. [28] Briefly, a series of 37 asset and other indicator variables were used in a dichotomous hierarchical ordered probit model to derive a latent variable which denotes the permanent income of household to be incorporated into the Ogumaniha MPI (Figure 2, orange line) [12]. The adjusted headcount is decomposed by dimension for Morrumbala, Alto Molócuè, Namacurra and all three districts combined. Data that are overlaid include percent of households in the lowest quintile for permanent income wealth and % of households making less than USD$1.25/day. MZN = Metical. Participation in the household survey was completely voluntary, no incentive was provided for participation. At enrollment written informed consent was obtained. The protocol for data collection and consent forms were approved by the Mozambican National Bioethics Committee for Health (Comité Nacional de Bioética em Saúde [CNBS]) and the Institutional Review Board of Vanderbilt University.
N/A
