Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Measuring socioeconomic status in multicountry studies: results from the eight-country MAL-ED study

Measuring socioeconomic status in multicountry studies: results from the eight-country MAL-ED study Background: There is no standardized approach to comparing socioeconomic status (SES) across multiple sites in epidemiological studies. This is particularly problematic when cross-country comparisons are of interest. We sought to develop a simple measure of SES that would perform well across diverse, resource-limited settings. Methods: A cross-sectional study was conducted with 800 children aged 24 to 60 months across eight resource-limited settings. Parents were asked to respond to a household SES questionnaire, and the height of each child was measured. A statistical analysis was done in two phases. First, the best approach for selecting and weighting household assets as a proxy for wealth was identified. We compared four approaches to measuring wealth: maternal education, principal components analysis, Multidimensional Poverty Index, and a novel variable selection approach based on the use of random forests. Second, the selected wealth measure was combined with other relevant variables to form a more complete measure of household SES. We used child height-for-age Z-score (HAZ) as the outcome of interest. Results: Mean age of study children was 41 months, 52% were boys, and 42% were stunted. Using cross-validation, we found that random forests yielded the lowest prediction error when selecting assets as a measure of household wealth. The final SES index included access to improved water and sanitation, eight selected assets, maternal education, and household income (the WAMI index). A 25% difference in the WAMI index was positively associated with a difference of 0.38 standard deviations in HAZ (95% CI 0.22 to 0.55). Conclusions: Statistical learning methods such as random forests provide an alternative to principal components analysis in the development of SES scores. Results from this multicountry study demonstrate the validity of a simplified SES index. With further validation, this simplified index may provide a standard approach for SES adjustment across resource-limited settings. Keywords: Socioeconomic status, Child growth, Classification, Measurement * Correspondence: wcheckl1@jhmi.edu Fogarty International Center, National Institutes of Health, Bethesda, USA Program in Global Disease Epidemiology and Control, Department of International Health, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, USA Full list of author information is available at the end of the article © 2014 Psaki et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. Psaki et al. Population Health Metrics 2014, 12:8 Page 2 of 11 http://www.pophealthmetrics.com/content/12/1/8 Introduction whether variables that measure SES, such as ownership Socioeconomic status (SES) is a theoretical construct of specific assets, have the same meaning across popula- encompassing individual, household, and/or community tions. The DHS wealth index is derived using country- access to resources. It is commonly conceptualized as a specific data rather than globally pooled data. One combination of economic, social, and work status, mea- result is that a household in the poorest wealth quintile sured by income or wealth, education, and occupation, in Egypt might be wealthier than a household in the respectively [1,2]. SES has been linked to a wide range of richest wealth quintile in Ethiopia. Therefore, control- health-related exposures and outcomes, including child ling for SES in pooled analyses using this approach, undernutrition, chronic disease, and infection [2,3]. In a either by raw score or wealth quintile, is inappropriate. review of risk factors for adverse outcomes in child cogni- The United Nations Development Programme (UNDP) tive development, Walker and colleagues [4] conceptual- sought to overcome the challenge of comparing household ized poverty as underlying more proximal psychological SES across countries through the Multidimensional Pov- and biological risk factors, including maternal depression erty Index (MPI), introduced as an experimental measure and nutrient deficiencies. More recently, researchers have in the 2010 Human Development Report. The MPI in- highlighted connections between childhood SES and life- cludes three equally weighted dimensions of household time health outcomes, such as heart disease and chronic SES: education (years of schooling, school attendance), obstructive pulmonary disease [5]. health (child mortality, nutrition), and standard of living Literature on SES measurement distinguishes between (household attributes, asset ownership) [12]. wealth, or accumulated financial resources, and income, a The Malnutrition and Enteric Infections: Consequences measure of shorter-term access to capital [2]. Researchers for Child Health and Development (MAL-ED) study seeks have identified challenges in collecting income data, par- to explore relationships between early exposures to mal- ticularly in low-income settings, due to monthly fluctua- nutrition and enteric infections and their consequences tions, informal work, and reporting biases [6]. Recent for child growth and cognitive development across eight empirical work has drawn attention to the approach of sites. Geographic, cultural, and socioeconomic differences supplementing or replacing information on income with between these sites present an added challenge to develop- direct measures of wealth, such as household assets [7]. ing a measure of SES that is relevant in all sites. We Perhaps the most widespread approach to direct measure- sought to compare different approaches to measuring SES ment of household wealth is that used by the Demo- in resource-limited settings, and provide guidance for graphic and Health Surveys (DHS), implemented in more measuring SES accurately and simply in epidemiologic than 90 countries since 1984 [8]. Using nationally repre- studies of diverse populations. sentative data from India, Filmer and Pritchett [7] created an index based on household ownership of assets and Materials and methods housing materials to serve as a proxy for wealth. The Study setting resulting index was internally valid and coherent, and ro- This study took place at the eight field sites of the MAL- bust to the choice of assets. Using additional data sets ED study (see Table 1). Study sites are located in a from Indonesia, Nepal, and Pakistan, they further argued mix of rural, urban, and peri-urban areas of: Dhaka, that a composite asset index is as reliable as data on Bangladesh (BGD); Fortaleza, Brazil (BRF); Vellore, India household consumption and is less subject to measure- (INV); Bhaktapur, Nepal (NEB); Naushahro Feroze, ment error [7]. Their statistical approach, using princi- Pakistan (PKN); Loreto, Peru (PEL); Venda, South Africa pal components analysis (PCA), has since been adapted (SAV); and Haydom, Tanzania (TZH). Sites used a stan- to create a household wealth index in each DHS dataset dardized protocol for data collection. [8]. Concerns about this approach include its over- representation of urban settings, and its failure to dis- Study design tinguish between the poorest of the poor, particularly in Prior to beginning the ongoing cohort study, we con- rural areas [9]. Furthermore, this approach requires ducted a cross-sectional feasibility study to identify the lengthy surveys of household assets. Several studies optimal approach to measuring household SES. We ad- have found that rapid wealth appraisals requiring as few ministered a standardized survey including demographic, as four survey questions perform as well as the DHS socioeconomic status, and food insecurity questions to wealth index in categorizinghouseholds [10] and pre- 100 households in each of the eight field sites between dicting mortality [11]. September 2009 and August 2010. Households were ran- Multicountry studies pose an added challenge to domly selected from census results collected within the measuring SES. While approaches focused on asset previous year at each site. Households were eligible to par- ownership are often sufficient for homogenous popula- ticipate if they were located within the MAL-ED catch- tions, studies in more diverse populations must explore ment area and if a child aged 24 to 60 months lived in the Psaki et al. Population Health Metrics 2014, 12:8 Page 3 of 11 http://www.pophealthmetrics.com/content/12/1/8 Table 1 Description of MAL-ED study sites and mean WAMI scores Country Urban/rural Site description Mean (SD) WAMI score Brazil Urban Parque Universitário is an urban community inhabited by poor and middle class families. 0.80 (0.08) The community has approximately 33,000 people with 12% less than 5 years old. Of 288 children ≤3 years old, 31% have < −1 and 9% < −2 HAZ. Peru Peri-urban The site is peri-urban with an economic base in agriculture, extraction of forest products, 0.71 (0.11) and fishing. South Africa Rural The Dzimauli site is rural and mountainous, characterized by agricultural livelihood, low 0.70 (0.16) socioeconomic status, and poor infrastructure with waterfalls and many rivers across the villages. It is situated 25 km from the central business district. Nepal Peri-urban The study site Bhaktapur Municipality and adjoining villages are peri-urban areas, with safe 0.69 (0.12) drinking water and toilet facilities. The main economic base is agriculture. Bangladesh Urban Mirpur, an underprivileged community in Dhaka, is inhabited by poor and middle-class 0.55 (0.12) families. Residential and sanitary conditions are typical of any congested urban settlement. The investigators have ongoing research activities in the area. Pakistan Rural The Molhan study site in district Naushahro Feroze is in the southern Sindh province. The 0.52 (0.17) site is surrounded by fertile plains near the Indus river, predominantly rural communities, agricultural occupations, low socioeconomic class, and poor infrastructure, including mud houses. India Urban The study site is situated in a slum area in Vellore, which is a small city in Tamil Nadu in 0.43 (0.10) southern India. It is predominantly inhabited by poor families. The major occupation is manual labor in the market or construction work. Tanzania Rural The Haydom area is ethnically and geographically diverse, situated at approximately 1700 0.22 (0.11) meters above sea level and 300 kilometers from the nearest urban center. The study population is mainly agro-pastoralists. OVERALL 0.58 (0.22) The WAMI score (range 0 to 1) measures household socioeconomic status, including access to improved Water/sanitation, Assets, Maternal education, and Income. household. In households with multiple children in this Anthropometry age range, we randomly selected only one eligible child. Field workers measured the selected child aged 24 to Data collection lasted two to four weeks in each site. We 60 months for height and weight in each participating obtained ethical approval from the Institutional Review household. Trained field staff used a locally produced Boards at each of the participating research sites, the platform with sliding headboard to measure standing Johns Hopkins Bloomberg School of Public Health (Balti- height to the nearest 0.1 cm. They used digital scales to more, USA) and the University of Virginia School of measure weight to the nearest 100 grams. We used the Medicine (Charlottesville, USA). 2006 World Health Organization Multi-Country Growth Reference Study (WHO MGRS) to calculate height- for-age Z-scores (HAZ). Based on these standards, we Socioeconomic status survey defined stunting as a HAZ less than two standard devia- We adapted demographic and SES questions from the tions below the global median [15]. most recent DHS questionnaires [13]. Improved water and sanitation were based on World Health Organization definitions [14]. Site investigators reviewed question- Biostatistical methods naires and identified items that were problematic in their Our statistical analyses comprised two phases. First, we sites. Each site approved a final list of questions and identified the best approach to selecting and weight response categories and the associated data collection household assets as a proxy for wealth. Second, we com- procedures. Final demographic questions focused on age bined our wealth measure with other relevant variables and education of the head of household and child’s to form a more complete measure of household SES. In mother, as well as mother’s fertility history. The SES both phases we assessed the associations between SES/ section assessed household assets, housing materials, wealth measures and child HAZ for two reasons: 1) we water source and sanitation facilities, and ownership of were interested in directly comparing the predictive land or livestock. The survey also included a question on power of wealth/SES measures, and 2) assessing associa- monthly household income in local currency. The ques- tions between a construct of interest and other con- tionnaire was developed in English and translated into structs that are believed to be related theoretically or local languages as appropriate and back-translated for empirically is one way of assessing construct validity quality assurance. [16]. We chose HAZ rather than weight-for-height Psaki et al. Population Health Metrics 2014, 12:8 Page 4 of 11 http://www.pophealthmetrics.com/content/12/1/8 because the former is a better measure of chronic regression models. Leave-one-out cross-validation uses all deprivation, while the latter commonly indicates a com- observations except one to identify important variables for posite of acute and chronic deprivation [3]. In both classification, while the remaining observation is used as phases of analyses we were guided by a desire to identify the test set to measure the predictive error. This process is the simplest valid measure of wealth or SES in terms of repeated using each observation as the test set to calculate variables and computation required. the mean squared error (MSE) [17]. The approach with We compared four approaches to selecting and the smallest MSE predicts HAZ the most accurately. weighting indicators to measure household wealth: ma- We also calculated 10-fold cross validation (results not ternal education, PCA, MPI, and a novel variable selec- shown), which produced similar findings to leave-one- tion method based on the use of conditional random out cross-validation. The coefficient of determination forests [17]. We used maternal education as a baseline R represents the proportion of variability explained by to assess the added value of assets beyond this com- a statistical model. The approach with the largest coeffi- monly used proxy for household wealth [18,19]. Mater- cient of determination captures the most variability in nal education was constructed as a simple continuous HAZ [21]. The effect size represents the estimated measure of years of education completed by the child’s change in HAZ for each one-unit change in household mother at the time of the survey. To construct the PCA- wealth. Since the scales of each approach vary, we com- based SES index, we first selected a subset of dichotom- pared the effect of a 25% increase in each measure of ous indicators, including assets, housing materials, and household wealth. facilities, using Cronbach’s coefficient alpha. PCA was We then examined associations between each wealth then conducted on the tetrachoric correlation matrix of measurement approach and monthly household income. selected indicators Additional file 1: Table S1, and we We converted household income to USD using January used the first principal component as the SES score 1, 2010 exchange rates. Given the expected association for each household [20]. The MPI index, adapted from between household wealth and income, these analyses the UNDP approach based on available data, included provided evidence of the construct validity of each ap- the following indicators: maternal education (years of proach to measuring household wealth [22]. Based on schooling); health (any child has died); and standard the cumulative evidence from these analyses, we selected of living (electricity, water, sanitation, flooring, cooking one approach to measuring household wealth. fuel, and ownership of more than one of seven assets). The second phase of our analyses sought to incorporate Although the UNDP includes child nutritional status, we several aspects of SES: access to improved Water and sani- excluded this variable because it was our outcome of tation, the selected approach to measuring household interest. We weighted these three areas equally to create wealth (Assets), Maternal education, and Income (i.e. the a household wealth score [12]. Random forests (RF) are WAMI index). We included improved water and sanitation an expansion on classification trees using bootstrapping in response to guidance that SES measures should be based methods to generate multiple trees [17]. The RF ap- on hypothesized causal pathways in a study [23]. We then proach to measuring wealth used the same initial indica- examined the predictive power of this composite measure tors as the PCA method to ensure comparability of of household SES relative to HAZ using the criteria de- results Additional file 1: Table S1, i.e., so that differences scribed above. We used R 2.10.1 (www.r-project.org) and in predictive power could be attributed to the method STATA 12.1 (STATA Corp., College Station, USA) for stat- rather than the selection of assets. We used unsuper- istical analysis. vised learning with random forests to calculate condi- tional variable importance using the cforest package in Results R, which produces a variable importance rank in terms Study sample characteristics of their predictive value of a specified outcome (i.e., We surveyed a total of 800 households across all sites. HAZ). Ownership of a subset of indicators was summed One child had missing anthropometry and 10 were ex- to create household wealth scores. cluded for extreme HAZ values, resulting in a final sam- We then compared the three approaches (PCI, MPI, ple size of 789 households (99% of original sample). All and RF) with maternal education to measure household remaining observations had complete data on the vari- wealth and the strength association with HAZ. The fol- ables used for these analyses. Mean age of sampled chil- lowing criteria were used to compare the three wealth dren was 41 months (SD = 10.4); 52% of children were measurement approaches vs. maternal education: 1) male, ranging from 59% in Tanzania to 44% in Pakistan. leave-one-out cross-validation; 2) coefficient of determin- Overall, 42% of children were stunted, ranging from 8% ation (R ) values based on linear regression models with to 55% by site (Figure 1). Differences across sites were each wealth measure as the predictor and indicator vari- evidenced by variations in maternal education (from ables for each site; and 3) scaled effect sizes from the same 3.3 years in Pakistan to 10.1 years in South Africa) and Psaki et al. Population Health Metrics 2014, 12:8 Page 5 of 11 http://www.pophealthmetrics.com/content/12/1/8 Figure 1 Proportion of stunted children (height-for-age < −2 Z-scores) aged 24 to 60 months by study site. Study sites are: Dhaka, Bangladesh (BGD); Fortaleza, Brazil (BRF); Vellore, India (INV); Bhaktapur, Nepal (NEB); Naushahro Feroze, Pakistan (PKN); Loreto, Peru (PEL); Venda, South Africa (SAV); and Haydom, Tanzania (TZH). proportion with a bank account (from 2% in Tanzania to significant drop in internal consistency reliability. The 76% in South Africa) (see Table 2). Nearly all house- final 16 assets included were: iron, mattress, chair, sofa, holds, with the exception of those in Tanzania, reported cupboard, table, radio, computer, TV, sewing machine, improved water and sanitation. Bivariate associations be- mobile phone, fridge, bank account, separate kitchen, elec- tween stunting and either demographic or wealth indica- tricity, and people per room. Based on an approach similar tors (e.g., age, water source, sanitation facility, maternal to the scree plot used in PCA, where the magnitude of the education, separate kitchen, and people per room) dem- change between each value is used to select a cutoff point, onstrated the expected associations (Table 3). we ordered the 16 variables by their importance and selected the top eight for the RF measure. We compared Household wealth measurement the three wealth measurement approaches to mother’s Drawing on the Cronbach’s alpha results showing internal education in terms of predictive value (MSE), explained consistency and reliability, we selected 16 assets to use in variability (R ), and effect size (Table 4). The RF and PCA the PCA and RF analyses (final alpha = 0.86). We elimi- approaches performed better in terms of predictive value, nated variables with low variation between households explained variability, and effect size, and all of the wealth (defined as fewer than 10% of households in one cat- measurement approaches performed better than maternal egory), and variables, the inclusion of which led to a education. Table 2 Selected socioeconomic characteristics of households overall and by country (n = 789) Overall Bangladesh Brazil India Nepal Pakistan Peru South Africa Tanzania Sample size 789 99 98 100 100 98 99 96 99 Separate room for a kitchen 50% 10% 87% 23% 73% 27% 85% 74% 21% Household bank account 31% 23% 21% 10% 62% 39% 15% 76% 2% Mattress 58% 66% 98% 1% 99% 13% 82% 66% 39% Refrigerator 31% 12% 88% 3% 24% 27% 21% 78% 0% Wealth indicators TV 63% 55% 97% 69% 90% 61% 68% 68% 0% People per room (mean) 1.7 3.7 1.3 3.9 2.5 5.5 1.6 1.2 1.7 Table 57% 29% 86% 21% 65% 50% 100% 74% 33% Chair or bench 61% 37% 94% 59% 68% 21% 95% 95% 16% Education Mean maternal education (years) 6.4 3.7 7.8 6.7 6.6 3.3 7.8 10.1 5.3 Improved water source 86% 100% 100% 100% 98% 100% 98% 65% 28% Hygiene Improved sanitation facility 72% 100% 100% 37% 100% 74% 84% 84% 1% Wealth indicators are listed in terms of variable importance (highest to lowest) in predicting height-for-age Z-score, based on the random forests approach. Psaki et al. Population Health Metrics 2014, 12:8 Page 6 of 11 http://www.pophealthmetrics.com/content/12/1/8 Table 3 Relationship between selected indicators and middle group including the India, Peru, and Nepal sites; child stunting and a higher group including the South Africa and Brazil N % stunted p-value sites. Mean HAZ in Brazil was higher than would be pre- dicted by the regression line, while the opposite was true Sex for South Africa (Figure 2). Male 407 42.1 0.95 Female 382 41.9 Income and wealth Age Each wealth measure was also significantly associated 24-35 months 284 41.2 with monthly household income (Figure 3). These as- 36-47 months 243 49.0 0.01 sociations were strongest for the PCA and RF ap- 48-60 months 262 36.3 proaches to measuring wealth. The ranking of sites by Water source mean monthly household income followed a similar Not improved 109 58.7 pattern to wealth overall, with some notable depar- <0.001 tures. For example, the Pakistan site ranked lowest in Improved 680 39.3 terms of mean number of years of mother’seducation, Sanitation facility but ranked fifth out of eight sites in terms of monthly Not improved 218 49.5 income. When comparing wealth measured by PCA <0.01 Improved 571 39.1 with household income, however, the rankings were Maternal education nearly identical. When wealth was measured using the None 135 57.0 RF method, the South Africa and Nepal sites had nearly the same mean wealth score, but the South 1-5 years 174 43.1 <0.001 African site ranked higher in terms of mean monthly > 5 years 480 37.3 income. The associations between wealth measure and Separate kitchen monthly income in each site provide evidence of the No 396 51.0 construct validity of each measure of wealth, while the <0.001 Yes 393 32.8 changes in site rankings demonstrate the importance People per room of including both wealth and income in a complete measure of household SES. < 2 433 35.3 <0.001 ≥ 2 356 50.0 Choice of wealth measure We chose the RF approach to measuring wealth. We All tested wealth measures were significantly associated ruled out the approach of using maternal education with HAZ, with slightly stronger associations for the PCA alone because it performed poorly relative to the other and RF measures. The ranking of mean wealth score by site measures, it was inconsistent with our theoretical under- followed a similar pattern across measures: a low group in- standing of wealth, and availability of education to cluding the Tanzania, Pakistan, and Bangladesh sites; a women is dependent on societal values and investments, Table 4 Results comparing three wealth measurement approaches and mother’s education in terms of value in predicting child HAZ (n = 789) Wealth method 1: Wealth method 2: Wealth method 3: Wealth method 4: Full SES Measure: maternal principal multidimensional asset selection by Water and sanitation, education components poverty Index random forests Assets, Maternal education, analysis Income (WAMI) index Mean squared error 1.39 1.37 1.39 1.37 1.37 (MSE) from leave-one-out cross validation Adjusted R 18.67% 19.98% 18.89% 19.86% 20.19% Effect size (95% CI) 0.123 (0.029-0.217) 0.290 (0.164-0.416) 0.149 (0.061-0.237) 0.220 (0.121-0.319) 0.384 (0.222-0.546) Change in HAZ p = 0.01 p < 0.001 p = 0.002 p < 0.001 p < 0.001 associated with a 25% increase in wealth score Number of variables 1 16 variables 14 variables Selected 8 out of 16 Summarized 12 variables summarized into 1 summarized into 1 variables, and summarized into 1 these 8 variables into 1 Psaki et al. Population Health Metrics 2014, 12:8 Page 7 of 11 http://www.pophealthmetrics.com/content/12/1/8 Figure 2 Relationship between height-for-age of children aged 24 to 60 months and four different types of socioeconomic status indices. Each data point represents the average score for either height-for-age or each socioeconomic status index at each study site. The red line indicates a linear regression fit of the relationship between height-for-age and socioeconomic status indices using site as the unit of analysis. Study sites are: Dhaka, Bangladesh (BGD); Fortaleza, Brazil (BRF); Vellore, India (INV); Bhaktapur, Nepal (NEB); Naushahro Feroze, Pakistan (PKN); Loreto, Peru (PEL); Venda, South Africa (SAV); and Haydom, Tanzania (TZH). which is highly heterogeneous across low- and middle- be used initially to identify important variables, which income countries such as those in these studies. PCA can then be combined using a simple approach. While and RF performed better than MPI in terms of predict- the PCA approach retains all variables, potentially result- ive validity and variation explained in HAZ (Table 4). ing in the inclusion of variables that are irrelevant in Both the PCA and RF approaches require statistical cal- some study sites, the RF approach keeps only the vari- culations. PCA requires use of different weights for each ables that are most closely related to the outcome of variable that have no inherent meaning, whereas RF can interest. Figure 3 Relationship between household income and four different types of socioeconomic status indices. Each data point represents the average household income or average score for each socioeconomic status index at each study site. The red line indicates a linear regression fit of the relationship between household income and socioeconomic status indices using site as the unit of analysis. Study sites are: Dhaka, Bangladesh (BGD); Fortaleza, Brazil (BRF); Vellore, India (INV); Bhaktapur, Nepal (NEB); Naushahro Feroze, Pakistan (PKN); Loreto, Peru (PEL); Venda, South Africa (SAV); and Haydom, Tanzania (TZH). Psaki et al. Population Health Metrics 2014, 12:8 Page 8 of 11 http://www.pophealthmetrics.com/content/12/1/8 Development of an integrated socioeconomic status of socioeconomic position: (a) actual resources, and (b) index status, meaning prestige- or rank-related characteristics.” We formulated a complete index of household SES, in- We use the term SES rather than socioeconomic pos- cluding the following components: access to improved ition because the latter is intended to explicitly include Water and sanitation, wealth measured by a set of prestige-based measures, as linked to social class. Our eight Assets, Maternal education, and monthly house- measure focuses only on the actual resources in a house- hold Income (i.e. WAMI index). In Table 5, we show hold because we did not collect prestige-based indicators our approach to combining these four components in our study. Hackman and Farah [25] emphasize the into a complete measure, and in Figure 4 we show as- importance of a clear conceptual framework driving the sociations between the WAMI index and HAZ. We measurement of socioeconomic status. In the MAL-ED compare the WAMI index with the wealth-only mea- study, which focuses on the relationships between en- sures in Table 4. While the WAMI index is compar- teric infections in infancy and subsequent growth and able to wealth alone in terms of predictive value and cognitive development, access to improved water and variation explained, there is a notable difference in the sanitation sources are important risk factors that are effect size. A 25% difference in the WAMI score is closely related to SES. Our final measure requires 12 positively associated with a difference of 0.38 SD in variables, most of which are easily collected across di- HAZ (95% CI 0.22 to 0.55). In contrast, a 25% differ- verse settings, which is a requirement in order for an ence in wealth score alone (RF approach) is only asso- SES measure to be applicable in a multicountry study. ciated with a 0.22 SD difference in HAZ (95% CI 0.12 Compared to maternal education alone, as well as to 0.32). more complete measures of wealth, the WAMI index demonstrated a significantly stronger association with Discussion HAZ across our eight study sites. These results provide We compared three approaches to measuring household evidence of the importance of using a robust measure of wealth with maternal education and selected the random household SES, even as an adjustment factor, rather than forests approach due to its consistent association with a measure of wealth or education alone. Based on ana- HAZ and simplicity of use and interpretation of selected lyses of a nationally representative sample in the United assets relative to PCA. Although maternal education was States, Braveman and colleagues [25] found that, de- the simplest approach as it only required one variable, it pending on the choice of income, education, or both as did not reflect theoretical models of household SES, an a control for SES, the ranking of racial groups by odds important consideration in choice of variables [24]. In of poor health changed dramatically. Many studies also addition, maternal education is strongly influenced by seek to simplify measurement of SES by identifying the culture. Our data support previous evidence that, while variables that are most strongly associated with their wealth and education are important components of SES, outcomes. Daly and colleagues [27] found that wealth in many settings they do not measure the same expo- and income indicators were more strongly associated sures [25]. with mortality than education and occupation indicators We then combined our selected wealth measure with in a United States cohort. In settings where collection of access to improved water and sanitation, maternal edu- income data is not feasible, an expanded measure of SES cation, and household income to form a complete SES including wealth, education, and water and sanitation is measure. Krieger and colleagues [26] argue that the term still a significant improvement over wealth or maternal SES “blurs distinctions between two different aspects education alone (results not shown). Table 5 Calculation of the Water/sanitation, Assets, Maternal education, and Income (WAMI) index Description Range Water/ Using World Health Organization definitions of access to improved water and improved sanitation, households with 0-8 sanitation access to improved water or improved sanitation are assigned a score of 4 for each. Households without access to improved water or improved sanitation are assigned a score of 0 for each. These scores were summed. Assets Eight priority assets were selected using random forests with HAZ as the outcome. For each asset, households were 0-8 assigned a 1 if they have the asset and 0 if they do not have the asset. These scores were summed. Maternal Each child’s mother provided the number of years of schooling she had completed, ranging from 0 to 16 years. This 0-8 education number was divided by 2. Income Monthly household income was converted to US dollars using the exchange rate from January 1, 2010. Income was 0-8 divided into octiles using the following scores and cutoffs: 1 (0–26), 2 (26.01-47), 3 (47.01-72), 4 (72.01-106), 5 (106.01-135), 6 (135.01-200), 7 (200.01-293), 8 (293+). TOTAL Scores in water and sanitation, assets, mother’s education, and income were summed then divided by 32. 0-1 Psaki et al. Population Health Metrics 2014, 12:8 Page 9 of 11 http://www.pophealthmetrics.com/content/12/1/8 Figure 4 Scatterplot of height-for-age and WAMI index score stratified by study site. The black line represents a linear regression fit of the relationship between height-for-age and WAMI index score at each site. Study sites are: Dhaka, Bangladesh (BGD); Fortaleza, Brazil (BRF); Vellore, India (INV); Bhaktapur, Nepal (NEB); Naushahro Feroze, Pakistan (PKN); Loreto, Peru (PEL); Venda, South Africa (SAV); and Haydom, Tanzania (TZH). Our study has a number of important strengths. It in- to compare our proposed measure, particularly in a mul- cludes data from households from eight country sites lo- ticountry study setting. However, associations between cated in South Asia, sub-Saharan Africa, and Latin America the WAMI index and HAZ demonstrate construct valid- to derive a multicountry index. The experienced staff in ity of the measure, since we expect these constructs to these eight sites used a standardized protocol with identical be associated theoretically. Another limitation of our questionnaires to collect demographic and SES information. study is that we do not have a measure of occupation. The survey included extensive questions related to SES, due Much of the work on socioeconomic status that includes to an a priori interest in identifying an appropriate approach occupation as a key component is based on research in to measuring this construct. Our analyses systematically high-income settings. In many low-income settings, this compared both commonly used and new approaches to concept does not distinguish between households as measuring wealth and SES. Our final selection of a measure readily, either due to homogeneity or instability of income of socioeconomic status balances statistical and theoretical sources. Alternatives include caste or religious group. strength with feasibility in a field research setting. While we collected data on these groupings, we did not The results of this study should be considered in light feel that they were sufficiently comparable across the eight of some limitations. Although the MAL-ED study sites study sites. Similarly, indicators of prestige or rank-related are located in eight diverse country settings, they are characteristics have been included in measures of socio- not nationally representative samples. Therefore, the re- economic position [26], but these are unavailable in our sults cannot be generalized to national comparisons or dataset. With the exception of data on access to electricity, compared to country-level indices such as the DHS. our available SES indicators were measured at the house- More broadly, the resulting components of an ideal hold level, rather than the community level. Previous re- measure of SES are likely to vary across settings and search has shown that the addition of community-level study objectives. Consistencies across our study sites in- variables to SES measures can be helpful in exploring dicate that it is possible to identify measures of SES that trends and inequalities in health outcomes [28]. are relevant in diverse settings. However, the actual vari- In summary, novel classification approaches such as ables that form an ideal measure will be informed by the random forests provide an alternative to the more widely study populations, settings, and research questions. Ra- used PCA for the measurement of household wealth. ther than selecting indicators to be used by all studies, However, assets alone are not sufficient to capture the our findings are intended to demonstrate the importance full domain of socioeconomic status. We developed a of developing a measure of socioeconomic status that is simplified socioeconomic index that combines measures theoretically sound and contextually relevant, especially of improved water and sanitation, assets, maternal educa- in multisite or multicountry studies. There is no gold tion, and household income that may be applicable to a standard measure of socioeconomic status against which multicountry setting. We believe that this measure is an Psaki et al. Population Health Metrics 2014, 12:8 Page 10 of 11 http://www.pophealthmetrics.com/content/12/1/8 improvement over the commonly used PCA-based wealth Vellore, India), Sushil John (Christian Medical College, Vellore, India), Sudhir Babji (Christian Medical College, Vellore, India), Mohan Venkata Raghava measurement approach for several reasons: 1) It is a robust (Christian Medical College, Vellore, India), Anuradha Rose (Christian Medical measure that more fully reflects a theoretical understanding College, Vellore, India), Beena Kurien (Christian Medical College, Vellore, of SES; 2) it reduces the data collection burden by India), Anuradha Bose (Christian Medical College, Vellore, India), Jayaprakash Muliyil (Christian Medical College, Vellore, India), Anup Ramachandran highlighting a priority set of indicators for measurement, in (Christian Medical College, Vellore, India); in Nepal: Carl J Mason (Armed contrast to commonly used PCA approaches, which require Forces Research Institute of Medical Sciences, Bangkok, Thailand), Prakash collecting data on a full set of indicators, even if some are Sunder Shrestha (Institute of Medicine, Tribuhvan University, Kathmandu, Nepal), Sanjaya Kumar Shrestha (Walter Reed/AFRIMS Research Unit, irrelevant; and 3) it is computationally simple to apply, once Kathmandu, Nepal), Ladaporn Bodhidatta (Armed Forces Research Institute of thepriorityassets havebeen selected usingthe random Medical Sciences, Bangkok, Thailand), Ram Krishna Chandyo (Institute of forests technique. With further validation, this simplified Medicine, Tribuhvan University, Kathmandu, Nepal), Rita Shrestha (Institute of Medicine, Tribuhvan University, Kathmandu, Nepal), Binob Shrestha (Walter WAMI index may provide a standardized approach for Reed/AFRIMS Research Unit, Kathmandu), Tor Strand (University of Bergen, adjustment across diverse study populations. Bergen, Norway), Manjeswori Ulak (Institute of Medicine, Tribuhvan University, Kathmandu, Nepal); in Pakistan: Zulfiqar A Bhutta (Aga Khan University, Naushahro Feroze, Pakistan), Anita K M Zaidi (Aga Khan University, Additional file Naushahro Feroze, Pakistan), Sajid Soofi (Aga Khan University, Naushahro Feroze, Pakistan), Ali Turab (Aga Khan University, Naushahro Feroze, Pakistan), Additional file 1: Table S1. Variables included in initial PCA and Didar Alam (Aga Khan University, Naushahro Feroze, Pakistan), Shahida random forests analyses. Sites are listed from left to right, starting with Qureshi (Aga Khan University, Naushahro Feroze, Pakistan), Aisha K Yousafzai the highest mean WAMI score (Brazil) and ending with the lowest mean (Aga Khan University, Naushahro Feroze, Pakistan), Asad Ali (Aga Khan WAMI score (Tanzania). University, Naushahro Feroze, Pakistan), Imran Ahmed (Aga Khan University, Naushahro Feroze, Pakistan), Sajad Memon (Aga Khan University, Naushahro Feroze, Pakistan), Muneera Rasheed (Aga Khan University, Naushahro Feroze, Competing interests Pakistan). North America: in the United States: Michael Gottlieb (Foundation The authors have no competing interests to declare. for the NIH, Bethesda, MD, USA), Mark Miller (Fogarty International Center/ National Institutes of Health, Bethesda, MD, USA), Karen H. Tountas Authors’ contribution (Foundation for the NIH, Bethesda, MD, USA), Rebecca Blank (Foundation for SP and WC contributed equally to the conception, design and analysis of the NIH, Bethesda, MD, USA), Dennis Lang (Fogarty International Center/ data, and interpretation of findings. SP led the writing of the manuscript. SP, National Institutes of Health, Bethesda, MD, USA), Stacey Knobler (Fogarty JS, MM, MG, ZB, TA, AS, PB, SJ, GK, MK, AL, PS, ES, and WC contributed International Center/National Institutes of Health, Bethesda, MD, USA), equally to study design and data acquisition. All authors read and approved Monica McGrath (Fogarty International Center/National Institutes of Health, the final manuscript. WC had ultimate oversight over the study design, data Bethesda, MD, USA), Stephanie Richard (Fogarty International Center/National analysis, and writing of this manuscript. Institutes of Health, Bethesda, MD, USA), Jessica Seidman (Fogarty International Center/National Institutes of Health, Bethesda, MD, USA), Zeba Acknowledgments Rasmussen (Fogarty International Center/National Institutes of Health, MAL-ED Network Investigators by region: Africa: in South Africa: Pascal Bethesda, MD, USA), Ramya Ambikapathi (Fogarty International Center/ Bessong (University of Venda, Thohoyandou, South Africa), Angelina Mapula National Institutes of Health, Bethesda, MD, USA), Benjamin McCormick (University of Venda, Thohoyandou, South Africa), Emanuel Nyathi (University (Fogarty International Center/National Institutes of Health, Bethesda, MD, of Venda, Thohoyandou, South Africa), Cloupas Mahopo (University of Venda, USA), Stephanie Psaki (Fogarty International Center/National Institutes of Thohoyandou, South Africa), Amidou Samie (University of Venda, Health, Bethesda, MD, USA), Vivek Charu (Fogarty International Center/ Thohoyandou, South Africa), Cebisa Nesamvuni (University of Venda, National Institutes of Health, Bethesda, MD, USA), Jhanelle Graham (Fogarty Thohoyandou, South Africa); in Tanzania: Erling Svensen (Haydom Lutheran International Center/National Institutes of Health, Bethesda, MD, USA), Hospital, University of Bergen, Norway), Estomih R. Mduma (Haydom Gaurvika Nayyar (Fogarty International Center/National Institutes of Health, Lutheran Hospital, Haydom, Tanzania), Crystal L. Patil (University of Illinois, Bethesda, MD, USA), Viyada Doan (Fogarty International Center/National Urbana-Champaign, IL, USA), Caroline Amour (Haydom Lutheran Hospital, Institutes of Health, Bethesda, MD, USA), Leyfou Dabo (Fogarty International Haydom, Tanzania). South America: in Brazil: Aldo A. M. Lima (Universidade Center/National Institutes of Health, Bethesda, MD, USA), Danny Carreon Federal do Ceara, Fortaleza, Brazil), Reinaldo B. Oriá (Universidade Federal do (Fogarty International Center/National Institutes of Health, Bethesda, MD, Ceara, Fortaleza, Brazil), Noélia L. Lima (Universidade Federal do Ceara, USA), Archana Mohale (Fogarty International Center/National Institutes of Fortaleza, Brazil), Alberto M. Soares, (Universidade Federal do Ceara, Fortaleza, Health, Bethesda, MD, USA), Christel Host (Fogarty International Center/ Brazil), Alexandre H. Bindá (Universidade Federal do Ceara, Fortaleza, Brazil), National Institutes of Health, Bethesda, MD, USA), Dick Guerrant (University of Ila F. N. Lima (Universidade Federal do Ceara, Fortaleza, Brazil), Josiane S. Virginia, Charlottesville, VA, USA), Bill Petri (University of Virginia, Charlottesville, Quetz (Universidade Federal do Ceara, Fortaleza, Brazil), Milena L. Moraes VA, USA), Eric Houpt (University of Virginia, Charlottesville, VA, USA), Jean Gratz (Universidade Federal do Ceara, Fortaleza, Brazil), Bruna L. L. Maciel (University of Virginia, Charlottesville, VA, USA), Leah Barrett (University of Virginia, (Universidade Federal do Ceara, Fortaleza, Brazil), Hilda Costa (Universidade Charlottesville, VA, USA), Rebecca Scharf (University of Virginia, Charlottesville, Federal do Ceara, Fortaleza, Brazil), Jose Quirino Filho (Universidade Federal VA, USA), Laura Caulfield (Johns Hopkins University, Baltimore, MD, USA), William do Ceara, Fortaleza, Brazil), Álvaro J. M. Leite (Universidade Federal do Ceara, Checkley (Johns Hopkins University, Baltimore, MD, USA), Margaret Kosek (Johns Fortaleza, Brazil), Francisco B. Mota (Universidade Federal do Ceara, Fortaleza, Hopkins University, Baltimore, MD, USA), Pablo Penataro Yori (Johns Hopkins Brazil), Alessandra F. Di Moura (Universidade Federal do Ceara, Fortaleza, University, Baltimore, MD, USA), Gwenyth Lee (Johns Hopkins University, Brazil); in Peru: Maribel Paredes Olortegui (A.B. PRISMA, Iquitos, Peru), Cesar Baltimore, MD, USA), Ping Chen (Johns Hopkins University, Baltimore, MD, USA), Banda Chavez (A.B. PRISMA, Iquitos, Peru), Dixner Rengifo Trigoso (A.B. Robert Black (Johns Hopkins University, Baltimore, MD, USA), Laura Murray-Kolb PRISMA, Iquitos, Peru), Julian Torres Flores (A.B. PRISMA, Iquitos, Peru), Angel (Pennsylvania State University, University Park, PA, USA), Barbara Schaefer Orbe Vasquez (A.B. PRISMA, Iquitos, Peru), Silvia Rengifo Pinedo (A.B. PRISMA, (Pennsylvania State University, University Park, PA, USA), William Pan (Duke Iquitos, Peru), Angel Mendez Acosta (A.B. PRISMA, Iquitos, Peru). South Asia: University, Durham, NC, USA). in Bangladesh: Tahmeed Ahmed (ICDDR-B, Dhaka, Bangladesh), Rashidul Haque (ICDDR-B, Dhaka, Bangladesh), AM Shamsir Ahmed (ICDDR-B, Dhaka, Bangladesh), Munirul Islam, (ICDDR-B, Dhaka, Bangladesh), Iqbal Hossain Funding (ICDDR-B, Dhaka, Bangladesh), Mustafa Mahfuz (ICDDR-B, Dhaka, Bangladesh), The Interactions of Malnutrition & Enteric Infections: Consequences for Child Dinesh Mondol (ICDDR-B, Dhaka, Bangladesh), Fahmida Tofail (ICDDR-B, Health and Development (MAL-ED) study is carried out as a collaborative Dhaka, Bangladesh); in India: Gagandeep Kang (Christian Medical College, project and supported by the Bill & Melinda Gates Foundation, the Psaki et al. Population Health Metrics 2014, 12:8 Page 11 of 11 http://www.pophealthmetrics.com/content/12/1/8 Foundation for the NIH, and the National Institutes of Health/Fogarty 16. Cronbach LJ, Meehl PE: Construct Validity in Psychological Tests. International Center. The authors thank the staff and participants of the Psychol Bull 1955, 52:281–302. MAL-ED Network Project for their important contributions. William Checkley 17. Hastie T, Tibshirani R, Friedman J: Elements of Statistical Learning: Data was further supported by a Pathway to Independence Award (R00HL096955) Mining, Inference, and Prediction. Secondth edition. New York, NY: from the National Heart, Lung and Blood Institute, National Institutes of Health. Springer-Verlag; 2009. 18. Monteiro CA, Conde WL, Popkin BM: Obesity and inequities in health in Author details the developing world. Int J Obes 2004, 28:1181–1186. Fogarty International Center, National Institutes of Health, Bethesda, USA. 19. Desai S, Alva S: Maternal education and child health: Is there a strong Program in Global Disease Epidemiology and Control, Department of causal relationship? Demography 1998, 35:71–81. International Health, Bloomberg School of Public Health, Johns Hopkins 20. Vyas S, Kumaranayake L: Constructing socio-economic status indices: how University, Baltimore, USA. Science Division, Foundation for the National to use principal components analysis. Health Policy Plan 2006, 21:459–468. Institutes of Health, Bethesda, USA. Division of Women and Child Health, 21. Pagano M, Gauvreau K: Principles of Biostatistics. 2nd edition. Pacific Grove, Aga Khan University, Karachi, Pakistan. Division of Nutrition and Food CA: Duxbury Thomson Learning; 2000. Security, International Centers for Diarrheal Disease Research, Matlab, 22. DeVellis R: Scale Development: theory and applications. 2nd edition. Bangladesh. HIV/AIDS and Global Health Research Programme, University of Thousand Oaks, CA: SAGE Publications; 2003. Venda, Thohoyandou, South Africa. Christian Medical College, Vellore, India. 23. Braveman P, Cubbin C, Marchi K, Egerter S, Chavez G: Measuring Clinical Research Unit and Institute of Biomedicine, Federal University of socioeconomic status/position in studies of racial/ethnic disparities: Ceara, Fortaleza, Brazil. Institute of Medicine, Tribhuvan University, maternal and infant health. Public Health Rep 2001, 116:449–463. Kathmandu, Nepal. Centre for International Health, University of Bergen, 24. Hackman DA, Farah MJ: Socioeconomic status and the developing brain. Bergen, Norway. Haydom Lutheran Hospital, Haydom, Tanzania. Trends Cogn Sci 2009, 13:65–73. 25. Braveman PA, Cubbin C, Egerter S, Chideya S, Marchi KS, Metzler M, Posner S: Received: 2 July 2013 Accepted: 14 February 2014 Socioeconomic status in health research: one size does not fit all. JAMA Published: 21 March 2014 2005, 294:2879–2888. 26. Krieger N, Williams DR, Moss NE: Measuring social class in US public health research: concepts, methodologies, and guidelines. Annu Rev Public Health 1997, 18:341–378. References 27. Daly MC, Duncan GJ, McDonough P, Williams DR: Optimal indicators of 1. Adler NE, Boyce T, Chesney MA, Cohen S, Folkman S, Kahn RL, Syme SL: socioeconomic status for health research. Am J Public Health 2002, Socioeconomic status and health. The challenge of the gradient. Am 92:1151–1157. Psychol 1994, 49:15–24. 28. Fosto JC, Kuate-Defo B: Measuring Socioeconomic Status in Health 2. Pollack CE, Chideya S, Cubbin C, Williams B, Dekker M, Braveman P: Should Research in Developing Countries: Should we be focusing on health studies measure wealth? A systematic review. Am J Prev Med 2007, households, communities or both? Soc Indicators Res 2005, 72:189–237. 33:250–264. 3. Black RE, Allen LH, Bhutta ZA, Caulfield LE, de Onis M, Ezzati M, Mathers C, doi:10.1186/1478-7954-12-8 Rivera J; Maternal and Child Undernutrition Study Group: Maternal and Cite this article as: Psaki et al.: Measuring socioeconomic status in child undernutrition: global and regional exposures and health multicountry studies: results from the eight-country MAL-ED study. consequences. Lancet 2008, 371:243–260. Population Health Metrics 2014 12:8. 4. Walker SP, Wachs TD, Gardner JM, Lozoff B, Wasserman GA, Pollitt E, Carter JA; International Child Development Steering Group: Child development: risk factors for adverse outcomes in developing countries. Lancet 2007, 369:145–157. 5. Galobardes B, Lynch JW, Davey Smith G: Childhood socioeconomic circumstances and cause-specific mortality in adulthood: systematic review and interpretation. Epidemiol Rev 2004, 26:7–21. 6. O’Donnell O, van Doorslaer E, Wagstaff A, Lindelow M: Analyzing Health Equity using Household Survey Data: a Guide to Techniques and their Implementation. Washington DC: World Development Institute, World Bank; 2007. www.worldbank.org/analyzinghealthequity (ISBN: 0-8213-6933-4). 7. Filmer D, Pritchett LH: Estimating wealth effects without expenditure data-or tears: an application to educational enrollments in states of India. Demography 2001, 38:115–132. 8. Rutstein SO, Johnson K: The DHS Wealth Index: DHS comparative reports No. 6. Calverton, MD: ORC Macro; 2004. 9. Rutstein SO: The DHS Wealth Index: Approaches for rural and urban areas. DHS Working Papers No. 60.. Calverton, MD: Macro International; 2008. 10. Pitchforth E, van Teijlingen E, Graham W, Fitzmaurice A: Development of a proxy wealth index for women utilizing emergency obstetric care in Bangladesh. Health Policy Plan 2007, 22:311–319. 11. van Bodegom D, May L, Kuningas M, Kaptijn R, Thomése F, Meij HJ, Submit your next manuscript to BioMed Central Amankwa J, Westendorp RG: Socio-economic status by rapid appraisal is and take full advantage of: highly correlated with mortality risks in rural Africa. Trans R Soc Trop Med Hyg 2009, 103:795–800. • Convenient online submission 12. Alkire S, Santos ME: Acute Multidimensional Poverty: A new index for • Thorough peer review developing countries. Oxford: Oxford Poverty & Human Development Initiative; 2010. OPHI Working Paper 38. • No space constraints or color figure charges 13. USAID: Demographic and Health Surveys. Available at: http://www. • Immediate publication on acceptance measuredhs.com/. Accessed December 10, 2012. 14. World Health Organization: Child Growth Standards: Anthropometry Macros. • Inclusion in PubMed, CAS, Scopus and Google Scholar 2011. Available at: http://www.who.int/childgrowth/software/en/index.htm. • Research which is freely available for redistribution 15. de Onis M, Garza C, Victora CG: The WHO Multicentre Growth Reference Study: strategy for developing a new international growth reference. Submit your manuscript at Forum Nutr 2003, 56:238–240. www.biomedcentral.com/submit http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Population Health Metrics Springer Journals

Loading next page...
 
/lp/springer-journals/measuring-socioeconomic-status-in-multicountry-studies-results-from-OXWry6X9xL

References (42)

Publisher
Springer Journals
Copyright
Copyright © 2014 by Psaki et al.; licensee BioMed Central Ltd.
Subject
Medicine & Public Health; Public Health; Epidemiology
eISSN
1478-7954
DOI
10.1186/1478-7954-12-8
pmid
24656134
Publisher site
See Article on Publisher Site

Abstract

Background: There is no standardized approach to comparing socioeconomic status (SES) across multiple sites in epidemiological studies. This is particularly problematic when cross-country comparisons are of interest. We sought to develop a simple measure of SES that would perform well across diverse, resource-limited settings. Methods: A cross-sectional study was conducted with 800 children aged 24 to 60 months across eight resource-limited settings. Parents were asked to respond to a household SES questionnaire, and the height of each child was measured. A statistical analysis was done in two phases. First, the best approach for selecting and weighting household assets as a proxy for wealth was identified. We compared four approaches to measuring wealth: maternal education, principal components analysis, Multidimensional Poverty Index, and a novel variable selection approach based on the use of random forests. Second, the selected wealth measure was combined with other relevant variables to form a more complete measure of household SES. We used child height-for-age Z-score (HAZ) as the outcome of interest. Results: Mean age of study children was 41 months, 52% were boys, and 42% were stunted. Using cross-validation, we found that random forests yielded the lowest prediction error when selecting assets as a measure of household wealth. The final SES index included access to improved water and sanitation, eight selected assets, maternal education, and household income (the WAMI index). A 25% difference in the WAMI index was positively associated with a difference of 0.38 standard deviations in HAZ (95% CI 0.22 to 0.55). Conclusions: Statistical learning methods such as random forests provide an alternative to principal components analysis in the development of SES scores. Results from this multicountry study demonstrate the validity of a simplified SES index. With further validation, this simplified index may provide a standard approach for SES adjustment across resource-limited settings. Keywords: Socioeconomic status, Child growth, Classification, Measurement * Correspondence: wcheckl1@jhmi.edu Fogarty International Center, National Institutes of Health, Bethesda, USA Program in Global Disease Epidemiology and Control, Department of International Health, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, USA Full list of author information is available at the end of the article © 2014 Psaki et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. Psaki et al. Population Health Metrics 2014, 12:8 Page 2 of 11 http://www.pophealthmetrics.com/content/12/1/8 Introduction whether variables that measure SES, such as ownership Socioeconomic status (SES) is a theoretical construct of specific assets, have the same meaning across popula- encompassing individual, household, and/or community tions. The DHS wealth index is derived using country- access to resources. It is commonly conceptualized as a specific data rather than globally pooled data. One combination of economic, social, and work status, mea- result is that a household in the poorest wealth quintile sured by income or wealth, education, and occupation, in Egypt might be wealthier than a household in the respectively [1,2]. SES has been linked to a wide range of richest wealth quintile in Ethiopia. Therefore, control- health-related exposures and outcomes, including child ling for SES in pooled analyses using this approach, undernutrition, chronic disease, and infection [2,3]. In a either by raw score or wealth quintile, is inappropriate. review of risk factors for adverse outcomes in child cogni- The United Nations Development Programme (UNDP) tive development, Walker and colleagues [4] conceptual- sought to overcome the challenge of comparing household ized poverty as underlying more proximal psychological SES across countries through the Multidimensional Pov- and biological risk factors, including maternal depression erty Index (MPI), introduced as an experimental measure and nutrient deficiencies. More recently, researchers have in the 2010 Human Development Report. The MPI in- highlighted connections between childhood SES and life- cludes three equally weighted dimensions of household time health outcomes, such as heart disease and chronic SES: education (years of schooling, school attendance), obstructive pulmonary disease [5]. health (child mortality, nutrition), and standard of living Literature on SES measurement distinguishes between (household attributes, asset ownership) [12]. wealth, or accumulated financial resources, and income, a The Malnutrition and Enteric Infections: Consequences measure of shorter-term access to capital [2]. Researchers for Child Health and Development (MAL-ED) study seeks have identified challenges in collecting income data, par- to explore relationships between early exposures to mal- ticularly in low-income settings, due to monthly fluctua- nutrition and enteric infections and their consequences tions, informal work, and reporting biases [6]. Recent for child growth and cognitive development across eight empirical work has drawn attention to the approach of sites. Geographic, cultural, and socioeconomic differences supplementing or replacing information on income with between these sites present an added challenge to develop- direct measures of wealth, such as household assets [7]. ing a measure of SES that is relevant in all sites. We Perhaps the most widespread approach to direct measure- sought to compare different approaches to measuring SES ment of household wealth is that used by the Demo- in resource-limited settings, and provide guidance for graphic and Health Surveys (DHS), implemented in more measuring SES accurately and simply in epidemiologic than 90 countries since 1984 [8]. Using nationally repre- studies of diverse populations. sentative data from India, Filmer and Pritchett [7] created an index based on household ownership of assets and Materials and methods housing materials to serve as a proxy for wealth. The Study setting resulting index was internally valid and coherent, and ro- This study took place at the eight field sites of the MAL- bust to the choice of assets. Using additional data sets ED study (see Table 1). Study sites are located in a from Indonesia, Nepal, and Pakistan, they further argued mix of rural, urban, and peri-urban areas of: Dhaka, that a composite asset index is as reliable as data on Bangladesh (BGD); Fortaleza, Brazil (BRF); Vellore, India household consumption and is less subject to measure- (INV); Bhaktapur, Nepal (NEB); Naushahro Feroze, ment error [7]. Their statistical approach, using princi- Pakistan (PKN); Loreto, Peru (PEL); Venda, South Africa pal components analysis (PCA), has since been adapted (SAV); and Haydom, Tanzania (TZH). Sites used a stan- to create a household wealth index in each DHS dataset dardized protocol for data collection. [8]. Concerns about this approach include its over- representation of urban settings, and its failure to dis- Study design tinguish between the poorest of the poor, particularly in Prior to beginning the ongoing cohort study, we con- rural areas [9]. Furthermore, this approach requires ducted a cross-sectional feasibility study to identify the lengthy surveys of household assets. Several studies optimal approach to measuring household SES. We ad- have found that rapid wealth appraisals requiring as few ministered a standardized survey including demographic, as four survey questions perform as well as the DHS socioeconomic status, and food insecurity questions to wealth index in categorizinghouseholds [10] and pre- 100 households in each of the eight field sites between dicting mortality [11]. September 2009 and August 2010. Households were ran- Multicountry studies pose an added challenge to domly selected from census results collected within the measuring SES. While approaches focused on asset previous year at each site. Households were eligible to par- ownership are often sufficient for homogenous popula- ticipate if they were located within the MAL-ED catch- tions, studies in more diverse populations must explore ment area and if a child aged 24 to 60 months lived in the Psaki et al. Population Health Metrics 2014, 12:8 Page 3 of 11 http://www.pophealthmetrics.com/content/12/1/8 Table 1 Description of MAL-ED study sites and mean WAMI scores Country Urban/rural Site description Mean (SD) WAMI score Brazil Urban Parque Universitário is an urban community inhabited by poor and middle class families. 0.80 (0.08) The community has approximately 33,000 people with 12% less than 5 years old. Of 288 children ≤3 years old, 31% have < −1 and 9% < −2 HAZ. Peru Peri-urban The site is peri-urban with an economic base in agriculture, extraction of forest products, 0.71 (0.11) and fishing. South Africa Rural The Dzimauli site is rural and mountainous, characterized by agricultural livelihood, low 0.70 (0.16) socioeconomic status, and poor infrastructure with waterfalls and many rivers across the villages. It is situated 25 km from the central business district. Nepal Peri-urban The study site Bhaktapur Municipality and adjoining villages are peri-urban areas, with safe 0.69 (0.12) drinking water and toilet facilities. The main economic base is agriculture. Bangladesh Urban Mirpur, an underprivileged community in Dhaka, is inhabited by poor and middle-class 0.55 (0.12) families. Residential and sanitary conditions are typical of any congested urban settlement. The investigators have ongoing research activities in the area. Pakistan Rural The Molhan study site in district Naushahro Feroze is in the southern Sindh province. The 0.52 (0.17) site is surrounded by fertile plains near the Indus river, predominantly rural communities, agricultural occupations, low socioeconomic class, and poor infrastructure, including mud houses. India Urban The study site is situated in a slum area in Vellore, which is a small city in Tamil Nadu in 0.43 (0.10) southern India. It is predominantly inhabited by poor families. The major occupation is manual labor in the market or construction work. Tanzania Rural The Haydom area is ethnically and geographically diverse, situated at approximately 1700 0.22 (0.11) meters above sea level and 300 kilometers from the nearest urban center. The study population is mainly agro-pastoralists. OVERALL 0.58 (0.22) The WAMI score (range 0 to 1) measures household socioeconomic status, including access to improved Water/sanitation, Assets, Maternal education, and Income. household. In households with multiple children in this Anthropometry age range, we randomly selected only one eligible child. Field workers measured the selected child aged 24 to Data collection lasted two to four weeks in each site. We 60 months for height and weight in each participating obtained ethical approval from the Institutional Review household. Trained field staff used a locally produced Boards at each of the participating research sites, the platform with sliding headboard to measure standing Johns Hopkins Bloomberg School of Public Health (Balti- height to the nearest 0.1 cm. They used digital scales to more, USA) and the University of Virginia School of measure weight to the nearest 100 grams. We used the Medicine (Charlottesville, USA). 2006 World Health Organization Multi-Country Growth Reference Study (WHO MGRS) to calculate height- for-age Z-scores (HAZ). Based on these standards, we Socioeconomic status survey defined stunting as a HAZ less than two standard devia- We adapted demographic and SES questions from the tions below the global median [15]. most recent DHS questionnaires [13]. Improved water and sanitation were based on World Health Organization definitions [14]. Site investigators reviewed question- Biostatistical methods naires and identified items that were problematic in their Our statistical analyses comprised two phases. First, we sites. Each site approved a final list of questions and identified the best approach to selecting and weight response categories and the associated data collection household assets as a proxy for wealth. Second, we com- procedures. Final demographic questions focused on age bined our wealth measure with other relevant variables and education of the head of household and child’s to form a more complete measure of household SES. In mother, as well as mother’s fertility history. The SES both phases we assessed the associations between SES/ section assessed household assets, housing materials, wealth measures and child HAZ for two reasons: 1) we water source and sanitation facilities, and ownership of were interested in directly comparing the predictive land or livestock. The survey also included a question on power of wealth/SES measures, and 2) assessing associa- monthly household income in local currency. The ques- tions between a construct of interest and other con- tionnaire was developed in English and translated into structs that are believed to be related theoretically or local languages as appropriate and back-translated for empirically is one way of assessing construct validity quality assurance. [16]. We chose HAZ rather than weight-for-height Psaki et al. Population Health Metrics 2014, 12:8 Page 4 of 11 http://www.pophealthmetrics.com/content/12/1/8 because the former is a better measure of chronic regression models. Leave-one-out cross-validation uses all deprivation, while the latter commonly indicates a com- observations except one to identify important variables for posite of acute and chronic deprivation [3]. In both classification, while the remaining observation is used as phases of analyses we were guided by a desire to identify the test set to measure the predictive error. This process is the simplest valid measure of wealth or SES in terms of repeated using each observation as the test set to calculate variables and computation required. the mean squared error (MSE) [17]. The approach with We compared four approaches to selecting and the smallest MSE predicts HAZ the most accurately. weighting indicators to measure household wealth: ma- We also calculated 10-fold cross validation (results not ternal education, PCA, MPI, and a novel variable selec- shown), which produced similar findings to leave-one- tion method based on the use of conditional random out cross-validation. The coefficient of determination forests [17]. We used maternal education as a baseline R represents the proportion of variability explained by to assess the added value of assets beyond this com- a statistical model. The approach with the largest coeffi- monly used proxy for household wealth [18,19]. Mater- cient of determination captures the most variability in nal education was constructed as a simple continuous HAZ [21]. The effect size represents the estimated measure of years of education completed by the child’s change in HAZ for each one-unit change in household mother at the time of the survey. To construct the PCA- wealth. Since the scales of each approach vary, we com- based SES index, we first selected a subset of dichotom- pared the effect of a 25% increase in each measure of ous indicators, including assets, housing materials, and household wealth. facilities, using Cronbach’s coefficient alpha. PCA was We then examined associations between each wealth then conducted on the tetrachoric correlation matrix of measurement approach and monthly household income. selected indicators Additional file 1: Table S1, and we We converted household income to USD using January used the first principal component as the SES score 1, 2010 exchange rates. Given the expected association for each household [20]. The MPI index, adapted from between household wealth and income, these analyses the UNDP approach based on available data, included provided evidence of the construct validity of each ap- the following indicators: maternal education (years of proach to measuring household wealth [22]. Based on schooling); health (any child has died); and standard the cumulative evidence from these analyses, we selected of living (electricity, water, sanitation, flooring, cooking one approach to measuring household wealth. fuel, and ownership of more than one of seven assets). The second phase of our analyses sought to incorporate Although the UNDP includes child nutritional status, we several aspects of SES: access to improved Water and sani- excluded this variable because it was our outcome of tation, the selected approach to measuring household interest. We weighted these three areas equally to create wealth (Assets), Maternal education, and Income (i.e. the a household wealth score [12]. Random forests (RF) are WAMI index). We included improved water and sanitation an expansion on classification trees using bootstrapping in response to guidance that SES measures should be based methods to generate multiple trees [17]. The RF ap- on hypothesized causal pathways in a study [23]. We then proach to measuring wealth used the same initial indica- examined the predictive power of this composite measure tors as the PCA method to ensure comparability of of household SES relative to HAZ using the criteria de- results Additional file 1: Table S1, i.e., so that differences scribed above. We used R 2.10.1 (www.r-project.org) and in predictive power could be attributed to the method STATA 12.1 (STATA Corp., College Station, USA) for stat- rather than the selection of assets. We used unsuper- istical analysis. vised learning with random forests to calculate condi- tional variable importance using the cforest package in Results R, which produces a variable importance rank in terms Study sample characteristics of their predictive value of a specified outcome (i.e., We surveyed a total of 800 households across all sites. HAZ). Ownership of a subset of indicators was summed One child had missing anthropometry and 10 were ex- to create household wealth scores. cluded for extreme HAZ values, resulting in a final sam- We then compared the three approaches (PCI, MPI, ple size of 789 households (99% of original sample). All and RF) with maternal education to measure household remaining observations had complete data on the vari- wealth and the strength association with HAZ. The fol- ables used for these analyses. Mean age of sampled chil- lowing criteria were used to compare the three wealth dren was 41 months (SD = 10.4); 52% of children were measurement approaches vs. maternal education: 1) male, ranging from 59% in Tanzania to 44% in Pakistan. leave-one-out cross-validation; 2) coefficient of determin- Overall, 42% of children were stunted, ranging from 8% ation (R ) values based on linear regression models with to 55% by site (Figure 1). Differences across sites were each wealth measure as the predictor and indicator vari- evidenced by variations in maternal education (from ables for each site; and 3) scaled effect sizes from the same 3.3 years in Pakistan to 10.1 years in South Africa) and Psaki et al. Population Health Metrics 2014, 12:8 Page 5 of 11 http://www.pophealthmetrics.com/content/12/1/8 Figure 1 Proportion of stunted children (height-for-age < −2 Z-scores) aged 24 to 60 months by study site. Study sites are: Dhaka, Bangladesh (BGD); Fortaleza, Brazil (BRF); Vellore, India (INV); Bhaktapur, Nepal (NEB); Naushahro Feroze, Pakistan (PKN); Loreto, Peru (PEL); Venda, South Africa (SAV); and Haydom, Tanzania (TZH). proportion with a bank account (from 2% in Tanzania to significant drop in internal consistency reliability. The 76% in South Africa) (see Table 2). Nearly all house- final 16 assets included were: iron, mattress, chair, sofa, holds, with the exception of those in Tanzania, reported cupboard, table, radio, computer, TV, sewing machine, improved water and sanitation. Bivariate associations be- mobile phone, fridge, bank account, separate kitchen, elec- tween stunting and either demographic or wealth indica- tricity, and people per room. Based on an approach similar tors (e.g., age, water source, sanitation facility, maternal to the scree plot used in PCA, where the magnitude of the education, separate kitchen, and people per room) dem- change between each value is used to select a cutoff point, onstrated the expected associations (Table 3). we ordered the 16 variables by their importance and selected the top eight for the RF measure. We compared Household wealth measurement the three wealth measurement approaches to mother’s Drawing on the Cronbach’s alpha results showing internal education in terms of predictive value (MSE), explained consistency and reliability, we selected 16 assets to use in variability (R ), and effect size (Table 4). The RF and PCA the PCA and RF analyses (final alpha = 0.86). We elimi- approaches performed better in terms of predictive value, nated variables with low variation between households explained variability, and effect size, and all of the wealth (defined as fewer than 10% of households in one cat- measurement approaches performed better than maternal egory), and variables, the inclusion of which led to a education. Table 2 Selected socioeconomic characteristics of households overall and by country (n = 789) Overall Bangladesh Brazil India Nepal Pakistan Peru South Africa Tanzania Sample size 789 99 98 100 100 98 99 96 99 Separate room for a kitchen 50% 10% 87% 23% 73% 27% 85% 74% 21% Household bank account 31% 23% 21% 10% 62% 39% 15% 76% 2% Mattress 58% 66% 98% 1% 99% 13% 82% 66% 39% Refrigerator 31% 12% 88% 3% 24% 27% 21% 78% 0% Wealth indicators TV 63% 55% 97% 69% 90% 61% 68% 68% 0% People per room (mean) 1.7 3.7 1.3 3.9 2.5 5.5 1.6 1.2 1.7 Table 57% 29% 86% 21% 65% 50% 100% 74% 33% Chair or bench 61% 37% 94% 59% 68% 21% 95% 95% 16% Education Mean maternal education (years) 6.4 3.7 7.8 6.7 6.6 3.3 7.8 10.1 5.3 Improved water source 86% 100% 100% 100% 98% 100% 98% 65% 28% Hygiene Improved sanitation facility 72% 100% 100% 37% 100% 74% 84% 84% 1% Wealth indicators are listed in terms of variable importance (highest to lowest) in predicting height-for-age Z-score, based on the random forests approach. Psaki et al. Population Health Metrics 2014, 12:8 Page 6 of 11 http://www.pophealthmetrics.com/content/12/1/8 Table 3 Relationship between selected indicators and middle group including the India, Peru, and Nepal sites; child stunting and a higher group including the South Africa and Brazil N % stunted p-value sites. Mean HAZ in Brazil was higher than would be pre- dicted by the regression line, while the opposite was true Sex for South Africa (Figure 2). Male 407 42.1 0.95 Female 382 41.9 Income and wealth Age Each wealth measure was also significantly associated 24-35 months 284 41.2 with monthly household income (Figure 3). These as- 36-47 months 243 49.0 0.01 sociations were strongest for the PCA and RF ap- 48-60 months 262 36.3 proaches to measuring wealth. The ranking of sites by Water source mean monthly household income followed a similar Not improved 109 58.7 pattern to wealth overall, with some notable depar- <0.001 tures. For example, the Pakistan site ranked lowest in Improved 680 39.3 terms of mean number of years of mother’seducation, Sanitation facility but ranked fifth out of eight sites in terms of monthly Not improved 218 49.5 income. When comparing wealth measured by PCA <0.01 Improved 571 39.1 with household income, however, the rankings were Maternal education nearly identical. When wealth was measured using the None 135 57.0 RF method, the South Africa and Nepal sites had nearly the same mean wealth score, but the South 1-5 years 174 43.1 <0.001 African site ranked higher in terms of mean monthly > 5 years 480 37.3 income. The associations between wealth measure and Separate kitchen monthly income in each site provide evidence of the No 396 51.0 construct validity of each measure of wealth, while the <0.001 Yes 393 32.8 changes in site rankings demonstrate the importance People per room of including both wealth and income in a complete measure of household SES. < 2 433 35.3 <0.001 ≥ 2 356 50.0 Choice of wealth measure We chose the RF approach to measuring wealth. We All tested wealth measures were significantly associated ruled out the approach of using maternal education with HAZ, with slightly stronger associations for the PCA alone because it performed poorly relative to the other and RF measures. The ranking of mean wealth score by site measures, it was inconsistent with our theoretical under- followed a similar pattern across measures: a low group in- standing of wealth, and availability of education to cluding the Tanzania, Pakistan, and Bangladesh sites; a women is dependent on societal values and investments, Table 4 Results comparing three wealth measurement approaches and mother’s education in terms of value in predicting child HAZ (n = 789) Wealth method 1: Wealth method 2: Wealth method 3: Wealth method 4: Full SES Measure: maternal principal multidimensional asset selection by Water and sanitation, education components poverty Index random forests Assets, Maternal education, analysis Income (WAMI) index Mean squared error 1.39 1.37 1.39 1.37 1.37 (MSE) from leave-one-out cross validation Adjusted R 18.67% 19.98% 18.89% 19.86% 20.19% Effect size (95% CI) 0.123 (0.029-0.217) 0.290 (0.164-0.416) 0.149 (0.061-0.237) 0.220 (0.121-0.319) 0.384 (0.222-0.546) Change in HAZ p = 0.01 p < 0.001 p = 0.002 p < 0.001 p < 0.001 associated with a 25% increase in wealth score Number of variables 1 16 variables 14 variables Selected 8 out of 16 Summarized 12 variables summarized into 1 summarized into 1 variables, and summarized into 1 these 8 variables into 1 Psaki et al. Population Health Metrics 2014, 12:8 Page 7 of 11 http://www.pophealthmetrics.com/content/12/1/8 Figure 2 Relationship between height-for-age of children aged 24 to 60 months and four different types of socioeconomic status indices. Each data point represents the average score for either height-for-age or each socioeconomic status index at each study site. The red line indicates a linear regression fit of the relationship between height-for-age and socioeconomic status indices using site as the unit of analysis. Study sites are: Dhaka, Bangladesh (BGD); Fortaleza, Brazil (BRF); Vellore, India (INV); Bhaktapur, Nepal (NEB); Naushahro Feroze, Pakistan (PKN); Loreto, Peru (PEL); Venda, South Africa (SAV); and Haydom, Tanzania (TZH). which is highly heterogeneous across low- and middle- be used initially to identify important variables, which income countries such as those in these studies. PCA can then be combined using a simple approach. While and RF performed better than MPI in terms of predict- the PCA approach retains all variables, potentially result- ive validity and variation explained in HAZ (Table 4). ing in the inclusion of variables that are irrelevant in Both the PCA and RF approaches require statistical cal- some study sites, the RF approach keeps only the vari- culations. PCA requires use of different weights for each ables that are most closely related to the outcome of variable that have no inherent meaning, whereas RF can interest. Figure 3 Relationship between household income and four different types of socioeconomic status indices. Each data point represents the average household income or average score for each socioeconomic status index at each study site. The red line indicates a linear regression fit of the relationship between household income and socioeconomic status indices using site as the unit of analysis. Study sites are: Dhaka, Bangladesh (BGD); Fortaleza, Brazil (BRF); Vellore, India (INV); Bhaktapur, Nepal (NEB); Naushahro Feroze, Pakistan (PKN); Loreto, Peru (PEL); Venda, South Africa (SAV); and Haydom, Tanzania (TZH). Psaki et al. Population Health Metrics 2014, 12:8 Page 8 of 11 http://www.pophealthmetrics.com/content/12/1/8 Development of an integrated socioeconomic status of socioeconomic position: (a) actual resources, and (b) index status, meaning prestige- or rank-related characteristics.” We formulated a complete index of household SES, in- We use the term SES rather than socioeconomic pos- cluding the following components: access to improved ition because the latter is intended to explicitly include Water and sanitation, wealth measured by a set of prestige-based measures, as linked to social class. Our eight Assets, Maternal education, and monthly house- measure focuses only on the actual resources in a house- hold Income (i.e. WAMI index). In Table 5, we show hold because we did not collect prestige-based indicators our approach to combining these four components in our study. Hackman and Farah [25] emphasize the into a complete measure, and in Figure 4 we show as- importance of a clear conceptual framework driving the sociations between the WAMI index and HAZ. We measurement of socioeconomic status. In the MAL-ED compare the WAMI index with the wealth-only mea- study, which focuses on the relationships between en- sures in Table 4. While the WAMI index is compar- teric infections in infancy and subsequent growth and able to wealth alone in terms of predictive value and cognitive development, access to improved water and variation explained, there is a notable difference in the sanitation sources are important risk factors that are effect size. A 25% difference in the WAMI score is closely related to SES. Our final measure requires 12 positively associated with a difference of 0.38 SD in variables, most of which are easily collected across di- HAZ (95% CI 0.22 to 0.55). In contrast, a 25% differ- verse settings, which is a requirement in order for an ence in wealth score alone (RF approach) is only asso- SES measure to be applicable in a multicountry study. ciated with a 0.22 SD difference in HAZ (95% CI 0.12 Compared to maternal education alone, as well as to 0.32). more complete measures of wealth, the WAMI index demonstrated a significantly stronger association with Discussion HAZ across our eight study sites. These results provide We compared three approaches to measuring household evidence of the importance of using a robust measure of wealth with maternal education and selected the random household SES, even as an adjustment factor, rather than forests approach due to its consistent association with a measure of wealth or education alone. Based on ana- HAZ and simplicity of use and interpretation of selected lyses of a nationally representative sample in the United assets relative to PCA. Although maternal education was States, Braveman and colleagues [25] found that, de- the simplest approach as it only required one variable, it pending on the choice of income, education, or both as did not reflect theoretical models of household SES, an a control for SES, the ranking of racial groups by odds important consideration in choice of variables [24]. In of poor health changed dramatically. Many studies also addition, maternal education is strongly influenced by seek to simplify measurement of SES by identifying the culture. Our data support previous evidence that, while variables that are most strongly associated with their wealth and education are important components of SES, outcomes. Daly and colleagues [27] found that wealth in many settings they do not measure the same expo- and income indicators were more strongly associated sures [25]. with mortality than education and occupation indicators We then combined our selected wealth measure with in a United States cohort. In settings where collection of access to improved water and sanitation, maternal edu- income data is not feasible, an expanded measure of SES cation, and household income to form a complete SES including wealth, education, and water and sanitation is measure. Krieger and colleagues [26] argue that the term still a significant improvement over wealth or maternal SES “blurs distinctions between two different aspects education alone (results not shown). Table 5 Calculation of the Water/sanitation, Assets, Maternal education, and Income (WAMI) index Description Range Water/ Using World Health Organization definitions of access to improved water and improved sanitation, households with 0-8 sanitation access to improved water or improved sanitation are assigned a score of 4 for each. Households without access to improved water or improved sanitation are assigned a score of 0 for each. These scores were summed. Assets Eight priority assets were selected using random forests with HAZ as the outcome. For each asset, households were 0-8 assigned a 1 if they have the asset and 0 if they do not have the asset. These scores were summed. Maternal Each child’s mother provided the number of years of schooling she had completed, ranging from 0 to 16 years. This 0-8 education number was divided by 2. Income Monthly household income was converted to US dollars using the exchange rate from January 1, 2010. Income was 0-8 divided into octiles using the following scores and cutoffs: 1 (0–26), 2 (26.01-47), 3 (47.01-72), 4 (72.01-106), 5 (106.01-135), 6 (135.01-200), 7 (200.01-293), 8 (293+). TOTAL Scores in water and sanitation, assets, mother’s education, and income were summed then divided by 32. 0-1 Psaki et al. Population Health Metrics 2014, 12:8 Page 9 of 11 http://www.pophealthmetrics.com/content/12/1/8 Figure 4 Scatterplot of height-for-age and WAMI index score stratified by study site. The black line represents a linear regression fit of the relationship between height-for-age and WAMI index score at each site. Study sites are: Dhaka, Bangladesh (BGD); Fortaleza, Brazil (BRF); Vellore, India (INV); Bhaktapur, Nepal (NEB); Naushahro Feroze, Pakistan (PKN); Loreto, Peru (PEL); Venda, South Africa (SAV); and Haydom, Tanzania (TZH). Our study has a number of important strengths. It in- to compare our proposed measure, particularly in a mul- cludes data from households from eight country sites lo- ticountry study setting. However, associations between cated in South Asia, sub-Saharan Africa, and Latin America the WAMI index and HAZ demonstrate construct valid- to derive a multicountry index. The experienced staff in ity of the measure, since we expect these constructs to these eight sites used a standardized protocol with identical be associated theoretically. Another limitation of our questionnaires to collect demographic and SES information. study is that we do not have a measure of occupation. The survey included extensive questions related to SES, due Much of the work on socioeconomic status that includes to an a priori interest in identifying an appropriate approach occupation as a key component is based on research in to measuring this construct. Our analyses systematically high-income settings. In many low-income settings, this compared both commonly used and new approaches to concept does not distinguish between households as measuring wealth and SES. Our final selection of a measure readily, either due to homogeneity or instability of income of socioeconomic status balances statistical and theoretical sources. Alternatives include caste or religious group. strength with feasibility in a field research setting. While we collected data on these groupings, we did not The results of this study should be considered in light feel that they were sufficiently comparable across the eight of some limitations. Although the MAL-ED study sites study sites. Similarly, indicators of prestige or rank-related are located in eight diverse country settings, they are characteristics have been included in measures of socio- not nationally representative samples. Therefore, the re- economic position [26], but these are unavailable in our sults cannot be generalized to national comparisons or dataset. With the exception of data on access to electricity, compared to country-level indices such as the DHS. our available SES indicators were measured at the house- More broadly, the resulting components of an ideal hold level, rather than the community level. Previous re- measure of SES are likely to vary across settings and search has shown that the addition of community-level study objectives. Consistencies across our study sites in- variables to SES measures can be helpful in exploring dicate that it is possible to identify measures of SES that trends and inequalities in health outcomes [28]. are relevant in diverse settings. However, the actual vari- In summary, novel classification approaches such as ables that form an ideal measure will be informed by the random forests provide an alternative to the more widely study populations, settings, and research questions. Ra- used PCA for the measurement of household wealth. ther than selecting indicators to be used by all studies, However, assets alone are not sufficient to capture the our findings are intended to demonstrate the importance full domain of socioeconomic status. We developed a of developing a measure of socioeconomic status that is simplified socioeconomic index that combines measures theoretically sound and contextually relevant, especially of improved water and sanitation, assets, maternal educa- in multisite or multicountry studies. There is no gold tion, and household income that may be applicable to a standard measure of socioeconomic status against which multicountry setting. We believe that this measure is an Psaki et al. Population Health Metrics 2014, 12:8 Page 10 of 11 http://www.pophealthmetrics.com/content/12/1/8 improvement over the commonly used PCA-based wealth Vellore, India), Sushil John (Christian Medical College, Vellore, India), Sudhir Babji (Christian Medical College, Vellore, India), Mohan Venkata Raghava measurement approach for several reasons: 1) It is a robust (Christian Medical College, Vellore, India), Anuradha Rose (Christian Medical measure that more fully reflects a theoretical understanding College, Vellore, India), Beena Kurien (Christian Medical College, Vellore, of SES; 2) it reduces the data collection burden by India), Anuradha Bose (Christian Medical College, Vellore, India), Jayaprakash Muliyil (Christian Medical College, Vellore, India), Anup Ramachandran highlighting a priority set of indicators for measurement, in (Christian Medical College, Vellore, India); in Nepal: Carl J Mason (Armed contrast to commonly used PCA approaches, which require Forces Research Institute of Medical Sciences, Bangkok, Thailand), Prakash collecting data on a full set of indicators, even if some are Sunder Shrestha (Institute of Medicine, Tribuhvan University, Kathmandu, Nepal), Sanjaya Kumar Shrestha (Walter Reed/AFRIMS Research Unit, irrelevant; and 3) it is computationally simple to apply, once Kathmandu, Nepal), Ladaporn Bodhidatta (Armed Forces Research Institute of thepriorityassets havebeen selected usingthe random Medical Sciences, Bangkok, Thailand), Ram Krishna Chandyo (Institute of forests technique. With further validation, this simplified Medicine, Tribuhvan University, Kathmandu, Nepal), Rita Shrestha (Institute of Medicine, Tribuhvan University, Kathmandu, Nepal), Binob Shrestha (Walter WAMI index may provide a standardized approach for Reed/AFRIMS Research Unit, Kathmandu), Tor Strand (University of Bergen, adjustment across diverse study populations. Bergen, Norway), Manjeswori Ulak (Institute of Medicine, Tribuhvan University, Kathmandu, Nepal); in Pakistan: Zulfiqar A Bhutta (Aga Khan University, Naushahro Feroze, Pakistan), Anita K M Zaidi (Aga Khan University, Additional file Naushahro Feroze, Pakistan), Sajid Soofi (Aga Khan University, Naushahro Feroze, Pakistan), Ali Turab (Aga Khan University, Naushahro Feroze, Pakistan), Additional file 1: Table S1. Variables included in initial PCA and Didar Alam (Aga Khan University, Naushahro Feroze, Pakistan), Shahida random forests analyses. Sites are listed from left to right, starting with Qureshi (Aga Khan University, Naushahro Feroze, Pakistan), Aisha K Yousafzai the highest mean WAMI score (Brazil) and ending with the lowest mean (Aga Khan University, Naushahro Feroze, Pakistan), Asad Ali (Aga Khan WAMI score (Tanzania). University, Naushahro Feroze, Pakistan), Imran Ahmed (Aga Khan University, Naushahro Feroze, Pakistan), Sajad Memon (Aga Khan University, Naushahro Feroze, Pakistan), Muneera Rasheed (Aga Khan University, Naushahro Feroze, Competing interests Pakistan). North America: in the United States: Michael Gottlieb (Foundation The authors have no competing interests to declare. for the NIH, Bethesda, MD, USA), Mark Miller (Fogarty International Center/ National Institutes of Health, Bethesda, MD, USA), Karen H. Tountas Authors’ contribution (Foundation for the NIH, Bethesda, MD, USA), Rebecca Blank (Foundation for SP and WC contributed equally to the conception, design and analysis of the NIH, Bethesda, MD, USA), Dennis Lang (Fogarty International Center/ data, and interpretation of findings. SP led the writing of the manuscript. SP, National Institutes of Health, Bethesda, MD, USA), Stacey Knobler (Fogarty JS, MM, MG, ZB, TA, AS, PB, SJ, GK, MK, AL, PS, ES, and WC contributed International Center/National Institutes of Health, Bethesda, MD, USA), equally to study design and data acquisition. All authors read and approved Monica McGrath (Fogarty International Center/National Institutes of Health, the final manuscript. WC had ultimate oversight over the study design, data Bethesda, MD, USA), Stephanie Richard (Fogarty International Center/National analysis, and writing of this manuscript. Institutes of Health, Bethesda, MD, USA), Jessica Seidman (Fogarty International Center/National Institutes of Health, Bethesda, MD, USA), Zeba Acknowledgments Rasmussen (Fogarty International Center/National Institutes of Health, MAL-ED Network Investigators by region: Africa: in South Africa: Pascal Bethesda, MD, USA), Ramya Ambikapathi (Fogarty International Center/ Bessong (University of Venda, Thohoyandou, South Africa), Angelina Mapula National Institutes of Health, Bethesda, MD, USA), Benjamin McCormick (University of Venda, Thohoyandou, South Africa), Emanuel Nyathi (University (Fogarty International Center/National Institutes of Health, Bethesda, MD, of Venda, Thohoyandou, South Africa), Cloupas Mahopo (University of Venda, USA), Stephanie Psaki (Fogarty International Center/National Institutes of Thohoyandou, South Africa), Amidou Samie (University of Venda, Health, Bethesda, MD, USA), Vivek Charu (Fogarty International Center/ Thohoyandou, South Africa), Cebisa Nesamvuni (University of Venda, National Institutes of Health, Bethesda, MD, USA), Jhanelle Graham (Fogarty Thohoyandou, South Africa); in Tanzania: Erling Svensen (Haydom Lutheran International Center/National Institutes of Health, Bethesda, MD, USA), Hospital, University of Bergen, Norway), Estomih R. Mduma (Haydom Gaurvika Nayyar (Fogarty International Center/National Institutes of Health, Lutheran Hospital, Haydom, Tanzania), Crystal L. Patil (University of Illinois, Bethesda, MD, USA), Viyada Doan (Fogarty International Center/National Urbana-Champaign, IL, USA), Caroline Amour (Haydom Lutheran Hospital, Institutes of Health, Bethesda, MD, USA), Leyfou Dabo (Fogarty International Haydom, Tanzania). South America: in Brazil: Aldo A. M. Lima (Universidade Center/National Institutes of Health, Bethesda, MD, USA), Danny Carreon Federal do Ceara, Fortaleza, Brazil), Reinaldo B. Oriá (Universidade Federal do (Fogarty International Center/National Institutes of Health, Bethesda, MD, Ceara, Fortaleza, Brazil), Noélia L. Lima (Universidade Federal do Ceara, USA), Archana Mohale (Fogarty International Center/National Institutes of Fortaleza, Brazil), Alberto M. Soares, (Universidade Federal do Ceara, Fortaleza, Health, Bethesda, MD, USA), Christel Host (Fogarty International Center/ Brazil), Alexandre H. Bindá (Universidade Federal do Ceara, Fortaleza, Brazil), National Institutes of Health, Bethesda, MD, USA), Dick Guerrant (University of Ila F. N. Lima (Universidade Federal do Ceara, Fortaleza, Brazil), Josiane S. Virginia, Charlottesville, VA, USA), Bill Petri (University of Virginia, Charlottesville, Quetz (Universidade Federal do Ceara, Fortaleza, Brazil), Milena L. Moraes VA, USA), Eric Houpt (University of Virginia, Charlottesville, VA, USA), Jean Gratz (Universidade Federal do Ceara, Fortaleza, Brazil), Bruna L. L. Maciel (University of Virginia, Charlottesville, VA, USA), Leah Barrett (University of Virginia, (Universidade Federal do Ceara, Fortaleza, Brazil), Hilda Costa (Universidade Charlottesville, VA, USA), Rebecca Scharf (University of Virginia, Charlottesville, Federal do Ceara, Fortaleza, Brazil), Jose Quirino Filho (Universidade Federal VA, USA), Laura Caulfield (Johns Hopkins University, Baltimore, MD, USA), William do Ceara, Fortaleza, Brazil), Álvaro J. M. Leite (Universidade Federal do Ceara, Checkley (Johns Hopkins University, Baltimore, MD, USA), Margaret Kosek (Johns Fortaleza, Brazil), Francisco B. Mota (Universidade Federal do Ceara, Fortaleza, Hopkins University, Baltimore, MD, USA), Pablo Penataro Yori (Johns Hopkins Brazil), Alessandra F. Di Moura (Universidade Federal do Ceara, Fortaleza, University, Baltimore, MD, USA), Gwenyth Lee (Johns Hopkins University, Brazil); in Peru: Maribel Paredes Olortegui (A.B. PRISMA, Iquitos, Peru), Cesar Baltimore, MD, USA), Ping Chen (Johns Hopkins University, Baltimore, MD, USA), Banda Chavez (A.B. PRISMA, Iquitos, Peru), Dixner Rengifo Trigoso (A.B. Robert Black (Johns Hopkins University, Baltimore, MD, USA), Laura Murray-Kolb PRISMA, Iquitos, Peru), Julian Torres Flores (A.B. PRISMA, Iquitos, Peru), Angel (Pennsylvania State University, University Park, PA, USA), Barbara Schaefer Orbe Vasquez (A.B. PRISMA, Iquitos, Peru), Silvia Rengifo Pinedo (A.B. PRISMA, (Pennsylvania State University, University Park, PA, USA), William Pan (Duke Iquitos, Peru), Angel Mendez Acosta (A.B. PRISMA, Iquitos, Peru). South Asia: University, Durham, NC, USA). in Bangladesh: Tahmeed Ahmed (ICDDR-B, Dhaka, Bangladesh), Rashidul Haque (ICDDR-B, Dhaka, Bangladesh), AM Shamsir Ahmed (ICDDR-B, Dhaka, Bangladesh), Munirul Islam, (ICDDR-B, Dhaka, Bangladesh), Iqbal Hossain Funding (ICDDR-B, Dhaka, Bangladesh), Mustafa Mahfuz (ICDDR-B, Dhaka, Bangladesh), The Interactions of Malnutrition & Enteric Infections: Consequences for Child Dinesh Mondol (ICDDR-B, Dhaka, Bangladesh), Fahmida Tofail (ICDDR-B, Health and Development (MAL-ED) study is carried out as a collaborative Dhaka, Bangladesh); in India: Gagandeep Kang (Christian Medical College, project and supported by the Bill & Melinda Gates Foundation, the Psaki et al. Population Health Metrics 2014, 12:8 Page 11 of 11 http://www.pophealthmetrics.com/content/12/1/8 Foundation for the NIH, and the National Institutes of Health/Fogarty 16. Cronbach LJ, Meehl PE: Construct Validity in Psychological Tests. International Center. The authors thank the staff and participants of the Psychol Bull 1955, 52:281–302. MAL-ED Network Project for their important contributions. William Checkley 17. Hastie T, Tibshirani R, Friedman J: Elements of Statistical Learning: Data was further supported by a Pathway to Independence Award (R00HL096955) Mining, Inference, and Prediction. Secondth edition. New York, NY: from the National Heart, Lung and Blood Institute, National Institutes of Health. Springer-Verlag; 2009. 18. Monteiro CA, Conde WL, Popkin BM: Obesity and inequities in health in Author details the developing world. Int J Obes 2004, 28:1181–1186. Fogarty International Center, National Institutes of Health, Bethesda, USA. 19. Desai S, Alva S: Maternal education and child health: Is there a strong Program in Global Disease Epidemiology and Control, Department of causal relationship? Demography 1998, 35:71–81. International Health, Bloomberg School of Public Health, Johns Hopkins 20. Vyas S, Kumaranayake L: Constructing socio-economic status indices: how University, Baltimore, USA. Science Division, Foundation for the National to use principal components analysis. Health Policy Plan 2006, 21:459–468. Institutes of Health, Bethesda, USA. Division of Women and Child Health, 21. Pagano M, Gauvreau K: Principles of Biostatistics. 2nd edition. Pacific Grove, Aga Khan University, Karachi, Pakistan. Division of Nutrition and Food CA: Duxbury Thomson Learning; 2000. Security, International Centers for Diarrheal Disease Research, Matlab, 22. DeVellis R: Scale Development: theory and applications. 2nd edition. Bangladesh. HIV/AIDS and Global Health Research Programme, University of Thousand Oaks, CA: SAGE Publications; 2003. Venda, Thohoyandou, South Africa. Christian Medical College, Vellore, India. 23. Braveman P, Cubbin C, Marchi K, Egerter S, Chavez G: Measuring Clinical Research Unit and Institute of Biomedicine, Federal University of socioeconomic status/position in studies of racial/ethnic disparities: Ceara, Fortaleza, Brazil. Institute of Medicine, Tribhuvan University, maternal and infant health. Public Health Rep 2001, 116:449–463. Kathmandu, Nepal. Centre for International Health, University of Bergen, 24. Hackman DA, Farah MJ: Socioeconomic status and the developing brain. Bergen, Norway. Haydom Lutheran Hospital, Haydom, Tanzania. Trends Cogn Sci 2009, 13:65–73. 25. Braveman PA, Cubbin C, Egerter S, Chideya S, Marchi KS, Metzler M, Posner S: Received: 2 July 2013 Accepted: 14 February 2014 Socioeconomic status in health research: one size does not fit all. JAMA Published: 21 March 2014 2005, 294:2879–2888. 26. Krieger N, Williams DR, Moss NE: Measuring social class in US public health research: concepts, methodologies, and guidelines. Annu Rev Public Health 1997, 18:341–378. References 27. Daly MC, Duncan GJ, McDonough P, Williams DR: Optimal indicators of 1. Adler NE, Boyce T, Chesney MA, Cohen S, Folkman S, Kahn RL, Syme SL: socioeconomic status for health research. Am J Public Health 2002, Socioeconomic status and health. The challenge of the gradient. Am 92:1151–1157. Psychol 1994, 49:15–24. 28. Fosto JC, Kuate-Defo B: Measuring Socioeconomic Status in Health 2. Pollack CE, Chideya S, Cubbin C, Williams B, Dekker M, Braveman P: Should Research in Developing Countries: Should we be focusing on health studies measure wealth? A systematic review. Am J Prev Med 2007, households, communities or both? Soc Indicators Res 2005, 72:189–237. 33:250–264. 3. Black RE, Allen LH, Bhutta ZA, Caulfield LE, de Onis M, Ezzati M, Mathers C, doi:10.1186/1478-7954-12-8 Rivera J; Maternal and Child Undernutrition Study Group: Maternal and Cite this article as: Psaki et al.: Measuring socioeconomic status in child undernutrition: global and regional exposures and health multicountry studies: results from the eight-country MAL-ED study. consequences. Lancet 2008, 371:243–260. Population Health Metrics 2014 12:8. 4. Walker SP, Wachs TD, Gardner JM, Lozoff B, Wasserman GA, Pollitt E, Carter JA; International Child Development Steering Group: Child development: risk factors for adverse outcomes in developing countries. Lancet 2007, 369:145–157. 5. Galobardes B, Lynch JW, Davey Smith G: Childhood socioeconomic circumstances and cause-specific mortality in adulthood: systematic review and interpretation. Epidemiol Rev 2004, 26:7–21. 6. O’Donnell O, van Doorslaer E, Wagstaff A, Lindelow M: Analyzing Health Equity using Household Survey Data: a Guide to Techniques and their Implementation. Washington DC: World Development Institute, World Bank; 2007. www.worldbank.org/analyzinghealthequity (ISBN: 0-8213-6933-4). 7. Filmer D, Pritchett LH: Estimating wealth effects without expenditure data-or tears: an application to educational enrollments in states of India. Demography 2001, 38:115–132. 8. Rutstein SO, Johnson K: The DHS Wealth Index: DHS comparative reports No. 6. Calverton, MD: ORC Macro; 2004. 9. Rutstein SO: The DHS Wealth Index: Approaches for rural and urban areas. DHS Working Papers No. 60.. Calverton, MD: Macro International; 2008. 10. Pitchforth E, van Teijlingen E, Graham W, Fitzmaurice A: Development of a proxy wealth index for women utilizing emergency obstetric care in Bangladesh. Health Policy Plan 2007, 22:311–319. 11. van Bodegom D, May L, Kuningas M, Kaptijn R, Thomése F, Meij HJ, Submit your next manuscript to BioMed Central Amankwa J, Westendorp RG: Socio-economic status by rapid appraisal is and take full advantage of: highly correlated with mortality risks in rural Africa. Trans R Soc Trop Med Hyg 2009, 103:795–800. • Convenient online submission 12. Alkire S, Santos ME: Acute Multidimensional Poverty: A new index for • Thorough peer review developing countries. Oxford: Oxford Poverty & Human Development Initiative; 2010. OPHI Working Paper 38. • No space constraints or color figure charges 13. USAID: Demographic and Health Surveys. Available at: http://www. • Immediate publication on acceptance measuredhs.com/. Accessed December 10, 2012. 14. World Health Organization: Child Growth Standards: Anthropometry Macros. • Inclusion in PubMed, CAS, Scopus and Google Scholar 2011. Available at: http://www.who.int/childgrowth/software/en/index.htm. • Research which is freely available for redistribution 15. de Onis M, Garza C, Victora CG: The WHO Multicentre Growth Reference Study: strategy for developing a new international growth reference. Submit your manuscript at Forum Nutr 2003, 56:238–240. www.biomedcentral.com/submit

Journal

Population Health MetricsSpringer Journals

Published: Mar 21, 2014

There are no references for this article.