Scholarly article on topic 'Association between economic growth and early childhood undernutrition: evidence from 121 Demographic and Health Surveys from 36 low-income and middle-income countries'

Association between economic growth and early childhood undernutrition: evidence from 121 Demographic and Health Surveys from 36 low-income and middle-income countries Academic research paper on "Health sciences"

CC BY-NC-ND
0
0
Share paper
Academic journal
The Lancet Global Health
OECD Field of science
Keywords
{}

Abstract of research paper on Health sciences, author of scientific article — Sebastian Vollmer, Kenneth Harttgen, Malavika A Subramanyam, Jocelyn Finlay, Stephan Klasen, et al.

Summary Background Economic growth is widely regarded as a necessary, and often sufficient, condition for the improvement of population health. We aimed to assess whether macroeconomic growth was associated with reductions in early childhood undernutrition in low-income and middle-income countries. Methods We analysed data from 121 Demographic and Health Surveys from 36 countries done between Jan 1, 1990, and Dec 31, 2011. The sample consisted of nationally representative cross-sectional surveys of children aged 0–35 months, and the outcome variables were stunting, underweight, and wasting. The main independent variable was per-head gross domestic product (GDP) in constant prices and adjusted for purchasing power parity. We used logistic regression models to estimate the association between changes in per-head GDP and changes in child undernutrition outcomes. Models were adjusted for country fixed effects, survey-year fixed effects, clustering, and demographic and socioeconomic covariates for the child, mother, and household. Findings Sample sizes were 462 854 for stunting, 485 152 for underweight, and 459 538 for wasting. Overall, 35·6% (95% CI 35·4–35·9) of young children were stunted (ranging from 8·7% [7·6–9·7] in Jordan to 51·1% [49·1–53·1] in Niger), 22·7% (22·5–22·9) were underweight (ranging from 1·8% [1·3–2·3] in Jordan to 41·7% [41·1–42·3] in India), and 12·8% (12·6–12·9) were wasted (ranging from 1·2% [0·6–1·8] in Peru to 28·8% [27·5–30·0] in Burkina Faso). At the country level, no association was seen between average changes in the prevalence of child undernutrition outcomes and average growth of per-head GDP. In models adjusted only for country and survey-year fixed effects, a 5% increase in per-head GDP was associated with an odds ratio (OR) of 0·993 (95% CI 0·989–0·995) for stunting, 0·986 (0·982–0·990) for underweight, and 0·984 (0·981–0·986) for wasting. ORs after adjustment for the full set of covariates were 0·996 (0·993–1·000) for stunting, 0·989 (0·985–0·992) for underweight, and 0·983 (0·979–0·986) for wasting. These findings were consistent across various subsamples and for alternative variable specifications. Notably, no association was seen between per-head GDP and undernutrition in young children from the poorest household wealth quintile. ORs for the poorest wealth quintile were 0·997 (0·990–1·004) for stunting, 0·999 (0·991–1·008) for underweight, and 0·991 (0·978–1·004) for wasting. Interpretation A quantitatively very small to null association was seen between increases in per-head GDP and reductions in early childhood undernutrition, emphasising the need for direct health investments to improve the nutritional status of children in low-income and middle-income countries. Funding None.

Academic research paper on topic "Association between economic growth and early childhood undernutrition: evidence from 121 Demographic and Health Surveys from 36 low-income and middle-income countries"

Articles

Association between economic growth and early childhood undernutrition: evidence from 121 Demographic and Health Surveys from 36 low-income and middle-income countries

Sebastian Vollmer, Kenneth Harttgen, Malavika A Subramanyam, Jocelyn Finlay, Stephan Klasen, S VSubramanian

Summary

Background Economic growth is widely regarded as a necessary, and often sufficient, condition for the improvement of population health. We aimed to assess whether macroeconomic growth was associated with reductions in early childhood undernutrition in low-income and middle-income countries.

Methods We analysed data from 121 Demographic and Health Surveys from 36 countries done between Jan 1, 1990, and Dec 31, 2011. The sample consisted of nationally representative cross-sectional surveys of children aged 0-35 months, and the outcome variables were stunting, underweight, and wasting. The main independent variable was per-head gross domestic product (GDP) in constant prices and adjusted for purchasing power parity. We used logistic regression models to estimate the association between changes in per-head GDP and changes in child undernutrition outcomes. Models were adjusted for country fixed effects, survey-year fixed effects, clustering, and demographic and socioeconomic covariates for the child, mother, and household.

Findings Sample sizes were 462 854 for stunting, 485 152 for underweight, and 459 538 for wasting. Overall, 35-6% (95% CI 35-4-35-9) of young children were stunted (ranging from 8-7% [7-6-9-7] in Jordan to 51-1% [49-1-53-1] in Niger), 22-7% (22-5-22-9) were underweight (ranging from 1-8% [1-3-2-3] in Jordan to 41-7% [41-1-42-3] in India), and 12-8% (12-6-12-9) were wasted (ranging from 1-2% [0-6-1-8] in Peru to 28-8% [27-5-30-0] in Burkina Faso). At the country level, no association was seen between average changes in the prevalence of child undernutrition outcomes and average growth of per-head GDP. In models adjusted only for country and survey-year fixed effects, a 5% increase in per-head GDP was associated with an odds ratio (OR) of 0-993 (95% CI 0-989-0-995) for stunting, 0-986 (0-982-0-990) for underweight, and 0-984 (0-981-0-986) for wasting. ORs after adjustment for the full set of covariates were 0-996 (0-993-1-000) for stunting, 0-989 (0-985-0-992) for underweight, and 0-983 (0-979-0-986) for wasting. These findings were consistent across various subsamples and for alternative variable specifications. Notably, no association was seen between per-head GDP and undernutrition in young children from the poorest household wealth quintile. ORs for the poorest wealth quintile were 0-997 (0-990-1-004) for stunting, 0-999 (0-991-1-008) for underweight, and 0-991 (0-978-1-004) for wasting.

Interpretation A quantitatively very small to null association was seen between increases in per-head GDP and reductions in early childhood undernutrition, emphasising the need for direct health investments to improve the nutritional status of children in low-income and middle-income countries.

Funding None.

Copyright © Vollmer et al. Open Access article distributed under the terms of CC BY.

* CrossMark

Introduction

Increasing economic growth, as measured through increases in per-head gross domestic product (GDP), is the cornerstone of development policy for most national governments.1 In countries with low per-head GDP, increasing the rate of economic growth is often justified as a key policy instrument for improving population health and nutrition.2-4 The rationale is that increases in economic growth will lead to increases in average income, especially improving the incomes of poor people, which in turn will improve access to, and consumption of, goods and services that improve nutritional status and health.5 Although such a growth-mediated strategy for improving population health and nutrition6 is plausible, the empirical evidence to

Lancet Glob Health 2014; 2: e225-34

See Comment page el85 Department of Economics and Courant Research Centre "Poverty, Equity and Growth in Developing Countries", University of Göttingen, Göttingen, Germany (Prof S Vollmer PhD, Prof S Klasen PhD); NADEL, ETH Zurich, Zurich, Switzerland (K Harttgen PhD); Social Sciences, Indian Institute of Technology Gandhinagar, Ahmedabad, Gujarat, India (M A Subramanyam DSc); and Harvard Center for Population and Development Studies, Harvard School of Public Health, Cambridge, MA, USA (S Vollmer, J Finlay PhD, Prof S V Subramanian PhD) Correspondence to: Prof S V Subramanian, Harvard Center for Population and Development Studies, Cambridge, MA 02138, USA svsubram@hsph.harvard.edu or

Prof Sebastian Vollmer, Department of Economics, University of Göttingen, 37073 Göttingen, Germany svollmer@uni-goettingen.de

support this strategy remains unclear. Smith and Haddad4 used aggregated data from 63 low-income and middle-income countries with measurements from 1970 to 1996 and reported a strong inverse association between national economic growth and childhood underweight. Headey7 estimated the effect of economic growth on changes in undernutrition and showed that economic growth leads to a small but significant reduction in stunting. Such ecological analyses assume that the risk of undernutrition is the same for every child within a country, which, if invalid, can lead to a biased estimation of the association. Moreover, ecological analyses limit the scope for analysis of more detailed subnational drivers of undernutrition and within-country heterogeneity.

We identified only two studies that have examined the multilevel association of changes in aggregate macroeconomic growth with changes in the risk of an individual child being undernourished. Subramanyam and colleagues8 reported that there was no consistent evidence that economic growth in various Indian states was associated with a reduction in childhood stunting, underweight, or wasting. Thus, at least for India, which accounts for the largest national share of the global burden of undernutrition,9 the substantial increases in economic growth that the country experienced over the past two decades did not translate into a similarly substantial reduction in childhood undernutrition.10 Harttgen and colleagues11 examined the effect of economic growth on child undernutrition in sub-Saharan Africa using Demographic and Health Surveys (DHS) from the 1990s and 2000s and reported that growth had a small, inverse association with childhood stunting, underweight, and wasting.

Using the largest available, nationally representative, and mutually comparable repeated cross-sectional samples from 121 surveys in 36 low-income to middle-income countries, with objective measurements of childhood anthropometry, we investigated whether changes in national economic growth were associated with reductions in the prevalence of early childhood stunting, underweight, and wasting.

Methods

Data sources and procedures

Data were from the DHS, which are nationally representative cross-sectional surveys that have been done by ICF International in more than 82 low-income

and middle-income countries at varying intervals since 1985. These surveys are designed to collect nationally representative health and welfare data for women of reproductive age, their children, and their households. We included surveys done between Jan 1, 1990, and Dec 31, 2011, in the analysis.

The DHS used a multistage stratified sampling design. In the first stage of sampling, each country was divided into regions, which are either political regions such as states or provinces, or geographical areas divided and labelled north, south, east, and west. Within these subnational regions, populations were stratified by urban and rural area of residence. Within these stratified areas, a random selection of enumeration areas taken from the most recent population census was drawn. These primary sampling units (clusters) were selected such that the probability of each cluster being selected was equal to the proportion that specific cluster's population contributed to the total population. In the second stage of sampling, all households within the cluster were listed and an average of 25 houses within a cluster were randomly selected for an interview by equal-probability systematic sampling. Detailed sampling plans are available from the final survey reports.12

For each sampled household, members were listed and women eligible for a child health interview were identified. The women interviewed were typically between the ages of 15 and 49 years, although in some surveys the age range was 10-49 years, and a few were restricted to married and previously married women. Heights and weights of children born in the past 3 or 5 years at the time of interview were also recorded. For consistency across all surveys, only data from children

Stunting

Underweight

Wasting

865 368

814 566

794 711

497285

469 893

50 802 had a single survey

19 855 had missing data for per-head GDP

297 426 had missing data

for outcome variable

27392 had miscoded data for outcome variable

7039 had missing data for covariates

462 854 included in the analysis

865 368

814 566

794 711

507 822

492 455

50 802 had a single survey

19 855 had missing data for per-head GDP

286 889 had missing data

for outcome variable

15 367 had miscoded data for outcome variable

7303 had missing data for covariates

485 152 included in the analysis

865 368

814 566

794 711

479 920

466 506

50 802 had a single survey

19 855 had missing data for per-head GDP

314 791 had missing data

for outcome variable

13 414 had miscoded data for outcome variable

6968 had missing data for covariates

459 538 included in the analysis

Figure 1: Sample selection

GDP=gross domestic product.

aged 0-35 months were included in the analysis. The DHS provided sampling weights for the calculation of nationally representative statistics.

We used national aggregate data for per-head GDP from the Penn World Tables 8.0.13 These tables provide data for real per-head GDP, adjusted for purchasing power parity exchange rates and reported in 2005 International Comparison Program prices (in international dollars). The adjustment for purchasing power parity between countries makes the level and growth of per-head GDP comparable

between countries and over time. We used per-head GDP measured in natural logarithmic units to model a potentially non-linear association with early childhood undernutrition. We merged the individual data from the DHS with the data for per-head GDP by country and year. Children in the same survey (ie, same country and year) were assigned the same per-head GDP, which is representative of the national average.

The DHS data-collection procedures were approved by the ICF International (Calverton, MD, USA) institutional

Survey year Stunting sample Underweight sample Wasting sample GDP per head* Economic growtht

N Proportion stunted (95% CI) N Proportion underweight (95% CI) N Proportion wasted (95% CI)

Armenia 2010 860 19-4% (16-7-22-0) 875 5-4% (3-9-6-9) 850 4-1% (2-8-5-4) $5002 2-8%

Bangladesh 2007 3238 40-3% (38-6-42-0) 3313 38-9% (37-2-40-6) 3218 19-4% (18-0-20-7) $1314 3-7%

Benin 2006 7956 42-2% (41-1-43-3) 8412 19-4% (18-6-20-3) 8105 10-2% (9-5-10-9) $1226 0-9%

Bolivia 2008 4709 25-5% (24-2-26-7) 4754 5-6% (5-0-6-3) 4675 2-2% (1-8-2-7) $3795 4-7%

Burkina Faso 2003 5172 38-6% (37-3-40-0) 5360 37-6% (36-3-38-9) 5012 28-8% (27-5-30-0) $799 2-2%

Cameroon 2011 3309 29-8% (28-3-31-4) 3368 14-8% (13-6-16-0) 3310 7-8% (6-9-8-7) $1858 0-7%

Chad 2004 2705 38-9% (37-0-40-7) 2784 33-3% (31-6-35-1) 2671 21-0% (19-5-22-6) $1471 7-6%

Colombia 2009 9469 13-3% (12-6-14-0) 9486 3-7% (3-3-4-1) 9453 1-2% (0-9-1-4) $7481 3-9%

Côte d'Ivoire 1998 1034 27-8% (25-1-30-5) 1054 17-4% (15-1-19-7) 1026 9-0% (7-3-10-8) $2069 1-1%

Dominican Republic 2007 5554 10-9% (10-1-11-8) 5623 3-6% (3-2-4-1) 5521 2-6% (2-2-3-0) $7671 3-8%

Egypt 2008 6247 31-0% (29-8-32-1) 6602 6-7% (6-1-7-4) 6178 9-0% (8-3-9-7) $4513 4-3%

Ethiopia 2011 5663 38-3% (37-0-39-6) 5824 26-9% (25-7-28-0) 5646 13-3% (12-4-14-1) $783 8-1%

Ghana 2008 1504 25-0% (22-8-27-2) 1593 14-9% (13-2-16-7) 1497 12-6% (10-9-14-3) $1988 3-6%

Guinea 2005 1692 36-1% (33-8-38-3) 1730 23-2% (21-2-25-2) 1688 14-4% (12-7-16-1) $909 -7-1%

India 2005 24 924 45-1% (44-5-45-7) 25 909 41-7% (41-1-42-3) 24 586 23-4% (22-9-24-0) $2415 5-2%

Jordan 2009 2736 8-7% (7-6-9-7) 2760 1-8% (1-3-2-3) 2731 1-9% (1-4-2-4) $5102 6-2%

Kazakhstan 1999 328 13-7% (9-9-17-4) 331 4-0% (1-9-6-1) 324 3-3% (1-3-5-2) $4710 -3-0%

Kenya 2009 3215 36-7% (35-0-38-4) 3346 14-9% (13-7-16-1) 3204 7-8% (6-9-8-7) $1232 1-0%

Lesotho 2009 1050 33-1% (30-2-35-9) 1069 13-3% (11-3-15-3) 1052 5-1% (3-8-6-4) $1315 3-7%

Madagascar 2004 2776 50-2% (48-3-52-0) 2875 34-5% (32-7-36-2) 2773 17-9% (16-5-19-3) $767 -0-8%

Malawi 2010 2892 46-8% (45-0-48-7) 3038 13-9% (12-7-15-1) 2855 5-4% (4-6-6-3) $795 6-5%

Mali 2006 7124 36-1% (35-0-37-2) 7289 28-7% (27-7-29-8) 6998 19-9% (19-0-20-9) $876 0-3%

Morocco 2003 3111 25-0% (23-5-26-5) 3192 8-6% (7-6-9-5) 3065 10-7% (9-6-11-7) $3226 -1-2%

Mozambique 2003 5005 43-9% (42-5-45-3) 5104 23-2% (22-1-24-4) 5017 7-1% (6-4-7-8) $523 5-1%

Namibia 2007 2440 29-5% (27-7-31-4) 2487 16-2% (14-7-17-6) 2421 8-2% (7-1-9-3) $4941 4-0%

Nepal 2011 520 50-2% (45-9-54-5) 522 38-2% (34-0-42-4) 517 16-9% (13-7-20-1) $1185 3-7%

Niger 2006 2361 51-1% (49-1-53-1) 2436 38-6% (36-6-40-5) 2362 16-5% (15-0-18-0) $535 -0-9%

Nigeria 2008 12 490 40-4% (39-6-41-3) 14 165 27-0% (26-2-27-7) 12 239 16-5% (15-8-17-1) $1896 18-7%

Peru 2004 1264 26-2% (23-8-28-7) 1290 5-4% (4-1-6-6) 1258 1-2% (0-6-1-8) $5266 4-1%

Rwanda 2010 2427 41-0% (39-0-42-9) 2440 11-9% (10-6-13-2) 2419 4-0% (3-2-4-8) $1135 5-3%

Senegal 2011 2379 28-1% (26-3-29-9) 2448 17-3% (15-8-18-8) 2380 10-0% (8-8-11-2) $1412 -0-4%

Tanzania 2004 4665 40-9% (39-4-42-3) 4720 16-0% (15-0-17-1) 4656 4-6% (4-0-5-2) $891 4-0%

Turkey 2003 2308 13-4% (12-0-14-7) 2368 2-7% (2-1-3-4) 2294 1-5% (1-0-2-0) $9173 -1-9%

Uganda 2011 1319 31-4% (28-9-33-9) 1342 15-2% (13-2-17-1) 1315 6-8% (5-5-8-2) $1187 3-0%

Zambia 2007 3308 43-8% (42-1-45-5) 3492 14-9% (13-7-16-0) 3310 7-3% (6-4-8-2) $1685 7-2%

Zimbabwe 2011 2924 32-1% (30-4-33-8) 2990 10-3% (9-2-11-4) 2908 4-3% (3-6-5-1) $4348 6-6%

Data are from the most recent Demographic and Health Survey (DHS) for each country included in the analysis. N is the sample size. GDP=gross domestic product. international dollars in constant prices (adjusted for purchasing power parity and inflation). fAverage annual growth rate of GDP per head between DHS surveys.

Table 1: Prevalence of child undernutrition, by country

review board and by the relevant human subjects committees in each country. Survey respondents provided oral informed consent. This study was assessed by the Harvard School of Public Health (Boston, MA, USA) institutional review board and ruled exempt from full review because it was based on an anonymised public-use dataset.

Outcomes

We analysed the DHS data for three outcomes: stunting, underweight, and wasting in children aged 0-35 months at the time of interview. We used anthropometric data to calculate whether a child was stunted, underweight, or wasted as defined by WHO standards and classifications.14 For stunting, we calculated a Z score as the child's height

0 10 20 Average annual growth rate of GDP per head (%)

Figure2: Correlation between prevalence of early childhood undernutrition outcomes and log of per-head GDP

n=121 surveys. GDP=gross domestic product.

Figure3: Correlation between the change in prevalence of early childhood undernutrition outcomes and growth of per-head GDP

n=85 surveys (121 minus 36). GDP=gross domestic product.

minus the median height for that child's age and sex in the WHO reference population,15 divided by the standard deviation of this group in the reference population.15 We did similar calculations to establish Z scores for weight (for underweight) and weight-for-height (for wasting, an indicator ofacute undernutrition). Stunting, underweight, and wasting were defined by Z scores of less than -2; Z scores of less than -3 were classified as severe. Biologically implausible values (defined by WHO for height as a Z score of less than -6 or greater than 6, for weight as a Z score of less than -6 or greater than 5, and for weight for height as a Z score of less than -5 or greater than 5)14 were excluded.

We used age, sex, and birth order of the child; the mother's age at birth, education, and relationship status; household wealth quintile; and urban or rural residence as covariates for the analysis. Wealth quintile is a within-country measure of the wealth of the household relative to other households in that survey, based on ownership of household assets.16 Although several of these factors could be mediators of the association between economic growth and early childhood undernutrition, they are also modifiable by direct investment in social programmes. We therefore examined the association with and without adjustment for these factors.

Statistical analysis

We specified a series of logistic regression models for stunting, underweight, and wasting as outcome variables. For each outcome, we adjusted the models for country and survey-year fixed effects. We clustered standard errors by country and by primary sampling unit. We also specified regression models for various subsamples and alternative variable specifications: the poorest and richest wealth quintiles; children aged 0-11, 12-23, and 24-35 months; sub-Saharan African countries, Asian countries, and Latin American countries; low-income, lower-middle-income, and upper-middle-income countries; severe stunting, severe underweight, and severe wasting as outcome variables; log of per-head GDP from the previous year instead of the concurrent year as an alternative independent variable. We excluded Egypt and Turkey from the regional analysis since they do not fit clearly into the regional categories. To allow interpretation, we report the odds ratios (ORs) for a 5% increase in per-head GDP. For all analyses, associations with a p value of less than 0-05 were regarded as significant. All statistical analyses were done with Stata 13.

We also did several sensitivity analyses based on a linear probability model. First, we reweighted the observations with the population size of the country. Second, we trimmed the sample to exclude extreme observations that might have an especially large effect on the results. Third, we used instrumental variable regressions with the investment share of GDP 5 years previous to any given year used in the analysis as an instrument for log of the

per-head GDP3,17 to address two potential statistical problems: measurement error, particularly in GDP, which could bias the results downwards; and the endogenous nature of GDP, which could bias the findings because of either reverse causality or the effect of unmeasured variables on the association between per-head GDP and undernutrition.

Role of the funding source

There was no funding source for this study.

Results

172 surveys done between Jan 1, 1990, and Dec 31, 2011, in 64 countries had recorded anthropometric data for 865 368 children born up to 3 years before the time of interview. Because our aim was to examine the effect of economic growth over time, we excluded children from countries for which data from only one survey were available. Other exclusions resulted from missing data for GDP, outcome variables, and covariates, resulting in a sample of 462 854 for the stunting analysis, 485 152 for the underweight analysis, and 459 538 for the wasting analysis (figure 1). These samples include data from 121 surveys in 36 countries.

Of the 36 countries included in the study, 34 had their most recent survey between 2003 and 2011 (table 1). The most recent survey for India was done in 2005, which with data for 24 924 children in the stunting analysis was the largest of the most recent surveys included. Across all of the most recent surveys included in the study, those for Kazakhstan (n=328) and Nepal (n=520) were by far the smallest. Six surveys had a sample size between 1000 and 1999; 17 surveys between 2000 and 4999; and eight between 5000 and 9999 (table 1).

Overall, 35-6% (95% CI 35-4-35-9) of young children were stunted, 22-7% (22-5-22-9) were underweight, and 12-8% (12-6-12-9) were wasted. Jordan had the lowest prevalence of stunting and underweight, and Peru the lowest prevalence for wasting. Niger had the highest prevalence of stunting, India of underweight, and Burkina Faso of wasting (table 1).

The average annual growth of per-head GDP between survey years varied substantially between countries (table 1). Nigeria had the strongest average growth rate of per-head GDP at 18-7% per year between the two most recent surveys in 2003 and 2008. Seven countries showed negative growth between survey years, but 16 had growth rates between 1% and 5% (table 1).

Using the most recent surveys from each country, we calculated the average child undernutrition outcomes by categorical covariates (appendix pp 1-2). Over time, the See Online for appendix prevalence of stunting, underweight, and wasting changed within countries, as did the per-head GDP and its associated growth rate. For most countries, the prevalence of early childhood undernutrition fell and per-head income increased (appendix pp 3-5). Across countries and time, if a child was stunted, that child lived in an

Adjusted Unadjusted

Stunted Wasted Underweight Stunted Wasted Underweight

Full sample

OR (95% CI) 0.996 (0-993-1-000) 0-983 (0-979-0-986) 0.989 (0.985-0-992) 0-993 (0-989-0-995) 0-984 (0-981-0-986) 0-986 (0.982-0.990)

p value 0-021 <0-0001 <0-0001 <0-0001 <0.0001 <0-0001

N 462854 459 538 485 152 462 854 459538 485152

Poorest wealth quintile

OR (95% CI) 0-997 (0-990-1-004) 0-991 (0-978-1-004) 0-999 (0-991-1-008) 0-995 (0-992-0-998) 0-985 (0-982-0-987) 0-988 (0.983-0-993)

p value 0-367 0-153 0-784 0-002 <0-0001 <0-0001

N 104 040 103 473 109 329 104040 103 473 109 329

Richest wealth quintile

OR (95% CI) 0-997 (0-992-1-001) 0-984 (0-981-0-987) 0-990 (0-987-0-993) 0-990 (0-987-0-993) 0-983 (0-981-0-986) 0-985 (0-980-0-989)

p value 0-086 <0-0001 <0-0001 <0-0001 <0-0001 <0-0001

N 74 575 73 902 77 883 74 575 73 907 77 888

Children aged 0-11 months

OR (95% CI) 0-991 (0.981-1.001) 0-998 (0-988-1-008) 0-994 (0-987-1-001) 0.989 (0.985-0-993) 0-985 (0-982-0-989) 0-987 (0-984-0-989)

p value 0-071 0-593 0-054 <0.0001 <0-0001 <0-0001

N 162048 158 770 170 633 162 048 158 770 170 633

Children aged 12-23 months

OR (95% CI) 0.989 (0-979-1-000) 0-989 (0-977-1-002) 0-996 (0-985-1-006) 0-993 (0-990-0-997) 0-983 (0-981-0-986) 0-985 (0-979-0-991)

p value 0-035 0-085 0-372 0-0002 <0.0001 <0-0001

N 155 071 155 437 162 378 155 071 155 437 162378

Children 24-35 months

OR (95% CI) 0-997 (0-994-1-000) 0-982 (0-979-0-986) 0-988 (0-984-0-992) 0-995 (0-993-0-997) 0-982 (0-979-0-985) 0-986 (0-982-0-990)

p value 0-008 <0-0001 <0-0001 <0-0001 <0-0001 <0-0001

N 145734 145 330 152140 145 735 145 331 152 141

Sub-Saharan Africa

OR (95% CI) 0-996 (0-994-0-997) 0-984 (0-981-0-986) 0-992 (0-990-0-994) 0-995 (0-994-0-996) 0-988 (0-986-0-991) 0-993 (0-992-0-995)

p value <0-0001 <0-0001 <0-0001 <0-0001 <0-0001 <0-0001

N 241 448 239 546 250 507 241448 239 547 250 508

OR (95% CI) 0-999 (0-999-1-000) 0-979 (0-978-0-979) 0-984 (0-984-0-985) 0-992 (0-992-0-992) 0-978 (0-978-0-978) 0-980 (0-980-0-980)

p value 0-0002 <0-0001 <0-0001 <0-0001 <0-0001 <0-0001

N 112 342 111 587 123 619 112 342 111 587 123619

(Table 2 continues on next page)

environment with an average per-head GDP of $2055, compared with $2896 for a child without stunting. A similar pattern was seen for underweight and wasting. As expected, an inverse cross-sectional ecological association was seen between the average child undernutrition outcomes and the per-head GDP of the country (figure 2); however, no association was seen between average changes in the prevalence of child undernutrition outcomes and average growth of per-head GDP (figure 3).

Table 2 shows the adjusted and unadjusted ORs for the full sample, subsamples, and alternative variable specifications. In models adjusted only for country and

survey-year fixed effects, a 5% increase in per-head GDP was associated with a 0-7% decrease in the odds of being stunted (p<0-0001), a 1-4% decrease in the odds of being underweight (p<0-0001), and a 1-6% decrease in the odds of being wasted (p<0-0001). The respective figures from models adjusted for a full set of covariates were 0-4% for the odds of being stunted (p=0-021), 1-1% for the odds of being underweight (p<0-0001), and 1-7% for the odds of being wasted (p<0-0001). Results for all covariates are reported in the appendix (p 6).

For the various subsamples and alternative variable specifications, we noted several findings. First, the

Adjusted Unadjusted

Stunted Wasted Underweight Stunted Wasted Underweight

(Continued from previous page)

Latin America

OR (95% CI) 0-996 0-977 0-985 0-994 0-979 0-984

(0.994-0.997) (0-975-0-979) (0-982-0-987) (0-994-0-994) (0-979-0-979) (0-984-0-984)

p value <0-0001 <0-0001 <0-0001 <0-0001 <0-0001 <0-0001

N 66 710 66 321 67 504 66 710 66 362 67 504

Low-income countries

OR (95% CI) 1.003 0-998 0-985 1-000 0-999 0-989

(1-001-1-004) (0-996-1-000) (0-982-0-989) (0-999-1-002) (0-998-1-000) (0-986-0-992)

p value <0-0001 0-091 <0-0001 0-439 0-007 <0-0001

N 183 313 189490 182 084 183 313 189 490 182 084

Lower-middle-income countries

OR (95% CI) 0-994 0-990 0-981 0-990 0-987 0-982

(0-991-0-997) (0-985-0-995) (0-978-0-984) (0-989-0-992) (0-981-0-992) (0-979-0-984)

p value <0-0001 <0-0001 <0-0001 <0-0001 <0-0001 <0-0001

N 193 055 207 861 191 537 193 055 207 861 191 537

Upper-middle-income countries

OR (95% CI) 0-997 0-985 0-975 0-995 0-984 0-976

(0-996-0-998) (0-982-0-987) (0-972-0-978) (0-994-0-995) (0-983-0-986) (0-974-0-979)

p value <0-0001 <0-0001 <0-0001 <0-0001 <0-0001 <0-0001

N 86 486 87 801 85 876 86 486 87 801 85 876

Severe undernutrition as outcome variables

OR (95% CI) 0-988 0-977 0-985 0-985 0-978 0-981

(0-984-0-991) (0-972-0-981) (0-979-0-991) (0-982-0-988) (0-974-0-982) (0-975-0-988)

p value <0-0001 <0-0001 <0-0001 <0-0001 <0-0001 <0-0001

N 462 854 459538 485 152 462 854 459538 485 152

Log of per-head GDP from the previous year as an independent variabl

OR (95% CI) 0-996 0-983 0-989 0-993 0-984 0-987

(0-993-1-000) (0-980-0-986) (0-986-0-993) (0-990-0-995) (0-981-0-987) (0-983-0-990)

p value 0-017 <0-0001 <0-0001 <0-0001 <0-0001 <0-0001

N 462 854 459538 485 152 462 854 459 538 485 152

Data for per-head gross domestic product (GDP) were merged with Demographic and Health Survey data by survey year. SEs are clustered at the country level. Odds ratios (ORs) for the log of per-head GDP represent the difference in odds associated with a 5% increase in per-head GDP. All specifications include country and survey-year fixed effects. All ORs are rounded to three decimal places; thus an OR of 1-QQQ in the CI does not necessarily imply that the value 1 is included in the CI.

Table 2: Adjusted and unadjusted ORs for childhood undernutrition associated with the log of per-head GDP for various subsamples and alternative variable specifications

associations between the log of per-head GDP and stunting, underweight, and wasting were null for the poorest wealth quintile. For the richest wealth quintile, the associations were similar to those for the full sample (table 2). Second, for children aged 0-11 months, the associations between per-head GDP and the three outcome variables were not significant. For children aged 12-23 months, a 5% increase in per-head GDP was associated with reduced odds of being stunted, but the associations between per-head GDP and underweight and wasting were not significant. For children aged 24-35 months, a 5% increase in per-head GDP was associated with slightly reduced odds of stunting, underweight, and wasting. Third, the finding of a quantitatively very small association between the log of per-head GDP and stunting, underweight, and wasting holds for countries subdivided by region into sub-

Saharan Africa, Asia, and Latin America, as well as across World Bank classifications of low-income, lower-middle-income, and upper-middle-income countries. Fourth, a 5% increase in per-head GDP was associated with reduced odds of severe stunting, underweight, and wasting. Finally, the results were robust to the inclusion of the log of per-head GDP from the previous year instead of the concurrent year (table 2).

Table 3 shows the results from linear probability models and several sensitivity analyses fully adjusted for covariates, country fixed effects, and survey-year fixed effects. A 5% increase in per-head GDP was associated with a 0-3% reduced probability of stunting, a 0-2% reduced probability of underweight, and a 0-1% reduced probability of wasting. When we weighted the results by country population size, the probability decreased by 0-4% for stunting, 0-1% for underweight, and 0-1% for

Stunting Wasting Underweight

Un- Weighted by Excluding first, Instru- Un- Weighted by Excluding first, Instru- Un- Weighted by Excluding first, Instru-

weighted population second, third, mental weighted population second, third, mental weighted population second, third, mental

model size 98th, 99th, and 100th percentiles variable regression model size 98th, 99th, and 100th percentiles variable regression model size 98th, 99th, and 100th percentiles variable regression

Log of per-head GDP -0.0025 -0.0041 -0.0022 -0.0507 -0.0008 -0.0013 -0.0008 -0.0218 -0.0017 -0.0013 -0.0017 -0.0511

(SE) (0.0008) (0.0007) (0.0008) (0.1582) (0.0004) (0-0003) (0.0004) (0.0628) (0.0007) (0.0006) (0.0006) (0-1935)

p value 0.004 <0.0001 0.011 0.761 0-075 <0.0001 0.008 0-741 0.023 0-030 0.006 0-805

Number of 462854 462 854 435398 462 117 459 538 459 538 432 087 458803 485 152 485 152 457 055 484 403

observations

Data for per-head gross domestic product (GDP) were merged with Demographic and Health Survey data by survey year. All regressions are ordinary least squares and the instrumental variable regressions are two-stage least squares. All regression shown are adjusted for all covariates and for country and survey-year fixed effects. SEs are clustered at the country level. Coefficients for the log of per-head GDP represent a 5% increase in per-head GDP. In the instrumental variable regressions, we used the variable share of gross capital formation at present purchasing power parity (investment share of GDP) from the Penn World Tables 8.0,13 with a 5-year lag as an instrument for the log of the per-head GDP.

Table3: Estimates from linear probability models, models weighted for population size, models after trimming extreme observations, and instrumental variable regression

Panel: Research in context Systematic review

We first searched Google Scholar for articles published in English from Jan 1, 1990, to Oct 25, 2013, that included the search terms "undernutrition" and "income" and "growth"; "malnutrition" and "income" and "growth"; "undernutrition" and "economic" and "growth"; or "malnutrition" and "economic" and "growth" in the title. These four searches identified 29 unique entries. We searched PubMed using the same search strategy and did not identify any additional reports. We selected only empirical studies that were published in peer-reviewed journals and were not mainly focused on the assessment of specific programme interventions for our discussion of the scientific literature. Only three of the reports4811 identified from these searches were relevant, along with one additional report7 that was brought to our attention by a reviewer. Two studies47 used country-level data and showed an inverse association between economic growth and child undernutrition. A multilevel study8 that assessed state-level economic growth in India showed no consistent evidence for an association between economic growth and child undernutrition. Another multilevel study,11 which assessed country-level growth from the African continent, showed a small inverse association between economic growth and child undernutrition.

Interpretation

Our study is the first multilevel study to report estimates for the association between economic growth and early childhood stunting, underweight, and wasting for all low-income and middle-income countries for which nationally representative data are available and comparable across countries and over time. Our finding of a quantitatively very small to null association challenges the assumption that economic growth will automatically lead to reductions in child undernutrition.

wasting. When we excluded the first, second, third, 98th, 99th, and 100th percentiles of the outcome variables to examine whether and to what extent the results were biased by extreme observations, our main results were unaffected. In an instrumental variable regression that used the investment share of GDP from 5 years years previous to any given year used in the analysis as an instrument for log of the per-head GDP, the associations between the log of the per-head GDP and stunting, wasting, and underweight were not significant.

Discussion

Using data for child anthropometry from 121 surveys in 36 low-income and middle-income countries, we have shown that macroeconomic growth has a null to quantitatively very weak association with reductions in early childhood stunting, underweight, and wasting. This finding is robust for a wide variety ofcovariate adjustments, modelling approaches, and subsample analyses (panel).

Several plausible explanations could account for this result. First, the growth in incomes could be unequally distributed—if poor people are excluded from the benefits of growth, the effect of increased prosperity on average could be low. Second, even if rising incomes reach most households, they might not necessarily be spent in ways that enhance the nutritional status of children. A positive association would depend on how resources are allocated between food and non-food items, the quality of food purchased, and the distribution of food within households.18-20 Third, rising average incomes could be poorly associated with improvements to public services that are essential to improve the nutritional status of the population (eg, vaccinations against diseases that can precipitate and maintain undernutrition, prenatal and postnatal care, clean water and sanitation, etc). As Dreze and Sen6 have argued, progress in undernutrition can be achieved in low-income settings through investments in these public services, as places such as Sri Lanka, Kerala (India), and Costa Rica have shown. Conversely, high-income growth does not guarantee the provision of these services, which are typically provided and financed by the state, and the willingness and ability of states to deliver these services can differ greatly. Many other factors besides average prosperity affect childhood undernutrition, some of which (eg, female education) depend largely on public action that might be unrelated to per-head GDP.

Our study has several limitations, which we have attempted to address. The first issue relates to the quality of the measures of undernutrition used. Improvement in the underweight measure might not represent improve-

ments in a child's nutritional status, but rather a transition to a diet that includes more sugar and animal fats.21 This issue is less of a problem for stunting, and we have addressed the issue by reporting results for stunting in addition to underweight and wasting. The results are qualitatively similar for all three indicators, increasing our confidence in the results. The second issue is that a single reference standard for undernutrition measures might not account for genetic differences in height and weight potential across different world regions.21 We have partly addressed this issue by including country fixed effects in the regression models, which can account for such differences between countries.

Other limitations are related to the selection of countries for inclusion in the DHS sample and the quality of the GDP data. The DHS sample includes an oversampling of economically successful countries. Some of the poorest African countries have neither the capacity nor the political stability to undertake a survey of the scale of the DHS. The quality of the per-head GDP data from the Penn World Tables is related to a country's level of economic development, with larger error margins for poorer countries than for wealthier countries.22 The inclusion of country fixed effects in our models and the use of instrumental variable analysis greatly absorbs this bias. Furthermore, despite these issues, these are the best data that exist and all other studies have the same limitations. Nevertheless, the external validity of our results for countries not included in our sample is limited.

The final issues are related to our statistical approach. Per-head GDP could itself be affected by child undernutrition, since child undernutrition might either directly affect economic development or be a proxy for other factors that affect economic development for which we might not have accounted. A direct effect of child health on economic development could be related to the time that parents spend caring for sick or weak children; however, we believe that this direct effect, if it exists, will be quite small. Of course, child health might have very important and strong long-term effects on economic development, but such long-term effects would not affect our analysis, which focuses on the short-term to medium-term association between income and undernutrition. The contribution of children to economic growth through labour is similarly irrelevant for our analysis, because the children in our sample are too young to work.

Moreover, nutritional status of children could be a proxy for overall health conditions. We know from several studies3,23,24 that a causal link exists between population health and economic development. We partly addressed this concern with the inclusion of country fixed effects, which absorb differences in population health. Furthermore, we reduced potential bias by adjusting for a rich set of household-level covariates. Lastly, we used instrumental variable regressions specifically to address reverse causality and unobserved heterogeneity.

In summary, the quantitatively very small to null association seen in our study suggests that the contribution of economic growth to the reduction in early childhood undernutrition in developing countries is very small, if it exists at all. This finding challenges the assumption that economic growth will automatically lead to reductions in child undernutrition. Our results therefore emphasise the need to focus on direct investments in health and nutrition25,26 and not to rely on the so-called trickle-down approach of a growth-mediated strategy to improve nutrition in children.

Contributors

SV and SVS conceptualised the study, developed the analytical strategy, and interpreted the results. SV wrote the first draft of the report and contributed to the statistical analysis. KH did the statistical analysis and contributed to the interpretation of results and writing of the report. MAS contributed to the interpretation of the results and writing of the report. JF contributed to the statistical analysis and writing of the report. SK contributed to the conceptualisation of the study, interpretation of the results, and writing of the report. SVS contributed to the writing of the report and provided overall supervision.

Declaration of interests

We declare that we have no competing interests. Acknowledgments

We acknowledge support from the Open Access Publication Funds of the University of Gottingen (Gottingen, Germany).

References

1 Stiglitz J, Sen A, Fitoussi JP. Mismeasuring our lives. New York: The New Press, 2010.

2 Preston SH. The changing relation between mortality and level of economic development. Popul Stud 1975; 29: 231-48.

3 Pritchett L, Summers LH. Wealthier is healthier. J Hum Resour 1994; 31: 841-68.

4 Smith LC, Haddad L. How potent is economic growth in reducing undernutrition? What are the pathways of impact?

New cross-country evidence. Econ Dev Cult Change 2002; 51: 55-76.

5 Ravallion M. Growth, inequality and poverty: looking beyond averages. World Dev 2001; 29: 1803-15.

6 Dreze J, Sen AK. Hunger and public action. Oxford: Clarendon Press, 1989.

7 Headey DD. Developmental drivers of nutritional change: a cross-country analysis. World Dev 2013; 42: 76-88.

8 Subramanyam MA, Kawachi I, Berkman LF, Subramanian SV. Is economic growth associated with reduction in child undernutrition in India? PLoS Med 2011; 8: e1000424.

9 WHO. WHO global database on child growth and malnutrition. Geneva: World Health Organization, 2013. http://www.who.int/ nutgrowthdb/en/ (accessed March 22, 2013).

10 Subramanian SV, Subramanyam MA. Economic growth & health of poor children in India. Indian J Med Res 2011; 133: 685-86.

11 Harttgen K, Klasen S, Vollmer S. Economic growth and child undernutrition in sub-Saharan Africa. Pop Dev Rev 2013; 39: 397-412.

12 ICF International. DHS final reports. Calverton: Demographic and Health Surveys, 2011. http://dhsprogram.com/ (accessed Oct 31, 2013).

13 Feenstra RC, Inklaar R, Timmer MP. The next generation of the Penn World Table (NBER Working Paper No 19255). Cambridge, MA: National Bureau of Economic Research, 2013.

14 WHO Multicentre Growth Reference Study Group. WHO child growth standards: length/ height-for-age, weight for-age, weight-for-length and body mass index for age: methods and development. Geneva: World Health Organization, 2006.

15 Borghi E, de Onis M, Garza C, et al. Construction of the World Health Organization child growth standards: selection of methods for attained growth curves. Stat Med 2006; 25: 247-65.

16 Filmer D, Pritchett LH. Estimating wealth effects without expenditure data--or tears: an application to educational enrollments in states of India. Demography 2001; 38: 115-32.

17 Acemoglu D, Johnson S, Robinson JA, Yared P. Income and democracy. Am Econ Rev 2008; 98: 808-42.

18 Behrman JR, Deolalikar AB. Will developing country nutrition improve with income? A case study for rural south India.

J Polit Econ 1983; 95: 492-5077

19 Behrman JR, Deolalikar AB. The intrahousehold demand for nutrients in rural south India: individual estimates, fixed effects, and permanent income. J Hum Resour 1990; 25: 665-96.

20 Subramanaian S, Deaton A. The demand for food and calories. J Polit Econ 1996; 104: 133-62.

21 de Haen H, Klasen S, Qaim M. What do we really know? Metrics for food insecurity and undernutrition. Food Policy 2011; 36: 760-69.

22 Harttgen K, Klasen S, Vollmer S. An African growth miracle? Or: what do asset indices tell us about trends in economic performance? Rev Income Wealth 2013; 59 (suppl S1): S37-61.

23 Acemoglu D, Johnson S. Disease and development: the effect of life expectancy on economic growth. J Polit Econ 2007; 115: 925-85.

24 Bloom DE, Canning D, Fink G. Disease and development revisited. J Polit Econ (in press).

25 Ruel MT, Alderman H, Maternal and Child Nutrition Study Group. Nutrition-sensitive interventions and programmes: how can they help to accelerate progress in improving maternal and child nutrition? Lancet 2013; 382: 536-51.

26 Gillespie S, Haddad L, Mannar V, Menon P, Nisbett N, Maternal and Child Nutrition Study Group. The politics of reducing malnutrition: building commitment and accelerating progress. Lancet 2013; 382: 552-69.