impact of a large u.s. government nutrition program on · 2019-12-11 · however,...

Confidential: For Review OnlyImpact of a Large U.S. Government Nutrition Program on Under-5 Nutrition in Sub-Saharan Africa: A Difference-in-

Differences Evaluation of Feed the Future

Journal: BMJ

Manuscript ID BMJ-2019-050816

Article Type: Research

BMJ Journal: BMJ

Date Submitted by the Author: 22-May-2019

Complete List of Authors: Ryckman, Theresa; Stanford University, School of Medicine, Department of Health Research & PolicyRobinson, Margot; Stanford University, School of MedicinePedersen, Courtney; Stanford University, School of MedicineBhattacharya, Jayanta; Stanford University, Center for Health Policy and the Center for Primary Care and Outcomes Research, Department of Medicine; National Bureau of Economic ResearchBendavid, Eran; Stanford University, Center for Health Policy and the Center for Primary Care and Outcomes Research, Department of Medicine; Stanford University, Division of Primary Care and Population Health, Department of Medicine

Keywords: Health Policy, Global Health, Stunting, Impact Evaluation, Child Nutrition, Undernutrition

https://mc.manuscriptcentral.com/bmj

BMJ

Confidential: For Review Only

1

Impact of a Large U.S. Government Nutrition Program on Under-5 Nutrition in Sub-Saharan

Africa: A Difference-in-Differences Evaluation of Feed the Future

Theresa Ryckman (0000-0003-3214-3228), Margot Robinson, Courtney Pedersen, Jay Bhattacharya, Eran

Bendavid

Theresa Ryckman, PhD Candidate, Department of Health Research & Policy, Stanford University School

of Medicine, Stanford, CA, United States, [email protected]

Margot Robinson, Medical Student, Stanford University School of Medicine, Stanford University,

Stanford, CA, United States, [email protected]

Courtney Pederson, Medical Student, Stanford University School of Medicine, Stanford University,


Jay Bhattacharya, Professor of Medicine, Center for Health Policy and the Center for Primary Care and

Outcomes Research, Department of Medicine, Stanford University School of Medicine, Stanford, CA,

United States, [email protected]

Eran Bendavid, Associate Professor of Medicine, Center for Health Policy and the Center for Primary

Care and Outcomes Research, Department of Medicine, Stanford University School of Medicine,


Corresponding author: Theresa Ryckman, Department of Health Research & Policy,

150 Governor’s Lane, HRP Redwood Building, Stanford School of Medicine, Stanford University,

Stanford CA 94305. Telephone: +1 920-840-4123 Email: [email protected].

Word Count: 3969

Page 1 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

mailto:[email protected]







2

SUMMARY BOX

Section 1: What is already known on this topic

Child nutritional outcomes are known to be rooted in factors such as agriculture, food security,

and other social determinants.

There is ongoing debate about the relative effectiveness of programs that address the underlying

determinants of undernutrition such as food security and agriculture versus those that target

nutrition outcomes directly.

Previous efforts to evaluate the effects of large-scale nutrition programs such as Feed the Future,

one of the world’s largest agriculture and nutrition initiatives, have been hindered by inconsistent

outcome measurements and a lack of controls.

Section 2: What this study adds

We employ standardized nationally-representative outcome data for both treated and untreated

individuals to estimate the impact that Feed the Future has had on child nutrition outcomes. To

our knowledge, ours is the first study to make use of quasi-experimental statistical techniques to

evaluate this multi-country, multi-sectorial program that aims to address many of the root causes

of poor nutrition.

Our analysis supports the value of a multi-sectorial approach to child nutrition by showing that

there were meaningful decreases in child stunting associated with the Feed the Future program,

providing much-needed evidence on how an urgent global health issue can be addressed at scale.

Page 2 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


3

ABSTRACT

Objective To evaluate the impact of the US Government’s Feed the Future program, one of the largest

nutrition and agriculture initiatives in the world, on child nutrition outcomes in Sub-Saharan Africa.

Design We analyzed individual-level cross-sectional data from nationally-representative surveys. We

applied a difference-in-differences approach that compared outcomes among children in intervention

countries after the program’s implementation to children before Feed the Future’s introduction and

children in non-intervention countries, controlling for relevant covariates, time-invariant national

differences, and time trends.

Setting Data are from households in 33 low- and lower-middle-income countries in Sub-Saharan Africa.

Participants We included observations for under-five children whose weight, height, and age were

recorded in 118 surveys in 33 countries conducted between 2000 and 2017.

Main outcome measures We estimated Feed the Future’s impact on stunting (height-for-age z-score less

than two standard deviations below a reference median), a key indicator of child undernutrition, as our

primary outcome. Wasting (low weight-for-height) and underweight (low weight-for-age) were secondary

outcomes.

Results Our final data set consisted of a total of 883,309 under-five children. 39% of children in our

sample were stunted, 10% were wasted, and 23% were underweight. Children in Feed the Future

countries exhibited a 2.4 (0.6 to 4.2) percentage point greater decline in stunting, a 1.7 (0.2 to 3.2)

percentage point greater decline in wasting, and a 2.4 (0.8 to 4.1) percentage point greater decline in

underweight levels compared to children in non-Feed the Future countries after Feed the Future’s

implementation. These decreases translate to approximately one million fewer children stunted, wasted,

and underweight.

Page 3 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


4

Conclusions Feed the Future’s activities were linked to improvements in three standard measures of

undernutrition. This finding highlights the effectiveness of a large, country-tailored program focused on

underlying determinants, which has important implications for nutrition interventions worldwide.

Page 4 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


5

INTRODUCTION

Undernutrition is the single greatest risk factor for child morbidity and mortality globally [1], contributing

to 56% of all deaths in children under five years old and accounting for an estimated 5.6 million deaths

(5,000 each day) in 2016 [2]. These figures are even more striking in Sub-Saharan Africa (SSA), where in

2016 the World Health Organization (WHO) estimated that 32% of under-five children were chronically

undernourished, or stunted (low height-for-age), and 8% were acutely undernourished, or wasted (low

weight-for-height) [3]. Undernutrition can have serious implications for physical and cognitive

development and health status, and has been associated with diminished lifetime earning potential [4].

The Food and Agriculture Organization of the United Nations places its worldwide economic cost at up to

2.1 trillion dollars annually [5].

While the direct causes of undernutrition include insufficient dietary intake and absorption of calories and

micronutrients, improvements in many so-called “nutrition-sensitive” sectors [6] - including agriculture,

water, sanitation, social protection, education, and women’s empowerment - also contribute to outcomes.

Previous work has identified correlations between nutrition improvements and some of these underlying

determinants, such as food security, socioeconomic status, and women’s decision-making power [7–10].

Over the past several decades, however, the field of nutrition has moved away from interventions

targeting many of these supporting sectors, particularly agriculture, towards “nutrition-specific” programs

that focus on more direct determinants, such as caloric intake, breastfeeding promotion, and

micronutrients [11]. The evidence that nutrition-specific interventions lead to improvements in child

growth is relatively robust [12], while empirical evidence about the effectiveness of broader nutrition-

sensitive approaches, such as agriculture, is more scarce [13,14]. However, nutrition-specific

interventions in isolation may not be sufficient at reducing stunting [15]; in 2013, papers presented in the

Lancet Series on Maternal and Child Health concluded that nutrition-specific approaches, even at 90%

coverage, need to be paired with nutrition-sensitive programs to reduce current burdens by more than

20% [12].

Page 5 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


6

In response to the 2008 global food price crisis, the U.S. Agency for International Development (USAID),

with support from ten other U.S. Government (USG) agencies, announced the launch of Feed the Future

(FTF) in 2010. The program aims to reduce both poverty and stunting among children under five by 20%

in its target geographies [16] using a hybrid of nutrition-sensitive and -specific approaches. The majority

of FTF’s investments target food security and agriculture [17], including the promotion of high-quality

agricultural inputs (e.g. more nutritious crops, weather-resistant seeds, fertilizer), agricultural and post-

harvest infrastructure (e.g. development of food storage technologies and irrigation systems), financial

services for farmers (e.g. creation of agricultural banking and lines of credit), and private sector

engagement (e.g. partnerships with in-country agri-businesses). FTF also supports nutrition-specific

activities including management of acute malnutrition, breastfeeding promotion, and micronutrient

supplementation. FTF additionally incorporates women’s education and support for female farmers and

entrepreneurs into its activities. Details of FTF’s programmatic activities and the pathways through which

these interventions may impact nutritional outcomes is discussed in greater detail elsewhere [18,19].

FTF was rolled out in 19 focus countries between 2011 and 2012, including 12 in SSA that are the focus

of this study. USAID’s allocations to FTF have totaled around one billion dollars annually since 2010,

with additional funds coming from other agencies and programs [20].

While FTF is in its eighth year, to our knowledge there have been no comparative evaluations into the

program’s impact on nutritional outcomes relative to outcomes in non-intervention countries. Moreover,

there are few evaluations of any large multi-country nutrition programs in the literature and a need for

more evidence [21]. Such assessments are crucial to improving program design, resource allocation, and

scalability for FTF and similar initiatives. Here, we present an impact analysis of one of the world’s

largest nutrition and agriculture programs on child nutritional outcomes.

Page 6 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


7

METHODS

We analyzed the effectiveness of FTF by comparing trends in nutrition outcomes among under-five

children before and after FTF’s implementation in both FTF focus countries and countries where FTF did

not operate, using a difference-in-differences (DID) design. Our analysis compares exposure to FTF at a

national level while controlling for individual-, household-, and country-level characteristics.

Data Sources

Our primary outcome of interest is a binary indicator of stunting (height-for-age z-score less than two

standard deviations below the median of a WHO reference population) [22]. Reducing stunting is one of

FTF’s high-level objectives [23], and stunting is both reflective of chronic undernutrition [24] and closely

associated with socioeconomic conditions and health over a child’s lifetime [25]. We additionally

analyzed two secondary undernutrition outcomes: wasting (weight-for-height z-score less than two

standard deviations below the median) and underweight (weight-for-age z-score less than two standard

deviations below the median), which are also included in FTF’s published child nutrition goals [23].

We drew upon nationally-representative Demographic and Health Surveys (DHS) and Multiple Indicator

Cluster Surveys (MICS), both of which collect anthropometric data along with relevant maternal and

household characteristics [26]. Our analysis was restricted to surveys conducted between 2000 and 2017

in low- and lower-middle-income countries in SSA because of FTF’s geographic focus, the high burden

of undernutrition, and the ready availability of survey data in these areas. We calculated z-scores and

excluded implausible observations based on WHO criteria [22].

Patient and Public Involvement

By nature of the design of this study, which draws upon secondary data collected as part of two large

survey programs and made available for academic research, patients and the public were not directly

involved.

Page 7 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


8

Statistical Approaches

We used a DID approach to compare covariate-adjusted trends among children living in FTF countries to

those living in non-FTF countries. This method isolates changes in outcomes that are related to FTF

exposure while controlling for pre-existing differences, time-invariant country-level differences, and

secular time effects common to all countries [27]. Inferences from DID models depend on two main

assumptions: (1) that outcomes would continue to follow pre-intervention trends absent the program, and

(2) that these pre-intervention trends are similar (e.g. parallel) between treated and control populations.

We implemented both unadjusted and adjusted linear regression models. In our unadjusted model, the

main predictor variable was an interaction term between binary indicators of whether the child resides in a

FTF country (FTF) and whether the observation was made after FTF’s implementation (post), controlling

for each separate indicator (Equation 1). The coefficient on this interaction term (β3) reflects the

unadjusted average difference in the probability of stunting, wasting, or underweight among children

living in FTF countries after FTF’s implementation relative to children living in control countries and

prior to FTF’s implementation. In the equation, the subscripts i, t, and c refer to an individual, time

period (e.g. year), and country, respectively. We used linear probability models in these analyses due to

their simple interpretability and lower computational requirements, but we also tested logistic regression

models in sensitivity analysis (Appendix).

[1]𝑌𝑖𝑡𝑐 = 𝛽0 + 𝛽1(𝐹𝑇𝐹𝑐) + 𝛽2(𝑝𝑜𝑠𝑡𝑡) + 𝛽3(𝐹𝑇𝐹𝑐 ∗ 𝑝𝑜𝑠𝑡𝑡) + 𝜖𝑖𝑡𝑐

In our fully-adjusted multivariable analysis (Equation 2), we additionally included individual-level

covariates ( from the DHS and MICS for the child’s sex, age in years, maternal education, mother’s 𝑋𝑖𝑡𝑐)

age at birth, number of under-five siblings, household setting (urban or rural), household size, whether the

survey was administered during the rainy season, and access to an improved drinking water source;

country-level covariates ( ) for Gross National Income (GNI) per capita (p.c.) [28], life expectancy at 𝑍𝑡𝑐

birth [28], a governance score calculated from the World Bank’s World Governance Indicators [29],

Page 8 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


9

Diphtheria-Tetanus-Pertussis (DTP3) vaccination coverage [30], and non-US nutrition and agriculture

Official Development Assistance (ODA) [31]; country indicator variables (αc); and time indicator

variables (αt). Country and time indicator variables control for time-invariant national differences and

secular time trends common to all countries, respectively. We selected these covariates based on expert

opinion, covariates used in similar studies, and FTF’s stated selection criteria for its focus countries

[18,32–34]. Importantly, we believe these covariates could affect undernutrition levels but would not be

affected by FTF itself. Alternative covariates and combinations of covariates are tested in sensitivity

analyses, including the omission of some covariates that which could plausibly be affected by FTF (such

as life expectancy), alternate governance and ODA terms, and variation in the duration of time indicator

variables.

[2]𝑌𝑖𝑡𝑐 = 𝛽0 + 𝛽1(𝐹𝑇𝐹𝑐 ∗ 𝑝𝑜𝑠𝑡𝑡) + 𝛽𝑛1𝑋𝑖𝑡𝑐 + 𝛽𝑛2𝑍𝑡𝑐 + 𝛼𝑐 + 𝛼𝑡 + 𝜖𝑖𝑡𝑐

The coefficient of interest in the fully-adjusted model is β1 in Equation 2. The non-interaction FTF and

post terms from Equation 1 are deliberately omitted from Equation 2 because of collinearity with country

and time indicator variables; such treatment of these variables is standard in DID analyses [35]. We also

implemented combinations of the unadjusted and fully-adjusted models, with and without covariates,

country indicator variables, and/or time indicator variables. We applied sample weights (see

Supplementary Methods for details) and estimated robust standard errors clustered at the level of strata, a

combination of sub-national region and a household’s urban/rural designation. Strata was chosen because

FTF operates in specific sub-national regions, called “Zones of Influence”, and because activities differ

markedly between rural and urban areas. We varied the level at which errors are clustered in additional

analyses (Appendix).

We assessed the robustness of our findings through a range of additional sensitivity analyses, including

falsification tests on outcomes and on the year of FTF’s implementation. Detailed descriptions of these

and other sensitivity analyses can be found in the appendix. All analyses were conducted in 15 Stata

Page 9 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


10

(StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX: StataCorp LP). Our

analysis and reporting conform to STROBE guidelines (eTable 1) for observational studies [36].

RESULTS

We assembled data from a total of 118 DHS and MICS surveys: 49 surveys from FTF focus countries and

69 from control countries (Table 1). The surveys covered all 12 FTF countries in SSA and 21 control

countries. The cleaned dataset included 883,309 under-five children. Table 1 shows the number of

surveyed children under five years of age across all surveys in each country (see eTables 2-3 for more

details). Overall, 39% of children in our sample were stunted, 10% were wasted, and 23% were

underweight. Table 2 shows FTF group-level statistics pre-2012 for the countries in our analysis. Baseline

differences in wasting prevalence, average age, maternal education, number of under-five siblings,

average family size, governance, and life expectancy at birth were statistically significant. All of these

characteristics were controlled for in our analysis.

Page 10 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


11

Table 1: Study Countries, Years, Sample Sizes, and Group Designation

Country Survey Years # Children < 5FTF Focus Countries

Ethiopia 2000, 2004, 2011, 2016 32,499Ghana 2003, 2006, 2008, 2011, 2014 19,237Kenya 2000, 2003, 2008/09, 2014 36,118Liberia 2006/07, 2013 7,875Malawi 2000, 2004/05, 2006, 2010, 2013/14, 2015/16 69,661Mali 2001, 2006, 2009/10, 2012/13, 2015 64,252Mozambique 2003/04, 2008, 2011 28,832Rwanda 2000, 2005, 2010/11, 2014/15 17,930Senegal 2000, 2005, 2010/11, 2012-14, 2015/16, 2017 50,885Tanzania 2004/05, 2009/10, 2015/16 23,360Uganda 2000/01, 2006, 2011, 2016 14,389Zambia 2001/02, 2007, 2013/14 23,014Total 388,052

Non-FTF CountriesBenin 2001, 2006, 2014 30,009Burkina Faso 2003, 2010 15,586Burundi 2010/11, 2016/17 9,573Cameroon 2004, 2006, 2011, 2014 21,393Central African Rep. 2000, 2006, 2010 31,625Chad 2000, 2004, 2010, 2014/15 33,332Congo, Rep. 2005, 2011/12, 2014/15 17,421Cote d’Ivoire 2006, 2011/12, 2016 20,573Congo, Dem. Rep. 2007, 2010, 2013/14 22,893Gambia 2005/06, 2013 9,781Guinea 2005, 2012, 2016 13,141Guinea-Bissau 2000, 2006, 2014 18,498Lesotho 2000, 2004, 2009, 2014 8,137Madagascar 2003/04, 2008/09 9,989Mauritania 2007, 2011 16,452Niger 2000, 2006, 2012 13,975Nigeria 2003, 2007, 2008, 2011, 2013, 2016/17 120,058Sierra Leone 2005, 2008, 2010, 2013, 2017 31,339eSwatini 2000, 2006/07, 2010, 2014 10,578Togo 2006, 2010, 2013/14 11,430Zimbabwe 2005/06, 2009, 2010/11, 2014, 2015 29,474Total 495,257

Page 11 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


12

Table 2: Baseline Comparison of FTF Focus Countries with Non-Focus Countries

Variable FTF Countries Non-FTF Countries P ValueStunted (%) 43.6 [40.5-46.6] 41.8 [37.1-46.4] .53Wasted (%) 8.2 [7.0-9.5] 12.0 [10.2-13.8] .001Underweight (%) 23.4 [19.2-27.6] 25.7 [20.6-30.8] .49Child’s Age (years) 1.90 [1.88-1.93] 1.86 [1.84-1.88] .002 % 0-1 years old 22.3 [21.7-22.9] 23.8 [23.1-24.6] .002 % 1-5 years old 77.7 [77.1-78.3] 76.2 [75.4-76.9] .002Gender (% male) 50.3 [49.8-50.7] 50.5 [50.0-51.0] .57Maternal Education (%) Less than Primary 45.8 [33.7-57.8] 44.7 [32.3-57.1] .90 Some Primary 43.1 [34.1-52.2] 28.4 [23.1-33.7] .007 Some Secondary or Higher 11.1 [7.4-14.8] 26.9 [18.7-35.1] < .001Under-5 Siblings (%) 1.04 [0.97-1.12] 1.25 [1.16-1.34] < .001Mother’s Age (at birth) 26.7 [26.5-26.9] 26.6 [26.4-26.9] .64Family Size (#) 6.6 [6.4-6.9] 7.2 [6.9-7.4] .003Urban (%) 16.8 [9.4-24.2] 30.0 [15.6-44.4] .080Rainy Month (%) 51.4 [40.5-62.2] 61.1 [54.9-67.4] .12Improved Water (%) 50.3 [45.0-55.7] 49.0 [41.2-56.8] .78GNI p.c. ($) 450 [269-631] 533 [350-716] .50Governance 1.9 [1.6-2.1] 1.4 [1.1-1.7] .003DTP3 Coverage (%) 70 [53-87] 58 [40-77] .31Life Expectancy (years) 57 [55-58] 52 [48-56] .004Nut. + Ag. ODA p.c. ($ Millions)

2.4 [1.5-3.3] 1.2 [0.2-2.2] .078

This table shows weighted means and p-values corresponding to a two-tailed t-test comparing differences in weighted means (with clustered standard errors) between FTF and non-FTF countries before FTF’s implementation (from surveys conducted between 2000 and 2011).

Page 12 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


13

Table 3 shows the unadjusted and adjusted regression results for both primary and secondary outcomes.

We found a statistically significant decrease in all three measures of undernutrition. When we do not

adjust for covariates, time periods, or countries (column 1 in Table 3), we find that FTF was associated

with a 9.7 percentage point decrease in stunting prevalence as well as a 3.3 percentage point decrease in

wasting prevalence and a 9.4 percentage point decrease in underweight prevalence. The coefficient of

interest remains significant under specifications with and without covariates, country indicator variables,

and time indicator variables (all columns in Table 3). We focus our reporting and interpretation on the

fully-adjusted regression results, corresponding to Equation 2 in the Methods (column 8 in Table 3;

Figure 1). After FTF was implemented, controlling for all covariates, the average child in a FTF focus

country was 2.4 percentage points (95% CI 0.6 to 4.2) less likely to be stunted, 1.7 percentage points

(95% CI 0.2 to 3.2) less likely to be wasted, and 2.4 percentage points (95% CI 0.8 to 4.1) less likely to

be underweight relative to children in non-FTF countries and children surveyed before FTF’s

implementation.

We used these results to estimate the number of children who were prevented from becoming stunted,

wasted, and underweight because of FTF’s activities (see Supplementary Methods for details). We find

that FTF has resulted in 1.3 million (95% CI 0.3 to 2.2) fewer children stunted, 0.9 million (95% CI 0.1 to

1.7) fewer children wasted, and 1.3 million (95% CI 0.4 to 2.2) fewer children underweight.

Given the observational nature of our study, we tested for possible effect misattributions using a wide

range of sensitivity analyses. Our findings remain stable under most of these analyses, including models

that use logistic regression; include, omit, or vary certain country-level covariates; exclude or add

additional surveys and countries; and cluster errors at the household level (eTables 4-6). However, there

are some exceptions. Focusing on stunting, we find that results remain relatively stable but lose some

statistical significance in a few sensitivity analyses, including clustering errors at the country level and

adding in domestic health expenditure per capita or total ODA as covariates. Our results pass falsification

tests in which we (a) treat pre-FTF years as treatment years (for stunting and wasting) and (b) test the

Page 13 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


14

impact of FTF on outcomes which it should not plausibly affect (eTable 7). We hypothesize that FTF’s

impact should be greater in later years, especially for chronic undernutrition outcomes such as stunting,

since children in later surveys would have been more exposed to the program. We find evidence that

FTF’s impact on stunting and underweight levels has increased over time, via the addition of an

interaction term between whether a country is a FTF focus country and the number of years after FTF’s

implementation (eTable 8). We also find that our results generally remain stable when single countries are

dropped from the analysis, indicating that our results are probably not being driven by improvements in a

single country (eTable 9), although there are some exceptions (e.g. stunting results become less

statistically significant when Democratic Republic of the Congo or Ghana are removed from the

analysis). A detailed discussion of all sensitivity analyses is included in the appendix.

Page 14 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


15

Table 3: Estimates of the association between FTF and undernutrition from unadjusted and adjusted regression models

(1) (2) (3) (4) (5) (6) (7) (8)-9.66*** -12.09*** -8.64*** -3.71*** -7.13*** -4.99*** -5.34*** -2.40***Stunting (1.29) (0.62) (2.73) (1.13) (0.80) (0.49) (1.10) (0.90)

N 858,937 858,937 858,937 858,937 794,236 794,236 794,236 794,236-3.33*** -1.58*** -5.93*** -4.29*** -2.33*** -0.97** -3.44*** -1.69**Wasting (1.20) (0.36) (1.53) (1.11) (0.49) (0.39) (0.74) (0.75)

N 850,872 850,872 850,872 850,872 786,498 786,498 786,498 786,498-9.41*** -7.47*** -10.80*** -5.65*** -5.95*** -2.43*** -7.09*** -2.43***Underweight (1.84) (0.84) (3.01) (1.49) (0.74) (0.48) (1.22) (0.84)

N 873,708 873,708 873,708 873,708 807,455 807,455 807,455 807,455Covariates No No No No Yes Yes Yes Yes

Time Indicators No No Yes Yes No No Yes Yes

Country Indicators No Yes No Yes No Yes No Yes

Estimated coefficients on the FTF x Post term are shown in percentage points. All regressions were run with weights and strata-clustered standard errors. Robust standard errors in parentheses; *** indicates p<0.01, ** indicates p<0.05, * indicates p<0.1

Page 15 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


16

The causal strength of a DID analysis depends on the assumption that outcomes would be similar between

treatment and control groups absent the intervention. While this assumption is not directly testable, one

way to probe it is to examine whether FTF and control countries were on similar undernutrition

trajectories prior to FTF’s implementation. We found that pre-FTF trends in outcomes were generally

parallel between treatment and control countries by using both graphical assessment and a regression of

each outcome on survey year, a FTF binary indicator, and an interaction between the two, using data from

pre-FTF years (eTable 10, eFigure 1).

DISCUSSION

This analysis examines the impact of FTF, a multi-billion-dollar program which aims to reduce poverty

and improve nutritional outcomes by focusing on nutrition, poverty, food security, agriculture, and the

status of women. FTF’s implementation is associated with approximately two percentage point

reductions in stunting, wasting, and underweight prevalence among under-five children in its 12 focus

countries in SSA. The magnitude and statistical significance of these findings, especially for stunting, the

primary outcome of interest, are robust to many but not all alternative model specifications and sensitivity

analyses. Our estimated effect sizes correspond to approximately one million fewer stunted, wasted, and

underweight children. We find compelling evidence that a multi-sectorial program focused predominantly

on agriculture is linked to significant improvements in child nutrition indicators within five to six years

after the full rollout of the program. This result has important implications for country governments and

global development partners in low- and middle-income countries.

We hypothesize that three features of FTF’s design - its country-tailored approach, its focus on underlying

drivers of undernutrition, and its large scale and volume of funding – may have made it a more effective

program. FTF combines a nutrition-specific approach with one that addresses agricultural production,

capacity, and markets; food security; poverty reduction; and women’s empowerment. These multi-

sectorial interventions are intuitively complementary: improved access to technological and financial

Page 16 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


17

instruments that facilitate food production and income generation increases the availability of nutritious

foods and the purchasing power of households, which promotes investments in child health and nutrition

[37]. This impact is enhanced by supporting women, who are uniquely focused on their children’s

wellbeing and whose health during pregnancy is linked to birthweights and early child growth. Previous

research has hypothesized that international nutrition efforts often fail because of inadequate funding and

a lack of focus on national priorities and capacity development [21], and has emphasized the importance

of context-specific solutions [15]. FTF works with in-country experts and stakeholders to identify gaps

and opportunities across multiple sectors country-by-country and address them at a large scale financially:

annual FTF funding is comparable with total ODA for all nutrition-specific programs [38,39].

The specific pathways of FTF’s estimated impacts are challenging to isolate both because its activities

vary by country and because the pathways connecting agriculture and nutrition are complex [40].

Nevertheless, future work could explore these pathways by testing the impact of FTF on intermediate

outcomes that were not available in both DHS and MICS surveys, such as gender equity, dietary

diversity, agricultural production, and poverty, which are themselves intrinsically important indicators.

Our results can be paired with FTF’s own monitoring and evaluation efforts to understand which

particular interventions have achieved success in individual countries [41]. Identifying which

programmatic elements are linked to improvements in nutritional outcomes could help FTF and other

nutrition initiatives improve their program designs and achieve sustained success going forward.

Understanding any unintended negative consequences that are masked by larger-scale improvements,

through shifts toward the production of cash crops or market effects on food prices, for instance, will be

critical as well.

It is important to understand our findings in the context of FTF’s policies and the broader global burden

of undernutrition. Although level of need is one of FTF’s country selection criteria, FTF also targets

countries based on projected capacity to absorb and benefit from its activities, using metrics such as

potential for in-country partnership and domestic resource availability [18]. FTF focus countries included

Page 17 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


18

in our analysis had, on average, higher life expectancies, better governance, and lower levels of wasting

than non-FTF countries. In the years included in this analysis, the program’s activities covered less than

40% of stunted children in SSA and FTF operated in only five of the ten SSA countries with the highest

burden of stunting. While we try to control for variables associated with FTF’s selection criteria in our

analysis (e.g. governance) as much as possible, it remains to be seen whether FTF’s approach is scalable

to other high-burden countries with less developed infrastructure, stability, or governmental commitment

to nutrition. Beyond generalizability, we cannot completely rule out the possibility that unobserved time-

varying factors, perhaps related to selection for the program, could be uniquely affecting nutritional

outcomes in FTF countries. The assumption that pre-FTF trends in nutrition would continue absent FTF is

not directly testable; however, country indicator variables control for non-time-varying attributes that

differentiate FTF countries from control countries. Furthermore, our stunting and wasting analyses pass

falsification tests and we find evidence of parallel pre-FTF trends in all three outcomes, indicating the gap

in outcomes between focus and control countries must have arisen around the time of FTF’s rollout.

We therefore reason that our analytical approach would threatened if other major changes occurred

around 2012 and affected primarily FTF or primarily control countries. Furthermore, these changes would

have had to affect nutrition levels but be unrelated to the extensive set of factors we control for: economic

performance; governance metrics, including institutions, corruption, and stability; donor involvement; and

strength of health systems. We conducted a qualitative review to identify any such potential threats to

validity – including ODA (eFigure 2), new country and donor health and nutrition programs, natural

disasters, droughts, armed conflicts, and disease outbreaks. While we control for non-FTF nutrition-

related ODA, we identified a major increase in funding to several focus countries from the Canadian

government around the time FTF began. We tested the extent to which this change could be influencing

results (eTable 7) and found it was not associated with significant effects on stunting and wasting;

however, it is possible that a combination of FTF and Canadian programs are influencing underweight

outcomes in a selection of countries where both programs operate. We did not identify any other

Page 18 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


19

programs of the same scope and time in the same countries, or any other alternate drivers that could be

uniquely improving nutrition outcomes in FTF countries starting around 2012 other than FTF itself.

Certain shocks have occurred in specific countries during this time period, but they are all countries for

which results remained robust when surveys from that country were dropped (eTable 9). However, we

remain cautious in attributing the observed nutritional improvements to FTF. We find that stunting results

lose significance at p<0.05 when total ODA or country-level health expenditure are included as covariates

in the regression. Although this may be due to missing data (country health expenditure data was missing

for several country-year combinations) and the fact that FTF support is included in total ODA (thus

double-counting FTF), this leaves open the possibility that non-FTF funding changes or governmental

commitments may have also played a role in reducing levels of undernutrition in FTF countries.

There are other limitations to our findings. While our analysis is conducted at the national level, FTF is

generally administered to a set of sub-national regions (“Zones of Influence”, ZOI) in each focus country.

We were unable to determine the precise locations of every ZOI from published documents and many

surveys in our sample do not contain GIS data that would allow us to match observations with ZOIs. Our

regressions thus average effects across treated and untreated areas within a focus country, meaning that

we may actually underestimate FTF’s true effect on the specific areas where it operates. Furthermore, our

analysis is weighted such that large countries have more impact on the effect sizes than small countries,

and the full distribution of effects across all countries may include null (or even negative) effects in some

countries that are not identifiable given our design. For instance, we observe distinct decreases in stunting

after FTF’s implementation in several FTF countries (e.g. Malawi, Mali, Senegal, Tanzania) but more

stable or flat-line trends in others (e.g. Ghana, Uganda) (eFigure 3). While these descriptive analyses are

not adjusted for relevant covariates, they do point to the potential for heterogeneous effects across

countries. Future work could investigate impact within specific countries at the level of ZOI.

Despite these limitations, our findings are a substantial addition to the current evidence base on the

impact of large-scale nutrition programs. Although the USG’s own evaluations of FTF show positive

Page 19 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


20

results [41], ours builds upon their efforts by pooling individual data across multiple countries,

controlling for potential confounders, and comparing impact to non-FTF countries. Notably, while USG’s

evaluation concludes that FTF has prevented 3.4 million children from becoming stunted, our estimated

impact of around 1 million fewer children stunted, while substantial, is much smaller. Studies have shown

the importance of addressing the indirect determinants of poor nutrition, such as agricultural production,

food security, socioeconomic status, and female empowerment [6–9], but to date there has been a lack of

large-scale, robust analyses evaluating multi-sectorial programs like FTF that focus on these indirect

determinants [6]. Most nutrition program evaluations focus on interventions implemented in one or two

countries at most [42,43]. FTF’s rollout across several countries offers an ideal natural experiment to infer

the effectiveness of a large multi-faceted approach at scale.

Our results are consistent with the hypothesis that the average child residing in a FTF focus country is

less likely to be undernourished than they would have been absent FTF. FTF is now concentrating its

activities on a smaller set of 12 focus countries worldwide, including eight in SSA. While our findings

indicate initial progress, it will be several years before we know whether FTF’s efforts were sustainable in

either its original or remaining focus countries. Although two to three percentage point decreases in

undernutrition outcomes in a subset of countries are noteworthy, many countries face stunting levels

above 30%. The World Health Assembly has specified six nutrition targets to meet globally by 2025,

including reducing stunting by 40% and reducing wasting levels to less than 5% [44], and the United

Nation’s Sustainable Development Goals include ending all forms of malnutrition by 2030 [45]. Based on

our analysis, FTF may be helping to move the needle, but its impact alone will not be enough to meet

these targets. While our study shows the promise of an approach like FTF’s, addressing the complexities

of undernutrition in the long term will require sustained effort across multiple sectors for years to come.

Page 20 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


21

DECLARATIONS AND ACKNOWLEDGEMENTS

Transparency Declaration: The lead author of this study, TR, affirms that the manuscript is transparent,

honest, and accurate and that no important aspects of the study have been omitted.

Dissemination Declaration: Dissemination to study participants and patient organizations is not

applicable.

Data Sharing: All data used in the analysis are available through the Demographic and Health Surveys

(DHS) and Multiple Indicator Cluster Surveys (MICS) websites for academic researchers.

Ethics Approval: As all data used in this study were publicly available, ethics approval was not required.

Competing Interests: All authors have completed the ICMJE uniform disclosure form at

http://www.icmje.org/coi_disclosure.pdf and declare: no support from any organization specifically for

the submitted work, no financial relationships with any organizations that might have an interest in the

submitted work in the previous three years, and no other relationships or activities that could appear to

have influenced the submitted work.

Contributorship: EB and TR were involved in initial conceptualization and methodology development.

TR and MR cleaned, curated, and analyzed the data. CP, MR, and TR contributed to the first draft of the

manuscript. EB, JB, TR, MR, and CP contributed to study design, interpretation of results, development

of visualizations, writing, and editing. EB and JB were involved in validation of results and conclusions.

All authors have read and approved the final draft of the manuscript. All authors had access to all data in

the study and had final responsibility for the decision to submit for publication. The corresponding author

attests that all listed authors meet authorship criteria and that no others meeting the criteria have been

omitted.

Exclusive License: The Corresponding Author has the right to grant on behalf of all authors and does

grant on behalf of all authors, a worldwide license

Page 21 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


22

(http://www.bmj.com/sites/default/files/BMJ%20Author%20Licence%20March%202013.doc) to the

Publishers and its licensees in perpetuity, in all forms, formats and media (whether known now or created

in the future), to i) publish, reproduce, distribute, display and store the Contribution, ii) translate the

Contribution into other languages, create adaptations, reprints, include within collections and create

summaries, extracts and/or, abstracts of the Contribution and convert or allow conversion into any format

including without limitation audio, iii) create any other derivative work(s) based in whole or part on the

on the Contribution, iv) to exploit all subsidiary rights that currently exist or as may exist in the future in

the Contribution, v) the inclusion of electronic links from the Contribution to third party material where-

ever it may be located; and, vi) license any third party to do any or all of the above. All research articles

will be made available on an open access basis (with authors being asked to pay an open access fee—see

http://www.bmj.com/about-bmj/resources-authors/forms-policies-and-checklists/copyright-open-access-

and-permission-reuse). The terms of such open access shall be governed by a Creative Commons

license—details as to which Creative Commons license will apply to the research article are set out in our

worldwide license referred to above.

Role of the Funding Source: The authors received no funding specifically for this work and have

declared that no competing interests exist. EB receives funding from the Doris Duke Charitable

Foundation (http://www.ddcf.org) and the National Institute for Allergy and Infectious Diseases

(https://www.niaid.nih.gov). JB was partially supported by the US National Institute on Aging

(https://www.nia.nih.gov) through the Stanford Center for Demography and Economics of Health and

Aging (https://cdeha.stanford.edu). TR receives support for her PhD research as a National Science

Foundation Graduate Research Fellow (https://www.nsfgrfp.org) and Stanford Graduate Fellow in

Science and Engineering (https://vpge.stanford.edu/fellowships-funding/sgf). The funders had no role in

study design, data collection and analysis, decision to publish, or preparation of the manuscript.

This material is based upon work supported by the National Science Foundation Graduate Research

Fellowship Program under Grant No. DGE-1656518. Any opinions, findings, and conclusions or

Page 22 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


23

recommendations expressed in this material are those of the authors and do not necessarily reflect the

views of the National Science Foundation.

Additional Acknowledgments: We are grateful for the advice of Dr. Kiersten Johnson, Monitoring and

Evaluation Advisor at the USAID Bureau for Food Security, who provided clarifications on the structure

and early implementation of FTF.

Page 23 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


24

REFERENCES

1 GBD Compare | IHME Viz Hub. http://vizhub.healthdata.org/gbd-compare (accessed 13 Mar 2018).

2 World Health Organization (WHO). Children: reducing mortality. Published Online First: 2016.http://www.who.int/mediacentre/factsheets/fs178/en/

3 United Nations Children’s Fund (UNICEF), World Health Organization (WHO), World Bank Group. Levels and Trends in Child Malnutrition: Joint Child Malnutriton Estimates. 2017. http://www.who.int/nutgrowthdb/jme_brochure2017.pdf (accessed 10 Oct 2017).

4 Almond D, Currie J. Killing Me Softly: The Fetal Origins Hypothesis. J Econ Perspect 2011;25:153–72. doi:10.1257/jep.25.3.153

5 The State of Food and Agriculture 2013 | FAO | Food and Agriculture Organization of the United Nations. http://www.fao.org/publications/sofa/2013/en/ (accessed 8 Jan 2019).

6 Ruel MT, Alderman H, Maternal, et al. Nutrition-sensitive interventions and programmes: how can they help to accelerate progress in improving maternal and child nutrition? Lancet 2013;382:536–51. doi:10.1016/S0140-6736(13)60843-0

7 Nguyen PH, Headey D, Frongillo EA, et al. Changes in Underlying Determinants Explain Rapid Increases in Child Linear Growth in Alive & Thrive Study Areas between 2010 and 2014 in Bangladesh and Vietnam. J Nutr 2017;147:462–9. doi:10.3945/jn.116.243949

8 Smith LC, Ramakrishnan K, Ndiaye A, et al. The Importance of Women’s Status for Child Nutrition in Developing Countries. Washington, D.C.: : International Food Policy Research Institute, 2003. http://ageconsearch.umn.edu/bitstream/16526/1/rr030131.pdf

9 Akseer N, Bhatti Z, Mashal T, et al. Geospatial inequalities and determinants of nutritional status among women and children in Afghanistan: an observational study. Lancet Glob Health Published Online First: 14 February 2018. doi:10.1016/S2214-109X(18)30025-1

10 Carletto G, Ruel M, Winters P, et al. Farm-Level Pathways to Improved Nutritional Status: Introduction to the Special Issue. J Dev Stud 2015;51:945–57.

11 Herforth A, Tanimichi-Hoberg Y. Learning from World Bank history: agriculture and food-based approaches for addressing malnutrition. Washington, DC: : World Bank Group 2014. http://documents.worldbank.org/curated/en/497241468168227810/Learning-from-World-Bank-history-agriculture-and-food-based-approaches-for-addressing-malnutrition

12 Bhutta ZA, Das JK, Rizvi A, et al. Evidence-based interventions for improvement of maternal and child nutrition: what can be done and at what cost? Lancet 2013;382:452–77. doi:10.1016/S0140-6736(13)60996-4

13 Pandey VL, Mahendra Dev S, Jayachandran U. Impact of agricultural interventions on the nutritional status in South Asia: A review. Food Policy 2016;62:28–40. doi:10.1016/j.foodpol.2016.05.002

14 Masset E, Haddad L, Cornelius A, et al. Effectiveness of agricultural interventions that aim to improve nutritional status of children: systematic review. BMJ 2012;344:d8222.

Page 24 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


25

15 Hossain M, Choudhury N, Adib Binte Abdullah K, et al. Evidence-based approaches to childhood stunting in low and middle income countries: a systematic review. Arch Child 2017;102:903–9. doi:10.1136/archdischild-2016-311050

16 United States Agency for International Development (USAID). Feed the Future Snapshot: Progress Through 2017. Washington, D.C.: : USAID 2017. https://feedthefuture.gov/sites/default/files/resource/files/2017%20Feed%20the%20Future%20Progress%20Snapshot.pdf (accessed 10 Oct 2017).

17 Elliott K, Dunning C. Assessing the US Feed the Future Initiative: A New Approach to Food Security. Center for Global Development 2016.

18 United States Agency for International Development (USAID). Feed the Future Guide. Washington, D.C.: : USAID 2010. https://feedthefuture.gov/sites/default/files/resource/files/FTF_Guide.pdf

19 Du L, Pinga V, Klein A, et al. Leveraging Agriculture for Nutrition Impact through the Feed the Future Initiative. Adv Food Nutr Res 2015;74:1–46. doi:10.1016/bs.afnr.2014.11.001

20 Lawson ML, Schnepf R, Cook N. The Obama Administration’s Feed the Future Initiative. Washington, D.C.: : Congressional Research Service 2016. https://fas.org/sgp/crs/row/R44216.pdf

21 Morris SS, Cogill B, Uauy R, et al. Effective international action against undernutrition: why has it proven so difficult and what can be done to accelerate progress? Lancet 2008;371:608–21. doi:10.1016/S0140-6736(07)61695-X

22 WHO | WHO Anthro (version 3.2.2, January 2011) and macros. WHO. http://www.who.int/childgrowth/software/en/ (accessed 4 Jan 2019).

23 United States Agency for International Development (USAID). M&E Guidance Series Volume 1: Monitoring and Evaluation Under Feed the Future. Washington, D.C.: : USAID https://www.feedthefuture.gov/sites/default/files/resource/files/ftf_guidance_volume1_overview_2015.pdf (accessed 10 Oct 2017).

24 Barrett CB. Measuring Food Insecurity. Science 2010;327:825–8. doi:10.1126/science.1182768

25 Hoddinott J, Alderman H, Behrman JR, et al. The economic rationale for investing in stunting reduction. Matern Child Nutr 2013;9:69–82. doi:10.1111/mcn.12080

26 Hancioglu A, Arnold F. Measuring coverage in MNCH: tracking progress in health for women and children using DHS and MICS household surveys. PLoS Med 2013;10:e1001391. doi:10.1371/journal.pmed.1001391

27 Basu S, Meghani A, Siddiqi A. Evaluating the Health Impact of Large-Scale Public Policy Changes: Classical and Novel Approaches. Annu Rev Public Health 2017;38:351–70. doi:10.1146/annurev-publhealth-031816-044208

28 World Bank. Indicators. https://data.worldbank.org/indicator

29 World Bank. World Governance Indicators. http://info.worldbank.org/governance/wgi/#home

Page 25 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


26

30 World Health Organization (WHO). Diptheria tetanus toxoid and pertussis (DTP3) Immunization coverage estimates by country. 2017.http://apps.who.int/gho/data/node.main.A827

31 OECD. Detailed aid statistics: Official bilateral commitments by sector. OECD Int. Dev. Stat. Database. 2018.http://dx.doi.org/10.1787/data-00073-en

32 Bendavid E, Holmes CB, Bhattacharya J, et al. HIV development assistance and adult mortality in Africa. JAMA 2012;307:2060–7. doi:10.1001/jama.2012.2001

33 Jakubowski A, Stearns SC, Kruk ME, et al. The US President’s Malaria Initiative and under-5 child mortality in sub-Saharan Africa: A difference-in-differences analysis. PLoS Med 2017;14:e1002319. doi:10.1371/journal.pmed.1002319

34 Lakkam M, Wager S, Wise PH, et al. Quantifying and exploiting the age dependence in the effect of supplementary food for child undernutrition. PLoS One 2014;9:e99632. doi:10.1371/journal.pone.0099632

35 Bertrand M, Duflo E, Mullainathan S. How Much Should We Trust Difference-in-Differences Estimates? Q J Econ 2002;119:249–75.

36 von Elm E, Altman DG, Egger M, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet 2007;370:1453–7. doi:10.1016/S0140-6736(07)61602-X

37 Herforth A, Jones A, Pinstrup-Andersen P. Prioritizing Nutrition in Agriculture and Rural Development: Guiding Principles for Operational Investments. Washington, DC: : World Bank 2012. https://openknowledge.worldbank.org/bitstream/handle/10986/13571/NonAsciiFileName0.pdf?sequence=1&isAllowed=y

38 Development Initiatives. Global Nutrition Report 2017: Nourishing the SDGs. Bristol, UK: : Development Initiatives 2017. https://www.globalnutritionreport.org/files/2017/11/Report_2017.pdf (accessed 23 Feb 2018).

39 Henry J Kaiser Family Foundation (KFF). U.S. Funding for International Nutrition Programs. KFF 2016. https://www.kff.org/global-health-policy/issue-brief/u-s-funding-for-international-nutrition-programs/ (accessed 31 Mar 2018).

40 Gillespie S, Harris J, Kadiyala S. The Agriculture-Nutrition Disconnect in India: What Do We Know? Washington, D.C.: 2012. http://ebrary.ifpri.org/cdm/ref/collection/p15738coll2/id/126958

41 Briggs L, Vondal P, Vijayakumar C, et al. Feed the Future Global Performance Evaluation Report. Washington, D.C.: : USAID 2016.

42 Kim SS, Rawat R, Mwangi EM, et al. Exposure to Large-Scale Social and Behavior Change Communication Interventions Is Associated with Improvements in Infant and Young Child Feeding Practices in Ethiopia. PLOS ONE 2016;11:e0164800. doi:10.1371/journal.pone.0164800

43 Menon P, Nguyen PH, Saha KK, et al. Impacts on Breastfeeding Practices of At-Scale Strategies That Combine Intensive Interpersonal Counseling, Mass Media, and Community Mobilization: Results of Cluster-Randomized Program Evaluations in Bangladesh and Viet Nam. PLOS Med 2016;13:e1002159. doi:10.1371/journal.pmed.1002159

Page 26 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


27

44 World Health Organization (WHO). Global nutrition targets 2025: policy brief series (WHO/NMH/NHD/14.2). Geneva: : World Health Organization 2014.

45 General Assembly resolution 70/1, Transforming our world: the 2030 Agenda for Sustainable Development, A/RES/70/1 (25 September 2015), available from undocs.org/A/RES/70/1.

Page 27 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Figure 1: Adjusted nutrition outcomes before and after FTF Notes: Overall differences, standard errors, and p-values were estimated from the fully-adjusted regression

model with covariates and country and time indicator variables.

158x482mm (150 x 150 DPI)

Page 28 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


1

APPENDIX: ONLINE SUPPLEMENT

Contents:

Supplementary MethodsSurvey WeightsMethods for Estimates of Total Reduction in BurdenDetailed Descriptions of Sensitivity Analyses

eTable 1: STROBE Checklist

eTable 2: Survey & Sample Size Details

eTable 3: Survey Availability by Year

eTable 4: Major Sensitivity Analysis Results

eTable 5: Development Assistance Sensitivity Analysis Results

eTable 6: Governance Sensitivity Analysis Results

eTable 7: Falsification Test Results

eTable 8: Alternate Start Year Specifications & Lagged Treatment Effect Models

eTable 9: Leave-One-Out Analysis Results

eTable 10: Parallel Trends Regression Results

eFigure 1: Pre-FTF Trends in Undernutrition

eFigure 2: Development Assistance to FTF and Control Countries

eFigure 3: Undernutrition Trends by country

Page 29 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


2

Supplementary Methods

Survey Weights

In all regressions, we applied sample weights for each child consisting of a combination of survey-specific weights that accounted for DHS and MICS sampling strategies together with investigator-created weights that accounted for survey size and country population [1]. Within-survey weights were taken from the MICS and DHS and were designed to account for unbalanced sampling within each survey. We scaled these to sum to one within each survey and then adjusted these weights using a between-survey weight equal to the ratio of the number of children in the survey to the number of children in the country. Between-survey weights account for the fact that each sample is representing a different proportion of the entire country’s population in each survey (e.g. a survey of 10,000 in a country of one million compared to a survey of 10,000 in a country of ten million) [1].

Methods for Estimates of Total Reductions in Burden

We estimated the reductions in total numbers of children stunted, wasted, and underweight that can be attributed to FTF by combining the percentage point effect sizes estimated from the fully-adjusted model with under-five population estimates from the WHO and survey-derived undernutrition prevalence estimates. For each FTF country, we obtained baseline undernutrition prevalence estimates from the most recently publicly available survey prior to 2011, using joint data compiled by the WHO, UNICEF, and the World Bank. We applied the percentage point reductions derived from our fully-adjusted analysis to estimate post-FTF prevalence rates. The product of the difference between before and after prevalence estimates and under-five population provides an approximate number of child undernutrition cases averted.

Detailed Descriptions of Sensitivity Analyses

eTable 4 shows the coefficient on the FTF-Post interaction term under several variations on the regression specification of equation 2 in the main text. First, results of the base case model specification using both linear (#1) and logistic (#2) models are displayed, since our outcomes are binary variables. The linear model (#1) is the same as Table 3 (#8) in the main text and is included for reference. We find similar results from the logistic model as the base case linear model. The remaining analyses are run using alternate specifications of the base case model; all were run with a linear regression model, the standard set of covariates, weights, country and time indicator variables, and robust standard errors, unless otherwise noted. We varied combinations of covariates to determine whether our results were robust to alternative plausible specifications (#3-#7). In particular, we were concerned that life expectancy and GNI p.c. were both proxies for a country’s levels of economic development or that they could be affected by FTF itself, and so we tested removing these covariates (#3 and #4). We also experimented with adding in additional country-level information, including measles first dose (MCV) coverage, current health expenditures per capita (CHE p.c.), and general government domestic health expenditures per capita (GGHE-D p.c.) to ensure that no major country-level factor associated with our outcomes had been missed (#5-#7). Results for all three undernutrition outcomes are robust to removing GNI p.c., life expectancy, and MCV coverage. Wasting and underweight results are robust to the addition of health expenditure variables; however, the stunting results become less significant when GGHE-D p.c. is added. Data on CHE p.c. and GGHE-D p.c. are missing for several countries in several years and had to be imputed; these variables were therefore omitted from the base case analysis, in part because of concern around imputation error. Thus, this finding is less concerning as it may be a result of inaccurately estimated imputations. GGHE-D p.c. may also be highly correlated with other covariates, such as GNI p.c. or DTP3 coverage. Nutrition ODA and governance data were also imputed for some years and countries, so we also tested a model specification in which observations with missing data are dropped, rather than using imputed data (#8) and find that results are robust to both methods of handling missing data.

Page 30 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


3

In our base case analysis, we used three-year indicator variables rather than single-year indicator variables because of our interest in longer-term trends and compositional differences in consecutive years, such that our sample of surveys had sufficient homogeneity within the FTF and non-FTF groups in a given three-year period, but not in a one-year period (see eTable 3). These country and time indicator variables are equivalent to fixed effects in econometric terminology; we chose fixed effects rather than random effects because random effects are biased in the presence of unobservable fixed differences between countries or time periods [2]. Regressions #9 and #10 include one-year and two-year indicator variables/fixed effects rather than three-year indicator variables. We believe three-year indicators are more accurate given our unbalanced sample, since one-year indicators would capture variation based on the heterogeneity of the small set of countries with a survey in a given year and not just time trends, but we included this model because it is generally standard to run a time-based DID with time unit indicator variables. Thus, we believe the lack of significant results for wasting and underweight when one-year and two-year indicators are included is due to this heterogeneity and not due to some sort of actual cross-country annual variation that is being missed in the other models.

We also tested a version of the linear core model that clustered standard errors at the more discrete household-level (#11), rather than using stratum-clustered errors. This model accounts for the fact that one additional child from one additional household gives us more useful information than one additional child from an already-included household, since nutrition outcomes are likely correlated within a household. Regression #12 clusters errors at the less discrete country-level. We hypothesize that this specification will be less accurate because FTF is implemented at the sub-national level, but we included regression #12 because similar difference-in-differences analyses of global health programs use country-clustered errors [3,4]. All results except for stunting and wasting (#12 only) remain significant under alternate handling of errors.

In our analysis, we calculate z-scores and the presence of stunting, wasting, and underweight manually based on WHO guidance and published code, in order to ensure consistency [5]. In regression #13, we use the z-scores from the DHS and MICS surveys. Regressions #14 and #15 tests the analysis using only DHS data and only MICS data, respectively. While MICS and DHS use similar techniques to measure and flag anthropometric data [6,7], we wanted to ensure that differences between the two surveys were not biasing our results. Our results are not robust to including MICS surveys only. However, upon further investigation of this results, we found that there were only two MICS surveys available in the post-FTF period in FTF countries (and 11 DHS surveys that are removed), thus we believe this finding is not representative of our sample as a whole. Regression #16 removes a few countries that did not conduct a survey both before and after FTF was rolled out. While country indicator variables prevent countries with only one survey year from affecting the results, some countries had only one survey that spanned two years or multiple surveys, but only before 2012 or only after 2011. We wanted to make sure we this variation was not somehow biasing our results. Finally, in regressions #17 and #18 we added in originally-excluded poor quality surveys and upper-middle-income country (UMIC) surveys, respectively. To improve precision, we had excluded from the base case results all or part of ten surveys based on poor anthropometric data quality, high levels of missing observations, or missing variables. These included the DHS for Benin 2011[8] and Burundi 2000, and the 2000 MICS for Cameroon, Cote d’Ivoire, Gambia, Madagascar, Rwanda, Sierra Leone, and Togo. All results remain stable and significant under these six specifications (#13-18), with the exception of #15, when only MICS data are used.

Additional tests using variations on regression covariates are shown in eTable 5 and eTable 6. In eTable 5 we test alternate overseas development assistance (ODA) covariates, including assistance for different sectors (nutrition, agriculture, health, all social sectors, and all sectors) and from different sources (OECD DAC database [11] and IHME’s Development Assistance for Health Database [12]). While we hypothesized that nutrition and agriculture ODA would be the most relevant variable to control for, ODA for these other sectors could indirectly impact nutrition outcomes as well. Including ODA in the regressions ensures that we do not misattribute an effect to FTF that is actually caused by an increase in aid to FTF countries (from sources other than FTF itself) starting around

Page 31 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


4

2012. We exclude US ODA in the base case because evidence indicates that the majority of nutrition funding from the US government comes from FTF [13], but we also test versions of ODA with the US included in the table. We find that are results are robust to the vast majority of ODA variations, with one exception (stunting and total ODA). eFigure 2 further shows how various ODA indicators varied between FTF and control countries over time.

eTable 6 similarly varies the governance covariate included in the regressions. Again, we hypothesized that all six elements of the World Bank’s World Governance Indicators – control of corporations, government effectiveness, stability and absence of violence, regulatory quality, rule of law, and voice and accountability – could be relevant to outcomes and so used a composite governance score in our base case (an unweighted average of scores on all six metrics). However, we find that including any one governance indicator by itself, including all six indicators separately, and excluding governance entirely do not substantially affect our results.

We ran several falsification tests on both outcomes and exposure to treatment, the results of which are shown in eTable 7. First, we conducted tests on outcomes that could not have possibly been affected by FTF, including mother’s age, mother’s height, mother’s education level (for mothers above the age of 20, since it is possible, although unlikely, that FTF had a substantial effect on the educational levels of younger mothers through its women’s empowerment initiatives), and low birthweight (for children born before FTF’s implementation but surveyed after). Mother’s height and birthweight data were only reliably available from DHS surveys, and birthweight data were missing for many individuals. However, we observe no significant effects from FTF on any of these pseudo-outcomes (regressions #1-6), which lends additional robustness to our conclusions.

Next, we conducted falsification tests that treated 2008, 2009, and 2010 as FTF start years (regressions #7-9). If the effects observed in our main regressions are indeed due to FTF, the results in these analyses should be insignificant, since FTF did not in fact begin the bulk of its programs until at least 2011. However, if the observed effects were due to pre-existing trends, we might observe effects starting in earlier years. The coefficients from these regressions are indeed insignificant, meaning that our findings pass these falsification tests. The exception is underweight for 2008. Although the effect of a version of FTF that starts in 2008 is stable, the coefficient is smaller in magnitude and the statistical significance is lower. Furthermore, when we investigated this result further by excluding observations from 2012 on, we found no statistical significance (coefficient of -0.2 [-2.9 - 2.5]), indicating that this result is likely driven by impact from observations in post-2012 years, but that the coefficient is stable enough that we still see some smaller effect in earlier years.

Because of substantial increases in nutrition program investment by the Canadian government in several of the same countries that FTF operates around the same time FTF started, we tested whether an effect could be observed in a similarly structured analysis framed around Canada treatment and control countries. The treatment countries we used in this analysis were those SSA countries specified as priority countries of Canada’s Maternal, Newborn, and Child Health Muskoka initiative and include Ethiopia, Malawi, Mali, Mozambique, Nigeria, and Tanzania.[9] The results of this regression are shown in row #10 of eTable 7. We find no association between this program and stunting or wasting trends, however a significant result is identified for underweight. Therefore, it is possible that the base case result we observed for underweight is driven by a combination of FTF and Canadian initiatives.

eTable 8 shows the FTF-Post coefficients from a set of analyses in which we varied the treatment of different years in the analysis to test for lagged effects or effects that increased over time. We also test the robustness of our assumption that FTF’s effects would have begun in 2012. We gathered information on the countries FTF operates within and the year the program was introduced from FTF reports [10]. Because the exact date of program initiation varies within and between countries, we assumed an implementation start date of 2012, the year when the commencement of program activities was announced. However, FTF multi-year strategies were put in place in 2011 and some activities may have begun even before that time. We assumed a lag time between the development of strategy documents and actual program implementation; however, we test this assumption in sensitivity analyses that

Page 32 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


5

vary the program’s start year. Because we were somewhat uncertain about which year to include as the FTF start year, regressions #2-4 test models that handle 2011 observations differently (as a treatment year or dropping all interviews or surveys that occurred in 2011 entirely); we assume 2011 is a control year in the base case. We find that effect sizes decrease and errors increase when 2011 is treated as the start year, but that results are relatively stable, which is consistent with our hypothesis that outcomes would not be affected in the first year when the program was just getting off the ground.

We also tested the effect of FTF over time by varying the start year and adding lagged treatment effects. Regressions #5-7 in eTable 8 treated 2013, 2014, and 2015, respectively, as the first year of FTF. Significant results for some of these regressions, particularly stunting and underweight, could imply that FTF had a greater effect in later years, as its activities ramped up. Regressions #8 and #9 further test this hypothesis via the inclusion of interactions between FTF-Post and year – either a linear time trend (#8) or dummy variables for each year (#9). A significant coefficient for regression #9 indicates that FTF had a unique effect in a particular year, while in regression #8 this indicates that FTF’s impact increased or decreased over time (depending on the sign of the coefficient). The results of these regressions are mostly inconclusive, although we do observe evidence of an increased impact of FTF on stunting and underweight in later years from regression #8. Regression #8 shows that absent FTF, these countries may have had slightly higher stunting and underweight levels than control countries in the post-2011 period, but that stunting and underweight are decreasing at a rate of 1.1-1.3 percentage points per year more in FTF countries than non-FTF countries. An increased effect over time is expected as the program ramps up and reaches more children, and as children are exposed to the program for longer durations, especially for chronic undernutrition outcomes like stunting.

eTable 9 shows the results of running our fully-adjusted base case analysis multiple times, each time leaving out survey observations from a single country. This analysis ensures that our results are not being driven by one particular country. The results of these regressions are largely consistent with our base case findings – in a few cases a result becomes statistically insignificant but at no point do we observe a change in the sign of the coefficients on the interaction term. The larger standard errors we observe for stunting and for wasting and underweight when Democratic Republic of the Congo and Nigeria are removed, respectively, is likely due to a lack of power, since these countries represent a large proportion of children in our sample and since regression coefficients remain relatively stable. The lack of significance when Ghana (stunting, wasting) and Niger (wasting) are removed merited further investigation into other potential changes in these countries around the time that FTF was implemented that could be measurably affecting nutritional outcomes. While we found a substantial increase in financial and strategic investment from the government of Niger around the time FTF began, much of the increased focus on nutrition in Niger was on chronic undernutrition, not wasting. We did not identify any substantial shifts in Ghana that could be influencing our results. Based on this review, we do not believe these analyses represent substantial threats to the validity of our findings.

Finally, as described in more detail in the main body of our study, eTable 10 and eFigure 1 show results of our pre-FTF parallel trends analysis. The lack of a significance on the FTF-Year term in eTable 10 demonstrates that trends in stunting, wasting, and underweight were not significantly different between FTF and control countries in pre-FTF years. In eFigure 1, more steeply sloped lines for the FTF group would indicate that the effects in our base regressions could be due to pre-FTF improving trends in nutritional outcomes rather than FTF itself. However, we observe similar or less steeply sloped (for underweight) trends in FTF countries, indicating that parallel trends are not a concern in this analysis.

Page 33 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


6

eTable 1: STROBE Checklist

STROBE Checklist for Cross-Sectional Studies

Item No Recommendation Location Addressed in

Manuscript(a) Indicate the study’s design with a commonly used term in the title or the abstract

Title and abstract 1 (b) Provide in the abstract an informative and balanced summary of what was done and what was found

Abstract

Introduction

Background/rationale 2 Explain the scientific background and rationale for the investigation being reported Introduction

Objectives 3 State specific objectives, including any prespecified hypotheses Introduction

Methods

Study design 4 Present key elements of study design early in the paper

Methods, “Statistical Approaches”

Setting 5Describe the setting, locations, and relevant dates, including periods of recruitment, exposure, follow-up, and data collection

Methods, “Data Sources”, Table 1, eTable 2, eTable

3

Participants 6 Give the eligibility criteria, and the sources and methods of selection of participants Methods, “Data Sources”

Variables 7Clearly define all outcomes, exposures, predictors, potential confounders, and effect modifiers. Give diagnostic criteria, if applicable

Methods, “Data Sources”

Data sources/ measurement 8

For each variable of interest, give sources of data and details of methods of assessment (measurement). Describe comparability of assessment methods if there is more than one group

Methods, “Data Sources”

Bias 9 Describe any efforts to address potential sources of bias

Methods, Discussion, Supplementary Methods

Study size 10Explain how the study size was arrived at Methods “Data Sources”,

Table 1, eTable 2, eTable 3

Quantitative variables 11Explain how quantitative variables were handled in the analyses. If applicable, describe which groupings were chosen and why

Methods “Statistical Approaches”

(a) Describe all statistical methods, including those used to control for confounding

Methods “Statistical Approaches”,

Supplementary Methods(b) Describe any methods used to examine subgroups and interactions

Methods “Statistical Approaches”

(c) Explain how missing data were addressed Methods “Data Sources” (d) If applicable, describe analytical methods taking account of sampling strategy

Methods “Statistical Approaches”,


Statistical methods 12

(e) Describe any sensitivity analyses Methods “Statistical Approaches” &


Page 34 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


7

Results(a) Report numbers of individuals at each stage of study—e.g. numbers potentially eligible, examined for eligibility, confirmed eligible, included in the study, completing follow-up, and analysed(b) Give reasons for non-participation at each stage

Participants 13

(c) Consider use of a flow diagram

Methods “Data Sources”, Results, Table 1, eTable 2

(a) Give characteristics of study participants (e.g. demographic, clinical, social) and information on exposures and potential confoundersDescriptive

data 14(b) Indicate number of participants with missing data for each variable of interest

Results, Table 1, Table 2, eTable 2

Outcome data 15 Report numbers of outcome events or summary measures Results, eTable 2(a) Give unadjusted estimates and, if applicable, confounder-adjusted estimates and their precision (e.g., 95% confidence interval). Make clear which confounders were adjusted for and why they were included(b) Report category boundaries when continuous variables were categorized

Main results 16

(c) If relevant, consider translating estimates of relative risk into absolute risk for a meaningful time period

Results (Table 3 & Figure 1)

Other analyses 17 Report other analyses done—e.g. analyses of subgroups and interactions, and sensitivity analyses

Results & Appendix (eTables 4-10 and eFigures 1-3)

DiscussionKey results 18 Summarise key results with reference to study objectives Discussion

Limitations 19Discuss limitations of the study, taking into account sources of potential bias or imprecision. Discuss both direction and magnitude of any potential bias

Discussion

Interpretation 20Give a cautious overall interpretation of results considering objectives, limitations, multiplicity of analyses, results from similar studies, and other relevant evidence

Discussion

Generalisability 21 Discuss the generalisability (external validity) of the study results Discussion

Other information

Funding 22Give the source of funding and the role of the funders for the present study and, if applicable, for the original study on which the present article is based

Acknowledgments

Page 35 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


8

eTable 2 below describes each survey used in our analysis in detail, excluding the poor-quality surveys and surveys administered in UMICs that we omitted from our base case analysis. Sudan and South Sudan are also excluded from this table and all analyses, because their division occurred in the middle of our study period. This table shows the numbers of children after data cleaning was completed.

eTable 2: Survey & Sample Size Details

Country Year Survey Children < 5, total

Children <5, excluding missing &

flagged dataStunted % Wasted % Underweight %

Feed the Future Countries2000 DHS 9,560 9,152 4,727 57% 1,230 13% 3,613 42%2005 DHS 4,455 4,254 1,846 50% 523 12% 1,401 34%2011 DHS 10,480 9,961 4,129 44% 1,153 10% 3,022 29%Ethiopia

2016 DHS 9,696 9,132 3,036 36% 1,091 10% 2,161 22%2003 DHS 3,400 3,276 1,144 35% 286 9% 661 19%2006 MICS 3,234 3,234 875 26% 238 7% 484 14%2008 DHS 2,685 2,561 683 28% 235 9% 372 14%2011 MICS 7,424 7,421 1,993 22% 553 6% 1,272 14%

Ghana

2014 DHS 2,782 2,745 529 18% 134 5% 299 11%2000 MICS 7,266 6,782 2,385 38% 512 8% 1,216 17%2003 DHS 5,189 4,955 1,653 36% 344 6% 806 16%

2008/09 DHS 5,490 5,371 1,799 35% 436 7% 899 16%Kenya

2014 DHS 19,344 19,010 5,127 26% 1,055 4% 2,571 11%2007 DHS 4,785 4,599 1,700 38% 348 8% 884 19%Liberia 2013 DHS 3,329 3,276 991 30% 210 6% 503 15%2000 DHS 9,983 9,843 4,863 54% 608 7% 1,951 21%2004 DHS 9,560 8,938 4,332 52% 534 6% 1,594 19%2006 MICS 22,171 22,164 10,771 49% 936 4% 3,250 15%2010 DHS 5,146 4,911 2,188 47% 196 4% 666 14%

2013/14 MICS 18,559 18,545 7,528 43% 705 4% 2,977 17%

Malawi

2015/16 DHS 5,384 5,260 1,593 32% 169 3% 571 11%2001 DHS 10,682 10,061 4,078 42% 1,269 13% 2,991 30%2006 DHS 12,024 11,674 4,316 38% 1,816 16% 3,275 28%

2009/10 MICS 22,939 22,935 6,356 28% 2,066 9% 4,448 19%2012/13 DHS 4,871 4,643 1,641 38% 586 13% 1,183 26%

Mali

2015 MICS 14,968 14,939 4,511 30% 2,025 14% 3,743 25%

Page 36 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


9




2003 DHS 8,830 8,335 3,723 47% 402 5% 1,635 22%2008 MICS 10,779 10,764 4,193 41% 397 5% 1,754 18%Mozambique2011 DHS 9,830 9,733 3,789 43% 524 6% 1,317 16%2000 DHS 6,632 6,408 2,856 47% 524 9% 1,227 20%2005 DHS 3,840 3,789 1,847 51% 178 5% 675 18%2010 DHS 4,133 4,128 1,787 44% 121 3% 482 12%Rwanda

2014/15 DHS 3,615 3,605 1,343 38% 81 2% 336 9%2000 MICS 9,064 8,687 2,538 30% 827 10% 1,795 21%2005 DHS 3,248 2,958 622 20% 267 9% 447 14%

2010/11 DHS 4,323 3,939 1,155 28% 376 10% 817 19%2012/14 DHS 12,398 12,169 2,513 19% 1,028 7% 2,023 14%2015/16 DHS 12,451 12,299 2,533 19% 952 8% 2,029 15%

Senegal

2017 DHS 11,127 10,833 1,765 14% 1,032 9% 1,580 13%2004/05 DHS 7,461 7,315 3,075 44% 317 4% 1,239 16%

2010 DHS 7,175 6,977 2,731 42% 451 5% 1,201 16%Tanzania2015/16 DHS 9,213 9,068 2,687 30% 441 5% 1,109 12%2000/01 DHS 5,888 5,384 2,271 45% 253 5% 955 19%

2006 DHS 2,517 2,429 928 38% 164 7% 426 16%2011 DHS 2,214 2,119 661 33% 115 5% 301 14%Uganda

2016 DHS 4,530 4,457 1,116 25% 168 4% 412 9%2001/02 DHS 5,805 5,662 2,937 53% 339 6% 1,330 23%

2007 DHS 5,621 5,437 2,317 46% 326 6% 799 15%Zambia2013/14 DHS 12,311 11,915 4,613 40% 750 6% 1,810 15%

Control Countries2001 DHS 4,582 4,525 1,736 38% 419 9% 979 21%2006 DHS 14,220 13,454 5,472 43% 1,117 8% 2,690 20%Benin2014 MICS 12,030 12,030 3,812 34% 551 4% 2,062 18%2003 DHS 9,131 8,842 3,644 43% 1,875 22% 3,191 35%Burkina Faso 2010 DHS 6,837 6,744 2,285 35% 1,016 16% 1,717 26%

2010/11 DHS 3,606 3,508 1,929 58% 210 6% 971 29%Burundi 2016/17 DHS 6,096 6,065 3,008 51% 309 5% 1,607 27%2004 DHS 3,487 3,358 1,163 36% 192 6% 467 15%2006 MICS 6,074 6,071 2,149 36% 434 8% 961 17%2011 DHS 5,286 5,199 1,635 32% 295 6% 730 15%Cameroon

2014 MICS 6,765 6,765 2,068 32% 309 5% 859 15%

Page 37 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


10




2000 MICS 14,300 13,685 5,711 44% 1,324 10% 2,841 22%2006 MICS 7,716 7,712 3,086 42% 869 13% 1,888 26%Central African

Rep. 2010 MICS 10,234 10,228 4,020 40% 682 7% 2,313 23%2000 MICS 5,384 5,298 1,983 39% 701 14% 1,503 29%2004 DHS 4,926 4,694 1,905 44% 785 16% 1,540 34%2010 MICS 12,859 12,850 4,941 38% 2,059 15% 4,070 30%Chad

2014/15 DHS 10,775 10,490 4,354 40% 1,495 14% 3,488 30%2005 DHS 4,190 4,093 1,152 30% 313 8% 452 12%

2011/12 DHS 4,621 4,537 1,203 23% 250 6% 601 11%Congo, Rep.2014/15 MICS 8,795 8,791 2,234 21% 568 8% 1,245 12%

2006 MICS 8,477 8,476 3,221 38% 749 9% 1,449 17%2011/12 DHS 3,488 3,311 970 29% 237 8% 512 16%Cote d’Ivoire

2016 MICS 8,786 8,786 2,097 21% 519 6% 1,215 13%2007 DHS 3,951 3,707 1,525 45% 352 11% 907 25%2010 MICS 10,782 10,778 4,536 43% 880 8% 2,603 24%Congo, Dem.

Rep. 2013/14 DHS 8,552 8,408 3,572 42% 673 8% 1,998 23%2005/06 MICS 6,387 6,386 1,710 26% 496 8% 1,031 16%Gambia 2013 DHS 3,640 3,395 831 25% 351 12% 624 17%

2005 DHS 2,845 2,765 1,061 39% 295 11% 618 23%2012 DHS 3,271 3,221 959 31% 331 10% 592 19%Guinea2016 MICS 7,157 7,155 2,280 32% 574 8% 1,309 18%2000 MICS 5,851 5,725 1,940 36% 641 12% 1,229 22%2006 MICS 5,340 5,303 2,144 47% 410 9% 876 17%Guinea-Bissau2014 MICS 7,472 7,470 2,010 28% 434 6% 1,237 17%2000 MICS 3,735 3,628 1,679 52% 215 7% 532 15%2004 DHS 1,584 1,481 632 43% 80 6% 274 18%2009 DHS 1,732 1,675 652 38% 76 4% 238 14%Lesotho

2014 DHS 1,381 1,353 466 33% 47 3% 157 11%2003/04 DHS 4,874 4,741 2,257 52% 671 16% 1,562 36%Madagascar* 2008/09 DHS 5,875 5,248 2,465 49% - - - -

2007 MICS 7,914 7,895 2,416 30% 1,114 14% 2,016 25%Mauritania 2011 MICS 8,560 8,557 2,491 30% 1,159 14% 2,131 25%2000 MICS 5,080 4,916 2,302 53% 726 16% 1,898 43%2006 DHS 4,038 3,880 1,872 55% 475 13% 1,373 40%Niger2012 DHS 5,576 5,179 2,041 43% 904 18% 1,931 38%

Page 38 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


11




2003 DHS 4,981 4,814 1,891 43% 523 12% 1,272 27%2007 MICS 13,254 13,215 4,805 38% 1,666 13% 3,076 23%2008 DHS 24,358 23,317 8,672 40% 3,009 15% 6,455 27%2011 MICS 24,227 24,225 9,417 36% 2,487 10% 6,287 24%2013 DHS 27,571 26,989 9,141 36% 4,239 18% 7,711 31%

Nigeria

2016/17 MICS 27,511 27,498 11,116 43% 2,679 11% 7,851 31%2005 MICS 4,385 4,378 1,819 44% 452 11% 1,142 26%2008 DHS 2,422 2,287 793 37% 240 11% 455 21%2010 MICS 8,229 8,199 3,445 44% 627 9% 1,696 22%2013 DHS 4,969 4,754 1,617 37% 409 10% 820 18%

Sierra Leone

2017 MICS 11,764 11,721 3,728 31% 783 7% 1,932 16%2000 MICS 3,423 3,238 1,157 36% 53 2% 290 9%

2006/07 DHS 2,226 2,111 558 28% 62 3% 115 6%2010 MICS 2,575 2,574 792 31% 19 1% 145 6%eSwatini

2014 MICS 2,656 2,655 686 25% 47 2% 158 6%2006 MICS 3,547 3,545 1,068 28% 672 17% 967 24%2010 MICS 4,649 4,647 1,537 30% 254 5% 906 17%Togo

2013/14 DHS 3,274 3,238 907 27% 237 7% 558 16%2005/06 DHS 4,536 4,251 1,356 33% 276 7% 562 13%

2009 MICS 6,161 6,157 2,011 33% 179 3% 741 12%2010/11 DHS 4,762 4,444 1,398 32% 157 3% 474 10%

2014 MICS 9,594 9,593 2,704 28% 313 3% 1,106 11%Zimbabwe

2015 DHS 5,253 5,029 1,131 23% 180 4% 337 7%*Weight data was missing from Madagascar’s 2008 DHS

Page 39 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


12

eTable 3 shows the number of surveys that were available for each year of the analysis (from 2000 to 2017). The information displayed in this table justify our statistical approach to control for time trends; few surveys are available in each year, demonstrating substantial compositional heterogeneity in each one-year period and far more homogeneity in three-year periods. Three-year periods were chosen to avoid straddling both treatment and control years.

eTable 3: Survey Availability by Year

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017Burundi D D D DBenin D D MBurkina Faso D DCAR M M MChad M D M D DCote d’Ivoire M D D MCameroon D M D MCongo, DRC D M D DCongo, Rep. D D D M MEthiopia D D D DGhana D M D M DGuinea D D MGambia M DGuinea-Biss. M M MKenya M D D D DLiberia D D DLesotho M D D DMadagascar D D D DMali D D M M D D MMozambique D M DMauritania M MMalawi D D M M D M M DNiger M D D

Page 40 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


13

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017Nigeria D M D M D M MRwanda D D D D D DSenegal M D D D D D D D D DSierra Leone M D M D MeSwatini M D D M MTogo M M D DTanzania D D D D D DUganda D D D D DZambia D D D D DZimbabwe D D M D D M DTotal (1 year) 11 3 1 5 6 7 13 5 4 4 13 9 4 7 10 8 6 2 FTF 5 2 1 3 2 3 4 2 2 1 4 5 1 3 3 5 2 1 Control 6 1 0 2 4 4 9 3 2 3 9 4 3 4 7 3 4 1Total (2 year) 14 6 13 18 8 22 11 18 8 FTF 7 4 5 6 3 9 4 8 3 Control 7 2 8 12 5 13 7 10 5Total (3 year) 15 18 22 26 21 16 FTF 8 8 8 10 7 8 Control 7 10 14 16 14 8

FTF countries are shown in bold font. D indicates a DHS survey was available for a certain country in a certain year; M designates a MICS survey.

Page 41 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


14

eTable 4: Major Sensitivity Analysis Results

Model Stunting Wasting Underweight-2.40*** -1.69** -2.43***1 Base Case (linear fully adjusted – for reference) (0.90) (0.75) (0.84)-3.28*** -1.70** -3.48***2 Logit (marginal effects shown) (1.06) (0.72) (0.90)

Covariates Added & Removed-2.28*** -3.31*** -4.01***3 GNI p.c. removed (0.87) (0.96) (1.18)-2.88*** -1.79** -2.40***4 Life Expectancy removed (0.87) (0.75) (0.85)-2.23** -2.19*** -2.75***5 MCV Coverage added (0.97) (0.79) (0.83)-1.98** -1.91** -2.31***6 CHE p.c. added (0.96) (0.77) (0.88)-1.81* -1.84** -2.25***7 GGHE-D p.c. added (1.02) (0.81) (0.84)

-2.65*** -1.80** -2.47***8 No country-level control data imputed (kept as missing) (0.92) (0.76) (0.86)-2.15** -0.79 -1.249 1-Year Fixed Effects added (0.99) (0.72) (0.95)-2.22** -1.04 -1.63*10 2-Year Fixed Effects added (1.06) (0.79) (0.95)

Alternative Handling of Errors -2.40*** -1.69*** -2.43***11 HH-clustered errors (0.67) (0.39) (0.58)-2.40* -1.69 -2.43***12 Country-clustered errors (1.29) (1.06) (0.56)

Variation in Surveys & Measures Included-2.17** -1.48** -2.46***13 Survey-generated Z-scores (0.89) (0.65) (0.85)-2.40*** -1.69** -2.43***14 DHS Data only (no MICS) (0.90) (0.75) (0.84)

-6.36 8.24*** 3.8315 MICS Data only (no DHS) (4.16) (2.22) (3.31)-2.22** -1.70** -2.28***16 Only countries w/ both a pre- & post-FTF survey included (0.91) (0.75) (0.84)-2.69*** -2.19*** -2.79***17 Poor quality surveys added (0.88) (0.73) (0.84)-2.29** -1.91** -2.53***18 UMIC surveys added (0.89) (0.76) (0.86)

Estimated coefficients on the FTF x Post term are shown in percentage points. Robust standard errors in parentheses; *** indicates p<0.01, ** indicates p<0.05, * indicates p<0.1

Page 42 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


15

eTable 5: Development Assistance Sensitivity Analysis Results

Model Stunting Wasting UnderweightOECD Development Assistance Data

-2.21** -1.78** -2.56***1 Not controlling for ODA (0.89) (0.75) (0.84)-2.21** -1.76** -2.54***2 Nutrition ODA (no US) (0.89) (0.75) (0.84)-2.24** -1.63** -2.58***3 Nutrition ODA (0.93) (0.74) (0.85)-2.42*** -1.70** -2.42***4 Agriculture ODA (no US) (0.90) (0.75) (0.83)-2.95*** -1.59** -2.53***5 Agriculture ODA (0.95) (0.76) (0.87)-2.40*** -1.69** -2.43***6 Nut. + Ag. ODA (no US - Base Case) (0.90) (0.75) (0.84)-2.21** -1.58** -2.35***7 Nut. + Ag. ODA (no US, smoothed over time) (0.90) (0.74) (0.84)-2.93*** -1.55** -2.54***8 Nut + Ag. ODA (0.96) (0.76) (0.87)-2.25** -1.90*** -2.60***9 Health ODA (0.90) (0.73) (0.84)-2.19** -1.87** -2.83***10 Social Sector ODA (0.90) (0.75) (0.84)-1.69* -1.89** -2.54***11 Total ODA (0.92) (0.79) (0.83)

IHME Development Assistance for Health Data-2.26** -1.81** -2.64***12 Child Nut. DAH (0.92) (0.75) (0.87)-2.51*** -1.80** -2.57***13 Total DAH (0.91) (0.79) (0.86)


Page 43 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


16

eTable 6: Governance Sensitivity Analysis Results

Model Stunting Wasting Underweight-2.42*** -1.53** -2.33***1 No Governance (0.88) (0.75) (0.83)-2.40*** -1.69** -2.43***2 Composite Governance Measure (Base Case) (0.90) (0.75) (0.84)-2.44*** -1.48** -2.39***3 Control of Corporations (0.89) (0.74) (0.83)-2.78*** -1.75** -2.94***4 Government Effectiveness (0.92) (0.76) (0.85)-2.26** -1.71** -2.21***5 Stability & Absence of Violence (0.89) (0.75) (0.82)-2.35** -1.57** -2.58***6 Regulatory Quality (0.95) (0.77) (0.90)-2.42*** -1.56** -2.36***7 Rule of Law (0.89) (0.75) (0.82)-2.42*** -1.51** -2.21**8 Voice & Accountability (0.88) (0.76) (0.86)-2.42** -1.98*** -2.64***9 All 6 Governance Indicators (0.97) (0.74) (0.92)


Page 44 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


17

eTable 7: Falsification Tests

Falsification Tests on the OutcomeOutcome Effect # of Observations

-0.011 Mother’s Age > 30

0.01816,600

0.012 Mother has less than Primary Education

(0.02)726,286

0.013 Mother has Secondary Education or above

(0.01)726,286

-0.034 Mother’s Height (cm)*

(0.20)421,915

0.005 Mother’s Height > Median (1.58 m)*

(0.01)421,915

0.016

Low Birthweight* (for children born pre-2012 only) (0.01)

144,595

Falsification Tests on the ExposureModel Stunting Wasting Underweight

-1.19 -0.60 -2.03**7 2008 Start Year

(1.23) (0.85) (0.92)-0.77 -0.34 -1.48*

8 2009 Start Year(1.02) (0.79) (0.85)-1.48 -0.44 -1.56*

9 2010 Start Year(1.02) (0.79) (0.85)-1.86 0.35 -2.49**

10 Canadian Focus vs. Control Countries(1.34) (0.85) (1.13)

*Indicates the outcome was reliably available in DHS surveys only. Estimated coefficients on the FTF x Post term are shown in percentage points. Robust standard errors in parentheses; *** indicates p<0.01, ** indicates p<0.05, * indicates p<0.1

Page 45 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


18

eTable 8: Alternate Start Year Specifications & Lagged Treatment Effect Models

Model Stunting Wasting Underweight-2.40*** -1.69** -2.43***1 Base Case – 2012 start year (0.90) (0.75) (0.84)

Alternate Start Years – Robustness Checks-2.09** -1.31* -2.37***2 2011 Start Year (0.96) (0.73) (0.78)-2.27** -1.94** -2.52***3 2011 observations removed (0.96) (0.80) (0.86)-2.50** -2.05** -2.72***4 2011 surveys removed (1.04) (0.88) (0.99)

Tests for Lagged & Time-Dependent Treatment Effects-3.47*** -0.87 -2.74***5 2013 Start Year (0.93) (0.86) (0.91)-4.55*** -1.05 -3.29***6 2014 Start Year (1.14) (0.85) (1.10)-3.94*** 0.38 -2.50*7 2015 Start Year (1.36) (0.74) (1.47)

Lagged Tx Effects I1.95 -2.49* 1.39 FTF x Post (2.04) (1.47) (1.84)

-1.30** 0.24 -1.14**8

FTF x Post x Year (0.51) (0.31) (0.48)Lagged Tx Effects II

-5.12*** 2.00** -0.68 FTF x Post (1.60) (0.85) (1.62)7.01*** -2.22 2.79 FTF x Post x 2012 (2.46) (1.48) (2.27)4.82** -3.66*** -0.76 FTF x Post x 2013 (2.34) (1.34) (2.07)2.69 -5.24*** -2.09 FTF x Post x 2014 (2.03) (1.34) (1.99)1.73 -1.45** 0.21 FTF x Post x 2015 (1.28) (0.65) (0.89)0.95 -1.98*** -3.09***

9

FTF x Post x 2016 (1.13) (0.72) (0.88)Estimated coefficients on the FTF x Post term are shown in percentage points. Robust standard errors in parentheses; *** indicates p<0.01, ** indicates p<0.05, * indicates p<0.1

Page 46 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


19

eTable 9: Leave-One-Country-Out Analysis Results

Country Left Out Stunting Wasting Underweight-2.40*** -1.69** -2.42***Benin (0.90) (0.75) (0.84)-2.38*** -1.66** -2.39***Burkina Faso (0.91) (0.75) (0.84)-2.55*** -1.90** -2.36**Burundi (0.98) (0.82) (0.93)-2.44*** -1.66** -2.47***Cameroon (0.91) (0.75) (0.84)-2.40*** -1.69** -2.43***Central African Rep. (0.90) (0.75) (0.84)-1.95** -1.98** -2.30***Chad (0.92) (0.78) (0.80)-2.37*** -1.66** -2.41***Congo, Rep. (0.91) (0.75) (0.84)-2.15** -1.73** -2.46***Cote d’Ivoire (0.90) (0.76) (0.83)-1.79* -2.16*** -2.40***Congo, Dem. Rep. (1.04) (0.74) (0.93)-2.40** -1.76** -1.82**Ethiopia (0.94) (0.75) (0.89)-2.40*** -1.69** -2.43***Gambia (0.90) (0.75) (0.84)

-1.42 -0.88 -2.11**Ghana (0.95) (0.72) (0.85)-2.62*** -1.74** -2.63***Guinea (0.94) (0.78) (0.87)-2.40*** -1.69** -2.43***Guinea-Bissau (0.90) (0.75) (0.84)-2.51*** -1.69** -2.36***Kenya (0.94) (0.74) (0.85)-2.40*** -1.71** -2.45***Lesotho (0.91) (0.76) (0.84)-2.41*** -1.72** -2.49***Liberia (0.91) (0.75) (0.84)-2.27** -1.71** -2.39***Madagascar (0.91) (0.75) (0.83)-2.73*** -1.67** -2.23***Malawi (0.89) (0.74) (0.84)-2.99*** -1.75** -2.61***Mali (0.88) (0.76) (0.86)-2.40*** -1.69** -2.43***Mauritania (0.90) (0.75) (0.84)-2.38*** -1.71** -2.37***Mozambique (0.90) (0.75) (0.84)-3.20*** -0.84 -2.26***Niger (0.89) (0.79) (0.87)-2.56** -0.98 -1.63*Nigeria (1.01) (0.77) (0.88)-2.49*** -1.73** -2.47***Rwanda (0.92) (0.75) (0.85)

Page 47 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


20

-2.35** -1.71** -2.53***Senegal (0.91) (0.75) (0.87)-2.25** -1.64** -2.39***Sierra Leone (0.92) (0.77) (0.85)-2.40*** -1.69** -2.43***eSwatini (0.90) (0.75) (0.84)-2.43*** -1.89** -2.65***Tanzania (0.91) (0.76) (0.84)-2.40*** -1.69** -2.43***Togo (0.90) (0.75) (0.84)-2.79*** -1.59** -2.71***Uganda (0.96) (0.76) (0.87)-2.61*** -1.82** -2.47***Zambia (0.93) (0.75) (0.84)-2.26** -2.08** -2.83***Zimbabwe (1.11) (0.88) (0.99)


Page 48 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


21

eTable 10: Parallel Trends Regression Results

Stunted Wasted Underweight-0.24 -0.25 -0.43FTF x Year (0.39) (0.36) (0.35)

9.83*** -1.71 0.78FTF (3.44) (1.88) (2.47)-0.51 0.27 -0.25Year (0.36) (0.35) (0.31)

43.30*** 7.64*** 21.71***Constant (2.43) (1.76) (2.10)Observations 449,510 441,915 457,849

Estimated coefficients on all terms are shown in percentage points. Country indicator variables, weights, and strata-clustered errors are included. Robust standard errors in parentheses; *** indicates p<0.01, ** indicates p<0.05, * indicates p<0.1

Page 49 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


22

eFigure 1: Pre-FTF Trends in Undernutrition

Page 50 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


23

eFigure 2: Development Assistance to FTF and Control Countries

Note: ODA data are from the OECD DAC database[11]; DAH data are from IHME[12]

Page 51 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


24

eFigure 3A: Undernutrition Trends by Country – Stunting

Note: FTF countries are shown in blue, non-FTF countries are shown in red.

Page 52 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


25

eFigure 3B: Undernutrition Trends by Country – Wasting

Note: FTF countries are shown in blue, non-FTF countries are shown in red. Madagascar’s 2009 DHS was missing weight data.

Page 53 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


26

eFigure 3C: Undernutrition Trends by Country – Underweight

Note: FTF countries are shown in blue, non-FTF countries are shown in red. Madagascar’s 2009 DHS was missing weight data.

Page 54 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


27

Appendix References

1 Ren R. Note on DHS standard weight de-normalization. DHS Program User Forum. 2013.https://userforum.dhsprogram.com/index.php?t=getfile&id=4&

2 Hausman J. Specification Tests in Econometrics. Econometrica 1978;46:1251–71.

3 Bendavid E, Holmes CB, Bhattacharya J, et al. HIV development assistance and adult mortality in Africa. JAMA 2012;307:2060–7. doi:10.1001/jama.2012.2001

4 Jakubowski A, Stearns SC, Kruk ME, et al. The US President’s Malaria Initiative and under-5 child mortality in sub-Saharan Africa: A difference-in-differences analysis. PLoS Med 2017;14:e1002319. doi:10.1371/journal.pmed.1002319

5 World Health Organization (WHO). WHO child growth standards : length/height-for-age, weight-for-age, weight-for-length, weight-for- height and body mass index-for-age : methods and development. Geneva: : WHO 2006.

6 United States Agency for International Development (USAID). Anthropometric Data in Population-Based Surveys, Meeting Report, July 14-15, 2015. Washington, D.C.: : FHI 360/FANTA 2016.

7 Corsi DJ, Perkins JM, Subramanian SV. Child anthropometry data quality from Demographic and Health Surveys, Multiple Indicator Cluster Surveys, and National Nutrition Surveys in the West Central Africa region: are we comparing apples and oranges? Glob Health Action 2017;10:1328185. doi:10.1080/16549716.2017.1328185

8 Assaf S, Kothari MT, Pullum T. An assessment of the quality of DHS anthropometric data, 2005-2014. Rockville, Maryland, USA: : ICF International 2015. http://dhsprogram.com/pubs/pdf/MR16/MR16.pdf

9 Bhushan A. The Muskoka Initiative and Global Health Financing. 2014.http://www.nsi-ins.ca/wp-content/uploads/2014/05/Muskoka-Final.pdf (accessed 8 Nov 2018).

10 United States Agency for International Development (USAID). Feed the Future Snapshot: Progress Through 2017. Washington, D.C.: : USAID 2017. https://feedthefuture.gov/sites/default/files/resource/files/2017%20Feed%20the%20Future%20Progress%20Snapshot.pdf (accessed 10 Oct 2017).

11 OECD. Detailed aid statistics: Official bilateral commitments by sector. OECD Int. Dev. Stat. Database. 2018.http://dx.doi.org/10.1787/data-00073-en

12 (IHME) I for HM and E. Development Assistance for Health Database 1990-2017. 2018.http://ghdx.healthdata.org/record/development-assistance-health-database-1990-2017 (accessed 1 Jan 2018).

13 Elliott K, Dunning C. Assessing the US Feed the Future Initiative: A New Approach to Food Security. Center for Global Development 2016.

Page 55 of 55


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

impact of a large u.s. government nutrition program on · 2019-12-11 · however,...

Documents