Scholarly article on topic 'Nutrition indicators in agriculture projects: Current measurement, priorities, and gaps'

Nutrition indicators in agriculture projects: Current measurement, priorities, and gaps Academic research paper on "Agriculture, forestry, and fisheries"

CC BY-NC-ND
0
0
Share paper
Academic journal
Global Food Security
OECD Field of science
Keywords
{"Nutrition-sensitive agriculture" / Agriculture-nutrition / "Impact evaluation" / "Nutrition outcomes" / Indicators}

Abstract of research paper on Agriculture, forestry, and fisheries, author of scientific article — Anna Herforth, Terri J. Ballard

Abstract How agriculture can improve human nutrition is a topic of debate. Recent reviews demonstrate little impact on nutritional status but do not critically examine the choice of appropriate outcome indicators. This paper reviews which nutrition impact indicators are currently used in agriculture-nutrition projects, and highlights priorities and gaps in measurement. Many project evaluations are statistically underpowered to observe impact on nutritional status, but appear to be powered to observe impacts on food consumption and dietary quality, which we conclude are an appropriate level of impact of agriculture-nutrition projects. To improve the evidence base, there is a need to develop indicators of outcomes that are not being fully measured, including dietary quality and food security, women's empowerment, health environments, and food environments.

Academic research paper on topic "Nutrition indicators in agriculture projects: Current measurement, priorities, and gaps"

ELSEVIER

Nutrition indicators in agriculture projects: Current measurement, priorities, and gaps

Anna Herforth *, Terri J. Ballard

Food and Agriculture Organization (FAO), Rome, Italy

ARTICLE INFO ABSTRACT

How agriculture can improve human nutrition is a topic of debate. Recent reviews demonstrate little impact on nutritional status but do not critically examine the choice of appropriate outcome indicators. This paper reviews which nutrition impact indicators are currently used in agriculture-nutrition projects, and highlights priorities and gaps in measurement. Many project evaluations are statistically underpowered to observe impact on nutritional status, but appear to be powered to observe impacts on food consumption and dietary quality, which we conclude are an appropriate level of impact of agriculture-nutrition projects. To improve the evidence base, there is a need to develop indicators of outcomes that are not being fully measured, including dietary quality and food security, women's empowerment, health environments, and food environments.

© 2016 Published by Elsevier B.V.

Contents lists available at ScienceDirect

Global Food Security

journal homepage: www.elsevier.com/locate/gfs

CrossMark

Article history: Received 27 October 2015 Received in revised form 16 June 2016 Accepted 15 July 2016

Keywords:

Nutrition-sensitive agriculture Agriculture-nutrition Impact evaluation Nutrition outcomes Indicators

1. Introduction

How agriculture contributes to improving nutrition of populations and vulnerable sub-groups is a topic of debate. Recent literature reviews, summarized in a review by Webb and Kennedy (2014), point to a lack of empirical evidence on nutritional status outcomes from agriculture, primarily due to methodological deficiencies in study design (Webb and Kennedy, 2014; Ruel and Alderman, 2013; Masset et al., 2011 ; Webb Girard et al., 2012). These reviews have focused on nutritional status indicators to measure impact, but the choice and appropriateness of outcome indicators have received less attention. Masset et al. (2011 ) found nutritional status outcomes to be insensitive to change: due to inadequate statistical power, none of the studies included in their review could have detected a small improvement in the prevalence of undernutrition (defined as a 2% reduction in stunting or underweight), and only half could have detected a large improvement (30% reduction).

In 2012, the Leverhulme Centre for Integrative Research on Agriculture and Health (LCIRAH) conducted a mapping study of current and planned research on agriculture for improved nutrition (hereafter called "agriculture-nutrition" projects) (Hawkes et al., 2012; Turner et al., 2013). One gap identified by the researchers was measurement of the full pathway of change from

* Corresponding author. E-mail addresses: anna@annaherforth.net (A. Herforth), terri.ballard@fao.org (T.J. Ballard).

http://dx.doi.org/10.1016Zj.gfs.2016.07.004 2211-9124/© 2016 Published by Elsevier B.V.

agricultural inputs and practices to nutrition outcomes in current research. Numerous conceptual frameworks have been elaborated to describe the pathways through which agriculture can improve nutrition outcomes (Hawkes et al., 2012; Turner et al., 2013; Webb, 2013; Herforth et al., 2012; Gillespie et al., 2012; Kennedy and Bouis, 1993). These frameworks share the common theme that agriculture can affect each of the underlying determinants of nutrition: access to adequate food (food security), care practices, health services and adequate health environments (UNICEF, 1990). Fig. 1 illustrates these pathways:

• food access through improved access to nutritious foods on-farm; increased availability and lower prices of diverse nutritious foods in markets; and income which can be spent on more diverse nutritious food if such food is available, affordable, and convenient.

• care practices through empowerment of women (particularly if they can control income, their time and labor), and through incorporating behavior change communication.

• health environments through management practices that protect natural resources (water in particular), and safeguard against health risks introduced by agricultural production (e.g. livestock, standing water, agrochemicals). Agricultural income can also affect health care access if health care is available, affordable, and convenient.

Prompted by the gap in understanding the range and appropriateness of indicators being used to measure agriculture-

Intervention

Fig. 1. Conceptual framework for nutrition interventions in agriculture. Source: Authors.

nutrition intervention outcomes, our aim is to review which indicators are currently being selected, in order to understand better how to strengthen the evidence base and to recommend what indicators should be used or need to be developed. We discuss how current measurement has advanced compared to previous literature, and what we can expect to learn from current agriculture-nutrition research based on indicators selected and power calculations. We use this information on current research as the basis for a broader discussion and recommendations around how nutrition measurement in agriculture projects can be strengthened.

2. Methods

In order to review the status of nutrition measurement in agriculture-nutrition research, we conducted a survey of investigators currently researching the links between agriculture projects and nutrition outcomes. Because a mapping study had recently been done of current research projects exploring the links between agriculture and nutrition, we drew our sample from the 151 studies that had been identified in that study (Hawkes et al., 2012; Turner et al., 2013). We included only those that explicitly listed nutrition improvement as an objective and that engaged in field research, and excluded secondary data analyses, formative research, unspecified research activities, and unfunded projects. Seventy-three intervention-based studies met the criteria. Principal investigators of the 73 eligible projects were surveyed on use of indicators relevant to nutrition outcomes via an online questionnaire using SurveyMonkey® (Supplementary materials 1). The survey questions were designed to reflect the pathways of how agriculture can affect nutrition. Respondents were asked to describe their project's nutrition-relevant goals and how project

activities would be expected to affect nutrition. They were asked to identify the indicators used in their projects in the categories of: nutritional status, diet and food consumption, food security or food access, economic outcomes, women's labor or empowerment, nutrition knowledge or behaviors, natural resource management or environmental safeguards, and "other." Respondents were also asked if their projects linked with health, water and sanitation, or social protection activities. Information was gathered on study design, including target population of projects and survey sample sizes, use of a comparison group, timing of surveys (baseline, midline, endline, other), if they were employed at the same time of year and if any related qualitative data were collected.

The survey was personally sent by e-mail to project principal investigators. Non-responders were followed up twice. Data were downloaded, cleaned, and coded, and frequencies were calculated using IBM SPSS software (IBM Corp, 2011).

To investigate the statistical power needed for detecting improvements in two distinct nutrition outcomes (reducing stunting and improved dietary quality), we estimated sample sizes that would be needed to have 80% probability of observing improvements in stunting and dietary diversity of young children at a significance level of 0.05, using an on-line sample size calculator (Rollin Brant's Sample Size Calculators, 2016). Sufficient sample size to statistically detect changes in impact indicators in an intervention population over time is an essential component of a rigorous evaluation design. In order to attribute observed changes to the intervention itself, the same outcomes need to be measured in a comparison population, that is comparable but that does not participate in the intervention. Our power calculations estimate sample sizes needed for each group (i.e. the intervention and the comparison group). Because most intervention studies cannot randomize participation in the intervention, alternative sampling designs to select a survey sample are often employed, such as

cluster sampling. There is a loss of sampling efficiency resulting from the use of cluster sampling instead of simple random sampling, therefore a design effect should be incorporated into the sample size formula to account for the need for a larger sample size, often doubling what is needed for a random sample (design effect of 2) (Magnani, 1999, p. 9). We calculated sample sizes at two different hypothetical baseline levels of stunting (40% and 30%), and for 3 different targets for impact based on the Feed the Future Initiative's target for reducing stunting by 20% over 5 years, which roughly corresponds to a 4% reduction per year (USAID, 2013). These baseline levels of stunting are not uncommon in sub-Saharan Africa and the Indian subcontinent (IFPRI, 2015). We chose the Minimum Diet Diversity (MDD) indicator, one component of the Minimum Adequate Diet (MAD) indicator of the Infant and Young Child Feeding Indicators (WHO, 2008), to represent a reasonable outcome indicator for nutrition-sensitive agricultural projects that aim to improve diets of vulnerable population groups. The MDD is a food group indicator defined as consuming 4 or more food groups out of 7 in a 24 h period. We calculated sample sizes at two different hypothetical baseline levels of achieving minimum dietary diversity (20% and 40%), and for 3 different targets for impact. These levels of baseline prevalence of MDD are close to the median MDD of 29%, based on data from 41 countries (IFPRI, 2016). The impact targets are based on the Feed the Future Initiative's program targets for improving child feeding indicators, which is to improve MAD prevalence by 100% if the baseline level is less than 40%, and by 80% if the baseline level is between 40% and 80% (USAID, 2013).

3. Results

3.1. Response rate, implementing organizations, and geographic location

Of the 73 agriculture-nutrition projects surveyed, principal investigators from 64 responded to the survey (88% response rate). Of those, four were excluded because data provided were incomplete, resulting in a total analyzable sample of 60 agriculture-nutrition projects (82% of eligible projects identified). Of those, 50 were agricultural interventions, 2 were interventions focused solely on nutrition behavior change within farmer households, and 8 were observational studies with the aim of using the results to design a future intervention. Roughly a third each were undertaken by universities (n=20), CGIAR agencies (n = 18), and nongovernmental organizations (n=21), and 1 was undertaken by a UN agency (FAO). Nearly two-thirds of all identified agriculture-nutrition projects were taking place in sub-Saharan Africa (62%), a quarter in South Asia (25%), followed by Latin America (12%) and Southeast Asia (10%). Very few projects were located in the Middle East region (3%) or in small island nations (3%). (The numbers total greater than 100% because many projects covered more than one country or region.) Within Africa projects, 60% were in East Africa, 31% in West Africa, and 8% in southern Africa. The five most researched countries were Kenya (11 projects), India (9 projects), Bangladesh, Malawi, and Uganda (7 projects each).

3.2. Indicators

The indicators reported by the projects are found in Table 1. The most-measured categories included diet and food consumption and food security, followed by anthropometry, nutrition knowledge or behaviors, and economic outcomes. Relatively fewer interventions measured women's empowerment, biochemical indicators, and natural resource management practices related to health environments.

Table 1

Indicators selected by research projects.3

Indicator % N

Anthropometry 72 43

• Stunting 52 31

• Underweight 48 29

• Wasting 33 20

• Maternal weight/underweight/BMl 22 13

• Specifically mentions obesity 7 4

Biochemical indicators 38 23

• Iron status and/or Anemia 35 21

• Vitamin A status 15 9

• Zinc status 3 2

• lodine status 2 1

• B12 status 2 1

Diet and food consumption 93 56

• Household Dietary Diversity Score (HDDS) 45 27

• Food Consumption Score (household-level indicator) 3 2

• Women's dietary diversity (usually WDDS) 40 24

• Individual Dietary Diversity Score (IDDS) - young child 33 20

• Minimum Acceptable Diet (MAD) - child under 2 20 12

• Unique food items/dietary variety 10 6

• Quantitative nutrient intakes 32 19

• Vitamin A-rich food intake 10 6

• lron-rich food intake 5 3

• Other food consumption indicator specific to project 32 19

• Consumption of specific target foods 32 19

• At least one indicator of total diet quality 87 52

Food security 80 48

• Household Food lnsecurity Access Scale (HFlAS) 32 19

• Household Hunger Scale (HHS) 13 8

• Coping strategies 13 8

• MAHFP or months of available food 15 9

• Meal frequency 7 4

• Food availability in general 17 10

• Other or non-specific 8 5

Nutrition knowledge or behaviors 72 43

Women's empowerment or labor 53 32

Economic outcomes 68 41

• Disaggregated by gender 42 25

Natural resource management 29 17

• Water quality and contamination 7 4

a Additional information on indicators of diet and food consumption, women's empowerment, nutrition knowledge and behaviors, economic outcomes, and natural resource management is available in Supplementary material 2.

Almost all (93%) employed indicators of diet or food consumption. Of 56 projects that measured food consumption, most aimed to assess diet quality: 46 used an indicator of diet quality at individual level, and 6 more used a proxy only at household level. The most commonly used indicators for assessing diet or food consumption were household dietary diversity, women's dietary diversity and young child dietary diversity, followed by the Minimum Adequate Diet indicator for young children under age 24 months (WHO, 2008), quantitative nutrient intakes, consumption of specific target foods, and other food consumption indicators specific to individual projects. Almost three-quarters (72%) measured anthropometry: stunting, underweight, and wasting in children, and underweight and BMI of women. Biochemical indicators of nutritional status were employed in 23 projects (38%). Anemia was measured in 21 projects (35%): 18 by hemoglobin only and 3 with additional biomarkers of anemia. Vitamin A status was measured in 11 projects (by serum retinol). Almost 60% of projects assessed food security at household level. A high proportion of projects also collected information on nutrition knowledge or behaviors (72%) and economic outcomes (68%). Women's empowerment or labor was measured in 53% of projects, covering information including time and labor, income, assets, decision-making, and knowledge. Three studies used the Women's Empowerment in Agriculture Index (Alkire et al., 2013). Natural resource management that could affect human health and

Table 2

Types of indicators measured by predominant intervention type.

Type of indicator All intervention pro- Biofortification projects Marketing or income Diversification projects Knowledge promo- Integrated pro-

jects (n =50) (n= 7) projects (n = 12) (n = 14) tion projects (n=3) jects (n= 14)

% % % % % %

Anthropometry 74

Biochemical 40

Diet and food 98

consumption

Food security experience 63

Economic outcomes 74

Women's empowerment/ 60 labor

Knowledge or behaviors 78

Natural resource 33 management

71 57 100

43 71 29

33 17 92

54 83 58

78 29 100

69 64 57

0 0 100

67 0 100

100 71 100

75 93 71

agricultural productivity (e.g. soil fertility, water quality) was measured in 29% of studies.

One striking result was the diversity of interventions by which projects addressed the goal of improving nutrition. Table 2 categorizes projects by type of primary approach (production diversification, biofortification, marketing or income generation, knowledge promotion, or integrated projects i.e. those that include health, sanitation, and/or environmental components) and shows the types of indicators used by the different intervention types.

3.3. Project goals and causal pathways

Of the 50 intervention projects, most indicated more than one goal or objective. The most commonly stated project goals and objectives were to improve nutrition/health/micronutrient status (58%); improve food consumption in general or consumption of specific foods (44%); improve food security (28%); develop value chains of specific products or improve incomes (32%); develop the capacity of institutions or to influence policy (22%); and empower women (12%).

In most studies, improving diets or child feeding was the main hypothesized pathway to improving nutrition. Of the 28 projects that took anthropometric measures in children < 5,12 depended entirely on impact through food on the expectation that producing certain nutritious foods, and/or providing nutrition education to favor adoption of those foods, would lead to improvement of diets or complementary feeding, improved child growth and reduced malnutrition. The remaining 16 projects using anthropometry had water, sanitation or social protection activities or linkages to address non-food causes of malnutrition.

Of those using diet, food security, and nutrition status indicators, 90% said the primary reason for the selection was their importance to project goals. The second reason for selecting indicators was "to evaluate impact of specific project activities", reported for over two-thirds of all diet, food security, and nutritional status indicators being used. Few indicators were selected because they were required by a funder (< 11%). Most projects did not select diet and food security indicators because they were government or country standard indicators, but 18% of nutritional status indicators were selected at least in part for that reason. Other reasons for selecting anthropometric indicators were that they were easier/less costly to use than other indicators; "just in case"; and to investigate associations with other variables.

3.4. Statistical power to detect improvements in nutritional outcomes from interventions

The main indicators used in intervention studies were those measuring diet/food consumption and child anthropometry. To

examine whether projects would have adequate power to observe changes in these indicators, we limited our analysis to the 50 agricultural intervention projects. Dietary/food consumption indicators were selected by 48 of the 50 agricultural intervention studies, the majority of which (34) measured food consumption in children < 5 years of age. Child anthropometric indicators were selected by 33 intervention studies; 24 measured both child anthropometry and diet/ food consumption among children < 5.

Respondents reported the planned sample sizes for their evaluation surveys and use of a comparison group. We categorized this information into six mutually exclusive groups: large, medium, or small, with or without a comparison group (Table 3). Table 4 provides an example of survey sample sizes that would be needed in both the intervention and comparison group to detect a statistically significant reduction in stunting at 95% level of confidence in the intervention group. The sample size estimates for this example were based on a design effect of 2 (i.e. doubling the sample needed for a random sample) for detecting differences in stunting prevalence. In this simplified example, sample sizes of over 1200 in each group (intervention and comparison) would be necessary to detect a 20% reduction in stunting at a baseline prevalence of 40%, and a considerably higher number where the baseline prevalence is 30%. The smaller the expected reduction in stunting, the larger the sample needed.

Comparing the sample size estimates to observe a change in stunting (Table 4) with the actual planned sample sizes in the studies surveyed (Table 3), only 6 studies with counterfactual measuring child stunting (25%) would have adequate power to observe a 20% decline in stunting, if planned activities actually can produce that large a decline. Of the 6 studies with large enough samples to observe a 20% stunting reduction, 4 are linked with other sector interventions such as health system strengthening, water and sanitation, and care practices promotion. No study appears powered to observe a decline in stunting of less than 15%.

There are many dietary indicators used in these studies. As an example using one of the most popular and validated indicators, Table 5 shows survey sample sizes that would be needed to detect a statistically significant increase in the percent of children achieving minimum dietary diversity (MDD), measured by percent of the sample consuming 4 out of 7 food groups (WHO, 2008). Detecting a doubling in the proportion of children achieving MDD, the target set by the Feed the Future Program (USAID, 2013), would require a sample of 46-164 per group (taking into account a design effect of 2), based on typical baseline MDD rates in sub-Saharan Africa (IFPRI, 2015). All studies in our sample with a counterfactual (Table 3) would be able to detect this difference in MDD (n=24). Most medium and large studies in our sample (n=17) would be able to detect a more modest 50% improvement in MDD, which would require approximately 194-588

Table 3

Planned sample sizes of all intervention projects and those measuring stunting and any type of diet/food consumption indicator among children < 5 years.

Sample size category by comparison All intervention projects Projects measuring Stunting in Projects measuring Diet/ food consumption in

group status (N — 50 total) children <5 (N—32 total) children < 5 (N—35 total)

N(%) N (%) N (%)

Large with comparison group (900 + per 9 (18) 6 (19) 7 (20)

group)a

Medium with comparison group (450- 8(16) 5 (16) 5 (14)

899 per group)b

Small with comparison group (150-449 7 (14) 4 (12) 6(17)

per group)c

Medium/large (450 +) without compar- 5 (10) 3 (9) 3 (9)

ison groupd

Small (150-450) without comparison 4 (8) 2 (6) 2 (6)

groupe

Very small ( < 150) with or without 7(14) 3(9) 4(11)

comparison groupf

Missing 10 (20) 9 (28) 8 (23)

a Total actual reported sample sizes of 1800-4000, intervention/comparison groups roughly balanced. b Total actual reported sample sizes 900-1300, intervention/comparison groups roughly balanced. c Total actual reported sample sizes 300-750, intervention/comparison groups roughly balanced. d Total actual reported sample sizes 500-1500 (single group). e Total actual reported sample sizes 190-400 (single group). f Total actual reported sample sizes 80-130.

Table 4

Sample sizes required to detect a statistically significant change in stunting prevalence in the intervention group.a

Baseline Desired target Desired pre- N in each N in each

prevalence for reduction in valence to group group being

stunting in in- achieve in in- being compared,

tervention tervention compared with Design

group group Effect —2

40% 4% 38.4% 14,614 29,228

40% 12% 35.2% 1598 3196

40% 20% 32.0% 564 1128

30% 4% 28.8% 22,626 45,252

30% 12% 26.4% 2452 4904

30% 20% 24.0% 1361 2722

a Using an on-line sample size calculator (Rollin Brant's Sample Size Calculators, 2016). For all sample size calculations, we used Alpha 0.05, Power 0.80, assuming balanced samples of both groups being compared.

Table 5

Sample size required to detect a statistically significant difference in the percent of children achieving minimum dietary diversitya in the intervention group.

Baseline prevalence Desired target for increase in MDD in intervention group Desired prevalence to achieve in intervention group N in each group being compared N in each group being compared, with Design Effect —2

20% 25% 25.0% 1094 2188

20% 50% 30.0% 294 588

20% 75% 35.0% 138 276

20% 100% 40.0% 82 164

40% 25% 50.0% 388 776

40% 50% 60.0% 97 194

40% 75% 70.0% 42 84

40% 100% 80.0% 23 46

aChildren < 2 years of age consuming foods from 4 or more food groups out of 7 based on the WHO/UNICEF Infant and Young Child Feeding minimum dietary diversity indicator (WHO, 2008), using an on-line sample size calculator (Rollin Brant's Sample Size Calculators, 2016).

observations in each group (intervention and comparison). Other dietary/food consumption indicators may require smaller or larger sample sizes than those shown in Table 5, depending on their baseline value and estimated magnitude of change.

3.5. Study design

Duration of the intervention projects in this sample ranged from 1 to 13 years, with an average project length of 4.2 years (median 4 years, mode 5 years). Of these 50 studies, 37 indicated use of a comparison group (74%): two used a true experimental design involving randomization to allocate participation and 35 used a quasi-experimental design where the comparison group was composed through other selection criteria (not investigated). Ten studies indicated no comparison group, and information was missing for another three. Most of those following an experimental or quasi-experimental design (76%) indicated that assessments would be carried out at both baseline and endline on both the intervention and the comparison group, an important element for attributing observed changes in the intervention group to the project itself and not to other, unobserved influencing factors.

4. Discussion

This study builds upon the previous mapping study (Hawkes et al., 2012; Turner et al., 2013), from which our sample was drawn, by showing the indicators each research project is using, by comparing these to the set of outcomes and impacts most likely to be affected by the projects based on their impact pathways, and by analyzing whether the studies appear to be adequately powered to observe changes in anthropometric and dietary outcomes. One strength of this review is that we draw on research projects that are on-going or in the final planning phases, thus having a snapshot of current use of indicators rather than relying on published investigations that may have taken place years before.

4.1. What is different in this generation of evidence compared to the past

This study shows a significant change in the current generation of evidence on agriculture-nutrition projects, compared to the

previously reviewed literature (Ruel and Alderman, 2013; Webb Girard et al., 2012; Masset et al., 2011). Almost all projects of our sample (87%) used one or more indicators of dietary quality, whereas most studies included in earlier reviews focused on consumption of specific foods rather than overall diet. In the intervention projects we surveyed, dietary diversity was commonly measured. Many projects measured both household and individual dietary diversity of women or young children as a way of examining food access at household level and micronutrient adequacy of the diets of vulnerable household members. As of 2008, no study investigating agricultural impact on nutrition had used dietary diversity scores to evaluate impact (Herforth, 2010). Diet-related conclusions in previous literature mostly concerned consumption of the specific foods promoted in projects, which does not necessarily translate into positive changes in diets as a whole. The current generation of agriculture/nutrition studies is starting to fill the gap in understanding how changes in agriculture may affect overall diet quality.

Given the high proportion of respondents who reported selecting indicators from each indicator class, projects appear to be applying a broader range of indicators than before, and these indicators relate well to at least parts of the conceptual framework (Fig. 1). This trend means that current projects have greater potential to assess their outcomes along program impact pathways. Our study clearly shows that newly developed indicators are being used, since many of the commonly employed indicators such as the Household Dietary Diversity Score (HDDS) (FAO, 2010), the Women's Dietary Diversity Score (WDDS) (FAO, 2010) and the Household Food Insecurity Access Scale (HFIAS) (Coates et al., 2007) were developed in the mid-2000s, and therefore were not available at the time older studies were conducted. Three projects addressing some aspect of women's empowerment are using the Women's Empowerment in Agriculture Index (Alkire et al., 2013), a newly developed indicator that is currently undergoing validation research (Malapit et al., 2014). The popularity of new indicators shows that there is a great demand for well-defined, valid and feasible indicators related to the agriculture-nutrition nexus.

4.2. Program impact pathways, and expectations for impact

Our study points out a number of gaps in the current generation of agriculture-nutrition projects related to indicator use and program theory, and perhaps unrealistic expectations of what agriculture projects can achieve to improve nutrition. A basic premise is that project activities should be based on program impact pathways leading to desired outcomes that can realistically be attributed to the interventions. Because this study focused on projects in progress, we are not able to determine how well the selected indicators have been able to capture changes that could be attributed to project activities. Based on responses describing project impact pathways, however, we can make some observations about their potential to achieve impact.

Encouragingly, a generalized theory of change linking agriculture to improved nutrition seems to be integrated into the projects: e.g. to increase production, income, women's empowerment, and/or nutrition knowledge in order to improve food security and dietary consumption, leading to improvements nutritional status of project beneficiaries. Of the most common indicator classes (diet, food security, nutrition knowledge and behaviors, women's empowerment, and economic) that potentially link agricultural interventions to improved nutritional outcomes, 62% of projects are selecting indicators across several of these classes, which could allow analysis along program impact pathways relevant to the project activities. Most projects are also collecting qualitative information using methods such as focus groups and in-depth interviews with beneficiaries and key

informants, which can greatly contribute to establishing plausible explanations of how each project brings about changes.

A potential risk is that researchers may be aiming for a standard impact model, culminating in improved nutritional status, regardless of the intervention's scope. While using standard indicators can facilitate comparison of results across studies, it is also important that indicators be selected for plausible and realistic outcomes for each project's objectives, activities, and statistical power. As shown in Table 2, the use of indicator types varied somewhat by project type, but in most cases the variation was relatively small; that is, regardless of the project's approach, similar outcomes were assessed. We would have expected biochemical indicators to be used in biofortification projects much more frequently than in integrated projects, for example, but this was not the case. Measuring nutritional status indicators across all types and sizes of agriculture-nutrition projects may set up expectations for impacts that certain projects cannot realistically deliver.

4.3. What will we learn about impact on nutritional status?

Similar to previous research, three-quarters of projects are assessing impact on nutritional status, but very few appear to have sufficient statistical power to detect an impact. There are two issues: first, whether the projects can actually result in a large reduction in stunting, and second, whether sample sizes are adequate to observe such a reduction.

Based on our sample size calculations, a 15-20% stunting reduction is the smallest magnitude of effect likely to be observed with available samples in the studies surveyed. Is it plausible for these projects to produce this level of reduction in stunting? How large a reduction in stunting is seen from other types of nutrition interventions?

Agriculture-nutrition projects that involve farming are always effectiveness trials, attenuated by internal and external factors such as implementation quality and context-appropriateness. They are therefore not comparable to the efficacy trials that have provided the vast majority of the evidence for direct nutrition interventions; for example, in the 2008 Lancet series on maternal and child undernutrition, over 97% of studies on which recommended interventions were based were efficacy trials (Bhutta et al., 2008). Therefore, we look to high-quality effectiveness trials of targeted nutrition interventions as a point of comparison. A review of USAID Title II-funded maternal and child health and nutrition (MCHN) programs found an average stunting reduction of 2.4% points per year (Swindale et al., 2004). If the baseline stunting prevalence were 45%, as in our example, this would translate to a 21.3% stunting reduction in 4 years. Other effectiveness trials for improved complementary feeding have shown a similar magnitude of effect (Caulfield et al., 1999); another evaluation found a 10% reduction in stunting attributable to a carefully-implemented preventive approach to malnutrition over three years in Haiti (Menon and Ruel, 2007). Therefore, if agriculture-nutrition projects result in a similar magnitude of effect as carefully-implemented targeted MCHN interventions, a 20% decline in stunting within 5 years is plausible. Most of the studies in our sample, including the 6 largest, had project durations of 2-5 years.

However, it is probably unreasonable to expect agriculture interventions to have as large a magnitude of effect on nutritional status as a direct nutrition intervention over the same time period. Nutrition-sensitive interventions target underlying causes of malnutrition, rather than immediate causes targeted by direct nutrition interventions (Ruel and Alderman, 2013). Most researchers reported that their projects aim to improve nutrition through the pathway of access to high-quality food - an underlying cause of malnutrition (UNICEF, 1990) - which is in turn expected to lead to

improved diets. While access to food is critical to nutrition and a basic human right, it is a step further away from nutritional status than health/disease status and nutrient intake targeted by direct nutrition interventions. Even among underlying causes, lack of access to food may not be the strongest limitation to child growth, depending on context.

Even if the interventions surveyed could produce a 20% decline in stunting, almost no project in our survey has statistical power to observe that level of impact, based on the survey sample sizes reported. In several cases, the required sample would be greater than the program's population coverage, pointing not to weak study design per se, but to unrealistic expectations of project designers about the kind of impact that can be observed. This echoes the finding of Masset et al. (2011) that necessary survey sample sizes for observing impact on anthropometry are often unobtainable. Therefore, the objective to reduce stunting in an agricultural project may be misguided when both resources and the types of interventions preclude large study populations, direct impact on immediate causes of malnutrition, and long intervention duration.

4.4. What will we learn about impact on diet and food security?

Many interventions in the survey reported program impact pathways primarily aiming to improve food consumption and diets of children. Based on program activities, changes in diet are closer to plausible impact pathways than changes in stunting, which has many non-food causes. Aligned with impact pathways, most studies are using indicators of diet quality. This is a major change from previous research and will contribute to knowledge on the potential impact of agricultural interventions to have a direct impact on diets. Because dietary diversity scores were not included as outcome indicators in most previous research (e.g. that included in previous reviews of the evidence base (Ruel and Alderman, 2013; Webb Girard et al., 2012)), it is difficult to estimate realistic targets for improvements in prevalence of minimum dietary diversity. We therefore based our estimates on published targets (described in the methods section) of USAID's Feed the Future, an initiative which includes dietary diversity as a key outcome indicator. If these targets are realistic, then many studies (at least the 24 in our sample) will have power to detect statistically significant improvements. This generation of research, however, may provide answers as to what magnitude of change in dietary diversity indicators is plausible.

Dietary diversity scores, which are the most widely-used indicators of diet quality in current research, correlate positively with micronutrient intakes and nutritional status at the individual level (FAO, 2010; Arimond et al., 2010). They do not capture all aspects of diet quality, however, which is important to note because some aspects of diet may be going unmeasured. For example, dietary diversity scores do not reflect dietary protections against obesity and non-communicable diseases (NCDs), increasingly common health problems in low- and middle-income countries (Steyn and Mchiza, 2014). Other indicators may be needed to capture these aspects of diets, along with research to understand their associations with nutritional status and NCDs. Furthermore, projects that promote one type of nutrient-rich food may not influence dietary diversity scores unless the particular food promoted adds a missing food group to diets. Clearly, many types of dietary and food consumption indicators are needed to fit the purpose of an intervention.

Many of the projects aiming to improve household food access were measuring food security, using a variety of different methods. Experience-based food insecurity tools can directly assess whether household or individual resources are sufficient for accessing food. Almost half of the studies used experience-based food security scales, such as the HFIAS (Coates et al., 2007) and Household

Hunger Scale (HHS) (Ballard et al., 2011) (which is a measure of severe food insecurity more appropriately used in emergency situations than agricultural interventions), while 25% described only measuring food availability in general or were not specific about how they intended to capture food security (Table 1). While there are many existing food security metrics (Jones et al., 2013), a suite of indicators that measures each dimension of food security (sufficiency, quality, acceptability, safety, certainty/stability) is not yet established (Coates, 2013).

Food security and diet are very important indicators for the agriculture-nutrition nexus and should be an integral part of program impact pathways. Researchers should be encouraged to use valid and standard indicators of food access and dietary diversity, such as the Food Insecurity Experience Scale (Ballard et al., 2013) and the Minimum Dietary Diversity for Women indicator (FAO and FHI 360, 2016), in order to allow comparability across studies for demonstrating effectiveness of agriculture on improving nutrition. (These indicators were both too recently validated to have been used in the research surveyed.) It would be useful for future research to contribute additional valid indicators of aspects of diet quality and food security not fully captured by existing indicators.

4.5. We will not learn enough about effects on women's empowerment, health/sanitation environments, and food environments from current research

Women's empowerment is a commonly emphasized pathway from agriculture activities to nutrition (Ruel and Alderman, 2013; Gillespie et al., 2012; Herforth and Harris, 2014), but it is a multi-faceted construct that is difficult to measure. Although 53% of studies surveyed are attempting to assess women's empowerment or labor, few indicators were reported or described that could be replicated in other settings, with the exception of the Women's Empowerment in Agriculture Index that was employed in three projects (Supplementary material 2). An index may be useful for overall assessment of women's status, but developing indicators of various aspects of empowerment separately (e.g. measures of women's income) could improve the project's ability to attribute improvement to project activities.

Current studies seem to pay little attention to natural resource management affecting health and sanitation environments. Only 4 projects were measuring water access or quality, although agriculture projects can have significant effects on water through effects on water for human consumption, habitats for disease vectors, and women's time and labor (Herforth et al., 2012). Few indicators of other ecological indicators related to health protection or disease risk were reported as being used. The dimensions of the health and sanitation environment most relevant to agriculture interventions include water quantity and quality, food safety, agrochemical exposure, risk of zoonotic or water vector-borne disease, and cleanliness of children's play areas.

There also appears to be low attention paid to food environments beyond producers' households. Most studies focused on on-farm food production as a route to diet, and measurement stopped at farm-gate. Food environment was identified along the impact pathway from production (or value chains) to diets in the mapping study (Hawkes et al., 2012; Turner et al., 2013). Although our survey did not ask respondents specifically about food environment indicators, a review of the original mapping study database showed that among the projects included in our survey sample, 9 projects (15% of our sample) are recorded as measuring aspects of the food environment, typically local food availability and/or prices. This low number may stem from a lack of established indicators that can be used to track availability and affordability of nutritious diets (Herforth and Ahmed, 2015).

4.6. Limitations of the study

This survey aimed to capture on-going agriculture-nutrition research comprehensively, but does not include all relevant projects to the agriculture-nutrition evidence base. A very high proportion of identified project principal investigators responded to the survey, indicating that we were able to capture most of the relevant research identified in the original LCIRAH mapping study. Therefore non-response bias is unlikely to have affected the scope of indicators reported here or the main conclusions regarding priorities and gaps in measurement (Hawkes et al., 2012; Turner et al., 2013). It is important to note that this does not capture an exhaustive sample but rather a convenience sample; for example the authors of the mapping study noted that studies that may be ongoing outside the English-speaking world were not captured well. Despite this limitation, research trends can be identified such as increased use of food consumption indicators and unresolved issues such as choosing stunting as a primary impact indicator.

We began with a database of on-going projects that researchers self-reported as being agriculture-nutrition projects and surveyed principal investigators of projects that (a) set out to affect nutrition, and (b) did so primarily via interventions whose focus was impact on farm households. Thus, this study excludes two important types of projects. One is research that explores effects of agriculture programs or policies on diets in the wider consumer population, representing significant potential for nutrition impact at scale. A systematic review found that there are very few studies of this type (Dangour et al., 2013). While measuring the effects of agricultural programs and policies on the wider consumer population is complex, not doing so represents serious gap in understanding the impact of agriculture on nutrition (Pinstrup-Ander-sen, 2013). The second type of intervention excluded in our study is larger agriculture programs or investments where nutrition is not necessarily the primary goal (i.e. the majority of investments). These usually do not include nutrition indicators at all (World Bank, 2014). Not all projects necessarily seek or need impact evaluations; some may measure trends simply to ensure objectives are being met or to track trends in the beneficiary population over the course of the project, rather than pre-post intervention/ comparison designs which can provide attribution of changes to the intervention (Habicht et al., 1999; FAO, 2016). Measuring nutrition-relevant indicators in such projects, however, would further contribute to understanding how agriculture investments can affect nutrition through intentional or unintentional effects (Le-vinson and Herforth, 2015).

While all but 10 respondents gave some indication of their sample sizes, the reported survey sample sizes for evaluation were often estimates, subject to change based on attrition, expansion of new phases, or decisions to pool data in separate projects/sites under the same funding umbrella. Therefore, our conclusions about ability of the surveyed projects' ability to observe an impact for certain indicators are based on approximate information. Further, little literature exists to suggest likely ranges of changes in dietary indicators. We used MDD as a sample dietary indicator because it is validated, and popular in our sample; however, it has only been used in impact evaluations within the last five years, so plausible improvements in prevalence are estimates. Despite the imprecision, we are confident in concluding that very few studies have sufficient statistical power to observe impact on child growth, while many appear to be powered to observe impact on at least some of the dietary outcomes of interest, which are also the projects' main impact pathways. It should also be noted that the sample size calculations are meant to show in general that large sample sizes are needed to detect a significantly significant decrease in stunting, but estimates may be different in the case of complex sampling designs such as cluster randomized trials and

where baseline prevalence is lower/higher. Each research team should undertake their own power calculations a priori.

The survey itself was not comprehensive in prompting respondents for indicators of food environments and health/sanitation environments. It is possible there may be a higher number of studies measuring these aspects than we captured. Based on the project aims and impact pathways reported, however, few projects appear to be addressing these factors even though they are important features in the pathway between agriculture and nutrition.

Finally, the survey is limited in understanding the full potential for project impact on identified outcomes because it did not cover any questions about quality of implementation or project context. These project features should be documented and characterized, and taken into consideration in the final analysis and interpretation of results.

5. Conclusions

The evidence base for impact of agriculture on nutrition is bounded by what is measured. Understanding the true impact of agriculture on nutrition has been limited by the scope of available indicators. When comparing the overall conceptual framework (Fig. 1) to the indicators reported in our survey, it is clear that not all important factors are being measured in the current generation of research. Furthermore, the focus of measurement on nutritional status may be unlikely to yield useful results. Our survey of current agriculture-nutrition projects finds that most projects are measuring nutritional status, but are unlikely to document improvements due to inadequate statistical power and/or project scopes unlikely to produce a large impact on nutritional status. Although three-quarters of current projects are measuring nutritional status, we estimate that only 6 have adequate power to observe a large (20%) decline in stunting, and that none have power to observe a decline of < 15%. Maintaining the expectation of impact on stunting prevalence when program activities are not designed to achieve it or impact evaluation surveys are too small to detect a meaningful improvement if there is one, may result in type II error and the perception that agriculture projects are ineffective for influencing nutrition.

It may be appropriate for some evaluations to measure changes in nutritional status, where the program is plausibly expected to improve these indicators and powered to observe the improvement. Nutritional status impact, however, should not be the basis for evaluating the evidence base in general, nor should it be the single benchmark of success expected of nutrition-sensitive investments by donors. For reasons of both program scope and statistical power in the majority of evaluations, the more appropriate outcomes to expect from agriculture-nutrition projects are improved food access and dietary consumption - those outcomes the majority of such projects are designed to affect. Most projects are measuring some aspect of diet quality, and researchers report that as the main impact pathway for improving nutrition.

Finally, most current projects focus their evaluations on farm households participating in the intervention, while their effects may in fact be much broader, affecting local markets and the wider consumer population. The effects of agriculture-nutrition interventions may be better understood by examining food environments and health environments. Moreover, the evidence base on agriculture-nutrition links should be broadened to the effects of policies and large-scale research investments on populations.

Our main conclusion is that agriculture-nutrition projects are measuring many important potential outcomes along program impact pathways, but that the focus needs to move away from nutritional status and toward the more proximal outcomes that such interventions can affect, such as indicators of diet and food

access (CGIAR Independent Science and Partnership Council, 2014). Impact indicators selected need to be appropriate to the intervention. Existing validated indicators of diet quality and food access (such as the Minimum Dietary Diversity for Women (FAO and FHI 360, 2016), and the Food Insecurity Experience Scale (Ballard et al., 2013) should be used, where appropriate, for comparability across studies. More and better indicators of diet quality, food security, food environments, women's empowerment, and health and sanitation environments need to be developed to more fully evaluate the outcomes that agriculture projects are best-suited and most likely to affect.

Competing interests

The authors declare that they have no competing interests.

Authors' contributions

AH and TB designed research; AH and TB conducted research; AH and TB analyzed data; and AH and TB wrote the paper. All authors read and approved the final manuscript.

The views expressed in this publication are those of the authors and do not necessarily reflect the views of FAO.

Acknowledgements

This work was carried out by the Nutrition and Food Systems Division of FAO with funding from the EU-FAO Improved Global Governance for Hunger Reduction Programme (grant number DCI-FOOD 2011/262-399) (2012-2015). We thank LCIRAH for their collaboration, especially Rachel Turner and Jeff Waage. We gratefully acknowledge all the researchers who took the time to respond to this survey.

Appendix A. Supplementary material

Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016Zj.gfs.2016.07.004.

References

Alkire S, Meinzen-Dick R, Peterman A, Quisumbing AR, Seymour G, Vaz A., 2013. The Women's Empowerment in Agriculture Index. OPHI Working Paper No. 58.

Arimond, M., Wiesmann, D., Becquey, E., Carriquiry, A., Daniels, M.C., Deitchler, M., Danou-Fogny, N., Joseph, M.L., Kennedy, G., Martin-Prevel, Y., et al., 2010. Simple food group diversity indicators predict micronutrient adequacy of women's diets in 5 diverse, resource-poor settings. J. Nutr. 140, 2059S-2069S.

Ballard T., Coates J., Swindale A., Deitchler M., 2011. Household Hunger Scale: Indicator Definition and Measurement Guide. FHI 360/FANTA; Washington, DC.

Ballard T.J., Kepple A.W., Cafiero C., 2013. The Food Insecurity Experience Scale: Development of a Global Standard for Monitoring Hunger Worldwide. FAO; Rome. Available at: (http://www.fao.org/3/a-as583e.pdf> (accessed 20.10.15).

Bhutta, Z.A., Ahmed, T., Black, R.E., Cousens, S., Dewey, K., Giugliani, E., Haider, B.A., Kirkwood, B., Morris, S.S., Sachdev, H.P.S., Shekar, M., 2008. What works? Interventions for maternal and child undernutrition and survival. Lancet 371, 417-440.

Caulfield, L.E., Huffman, S.L., Piwoz, E.G., 1999. Interventions to improve the intake of complementary foods by infants 6-12 months of age in developing countries: impact on growth and prevalence of malnutrition and potential contribution to child survival. Food Nutr. Bull. 20 (2), 183-200.

CGIAR Independent Science and Partnership Council. Nutrition and Health Outcomes: Targets for Agricultural Research. Brief No. 43, Sept 2014. (http://ispc. cgiar.org/sites/default/files/ISPC_MobilizeScience_SF2013_Brief_0.pdf> (accessed 14.01.16).

Coates, J., 2013. Build it back better: deconstructing food security for improved measurement and action. Glob. Food Secur. 2 (3), 188-194.

Coates J., Swindale A., Bilinsky P., 2007. Household Food Insecurity Access Scale (HFIAS) for Measurement of Household Food Access: Indicator Guide (v. 3). FHI 360/FANTA; Washington, DC.

Dangour, A.D., Hawkesworth, S., Shankar, B., Watson, L., Srinivasan, C.S., Morgan, E. H., Haddad, L., Waage, J., 2013. Can nutrition BE promoted through agriculture-LED Food price policies? A systematic review. bMj Open 3, e002937. http://dx. doi.org/10.1136/bmjopen-2013-002937.

EU, FAO, USAID, FANTA III, FHI 360, 2016. Introducing the Minimum Dietary Diversity - Women (MDD-W) Global Dietary Diversity Indicator for Women. Washington, DC: USAID/FANTA III/FHI 360. <http://www.fantaproject.org/sites/ default/files/resources/Introduce-MDD-W-indicator-brief-Sep2014.pdf> (accessed 11.06.15).

FAO. Assessing Impact of Development Programmes on Food Security - E-learning course. <http://www.fao.org/elearning/#/elc/en/course/IA> (accessed 11.01.16).

FAO, 2010. Guidelines for Measuring Household and Individual Dietary Diversity. FAO, Rome.

Gillespie S., Harris L., Kadiyala S., 2012. the Agriculture-Nutrition Disconnect in India, What Do We Know? Discussion Paper 01187. IFPRI; Washington, DC.

International Food Policy Research Institute. 2015. Global Nutrition Report 2015: Actions and Accountability to Advance Nutrition and Sustainable Development. IFPRI; Washington, DC.

Habicht, J.P., Victora, C.G., Vaughan, J.P., 1999. Evaluation designs for adequacy, plausibility and probability of public health programme performance and impact. Int. J. Epidemiol. 28, 10-18.

Hawkes, C., Turner, R., Waage, J., 2012. Current and Planned Research on Agriculture for Improved Nutrition: A Mapping and a Gap Analysis. Leverhulme Centre for Integrative Research on Agriculture and Health, London, Available at: <http://r4d.dfid.gov.uk/pdf/outputs/misc_susag/LCIRAH_mapping_and_gap_ana lysis_21Aug12.pdf> (accessed 11.06.15).

Herforth, A., 2010. Nutrition and the environment: fundamental to food security for Africa. In: Pinstrup-Andersen, P. (Ed.), The African Food System and Its Interaction with Human Health and Nutrition: Research and Policy Priorities. Cornell University Press, Ithaca.

Herforth, A., Ahmed, S., 2015. The food environment, its effects on dietary consumption, and potential for measurement within agriculture-nutrition interventions. Food Security 7 (3), 505-520.

Herforth A, Harris J., 2014. Understanding and applying primary pathways and principles. Brief #1. In Improving Nutrition through Agriculture Technical Brief series. USAID/SPRING Project; Arlington, VA.

Herforth A., Jones A., Pinstrup-Andersen P., 2012. Prioritizing Nutrition in Agriculture and Rural Development: Guiding Principles for Operational Investments. HNP Discussion Paper. The World Bank; Washington, DC.

IBM Corp., 2011. IBM SPSS Statistics for Windows, Version 20.0. IBM Corp, Armonk, NY. Released 2011.

Jones, A.D., Ngure, F.M., Pelto, G., Young, S.L., 2013. What are we assessing when we measure food security? A compendium and review of current metrics. Adv. Nutr. 4, 481-505.

Kennedy, E., Bouis, H., 1993. Linkages Between Agriculture and Nutrition: Implication for Policy and Research. IFPRI, Washington, DC.

Levinson F.J., Herforth A., 2013. Monitoring and evaluating the food security and nutrition effects of agricultural projects. Preparatory Paper for the Second International Conference on Nutrition. FAO and WHO; Rome. <http://www.fao. org/fileadmin/user_upload/agn/pdf/Levinson_Herforth_Paper.pdf> (accessed 30.01.15).

Magnani, R., 1999. Sampling Guide. FHI 360/FANTA; Washington, DC. Available at: <http://www.fantaproject.org/sites/default/files/resources/Sampling-1999-Ad dendum-2012-ENG_0.pdf>.

Malapit, H.J., Sproule, K., Kovarik, C., Meinzen-Dick, R., Quisumbing, A., Ramzan, F., Hogue, E., Alkire, S., 2014. Women's Empowerment in Agriculture Index: Baseline Report. IFPRI, Washington DC.

Masset, E., Haddad, L., Cornelius, A., Isaza-Castro, J., 2011. A Systematic Review of Agricultural Interventions that Aim to Improve Nutritional Status of Children. EPPI-Centre, Social Science Research Unit, Institute of Education, University of London, London.

Menon P., Ruel M., 2007. Prevention Is Better Than Cure: Final Report of the evaluation: Prevention of Cure? Comparing Preventive and Recuperative Approaches to Targeting Maternal and Child Health and Nutrition Programs. FANTA Project; Washington, DC.

Pinstrup-Andersen, P., 2013. Nutrition-sensitive food systems: From rhetoric to action. Lancet 382 (9890), 375-376.

Rollin Brant's Sample Size Calculators. University of British Columbia. Available at: <http://stat.ubc.ca/~rollin/stats/ssize/> (accessed 13.01.16).

Ruel, M.T., Alderman, H., 2013. Nutrition-sensitive interventions and programmes: how can they help to accelerate progress in improving maternal and child nutrition? Lancet 382, 536-551.

Steyn, N.P., Mchiza, Z.J., 2014. Obesity and the nutrition transition in Sub-Saharan, Africa. Afr. Ann. NY Acad. Sci. 1311, 88-101.

Swindale A., Deitchler M., Cogill B., Marchion T., 2004. The Impact of Title II Maternal and Child Health and Nutrition Programs on the Nutritional Status of Children. FHI 360/FANTA; Washington, DC.

Turner, R., Hawkes, C., Waage, J., Ferguson, E., Haseen, F., Homans, H., Hussein, J., Johnston, D., Marais, D., McNeill, G., et al., 2013. Agriculture for improved nutrition: the current research landscape. Food Nutr. Bull. 34 (4), 369-377.

United Nations, 1990. Children's Fund (UNICEF). Strategy for Improved Nutrition of Children and Women in Developing Countries. UNICEF, New York.

USAID, 2013. FTF Guidance for Setting Targets for Zone of Influence Population-

Based Indicators and for Percent Growth in Agricultural GDP Indicator. Available at: <https://agrilinks.org/sites/default/files/resource/files/Target%20setting %20guidance%20for%20FTF%20PBS%20and%20Ag%20GDP%20indicators%204.12. 13.docx>. (accessed May 2016).

Webb, P., Kennedy, E., 2014. Impacts of agriculture on nutrition: nature of the evidence and research gaps. Food Nutr. Bull. 35 (1), 126-132.

Webb Girard, A., Self, J.L., McAuliffe, C., Olude, O., 2012. The effects of household food production strategies on the health and nutrition outcomes of women and young children: a systematic review. Paediatr. Perinat. Epidemiol. 26 (Suppl. 1), S205-S222.

Webb P., 2013. Impact Pathways from Agricultural Research to Improved Nutrition and Health: A Literature Analysis and Recommendations for Research Priorities. Preparatory Paper for the Second International Conference on Nutrition. FAO and WHO; Rome.

World Bank, 2014. Learning from World Bank history: agriculture and food-based approaches for addressing malnutrition. Agriculture and Environmental Services Discussion Paper (no. 10). World Bank Group; Washington, DC.

World Health Organization, 2008. Indicators for Assessing Infant and Young Child Feeding Practices: Part1 Definitions. WHO, Geneva.