Review: Meta-analysis of the association between production diversity, diets, and nutrition in smallholder farm households

Food Policy 77 (2018) 1–18

Contents lists available at ScienceDirect

Food Policy journal homepage: www.elsevier.com/locate/foodpol

Review

Review: Meta-analysis of the association between production diversity, diets, and nutrition in smallholder farm households

T

⁎

Kibrom T. Sibhatua, , Matin Qaima,b a b

Department of Agricultural Economics and Rural Development, University of Goettingen, 37073 Goettingen, Germany Center of Biodiversity and Sustainable Land Use, University of Goettingen, 37073 Goettingen, Germany

A R T I C LE I N FO

A B S T R A C T

Keywords: Systematic review Meta-analysis Farm production Biodiversity Nutrition-sensitive agriculture

Undernutrition and low dietary diversity remain big problems in many developing countries. A large proportion of the people affected are smallholder farmers. Hence, it is often assumed that further diversifying small-farm production would be a good strategy to improve nutrition, but the evidence is mixed. We systematically review studies that have analyzed associations between production diversity, dietary diversity, and nutrition in smallholder households and provide a meta-analysis of estimated effects. We identified 45 original studies reporting results from 26 countries and using various indicators of diets and nutrition. While in the majority of these studies positive results are highlighted, less than 20% of the studies report consistently positive and significant associations between production diversity and dietary diversity and/or nutrition. Around 60% report positive associations only for certain subsamples or indicators, the rest found no significant associations at all. The average marginal effect of production diversity on dietary diversity is positive but small. The mean effect of 0.062 implies that farms would have to produce 16 additional crop or livestock species to increase dietary diversity by one food group. The mean effect is somewhat larger in Sub-Saharan Africa than in other regions, but even in Africa farms would have to produce around 9 additional species to increase dietary diversity by one food group. While results may look differently under very specific conditions, there is little evidence to support the assumption that increasing farm production diversity is a highly effective strategy to improve smallholder diets and nutrition in most or all situations.

1. Introduction Growth in agricultural productivity and food production has helped to reduce global hunger considerably over the last few decades (Gödecke et al., 2018; Khoury et al., 2014; Pingali, 2012). Nevertheless, nutritional deficiencies remain widespread, especially in Sub-Saharan Africa and Asia (FAO, 2017; IFPRI, 2017). Of particular concern are micronutrient deficiencies, which are less related to general food shortages than to low dietary quality and diversity (Headey and Ecker, 2013). In developing countries, nutritional deficiencies are still among the major causes of premature deaths, infectious diseases, physical and mental growth retardation in children, and other types of health problems (IFPRI, 2017). Eradicating malnutrition in all its forms is a fundamental part of the United Nations’ Sustainable Development Goals. Many of the world’s undernourished people are smallholder farmers (FAO, 2014). Therefore, the question as to how smallholder agriculture can be made more nutrition-sensitive is of central importance (Qaim, 2017; Sibhatu and Qaim, 2017; Frelat et al., 2016; Chamberlin et al., 2014; Ruel and Alderman, 2013). Given that smallholder farmers ⁎

typically consume a sizeable part of what they produce at home, increasing production diversity on their farms through introducing additional crop and livestock species is often seen as a promising strategy to improve dietary diversity and nutrition (Jones, 2017a, 2017b; Powell et al., 2015; Burlingame and Dernini, 2012; Fanzo et al., 2013). However, systematic evidence of this strategy’s effectiveness is limited. While several studies have analyzed the links between production diversity and dietary diversity, results are mixed and context-specific (Sibhatu and Qaim, 2018; Fanzo, 2017; Powell et al., 2015; Sibhatu et al., 2015). Powell et al. (2015) summarized 12 studies and noted that there may be a potential bias in terms of only reporting or highlighting significantly positive associations. Another recent review article found that 19 of the 21 original studies reviewed had reported positive associations between farm production diversity and dietary diversity and nutrition (Jones, 2017b). However, taking a closer look at the 21 original studies reveals that many of those that reported positive associations also included insignificant or even negative estimates for certain regions or for some of the indicators used. Mixed results within the

Corresponding author. E-mail addresses: [email protected] (K.T. Sibhatu), [email protected] (M. Qaim).

https://doi.org/10.1016/j.foodpol.2018.04.013 Received 27 October 2017; Received in revised form 18 April 2018; Accepted 21 April 2018 Available online 25 April 2018 0306-9192/ © 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).

Food Policy 77 (2018) 1–18

K.T. Sibhatu, M. Qaim

same original studies are not uncommon. For example, using data from four countries, Sibhatu et al. (2015) reported positive and significant associations between production diversity and dietary diversity in Indonesia and Malawi, but not in Ethiopia and Kenya. Using householdlevel data from Malawi, Jones et al. (2014) found positive associations between farm production diversity and aggregate dietary diversity scores, whereas some of the associations between production diversity and the frequency of consumption of certain healthy food groups (e.g., fruits, vegetables, meat, fish, dairy) were not statistically significant, or even negative and significant. Here, we review existing studies about the associations between farm production diversity and dietary diversity and/or nutrition in developing-country farm households, adding to the literature in three important ways. First, while previous review articles provided qualitative summaries of existing studies (Jones, 2017a, 2017b; Powell et al., 2015), we use a more structured approach and conduct a systematic review of published results and a meta-analysis of the estimated effects, effect sizes, and levels of statistical significance. Second, we explicitly consider all results reported in the original studies, not only those highlighted as main results. Third, we include a larger number of original studies than previous review articles. Using transparent search criteria, we identified 45 studies for inclusion in our systematic review. The main objective of our research is to shed additional light on the question whether further farm diversification can be considered a generally-applicable strategy to make smallholder agriculture more nutrition-sensitive.

dietary diversity, dietary quality, food security, and/or associated nutrition outcomes at household or individual level in farm households in developing countries. Specifically, a study was included if:

2. Methods

Using these criteria, we identified 45 studies that were included in the systematic review (a PRISMA flow diagram is shown in Fig. A1 in the Online Appendix). Of these 45 studies, 30 were published in academic journals, the other 15 were conference papers, dissertations, or studies published in books and institutional series. We decided to include studies not published in academic journals for at least three reasons. First, studies published in books, institutional series, and other formats often have a wide distribution and therefore also influence research and policy debates (Klümper and Qaim, 2014). Second, especially in the social sciences, conference papers and similar “grey literature” publications often report results of studies that are later published in academic journals. Third, a clear distinction between journal articles being peer-reviewed and other publications not being peer-reviewed is not always possible, because conferences, books, and institutional series often use a peer-review process as well (Rothstein and Hopewell, 2009). However, we control for the type of publication in meta-regressions. A full list of studies included is provided in Table 1. We also considered excluding studies with small sample sizes, as these are generally less likely to produce statistically significant results. For the field of medical research it was also shown that studies with small sample sizes are less likely to be published in academic journals (Schmucker et al., 2017). However, universally-applicable rules for minimum sample sizes do not exist, so introducing such an inclusion criterion would have been associated with a certain level of arbitrariness. Of the 45 studies included in our review, four had a sample size < 100 (Table 1). However, these four studies are all published in academic journals, so that we decided to leave them in. It can easily be verified that the general conclusions would not change if these four studies were excluded. In the meta-regressions, we use sample size as an explanatory variable.

• It investigated the association of at least one indicator of production

• •

2.1. Literature search We searched relevant documents according to PRISMA guidelines (Moher et al., 2009), with the objective to systematically review original studies that have analyzed associations between farm production diversity and household or individual diets and nutrition in developing countries. We used transparent, best-practice approaches for systematic reviews and meta-analyses (Stanley et al., 2013), following PICOS (Population, Intervention, Comparator, Outcomes, and Settings) criteria (Table A1 in the Online Appendix). Studies for inclusion in the systematic review were identified through keyword searches in the Web of Knowledge, Google Scholar, PubMed, Scopus, EconLit, AgEcon Search, Agris (a literature search platform of the Food and Agriculture Organization of the United Nations), and the International Food Policy Research Institute (IFPRI). We also screened reference lists of relevant documents that we found. Where full-text documents of studies were not publicly available, we contacted study authors and asked them to share their papers and additional details. We used English search terms without making restrictions concerning publication year and language. The key search terms we used for characterizing farm production diversity were “agrobiodiversity”, “production diversity”, “farm production diversity”, “crop diversity”, and “farm species richness”. These were combined with the following search terms to characterize diets and nutrition: “dietary diversity”, “dietary quality”, “food consumption”, “food variety”, “food security”, “nutrient consumption”, “nutrient intake”, “nutrition”, “body mass index”, “child anthropometrics”, “stunting”, “wasting”, and “underweight”. The search was completed on August 31, 2017. The authors of this review conducted literature searches and study classifications independently following the criteria described below; cases of disagreement were resolved with mutual consent.

diversity (e.g., crop species diversity, crop varietal diversity, livestock species diversity, crop and livestock species diversity, food groups produced, nutritional function diversity, crop species richness, crop species evenness) with at least one indicator of dietary diversity (e.g., household and individual dietary diversity scores, food variety scores, weighted food consumption scores, frequency of individual food items and food groups consumed, child feeding index), or an indicator of calorie and nutrient consumption (e.g., calories, protein, iron, zinc, vitamin A, nutrient adequacy ratios), or an indicator of nutrition outcomes (e.g., body mass index, child height-for-age, weight-for-age, weight-for-height, mid-upper-arm circumference), or an indicator of food security (e.g., hunger index, household food insecurity access scale), or nutrient traces in blood samples (e.g., serum iron and hemoglobin). It reported association coefficients and levels of statistical significance at household or individual level. We excluded studies conducted at village or higher levels. It reported farm production diversity data, referring to intentionally grown crops and livestock reared at the farm level. We excluded studies that focused only on homestead gardening and/or wild plants and trees.

2.3. Procedure of study classification for analysis Because of the large heterogeneity in the analytical methods and indicators used in the original studies to assess production diversity, diets, and nutrition, we start with a narrative approach for data synthesis (this is complemented by meta-analysis as described below). We classify each original study by assigning one of three possible labels, namely a “yes” label if positive associations are reported, a “no” label if

2.2. Inclusion criteria We included studies that explicitly attempted to associate at least one indicator of production diversity at farm scale with any indicator of 2

3

Philippines

Kenya Uganda Malawi

Remans et al. (2011)

Uganda Malawi

Lambrecht (2009)

Gonder (2011)

Kenya

Ekesa et al. (2008)

Kenya

Mali

Torheim et al. (2004)

DeClerck et al. (2011)

Mexico

Dewey (1981)

Kenya Tanzania

Country

Original study reference

Herforth (2010)

(2)

(1)

CS

170 HH

24 h recall, blood from W

24 h recall, CH anthrop.

CS

844 IN 246 CH 261 HH

24 h recall

24 h recall

7 d recall

7 d recall

24 h recall, CH anthrop

Type of diet/ nutrition data

(5)

Blood (unspecified target group)

CS

CS

CS

CS

CS

Data type

(4)

CS

30 Farms

376 HH (294 HH with CH)

120 HH

144 HH with CH

502 IN 319 HH

149 HH with CH

Sample size

(3)

significant associations between food group production • No and CDDS significant association between food group production • No and child height-for-age Z-scores

diversity score

significant association between crop diversity and HDDS • No significant association between livestock diversity and • No HDDS with higher crop diversity report higher food • Households insecurity that earn more money by selling crops are less • Households food insecure that buy more food items have higher dietary • Households diversity association between crop diversity and HFVS in • Significant Kenya significant association between crop diversity and HDDS, • No CDDS, and CFVS in Kenya association between crop diversity and HDDS, HFVS, • Positive CDDS, and CFVS in Tanzania significant association between livestock diversity and • No HFVS in Tanzania of purchased food greater than home-produced • Diversity foods species richness not associated with hemoglobin levels • Crop and significant association between crop functional • Positive diversity (based on iron content) and hemoglobin levels significant association between nutritional function • No diversity (food group production) and individual dietary

score

association between crop diversity and child height • Positive in one village, but not in another village No significant association between crop diversity and HDDS • and HFVS association between crop diversity and mean • Significant adequacy ratio adequacy ratio lower in women than in men • Mean association between farm production diversity (crop • Positive and livestock species combined) and children’s food variety

one village

significant association between crop diversity (below 5 • No crops on a farm) and HDDS (Shannon-Wiener function) in

but not in another village

association between crop diversity (above 5 crops on • Positive a farm) and HDDS (Shannon-Wiener function) in one village,

Summary of original study findings

(6)

Table 1 Overview of studies included in systematic review, main findings, and conclusions (sorted by year of publication).

Yes

No

No

Mixed

Yes

No

Mixed

Yes

No

Yes

Yes

No

Mixed

Mixed

(8) Conclusion considering all results reported

No

Yes

(7) Main conclusion of original study authors

Significantly positive association?

(continued on next page)

No

No

–

Mixed

No

Yes

No

Mixed

(9) Only considering dietary diversity results

K.T. Sibhatu, M. Qaim

Food Policy 77 (2018) 1–18

4

Tanzania

Powell (2012)

Tanzania

Tanzania

Keding et al. (2012)

Cleghorn (2014)

India

Bhagowalia et al. (2012)

Ecuador

Country

Original study reference

Oyarzun et al. (2013)

(2)

(1)

Table 1 (continued)

122 IN 424 CH

51 HH

274 CH; their MO & HH

252 W

19,000 CH & HH

Sample size

(3)

CS

CS

CS

CS

NR

Data type

(4)

24 h recall, CH anthrop.

24 h recall

24 h recall, 7 d recall

24 h recall

30 d recall

Type of diet/ nutrition data

(5)

in nutritional function of crops was observed • Redundancy association at village level (but focus of paper is at • Positive HH or individual level) association between crop diversity (number of crops • Positive grown) and HDDS and significant association between crop diversity • Negative and share of meat and milk consumption but insignificant association between crop diversity • Negative and vegetable and fruit consumption in higher income groups have higher dietary • Households diversity scores and higher budget shares of milk correlation of vegetable diversity with WDDS and • Negative FVS in March and April and significant association in June and July, and in • Positive November and December association between crop diversity and 7 d recall • Positive HDDS association between crop diversity and 24 h recall FVS • No and HDDS diversity positively correlated with children’s mean • Crop adequacy ratio diversity negatively correlated with percent of diet • Crop purchased diversity positively associated with percent of diet • Crop produced on farm association between crop diversity and richness • Positive (Margalef and Shannon Indices) and HFVS proportion of overall food items was purchased (only • Larger 33% from own farm)

(serum iron and retinol)

significant associations between nutritional function of • No edible crops diversity, HDDS, and blood nutritional outcomes

Summary of original study findings

(6)

No

Mixed

Yes

Mixed

Yes

Yes

Mixed

Mixed

(8) Conclusion considering all results reported

Yes

Yes

(7) Main conclusion of original study authors

Significantly positive association?

(continued on next page)

Mixed

Yes

Mixed

Mixed

Yes

(9) Only considering dietary diversity results

K.T. Sibhatu, M. Qaim

Food Policy 77 (2018) 1–18

Malawi

Jones et al. (2014)

a

Multiple

Country

Original study reference

Pellegrini and Tasciotti (2014)

(2)

(1)

Table 1 (continued)

33,119 HH

6623 HH

Sample size

(3)

NR

NR

Data type

(4)

Different recall periodsa

7 d recall

Type of diet/ nutrition data

(5)

5

•

•

•

•

food groups (cereals, meat and fish, dairy products, fruits), and with frequency of food groups consumption (cereals, meat and fish, dairy products, vegetables and fruits) Negative association between farm diversity (crop count) and items from specific food groups (dairy products) and frequency of food groups consumption (cereals, meat and fish, dairy products, fruits) Negative association between farm diversity (crop and livestock count) and frequency of food groups consumption (cereals, vegetables and fruits) Negative association between farm diversity (Simpson’s index) and items from specific food groups (cereals, dairy products, fruits) and frequency of food groups consumption (cereals, meat and fish, dairy products, fruits) Dedicating greater proportion of land to cash crops associated with higher dietary diversity

richness (Simpson’s index) not associated with food • Species consumption score magnitude significantly larger in woman-headed • Association than men-headed HH diversity (crop count, crop, livestock count, Simpson’s • Farm index) not significantly associated with items from specific

and food consumption score

selling associated with higher IDDS • Crop association between farm diversity (crop count, crop • Positive and livestock count) and richness (Simpson’s index), HDDS,

score, BMI-for-age Z-score and MUAC

village: • InCropanother and vegetable diversity (Shannon index) associated • with IDDS, CDDS, IFVS, and CFVS in adjusted models and vegetable diversity not associated with CDDS and • Crop CFVS in adjusted models and vegetable diversity negatively but insignificantly • Crop associated with height-for-age Z-score, weight-for-age Z-

score, BMI-for-age z score

• InCroponeandvillage: diversity (Shannon index) not associated • with IDDSvegetable and CDDS and vegetable diversity positively associated with IDDS, • Crop IFVS, and CFVS and vegetable diversity negatively associated with • Crop MUAC and vegetable diversity negatively but insignificantly • Crop associated with height-for-age Z-score, weight-for-age z

Summary of original study findings

(6)

Yes

Yes

(7) Main conclusion of original study authors

Mixed

Mixed

(8) Conclusion considering all results reported

Significantly positive association?

(continued on next page)

Mixed

Mixed

(9) Only considering dietary diversity results

K.T. Sibhatu, M. Qaim

Food Policy 77 (2018) 1–18

6

Nigeria

Bolivia

Jones (2015)

Bangla-desh

Sraboni et al. (2014)

Dillon et al. (2015)

Tanzania

Rajendran et al. (2014)

Peru

Country

Original study reference

Zander (2014)

(2)

(1)

Table 1 (continued)

251 HH with CH

2154 HH

180 RW

3273 HH 3150 MA 3263 FA

300 HH

Sample size

(3)

CS

NR

CS

NR

CS

Data type

(4)

24 h recall, 7 d recall, CH anthrop.

7 d recall

24 h recall

7 d recall, adult BMI

24 h recall

Type of diet/ nutrition data

(5)

• Increase in agricultural revenue changes diet composition

alternative specification

significant association between crop nutritional function • No diversity (excluding tree crops) and dietary diversity in

in instrumental variable estimations

association between crop nutritional functional • Positive diversity (food groups grown) and HDDS in linear estimation significant association between crop nutritional function • No diversity (count of food groups grown) and dietary diversity

estimations in rainy and post-harvest seasons

crop diversity insignificantly associated with calorie • Food consumption in instrumental variable regression crop diversity negatively associated with male BMI • Food crop diversity negatively but insignificantly associated • Food with female BMI association between crop variety richness, IDDS, and • Positive IFVS in all seasons association between variety of exotic crops, IDDS, • Positive and IFVS in all seasons significant association between indigenous crop varieties, • No IDDS, and IFVS in rainy, post-harvest, and farming seasons significant association between crop variety, IDDS, and • No IFVS when covariates (e.g., education) are added to the

credit

association between crop diversity, HDDS, and HFVS • Positive term of crop diversity negatively associated with • Square HDDS and HFVS significant associations in Albania and Panama • No diversity negatively associated with HFVS in Panama • Crop level of production diversity negatively associated • Higher with income diversity significantly and negatively associated with • Crop food count diversity significant association between crop count and HDDS • No significant association between Simpson’s index and • No HDDS crop diversity positively associated with calorie • Food consumption in ordinary least squares regression crop diversity positively associated with HDDS • Food crop diversity positively associated with HDDS in • Food instrumental variable regression crop diversity not associated with HDDS in • Food instrumental variable regression in women’s decisions on

Summary of original study findings

(6)

Yes

Yes

Mixed

Mixed

Mixed

Mixed

Yes

Yes

No

(8) Conclusion considering all results reported

No

(7) Main conclusion of original study authors

Significantly positive association?

(continued on next page)

Mixed

Mixed

Mixed

Mixed

No

(9) Only considering dietary diversity results

K.T. Sibhatu, M. Qaim

Food Policy 77 (2018) 1–18

7

30 HH

1769 CH

Kenya

Nepal

Indonesia, Kenya, Malawi, Ethiopia

Ng’endo et al. (2015)

Shively and Sunutnasuk (2015)

Sibhatu et al. (2015)

8230 HH

3332 HH 3076 MO 2817 CH

Nepal

CH in 525 HH

3040 HH

Malapit et al. (2015)

Zambia

Kumar et al. (2015)

Sample size

Kenya

Country

Original study reference

(3)

M’Kaibi et al. (2015)

(2)

(1)

Table 1 (continued)

CS & NR

NR

7 d recall

CH anthrop.

Hunger score

24 h recall, CH anthrop., W BMI

NR

CS

24 h recall

24 h recall, CH anthrop.

Type of diet/ nutrition data

(5)

CS

CS

Data type

(4)

significant association between crop count and child • No anthropometrics significant association between proportion of food • No consumed at home and child height-for-age Z-scores high degree of subsistence orientation puts children at a • Anutritional disadvantage

Shannon evenness) and household hunger scores

association between food plant species diversity (species • No richness, individual density, Shannon diversity index and

empowerment indicators

diversity not associated with 24 h recall dietary • Crop diversity indicator significant association between crop diversity (food • No groups diversity) and energy consumption in children between production diversity and child • Association anthropometrics not reported association between crop diversity, HDDS, and • Positive CDDS association between crop diversity and height-for• Positive age Z-scores of children aged 24–59 months association between crop diversity and height-for• Negative age Z-scores in children aged 6–23 months association between farm diversity and child stunting • No association between farm diversity and dietary diversity • No in younger children association between farm diversity indicators and weight• No for-height Z-score and livestock diversity not associated with calorie • Crop adequacy ratio association between crop and livestock diversity and • Positive child’s dietary adequacy of several micronutrients association between production diversity and WDDS • Mixed or negative association between production • Insignificant diversity and female BMI association between production diversity and CDDS • Mixed associations between production diversity index • Insignificant and child height-for-age Z-score association between production diversity index and • Mixed weight-for-height Z-score association between production diversity index • Insignificant and child weight-for-age Z-score under several women

observed)

diversity not associated with infant and child feeding • Crop index in low elevation areas (high production diversity

observed)

diversity positively associated with child feeding index • Crop score in high elevation areas (low production diversity

Summary of original study findings

(6)

Mixed

Yes

Mixed

No

No

Mixed

Yes

No

Mixed

Mixed

(8) Conclusion considering all results reported

Yes

Yes

(7) Main conclusion of original study authors

Significantly positive association?

(continued on next page)

Mixed

–

–

Mixed

–

Mixed

(9) Only considering dietary diversity results

K.T. Sibhatu, M. Qaim

Food Policy 77 (2018) 1–18

8

India

India

Uganda

Burkina Faso

Kavitha et al. (2016)

Linderhof et al. (2016)

Lourme-Ruiz et al. (2016)

Ghana

Argyropoulou (2016)

Chinnadurai et al. (2016)

Malawi

Snapp and Fisher (2015)

Benin

Country

Original study reference

Bellon et al. (2016)

(2)

(1)

Table 1 (continued)

579 HH

1722 HH

289 HH

13, 696 HH

880 MO

329 HH with CH

9189 HH

Sample size

(3)

CS

CS, (pan.)

CS

CS

CS

CS

NR

Data type

(4)

24 h recall

7 d recall

24 h recall

30 d recall

24 h recall

24 h recall, CH anthrop.

7 d recall

Type of diet/ nutrition data

(5)

count positively associated with HDDS, HFCS, and • Crop calorie consumption richness (Simpson’s index) positively associated with • Crop HDDS and HFCS richness (Simpson’s index) negatively associated with • Crop calorie consumption nutritional diversity not associated with HDDS and • Crop calorie consumption

factors

significant association between crop diversity (crop count • No and Simpson’s index) and HDDS when controlling for other

factors

associations between crop diversity (crop count, • No Shannon-Wiener index) and dietary diversity of children associations between crop diversity (crop count, • No Shannon-Wiener index) and nutrition status of children. index negatively associated with dietary • Shannon-Wiener diversity of children diversity positively associated with WDDS • Production than 70% of foods consumed by mothers were • More purchased diversity (Herfindahl index) not associated with HDDS • Crop of lower income groups Crop diversity (Herfindahl index) negatively associated with • HDDS in high-income HH Crop diversity (Herfindahl index) negatively associated with • HDDS in all income groups association between crop diversity (crop count and • Positive Simpson’s index) and HDDS when not controlling for other

HDDS than crop diversity

diversity positively associated with HDDS in • Production Indonesia and Malawi term of production diversity negatively associated • Square with HDDS access indicators have stronger association with • Market HDDS than farm production diversity association between production diversity and • Negative diversity of food purchased significant association between food crop diversity and • No HDDS in Kenya and Ethiopia diversity positively associated with HDDS and food • Crop consumption frequency in Poisson model estimation diversity not significantly but negatively associated • Crop with food consumption frequency in linear model estimation diversity positively associated with dietary • Livestock diversity income, market access, and availability of • Education, improved storage technologies had stronger influence on

Summary of original study findings

(6)

Mixed

Yes

No

No

Mixed

No

Yes

Yes

Yes

No

Mixed

(8) Conclusion considering all results reported

Yes

No

Yes

(7) Main conclusion of original study authors

Significantly positive association?

(continued on next page)

No

Mixed

Mixed

No

Yes

No

Mixed

(9) Only considering dietary diversity results

K.T. Sibhatu, M. Qaim

Food Policy 77 (2018) 1–18

(2)

Country

Kenya

Kenya

Indonesia, Kenya, Uganda

Bolivia

Ghana

Ethiopia

(1)

Original study reference

Ng’endo et al. (2016)

Romeo et al. (2016)

Sibhatu and Qaim (2016)

Vanek et al. (2016)

Ecker (2017)

Hirvonen and Hoddinott (2017)

Table 1 (continued)

9

3448 CH

834 HH (2005/ 06), 1219 HH (2012/13)

297 HH with 287 CH

1484 HH

1353 HH

30 HH, W

Sample size

(3)

CS

NR

CS

CS

CS

CS

Data type

(4)

7 d recall

30 d recall

24 h recall, HFIAS, CH anthrop.

7 d recall

7 d recall

7 d recall

Type of diet/ nutrition data

(5)

•

• • • • • •

• • •

• • • • • •

nutrient functional diversity, and combined crop and livestock count) not associated with HDDS, HFVS, WDDS, and WFCS Nutrient functional diversity positively associated with WDDS and HDDS Women with lower dietary diversity consume more from own production Nutritional function diversity (food group production) positively associated with HDDS Livestock ownership more strongly associated with HDDS than crop production diversity The association is attributed to income effect rather than more subsistence consumption Farm diversity (crop and livestock) positively associated with HDDS, calorie and micronutrients consumption in pooled sample Farm diversity (crop and livestock) not associated with calorie and micronutrients consumption in Kenya Farm diversity (crop and livestock) not associated with iron and vitamin A consumption in Uganda No association between crop nutritional function diversity (food groups grown) with HDDS, calorie and micronutrients consumption Market more important for dietary diversity than own production Negative association between crop diversity (number of crop species) and HFIAS Crop diversity not associated with HDDS Crop diversity not associated with infant and child feeding practices Crop diversity positively associated with child height-for-age Z-scores Production diversity (Simpson index) not associated with dietary diversity score in households with less than 2 ha of land Policies and programs that aim at improving dietary quality through farm production diversification may be ineffective among smallholder farmers

association between food crop diversity (excluding • No cotton) and HDDS association between animal species diversity and • Negative HDDS Farm diversity (species richness, Shannon index, Simpson’s • index, Shannon evenness, individual density, relative

Summary of original study findings

(6)

Mixed

No

No

Yes

Mixed

Mixed

Mixed

Yes

Yes

Mixed

(8) Conclusion considering all results reported

Yes

Yes

(7) Main conclusion of original study authors

Significantly positive association?

(continued on next page)

Mixed

No

No

Mixed

Yes

Mixed

(9) Only considering dietary diversity results

K.T. Sibhatu, M. Qaim

Food Policy 77 (2018) 1–18

10

Malawi

Kenya

Laos

Koppmair and Qaim (2017)

M’Kaibi et al. (2017)

Parvathi (2017)

Malawi

Jones (2017a)

Malawi

Country

Original study reference

Koppmair et al. (2017)

(2)

(1)

Table 1 (continued)

525 HH with CH

408 HH 519 CH 408 MO

408 HH 519 CH 408 MO

3000 HH (2012/ 13), 2556 HH (2010/11)

Sample size

(3)

CS (pan.)

CS

CS

7 d recall

24 h recall, CH anthrop.

24 h recall

24 h recall

7 d recall

NR

CS

Type of diet/ nutrition data

(5)

Data type

(4)

• •

• • • • • • • •

score and energy consumption when interaction term with distance to nearest population center is included in estimation Farm diversity (crop and livestock) positively associated with HDDS, WDDS, and CDDS No significant association between crop species count and HDDS and WDDS No significant association between crop species count and WDDS when market participation factors are controlled for Buying food, selling farm produce, and use of fertilizers more important for dietary diversity than diverse farm production Farm diversity (crop and livestock) positively associated with HDDS, WDDS, and CDDS Farm diversity (crop and livestock) not associated with CDDS (7 food groups) Farm diversity (crop and livestock) not associated with WDDS (9 food groups) Farm diversity (crop and livestock) not associated with CDDS and WDDS when market participation factors are controlled for No associations between farm diversity (crop and animals) and child anthropometrics Positive association between farm diversity and HDDS

diversity (species, varietal, and nutritional function • Crop richness) not significantly associated with dietary diversity

2010/11

production contributes 24% of total food consumption • Own in 2012/13 purchases contribute 65% of total food consumption • Market in 2012/13 species and varietal richness not significantly associated • Crop with protein, iron, and zinc consumption in 2010/11 nutritional functional richness not significantly • Crop associated with protein, vitamin A, and zinc consumptions in

HDDS, calorie, protein, and micronutrient consumption

integration more effective in reducing undernutrition • Market than production diversity crop diversity (species richness, crop varietal richness • Food and food groups production) positively associated with

in instrumental variable linear and Poisson estimation

controlling for market access, nutritional function • When diversity (food groups produced) not associated with CDDS

gender and age

function diversity (food groups produced) • Nutritional positively associated with CDDS function diversity (food groups produced) not • Nutritional associated with CDDS, when data disaggregated by children’s

Summary of original study findings

(6)

Yes

Yes

Mixed

Mixed

Mixed

Yes

Mixed

Mixed

(8) Conclusion considering all results reported

Mixed

Yes

(7) Main conclusion of original study authors

Significantly positive association?

(continued on next page)

Yes

Yes

Mixed

Mixed

Mixed

(9) Only considering dietary diversity results

K.T. Sibhatu, M. Qaim

Food Policy 77 (2018) 1–18

Food Policy 77 (2018) 1–18

No CH anthrop.

production diversity (crop and livestock count) • Farm positively associated with HFCS term of production diversity negatively associated • Square with HFCS crop-livestock farming system reduces household • Mixed dietary diversity association between farm diversity (crops and livestock) • No and height-for-age Z-score (stunting) of farmers who rely on own consumption more • Children stunted CS 335 CH Indonesia Purwestri et al. (2017)

556 HH (2012) 531 HH (2013)

Notes: HH, households. CH, children. IN, individuals. W, women. MO, mothers. RW, women of reproductive age CS, case study. NR, nationally representative. pan., panel data. h, hour. d, day. anthrop., anthropometric measurements. HDDS, household dietary diversity score. HFVS, household food variety score. HFCS, household food consumption score. CDDS, children dietary diversity score. CFVS, children food variety score. WDDS, women’s dietary diversity score. WFVS, women’s food variety score. IDDS, individual dietary diversity score. IFVS, individual food variety score. BMI, body mass index. HFIAS, household food insecurity access scale. a Albania (14 d), Indonesia (7 d), Malawi (7 d), Nepal (31 d), Nicaragua (15 d), Pakistan (14 d), Panama (15 d), Vietnam (365 d).

– No

(8) Conclusion considering all results reported (7) Main conclusion of original study authors Summary of original study findings Type of diet/ nutrition data Country Original study reference

Sample size

Data type

(6) (2) (1)

Table 1 (continued)

(3)

(4)

(5)

Significantly positive association?

(9) Only considering dietary diversity results

K.T. Sibhatu, M. Qaim

no positive associations are reported, and a “mixed” label if mixed evidence is reported. This labeling exercise is repeated three times, using different criteria, as explained in the following. In the first round of classification, we label according to the original study authors’ own main conclusion. If the original authors primarily highlight positive results in the abstract and conclusion section of their paper, we assign a “yes” label. If the original authors highlight in the abstract and conclusion section that they did not find any positive associations, we assign a “no” label, and if they highlight positive results in some situations but not in others, we assign a “mixed” label. In the second round of classification, we consider all results reported in the original studies. Based on the statistical significance levels of regression or correlation coefficients (5% level), we assign a “yes” label if the study reports only positive and significant associations. We assign a “mixed” label if the study presents mixed results, that is, positive and significant associations when using some indicators of production diversity and diets/nutrition, but not when using other indicators; positive and significant associations for some groups or household members (e.g., households as a whole, women, children of different age), but not for other groups; positive and significant associations in some geographical areas (e.g., villages, regions, countries), but not in other areas. We assign a “no” label if the study only reports statistically insignificant (or negative) associations. In the third round of classification, we use the same criteria as for the second round, but only consider results referring to a narrower set of dietary diversity indicators, namely dietary diversity scores (DDS, counting the number of different food groups consumed) and food variety scores (FVS, counting the number of different food items consumed). That is, we exclude results that refer to other diet and nutrition indicators, such as the hunger index, the consumption of specific food groups, the consumption of calories and specific nutrients, anthropometric measures, blood samples etc. While these other measures are commonly used to analyze specific aspects of nutrition, they are not necessarily indicators of dietary diversity. Including these other measures might therefore increase the number of studies with “mixed” or “no” labels, which we will test by comparing the frequencies of labels assigned in the three classifications. For the third round of classification, five studies that neither used DDS nor FVS had to be excluded (Table 1). 2.4. Procedure of meta-analysis Beyond the simple classification of studies, we use different approaches of meta-analysis to further scrutinize the evidence emerging from the different study estimates. First, we develop and estimate metaregressions to explain what study descriptors may influence whether or not production diversity is positively and significantly associated with dietary diversity. Second, we pool the results from the different studies to estimate the mean effect size of the association between production diversity and dietary diversity. Meta-analysis with quantitative tools is only meaningful when the original studies are conceptually similar and use comparable indicators. For the meta-analysis we therefore only include 29 studies that estimated the relationship between production diversity and dietary diversity with regression models, and that used either DDS or FVS as indicators of dietary diversity (see Table A2 in the Online Appendix for the 29 studies included). Many of the original studies report results from several regression models, for instance, referring to different geographic regions or using different model specifications. Each model estimate is considered as one observation, so the number of observations is larger than the number of studies included. Details of the meta-regressions and the effect size estimates are explained in the following. 2.5. Meta-regressions Meta-regressions are used to analyze what study descriptors may 11

Food Policy 77 (2018) 1–18

K.T. Sibhatu, M. Qaim

disaggregate the mean effect size by geographic region. For the regionally disaggregated analysis, the number of observations gets relatively small in some cases. As a robustness check, we also calculate mean Z-scores, allowing us to work with a larger sample because the different estimates for DDS and FVS can then be pooled. We calculate Z-scores by dividing each individual marginal effect estimate by the standard deviation of the dietary diversity indicator in the original study. Z-scores are designed to compare variables with different underlying distributions. This is also an advantage within the group of DDS estimates. The reason is that different types of DDS are used in the literature with slight variations in the number and composition of food groups considered (FAO, 2011). Hence, a robustness check with the use of Z-scores seems useful. The only disadvantage of Z-scores is that the interpretation is somewhat less intuitive than that of marginal effects. The Z-score describes how big the marginal effect of production diversity on dietary diversity is relative to mean dietary diversity observed in the sample and expressed in terms of standard deviations.

influence whether or not production diversity is significantly associated with dietary diversity. We follow an approach described by Card et al. (2010) in a different context and construct a dummy variable (D) that takes a value of one if a specific association i reported in original study j is positive and significant at the 5% level, and zero otherwise. Using D as dependent variable, we run the following meta-regression model.

Dij = α + β′Xij + εij where Xij is a vector of study descriptors, and εij is a random error term. As study descriptors we include variables such publication type (academic journal or different), geographic region from which the data were collected, dietary recall period used for data collection (24 h or different), and type of dietary diversity indicator employed (DDS or FVS). Studies also differ in terms of how production diversity is measured (Sibhatu and Qaim, 2018). Many count the number of food groups produced (FGP), others count the number of all crop and livestock species produced or calculate diversity indices. We control for the type of production diversity indicator used. Finally, we include the square root of samples size as explanatory variable to test whether larger samples are more likely to result in positive and significant estimates (Card et al., 2010). Given that the dependent variable (D) is a dummy, we use a probit specification for estimation. We start with a pooled probit model using 213 observations. A potential concern when using several observations from the same studies is that the error term may be heteroscedastic. To deal with this issue, we cluster standard errors at the study level (Klümper and Qaim, 2014). Another potential concern is that studies with household-level dietary data may potentially differ from studies with individual-level data, due to potential issues of intra-household food distribution. Therefore, in addition to the pooled-data model, we also estimate separate probit models for household-level (120 observations) and individual-level (93 observations) results.

3. Results Details of the methods and indicators used in the 45 original studies included in this review, as well as the main findings of these original studies, are summarized in Table 1. The studies included provide evidence from 26 countries: one from southeastern Europe (Albania), six from the Americas (Bolivia, Ecuador, Mexico, Nicaragua, Panama, and Peru), eight from Asia (Bangladesh, India, Indonesia, Laos, Nepal, Pakistan, the Philippines, and Vietnam), and eleven from Sub-Saharan Africa (Benin, Burkina Faso, Ethiopia, Ghana, Kenya, Malawi, Mali, Nigeria, Tanzania, Uganda, and Zambia). Six studies report results for two or more countries. The number of households or individuals surveyed in each study varied greatly, with sample sizes ranging from 30 to over 33,000 observations. Most of the available studies used crosssection data and analyzed associations rather than causality.

2.6. Analysis of mean effect sizes 3.1. Classification of studies The described meta-regressions with a dummy dependent variable are suitable to analyze the frequency of positive and significant associations and its correlates, but not to get a sense of the mean effect size. To calculate the mean effect size, we derive the marginal effect of production diversity on dietary diversity from the different regression models evaluated at original study sample mean values. This is a straightforward procedure for linear regression models but requires some calculation for non-linear models, taking into account the specific functional form. A few observations were lost, because details that would have been needed to calculate marginal effects were missing from the original studies. Furthermore, we exclude a few observations where production diversity was measured in terms of indices, because interpretation is different from more commonly used measures such as FGP or simple species counts (Sibhatu et al., 2015). The effect-size calculations are based on observations from 185 regression models (Table A2). We calculate mean marginal effects by averaging over all 185 observations. Two types of mean effects are reported: (i) unweighted means, (ii) weighted means, employing the inverse of the number of observations used from a study as the weight for each observation. This weighting procedure avoids that individual studies with several observations would possibly dominate the mean effect size calculations. Mean marginal effects are reported separately for DDS and FVS, as both indicators are not directly comparable. In addition, we test whether the choice of production diversity indicator influences the effect size estimates. We consider results for the pooled sample as well as separately for household-level and individual-level observations. In addition, we

As explained in the methods section, we use three types of study classification with different criteria for assigning “yes”, “no”, or “mixed” labels to address the question whether or not positive and significant associations between production diversity and dietary diversity/nutrition are reported. The three labels assigned to each study are shown in columns (7), (8), and (9) of Table 1. Summaries of the frequency of each label assigned are provided in Table 2. In the first round of classification, we use the original study authors’ own main conclusions. As can be seen in Table 2, in 31 of the 45 studies included (69%) the authors primarily highlight positive associations, in 10 studies (22%) no positive associations are highlighted, while in four studies (9%) mixed results are explicitly mentioned. This classification is based on our subjective reading of the original studies’ abstract and conclusion sections. A more objective classification is provided in the second labeling exercise, where all results reported in the original studies are taken into account (Table 2). In this alternative classification, we find that only five studies (11%) report consistently positive and significant associations between production diversity and diets/nutrition for all subsamples analyzed and with all indicators used. Twenty-nine studies (64%) report mixed results, meaning that positive and significant associations were observed only for some of the subsamples analyzed and some of the diet and nutrition indicators used. Eleven studies (24%) do not report any positive and significant associations at all. Interesting to observe is that there are also a few studies that highlight positive results in the abstract and conclusion sections, even though no statistically

12

Food Policy 77 (2018) 1–18

K.T. Sibhatu, M. Qaim

likely to be accepted in academic journals than insignificant results. Furthermore, the estimates in Table 4 suggest that studies building on 24-h dietary recall data (instead of longer recall periods) are more likely to result in positive and significant associations at the individual level. Finally, the individual-level estimates are more likely to be significant when referring to Sub-Saharan Africa than when referring to other geographic regions.

significant positive associations are reported in the studies (Table 1). The last column in Table 2 shows results of the classification exercise only considering results for dietary diversity, measured in terms of DDS or FVS. As explained above, we use this alternative classification to test whether the inclusion of other nutrition measures leads to a potential inflation in terms of the number of “mixed” or “no” labels. While slight changes in the frequency of the labels occur, the main conclusion is not affected by excluding other nutrition measures: more than 80% of the original studies report either mixed results or no positive and significant associations at all.

3.3. Mean effect sizes The study classifications and meta-regressions in the previous subsections have analyzed whether or not the different original study estimates are positive and significant, irrespective of the effect size. We now employ meta-analytical tools to estimate mean effect sizes, as described in the methods section. Mean marginal effects of the association estimates are shown in Table 5, separately for DDS and FVS. The estimated mean marginal effects are all positive, and most of them are statistically significant. In other words, a positive association between production diversity and dietary diversity is observed on average. While for the pooled and individual-level estimates, no difference between the weighted and unweighted mean effect sizes is observed, for the household-level estimates the weighted results are larger than the unweighted ones. Independent of weighting, the marginal effects of production diversity are larger when dietary diversity is measured in terms of FVS rather than DDS. This is plausible because the number of individual food items consumed is typically much larger than the number of different food groups consumed. Irrespective of the concrete indicator used, the marginal effects are very small. The mean marginal effect of 0.062 shown in Table 5 for DDS including all observations implies that increasing farm production diversity by one additional crop or livestock species would be associated with only a 0.062 increase in the number of food groups consumed. In other words, farms would have to produce 16 additional species in order to increase dietary diversity by one food group. It should also be mentioned that this simple extrapolation assumes a linear relationship. However, a few of the original studies showed that the relationship is actually not linear and that the magnitude of the effect decreases at higher levels of production diversity (Parvathi, 2017; Sibhatu et al., 2015; Pellegrini and Tasciotti, 2014). We conclude that further increasing production diversity is probably not the most effective way of improving dietary diversity in many situations. Table 6 further differentiates between the production diversity indicators used. For consistency, this analysis is confined to observations where dietary diversity was measured in terms of DDS. The results suggest that the marginal effects of production diversity on DDS are somewhat larger when production diversity is measured in terms of the

3.2. Meta-regressions We now discuss results from the meta-regressions that we use to analyze what study descriptors influence whether or not production diversity is significantly associated with dietary diversity. Summary statistics of the variables used in the meta-regressions are shown in Table 3. Of all the 213 regressions included, 51% show positive and significant associations between production diversity and dietary diversity. The other 49% show insignificant or negative associations. Interestingly, the proportion of positive and significant associations is somewhat lower for the individual-level than for the household-level estimates. Furthermore, Table 3 shows that 80% of the regression models were published in academic journals, and that over 90% of the results refer to either Sub-Saharan Africa or South and Southeast Asia. The results of the meta-regressions are shown in Table 4. In all three models (pooled, household level, individual level), the original study’s sample size increases the probability of observing a significantly positive association between farm production diversity and dietary diversity. This is plausible given that estimates with larger samples produce smaller standard errors. In the pooled model, and also in the model with only the household-level observations included, none of the estimates for the other study descriptors is significant. In other words, original study details such as the dietary recall period used, the choice of the dietary diversity indicator (DDS or FVS), or the choice of the production diversity indicator (FGP or otherwise) do not seem to systematically influence the statistical significance of the association estimates. In the individual-level meta-regression shown in the last column of Table 4, a few more of the study descriptors produce significant estimates. Publication of the original study in an academic journal increases the probability of a significant individual-level estimate. This is unrelated to sample size, because we control for sample size in the model. The significant effect may rather point at possible publication bias, meaning that results showing positive and significant associations between production diversity and dietary diversity are potentially more Table 2 Summary of systematic review results at individual study level.

Significantly positive association between production diversity and dietary diversity/nutrition?

Total number of original studies included in review Positive association between farm production diversity and diets/ nutrition (Yes) Mixed results (positive associations in some cases and insignificant or negative associations in others) No positive association reported (No)

Main conclusion of original study authors

Conclusion considering all results reported

Only considering dietary diversity results

45 studies (100%) 31 studies (68.9%) 4 studies (8.9%) 10 studies (22.2%)

45 studies (100%) 5 studies (11.1%) 29 studies (64.4%) 11 studies (24.4%)

40 studies (100%) 7 studies (17.5%) 23 studies (57.5%) 10 studies (25.0%)

Note: For conclusions on individual study results see Table 1.

13

Food Policy 77 (2018) 1–18

K.T. Sibhatu, M. Qaim

Table 3 Summary statistics of variables used in the meta-regressions

Association positive and significant (dummy) Published in academic journal (dummy) Dietary recall period was 24 h (dummy) Dietary diversity: DDS = 1; FVS = 0 (dummy) Production diversity: FGP = 1; otherwise = 0 (dummy) Sub-Saharan Africa (dummy) South and Southeast Asia (dummy) Square root of sample size Observations

All

Household level

Individual level

0.51 (0.50) 0.80 (0.40) 0.66 (0.48) 0.85 (0.36) 0.42 (0.49) 0.51 (0.50) 0.41 (0.49) 40.95 (25.45) 213

0.54 (0.50) 0.74 (0.44) 0.53 (0.50) 0.76 (0.42) 0.15 (0.36) 0.63 (0.48) 0.25 (0.43) 47.34 (29.96) 120

0.47 (0.50) 0.87 (0.34) 0.83 (0.38) 0.95 (0.20) 0.76 (0.43) 0.34 (0.48) 0.62 (0.49) 32.71 (14.49) 93

Notes: Mean values are shown with standard deviations in parentheses. For studies included, see Table A2 in the Online Appendix. DDS, dietary diversity score. FVS, food variety score. FGP, number of food groups produced.

Table 4 Factors influencing whether or not association between production diversity and dietary diversity is positive and significant (probit meta-regressions). Study descriptors

Published in academic journal (dummy) Dietary recall period was 24 h (dummy) Dietary diversity: DDS = 1; FVS = 0 (dummy) Production diversity: FGP = 1; otherwise = 0 (dummy) Sub-Saharan Africa (dummy) Square root of sample size Observations Pseudo R2

Table 5 Mean marginal effects of associations between production diversity and dietary diversity

Association positive and significant (dummy)

All

All

DDS

−0.012 (0.145) 0.038 (0.099) −0.114 (0.138) −0.059 (0.108) 0.010 (0.109) 0.006*** (0.001) 213 0.076

Household level −0.141 (0.146) −0.049 (0.107) −0.033 (0.161) −0.252 (0.157) 0.061 (0.118) 0.006*** (0.002) 120 0.140

Individual level Mean marginal effects (unweighted)

0.390** (0.194) 0.651*** (0.152) −0.327 (0.203) 0.007 (0.129) 0.381** (0.150) 0.013*** (0.002) 93 0.122

Mean marginal effects (weighted)a Observations

0.062

Household level FVS **

0.716

DDS **

Individual level

FVS

0.024

0.818

DDS ***

0.094

FVS ***

0.184**

(0.027) 0.062

(0.327) 0.716*

(0.055) 0.119*

(0.386) 2.195***

(0.013) 0.094**

(0.047) 0.184

(0.050) 160

(0.358) 25

(0.069) 75

(0.609) 21

(0.028) 85

(0.061) 4

Notes: Mean values are shown with standard errors in parenthesis. For studies included, see Table A2 in the Online Appendix. DDS, dietary diversity score. FVS, food variety score. a Each observation was weighted with the inverse of the number of observations used from the same study. * Significant at 10% level. ** Significant at 5% level. *** Significant at 1% level.

Notes: Marginal effects are shown with clustered standard errors in parentheses. For studies included, see Table A2 in the Online Appendix. DDS, dietary diversity score. FVS, food variety score. FGP, number of food groups produced. ** Significant at 5% level. *** Significant at 1% level.

0.115 for DDS observed in Sub-Saharan Africa implies that farms would have to produce 8.7 additional crop or livestock species in order to increase dietary diversity by one food group. As explained in the methods section, we also conducted the mean effect size analysis with Z-scores instead of marginal effects, resulting in larger subsamples because the estimates for DDS and FVS can then be combined. Results of the Z-score analysis are shown in Table A3 in the Online Appendix. These additional results confirm the main finding of positive but small mean effect sizes for the pooled sample and also for the different subsamples. The Z-score analysis also confirms that the mean effect size of production diversity on dietary diversity is larger in Sub-Saharan Africa than in other geographic regions.

number of food groups produced (FGP) than when it is measured in terms of a simple species count. This is plausible and expected especially in situations where subsistence consumption plays an important role (Sibhatu and Qaim, 2018). However, the marginal effects remain quite small also when using FGP, meaning that the main conclusion is not affected by the type of production diversity indicator used. Table 7 shows a disaggregation of the mean effect sizes by geographic region. For the FVS estimates, the number of observations in each of the regions is small, so that the results should be interpreted with caution. But for the DDS estimates, at least in Sub-Saharan Africa and Asia the numbers of observations are larger, so that the results and comparisons are more reliable. As can be seen, the largest mean marginal effect is observed in Africa, whereas in Asia the mean effect is very small. This geographic difference can probably be explained by lower market access and higher average subsistence orientation of farm households in Sub-Saharan Africa. Production diversity is expected to have a more direct effect on dietary diversity through the subsistence pathway (Sibhatu and Qaim, 2018). However, even the mean effect of

3.4. Focus on vulnerable groups In the analysis so far, we considered household-level and individuallevel estimates, yet without focusing on particular groups of household members. It is well known that children and women of reproductive age are particularly affected by nutritional deficiencies, so that a specific focus on these groups is warranted. To do this, we repeat our study

14

Food Policy 77 (2018) 1–18

K.T. Sibhatu, M. Qaim

production diversity was positively associated with child height-for-age Z-scores, but not with weight-for-age Z-scores, body-mass-index-for-age Z-scores, and mid-upper-arm circumference; some of the associations were even negative (Cleghorn, 2014). In Ghana, negative associations between crop diversity and child dietary diversity were reported (Argyropoulou, 2016). In Zambia, production diversity was positively associated with height-for-age of children aged 24–59 months, but not with several diet and nutrition indicators of children below and above 24 months of age (Kumar et al., 2015). Out of the 45 studies, 9 (20%) analyzed the relationship between production diversity and women’s diets and nutrition, but only two of them reported consistently positive associations (Zandar, 2014; Bellon et al., 2016). The other 7 studies reported mixed findings (Koppmair et al., 2017; Koppmair and Qaim, 2017; Ng’endo et al., 2016; Malapit et al., 2015; Sraboni et al., 2014; Keding et al., 2012; Powell, 2012). In Nepal, a negative association between production diversity and women’s body mass index was observed under certain conditions (Malapit et al., 2015). In Bangladesh, production diversity was also negatively associated with male and female body mass index (Sraboni et al., 2014). These results do not support the hypothesis that the role of production diversity is very different for the diets of vulnerable groups than it is for the diets of farm households in general.

Table 6 Mean marginal effects of associations between production diversity and dietary diversity scores by type of production diversity indicator used. All

Mean marginal effects (unweighted) Mean marginal effects (weighted)a Observations

Household level

Individual level

FGP

Species count

FGP

Species count

FGP

Species count

0.095*** (0.013)

0.020 (0.058)

0.076*** (0.024)

0.008 (0.073)

0.099*** (0.016)

0.069*** (0.017)

0.095** (0.029)

0.020 (0.106)

0.076 (0.042)

0.008 (0.131)

0.099 (0.040)

0.069** (0.024)

89

71

18

57

71

14

Notes: Mean values are shown with standard errors in parenthesis. FGP, number of food groups produced. a Each observation was weighted with the inverse of the number of observations used from the same study. ** Significant at 5% level. *** Significant at 1% level.

classification in terms of assigning “yes”, “no”, and “mixed” labels, but now only consider those results that refer specifically to children and women. As target group-specific analyses often use anthropometric data, we include all nutrition indicators for this exercise. Out of all 45 studies included in the review, 13 (29%) analyzed the association between production diversity and child anthropometrics (Table A4 in the Online Appendix). Of these 13 studies, no study reported consistently positive associations, seven reported mixed results (Vanek et al., 2016; Hirvonen and Hoddinott, 2017; Kumar et al., 2015; Malapit et al., 2015; Jones, 2015; Herforth, 2010; Dewey, 1981), and the remaining six did not find any statistically significant associations (M’Kaibi et al., 2017; Purwestri et al., 2017; Argyropoulou, 2016; Shively and Sununtnasuk, 2015; Cleghorn, 2014; Gonder, 2011). Vanek et al. (2016) reported a significantly positive association between production diversity and child height-for-age Z-scores, but production diversity was not significantly associated with the child feeding index or with household dietary diversity. In Tanzania, crop and vegetable

3.5. Role of markets Most of the reviewed studies did not only include production diversity as an explanatory variable in the dietary diversity/nutrition models, but also controlled for several other factors. Various studies showed that nutrition knowledge, education, gender, off-farm income, market access, use of certain farm and storage technologies, and other factors may influence dietary diversity directly and may also affect the association between production diversity and dietary diversity (Hirvonen and Hoddinott, 2017; Jones 2017a; Koppmair et al., 2017; Sibhatu et al., 2015; Snapp and Fisher, 2015; Jones et al., 2014; Pellegrini and Tasciotti, 2014; Keding et al., 2012). In many cases, the coefficient estimates for these other factors were larger than those for production diversity. The important role of markets to improve dietary quality was particularly highlighted in a number of studies (Ecker, 2017; Hirvonen and Hoddinott, 2017; Koppmair et al., 2017; Sibhatu and Qaim, 2016; Sibhatu et al., 2015; Snapp and Fisher, 2015; Cleghorn, 2014). This is not surprising. In spite of their subsistence orientation, smallholders often purchase more than half of all foods consumed from the market (Jones, 2017a; Sibhatu and Qaim, 2017; Bellon et al., 2016; Oyarzun et al., 2013). For higher-value nutritious foods the role of markets is often more important than for starchy staple foods (Sibhatu and Qaim, 2017; Luckett et al., 2015). The contribution of subsistence production to household diets decreases with closeness to markets. Good market access allows farmers to specialize on the production of the most profitable crops, including nonfood cash crops, leading to income gains and improved ability to purchase healthy diets. Indeed, several studies showed that dietary diversity was higher in market-oriented than in subsistence-oriented settings (Koppmair et al., 2017; Bellon et al., 2016; Sibhatu and Qaim, 2018; Jones et al., 2014; Sibhatu et al., 2015; Bhagowalia et al., 2012; Lambrecht, 2009). A few studies showed that production diversity is sometimes negatively associated with market orientation and household income (Sibhatu and Qaim, 2018; Pellegrini and Tasciotti, 2014). This could mean that commercialization of smallholder agriculture possibly leads to higher levels of specialization. However, stronger market orientation does not necessarily imply low on-farm diversity. Good infrastructure, efficient institutions, and well-functioning trading systems for various

Table 7 Mean marginal effects of associations between production diversity and dietary diversity by geographic region

Mean marginal effects (unweighted) Mean marginal effects (weighted)a Observations

Sub-Saharan Africa

Asia

DDS

DDS

FVS

DDS

FVS

0.006

1.655

0.108

0.565

0.115

FVS ***

0.308

**

Other regions

(0.018) 0.118***

(0.108) 0.523***

(0.052) 0.022

(1.004) 4.758***

(0.065) 0.451***

(1.32) 0.565

(0.017) 73

(0.156) 15

(0.105) 78

(1.126) 7

(0.094) 9

(1.320) 3

Notes: Mean values are shown with standard errors in parenthesis. DDS, dietary diversity score. FVS, food variety score. a Each observation was weighted with the inverse of the number of observations used from the same study. ** Significant at 5% level. *** Significant at 1% level.

15

Food Policy 77 (2018) 1–18

K.T. Sibhatu, M. Qaim

every farmer having to maximize diversity on his or her individual farm. Discussing a few limitations of the available studies may possibly encourage additional research. Available studies mostly rely on crosssection data, using simple statistical methods to identify associations. Hence, results should not be over-interpreted in a causal sense. Only two of the original studies used instrumental variable (IV) techniques to deal with endogeneity and allow more robust causal inference (Hirvonen and Hoddinott, 2017; Dillon et al., 2015). Both studies compared IV results with those from simpler regression models and found the IV effects to be somewhat larger. But even with the IV models, the effects of production diversity on dietary diversity remained fairly small. Additional research with panel data and improved statistical techniques could help to better understand causal mechanisms. Another limitation is related to seasonality, which was hardly addressed in most of the original studies included in this review. Analyzing seasonal differences in household diets and food sources would require data collection at higher frequency (Sibhatu and Qaim, 2017). Finally, the role of markets was often analyzed in a fairly crude way up till now. Agricultural markets and value chains are not only very heterogeneous but also rapidly transforming (Qaim, 2017). Further analysis is needed to better understand what types of markets and market transformations have what kind of nutritional effects in the developing country small-farm sector.

types of commodities can provide price incentives for farmers to produce a diverse range of products for market sales. Several studies showed that production diversity can be positively associated with the degree of commercialization and cash incomes under favorable market conditions (Jones, 2017a; Sibhatu and Qaim, 2018; Cleghorn, 2014; Pellegrini and Tasciotti, 2014; Torheim et al., 2004). Given the high importance of the income pathway, market-oriented farm diversification will have stronger positive effects on household diets than subsistence-oriented diversification. 4. Conclusion The systematic review has shown that farm production diversity is positively associated with household-level and individual-level dietary diversity and nutrition in some situations, but not in others. Similarly, a number of the original studies included in the review found positive associations when using certain indicators of diets and nutrition, but not when using other indicators. We argue that insignificant or negative results reported in the original studies should not be ignored, even when the same studies also report positive associations under certain conditions. Focusing only on the positive results may lead to biased policy conclusions. The meta-analysis has shown that the mean effect over all available estimates is positive and statistically significant, but very small in magnitude. Using dietary diversity scores (DDS) as the outcome indicator, the estimated mean marginal effect is 0.062, meaning that farms would have to produce 16 additional crop or livestock species in order to increase dietary diversity by one food group. The mean effect size is somewhat larger in Sub-Saharan Africa than in other geographic regions, but even in Africa farms would have to produce around 9 additional crop or livestock species in order to increase dietary diversity by one food group. Hence, there is little evidence to support the assumption that increasing farm production diversity is a highly effective strategy to improve smallholder diets and nutrition in most or all situations. This finding is in contrast to another recent review article that analyzed the associations in a qualitative way, but without considering effect sizes (Jones, 2017b). Increasing production diversity may have positive effects on smallholder nutrition in specific situations, but may have no effects or even negative effects in other situations. Negative effects can occur in particular when production diversity is already high; producing too many species can entail income losses through foregone gains from specialization. It may be argued that the positive nutrition effects of farm diversification are particularly large in subsistence-oriented settings. But subsistence-oriented farms are often quite diverse anyway. Further increasing production diversity in such situations may perpetuate subsistence and reduce market and development opportunities. Improving market access for subsistence farms seems to be a more promising development strategy. However, putting priority on improving market access should not be misunderstood as an argument against diverse production systems. When rural markets for diverse types of agricultural products are fostered through improvements in technology, infrastructure, storage, and logistics, production diversity and market orientation are not contradictory objectives. It should also be stressed that the diversity of food and agricultural systems is a question of scale. The studies reviewed here refer to diversity at the individual farm level. At higher scales (villages, districts, provinces etc.) sufficient diversity is important, because affordable access to diverse foods from the market requires that somebody produces these foods. If efficient markets for a wide range of products exist, food systems will become more diverse, even without

Competing interests The authors have declared that no competing interests exist. Acknowledgements This research was financially supported by the German Research Foundation (DFG), under grant number RTG 1666 (GlobalFood). The authors thank two anonymous reviewers and the editors of this journal for very useful comments on earlier versions of the manuscript. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.foodpol.2018.04.013. References Argyropoulou, S.E., 2016. The Association Between the Diversity of Crop Production and Nutritional Indicators of Rural Households in Northern Ghana. Master Thesis, Wageningen University, The Netherlands. Bellon, M.R., Ntandou-Bouzitou, G.D., Caracciolo, F., 2016. On-farm diversity and market participation are positively associated with dietary diversity of rural mothers in southern Benin, West Africa. PLoS ONE 11 (9), e0162535. Bhagowalia, P., Headey, D., Kadiyal, S., 2012. Agriculture, income, and nutrition linkages in India: Insights from a nationally representative survey. Discussion Paper 01195, IFPRI, Washington, DC. Burlingame, B., Dernini, S. (Eds.), 2012. Sustainable Diets and Biodiversity: Directions and Solutions for Policy, Research and Action. FAO and Bioversity International Rome. Card, D., Kluve, J., Weber, A., 2010. Active labour market policy evaluations: A metaanalysis. Econ. J. 120, F452–F477. Chamberlin, J., Jayne, T.S., Headey, D., 2014. Scarcity amidst abundance? Reassessing the potential for cropland expansion in Africa. Food Policy 48, 51–65. Chinnadurai, M., Karunakaran, K.R., Chandrasekaran, M., Balasubramanian, R., Umanath, M., 2016. Examining linkage between dietary pattern and crop diversification: an evidence from Tamil Nadu. Agric. Econ. Res. Rev. 29, 149–160. Cleghorn, C.L., 2014. Relationships Between Agrobiodiversity, Dietary Diversity and Nutritional Status in Tanzania. PhD Thesis, University of Leeds, UK.

16

Food Policy 77 (2018) 1–18

K.T. Sibhatu, M. Qaim

Burkina Faso. Cah. Agric. 25, 65001. Luckett, B.G., DeClerck, F.A.J., Fanzo, J., Mundorf, A.R., Rose, D., 2015. Application of the nutrition functional diversity indicator to assess food system contributions to dietary diversity and sustainable diets of Malawian households. Public Health Nutr. 18 (13), 2479–2487. M’Kaibi, F.K., Steyn, N.P., Ochola, S.A., Du Plessis, L., 2017. The relationship between agricultural biodiversity, dietary diversity, household food security, and stunting of children in rural Kenya. Food Sci. Nutr. 5, 243–254. M’Kaibi, F.K., Steyn, N.P., Ochola, S., Du Plessis, L., 2015. Effects of agricultural biodiversity and seasonal rain on dietary adequacy and household food security in rural areas of Kenya. BMC Public Health 15, 422. Malapit, H.J.L., Kadiyala, S., Quisumbing, A.R., Cunningham, K., Tyagi, P., 2015. Women’s empowerment mitigates the negative effects of low production diversity on maternal and child nutrition in Nepal. J. Dev. Stud. 51, 1097–1123. Moher, D., Liberati, A., Tetzlaff, J., Altman, D.G., 2009. Preferred reporting items for systematic reviews and meta-analyses: The PRISMA statement. Ann. Int. Med. 151, 264–269. Ng’endo, M., Bhagwat, S., Keding, G.B., 2016. Influence of seasonal on-farm diversity on dietary diversity: A case study of smallholder farming households in western Kenya. Ecol. Food Nutr. 55, 403–427. Ngendo, M., Keding, G.B., Bhagwat, S., Kehlenbeck, K., 2015. Variability of on-farm food plant diversity and its contribution to food security: A case study of smallholder farming households in Western Kenya. Agroecol. Sust. Food. 39, 1071–1103. Oyarzun, P.J., Borja, R.M., Sherwood, S., Parra, V., 2013. Making sense of agrobiodiversity, diet, and intensification of smallholder family farming in the highland Andes of Ecuador. Ecol. Food Nutr. 52, 515–541. Parvathi, P., 2017. Economic divorce mixed crop-livestock farming to increase dietary diversity? Evidence from smallholders in Lao PDR. Paper presented at the Third GlobalFood Symposium, April 28–29, University of Goettingen, Germany. Pellegrini, L., Tasciotti, L., 2014. Crop diversification, dietary diversity and agricultural income: Empirical evidence from eight developing countries. Can. J. Dev. Stud. 35, 211–227. Pingali, P., 2012. Green revolution: Impacts, limits, and the path ahead. Proc. Natl. Acad. Sci. U.S.A. 109, 2302–12308. Powell, B., 2012. Biodiversity and Human Nutrition in a Landscape Mosaic of Farms and Forests in the East Usambara Mountains. McGill University, Montreal, Canada, Tanzania. Powell, B., Thilsted, S.H., Ickowitz, A., Termote, C., Sunderland, T., Herforth, A., 2015. Improving diets with wild and cultivated biodiversity from across the landscape. Food Security 7, 535–554. Purwestri, R.C., Renz, L., Wirawan, N.N., Jati, I.R.A.P., Fahmi, I., Biesalski, H.K., 2017. Is agriculture connected with stunting in Indonesian children living in a rice surplus area? A case study in Demak Regency, Central Java. Food Security 9, 89–98. Qaim, M., 2017. Globalisation of agrifood systems and sustainable nutrition. Proc. Nutr. Soc. 76, 12–21. Rajendran, S., Afari-Sefa, V., Bekunda, M., Dominick, I., Lukumay, P.J., 2014. Does crop diversity contribute to dietary diversity? Evidence from integration of vegetables into maize farming systems in Tanzania. In: Paper presented at the 88th Annual Conference of the Agricultural Economics Society, AgroParisTech, April 9–11, Paris. Remans, R., Flynn, D.F.B., Mudiope, J., Mutuo, P.K., Nkhoma, P., Siriri, D., Sullivan, C., Palm, C.A., 2011. Assessing nutritional diversity of cropping systems in African villages. PLoS ONE 6, e21235. Romeo, A., Meerman, J., Demek, M., Scognamill, A., Asfaw, S., 2016. Linking farm diversification to household diet diversification: Evidence from a sample of Kenyan ultra-poor farmers. Food Security 8, 1069–1085. Rothstein, H.R., Hopewell, S., 2009. Grey literature. In: Cooper, H., Hegdes, L.V., Valentine, J.C. (Eds.), Handbook of Research Synthesis and Meta-Analysis. Russell Sage, New York, pp. 103–125. 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. Schmucker, C.M., Blümle, A., Schell, L.K., Schwarzer, G., Oeller, P., Cabrera, L., Von Elm, E., Briel, M., Meerpohl, J.J., on behalf of the OPEN consortium, 2017. Systematic review finds that study data not published in full text articles have unclear impact on meta-analyses results in medical research. PLoS ONE 12, e0176210. Shively, G., Sununtnasuk, C., 2015. Agricultural diversity and child stunting in Nepal. J. Dev. Stud. 51, 1078–1096. Sibhatu, K.T., Krishna, V.V., Qaim, M., 2015. Production diversity and dietary diversity in smallholder farm households. Proc. Natl. Acad. Sci. U.S.A. 112, 10657–10662. Sibhatu, K.T., Qaim, M., 2016. Farm production diversity and dietary quality: Linkages and measurement issues. GlobalFood Discussion Paper 80, University of Goettingen, Germany. Sibhatu, K.T., Qaim, M., 2017. Rural food security, subsistence agriculture, and seasonality. PLoS ONE 12, e0186406. Sibhatu, K.T., Qaim, M., 2018. Farm production diversity and dietary quality: Linkages and measurement issues. Food Security 10, 47–59. Snapp, S.S., Fisher, M., 2015. “Filling the maize basket” supports crop diversity and

DeClerck, F.A.J., Fanzo, J., Palm, C., Remans, R., 2011. Ecological approaches to human nutrition. Food Nutr. Bull. 32 (S1), S41–S50. Dewey, K.G., 1981. Nutritional consequences of the transformation from subsistence to commercial agriculture in Tabasco. Mexico. Hum. Ecol. 9 (2), 151–187. Dillon, A., McGee, K., Oseni, G., 2015. Agricultural production, dietary diversity and climate variability. J. Dev. Stud. 51 (8), 976–995. Ecker, O., 2017. Economic transformation and diets in rural Ghana: Does farm production diversity (still) matter? Paper presented at the Third GlobalFood Symposium, April 28–29. University of Goettingen, Germany. Ekesa, B.N., Walingo, M.K., Abukutsa-Onyango, M.O., 2008. Influence of agricultural biodiversity on dietary diversity of preschool children in Matungu Division, Western Kenya. Afr. J. Food Agric. Nutr. Dev. 8 (4), 390–404. Fanzo, J., Hunter, D., Borelli, T., Mattei, F. (Eds.), 2013. Diversifying Food and Diets: Using Agricultural Diversity to Improve Nutrition and Health. Routledge, London. Fanzo, J.C., 2017. Decisive decisions on production compared with market strategies to improve diets in rural Africa. J. Nutr. 147, 1–2. FAO, 2017. The State of Food Security and Nutrition in the World. FAO, Rome. FAO, 2014. The State of Food and Agriculture: Innovation in Family Farming. FAO, Rome. FAO, 2011. Guidelines for Measuring Household and Individual Dietary Diversity. FAO, Rome. Frelat, R., Lopez-Ridaura, S., Giller, K.E., Herrero, M., Douxchamps, S., Andersson, D.A., Erenstein, O., Henderson, B., Kassie, M., Paul, B.K., Rigolot, C., Ritzema, R.S., Rodriguez, D., van Asten, P.J., van Wijk, M.T., 2016. Drivers of household food availability in sub-Saharan Africa based on big data from small farms. Proc. Natl. Acad. Sci. U.S.A. 113, 458–463. Gödecke, T., Stein, A.J., Qaim, M., 2018. The global burden of chronic and hidden hunger: Trends and determinants. Global Food Security 17, 21–29. Gonder, C.E., 2011. Is Subsistence Enough? Examining the Impact of Household Farm Biodiversity on Dietary Diversity in Bukidnon, Philippines. Master Thesis, University of Illinois, Urbana-Champaign, IL. Headey, D., Ecker, O., 2013. Rethinking the measurement of food security: from first principles to best practice. Food Security 5, 327–343. Herforth, A., 2010. Promotion of Traditional African Vegetables in Kenya and Tanzania: A Case Study of an Intervention Representing Emerging Imperatives in Global Nutrition. PhD Thesis, Cornell University, Ithaca, NY. Hirvonen, K., Hoddinott, J., 2017. Agricultural production and children’s diets: evidence from rural Ethiopia. Agric. Econ. 48, 469–480. IFPRI, 2017. Global Nutrition Report. IFPRI, Washington, DC. Jones, A.D., 2015. The production diversity of subsistence farms in the Bolivian Andes is associated with the quality of child feeding practices as measured by a validated summary feeding index. Public Health Nutr. 18, 329–342. Jones, A.D., 2017a. On-farm crop species richness is associated with household diet diversity and quality in subsistence- and market-oriented farming households in Malawi. J. Nutr. 147, 86–96. Jones, A.D., 2017b. Critical review of the emerging research evidence on agricultural biodiversity, diet diversity, and nutritional status in low- and middle-income countries. Nutr. Rev. 75, 769–782. Jones, A.D., Shrinivas, A., Bezner-Kerr, R., 2014. Farm production diversity is associated with greater household dietary diversity in Malawi: Findings from nationally representative data. Food Policy 46, 1–12. Kavitha, K., Soumitra, P., Padmaja, R., 2016. Understanding the linkages between crop diversity and household dietary diversity in the semi-arid regions of India. Agric. Econ. Res. Rev. 29, 129–137. Keding, G.B., Msuya, J.M., Maass, B.L., Krawinkel, M.B., 2012. Relating dietary diversity and food variety scores to vegetable production and socio-economic status of women in rural Tanzania. Food Security 4, 129–140. Khoury, C.K., Khoury, C.K., Bjorkman, A.D., Dempewolf, H., Ramirez-Villegas, J., Guarino, L., Jarvis, A., Rieseberg, L.H., Struik, P.C., 2014. Increasing homogeneity in global food supplies and the implications for food security. Proc. Natl. Acad. Sci. U.S.A. 111, 4001–4006. Klümper, W., Qaim, M., 2014. A meta-analysis of the impacts of genetically modified crops. PLoS ONE 9, e111629. Koppmair, S., Kassie, M., Qaim, M., 2017. Farm production, market access and dietary diversity in Malawi. Public Health Nutr. 20, 325–335. Koppmair, S., Qaim, M., 2017. Farm production diversity and individual-level dietary diversity. Public Health Nutr. 20, 2070–2072. Kumar, N., Harris, J., Rawat, R., 2015. If they grow it, will they eat and grow? Evidence from Zambia on agricultural diversity and child undernutrition. J. Dev. Stud. 51, 1060–1077. Lambrecht, I., 2009. Linking Agro-Diversity and Nutrition: A Case Study in Two Millennium Villages. Master Thesis, KU Leuven, Belgium. Linderhof, V., Linderhof, V., Powell, J., Vignes, R., Ruben, R., 2016. The influence of household farming systems on dietary diversity and caloric intake: The case of Uganda. Paper presented at the Fifth Conference of the African Association of Agricultural Economists, September 23–26, Addis Ababa, Ethiopia. Lourme-Ruiz, A., Dury, S., Martin-Prével, Y., 2016. Do you eat what you sow? Linkages between farm production diversity, agricultural income and dietary diversity in

17

Food Policy 77 (2018) 1–18

K.T. Sibhatu, M. Qaim

determinants. Eur. J. Clin. Nutr. 58, 594–604. Vanek, S.J., Jones, A.D., Drinkwaterm, L.E., 2016. Coupling of soil regeneration, food security, and nutrition outcomes in Andean subsistence agroecosystems. Food Security 8, 727–742. Zandar, U.C., 2014. Agrobiodiversity, Cultural Factors and their Impact on Food and Nutrition Security: A Case-Study in the South-East Region of the Peruvian Andes. VVB Lauftersweiler Verlag, Giessen, Germany.

quality of household diet in Malawi. Food Security 7, 83–96. Sraboni, E., Malapit, H.J., Quisumbing, A.R., Ahmed, A.U., 2014. Women’s empowerment in agriculture: What role for food security in Bangladesh? World Dev. 61, 11–52. Stanley, T.D., Doucouliagos, H., Giles, M., Heckemeyer, J.H., Johnston, R.J., Laroche, P., Nelson, J.P., Paldam, M., Poot, J., Pugh, G., Rosenberger, R.S., Rost, K., 2013. Metaanalysis of economics research reporting guidelines. J. Econ. Surv. 27, 390–394. Torheim, L.E., Ouattara, F., Diarra, M.M., Thiam, F.D., Barikmo, I., Hatløy, A., Oshaug, A., 2004. Nutrient adequacy and dietary diversity in rural Mali: Association and

18