1. The ARENA research agenda on diet quality from 6 to 24 months: Evidence from 67,241 children in 39 countries
Samira Choudhury (PhD candidate, University of Adelaide),
Derek Headey (IFPRI) and William A. Masters (Tufts University)
Exploratory results for discussion with ARENA collaborators, 14 September 2017

2. DHS provides a under-exploited window into stylized facts about children’s diet quality
DHS data from Phase 5 & 6 include 24-hr recall whether child was fed each item on a standard food list
Dataset includes 47 surveys in 39 countries, totaling 67,241 children surveyed between 2006 and 2013
What can we learn when we zoom out to this scale of analysis?
Widest possible range of circumstances
Largest possible sample sizes
Our focus is on modifiable factors (health access, market access, breastfeeding and gender roles, and agricultural productivity)
Controlling for wealth, schooling, agroecological and demographic factors
With attention to functional forms, heterogeneity and interactions

3. DHS data of interest: A wide range of key variables
Mean
Std.Dev.
Min.
Max.
Child dietary diversity score (12 food groups) 6-23 months
3.24
2.34 12
Child dietary diversity score (7 food groups) 6-23 months
2.83
1.82 7
Child minimum dietary diversity (≥ 4 food groups) 6-23 months
0.31
0.46 1
Outcome of interest is CDD
Household wealth (quintile)
2.95
1.32 1 5
Maternal some primary schooling (1-6 years)
0.27
0.44
Maternal some secondary schooling (7-9 years)
0.14
0.34
Maternal some tertiary schooling (10-plus years)
0.23
0.42
Paternal some primary schooling (1-6 years)
0.26
Paternal some secondary schooling (7-9 years)
0.13
Paternal some tertiary schooling (10-plus years)
0.31
0.46
We will control for wealth and schooling
Father’s occupation is non-agricultural
0.59
0.49 1
Health access (antenatal care, medical birth, vaccinations)
0.27
0.44
Women can decide on own healthcare
0.61
Child was breastfed immediately
0.66
0.47
Health system access and some gender roles may be modifiable
Number of children aged 0-2
0.42
0.49 1
Number of children aged 3-5
0.14
0.35
Child is male
0.51
0.5
Child’s age in months
14.01
5.11 23
All results control for family size, age and sex

4. GIS data of interest: Children are not randomly located…
Mean
Std.Dev.
Min.
Max.
Child dietary diversity score (12 food groups) 6-23 months
3.24
2.34 12
Child dietary diversity score (7 food groups) 6-23 months
2.83
1.82 7
Child minimum dietary diversity (≥ 4 food groups) 6-23 months
0.31
0.46 1
Outcome of interest is CDD
Remote location (travel time>=1 hr to towns of >20k pop)
0.35
0.48 1
Night lights intensity index by cluster, middle tercile
0.11
0.32
Night lights intensity index by cluster, upper tercile
0.47
Population density, middle tercile
0.39
0.49
Population density, upper tercile
0.3
0.46
Distance to coastline, middle tercile
Distance to coastline, upper tercile
0.33
Dimensions of market access
Medium length of growing period in days
0.37
0.48 1
High length of growing period in days
0.3
0.46
Mean rainfall (mm) over , middle tercile
Mean rainfall (mm) over , upper tercile
0.32
0.47
Mean temperature (Celsius) over , middle tercile
0.31
Mean temperature (Celsius) over , upper tercile
0.4
0.49
Altitude (meters) by cluster, middle tercile
0.38
Altitude (meters) by cluster, upper tercile
Agro-climatic factors

5. The 6-23 month window is not just when stunting occurs, but also onset of micronutrient deficiencies
Age focus from
6 to 24 months
Notes: Total sample is 67,244 children in 39 countries, from DHS Phase 5 & 6 data

6. Diet quality, as measured by food group diversity, varies widely among surveyed children
WHO threshold:
4 of 7 food groups
Notes: Total sample is 67,244 children in 39 countries, from DHS Phase 5 & 6 data

7. Diet quality, as measured by food group diversity, varies widely among surveyed children
A 12-group classification has more variance
2 starchy staples (cereals; roots & tubers);
4 F&V groups (vit.A-rich veg., vit.A-rich fruits,
dark green leafy veg, other fruits & vegetables);
5 ASF groups (meat, organ meat, fish, eggs, & dairy);
1 legumes and nuts
Notes: Total sample is 67,244 children in 39 countries, from DHS Phase 5 & 6 data

8. Wealthier households reveal a clear preference for greater child diet diversity at all levels of wealth
Notes: Total sample is 67,244 children in 39 countries, from DHS Phase 5 & 6 data

9. CDD scores begin to differentiate around 6 months, and reach plateau (set by family diet?) around 18 months
6 months
18 months
Notes: Total sample is 67,244 children in 39 countries, from DHS Phase 5 & 6 data

10. Some determinants of CDD are clearly non-linear
Notes: Total sample is 67,244 children in 39 countries, from DHS Phase 5 & 6 data

11. Our aim is to investigate modifiable factors, like market access
Notes: Total sample is 67,244 children in 39 countries, from DHS Phase 5 & 6 data

12. Our aim is to investigate modifiable factors, like market access and electrification
Notes: Total sample is 67,244 children in 39 countries, from DHS Phase 5 & 6 data

13. Our aim is to investigate modifiable factors, while taking account of agroclimatic constraints
Notes: Total sample is 67,244 children in 39 countries, from DHS Phase 5 & 6 data

14. Our aim is to investigate modifiable factors, including heterogeneity to inform targeting
Exploratory regressions
Control for wealth, schooling, agroecological and demographic factors
Focus on health access, market access, breastfeeding and gender roles, and agricultural productivity
Results shown are from horserace-type exploratory regressions, to identify stylized facts that point to potentially causal relationships for further study
All errors are clustered at level of DHS enumeration area (villages)
All results control for child age, country and year fixed effects
Explore sensitive to child age range (as diets diversify more at older ages)

15. Correlates of child diet diversity across 47 DHS surveys in 39 countries, for 67,241 children from 2006 to 2013
  Child age:
6-23 mo.
(1)
Household wealth (quintile)
0.167***
Maternal some primary schooling (1-6 years)
0.142***
Maternal some secondary schooling (7-9 years)
0.317***
Maternal some tertiary schooling (10-plus years)
0.516***
Paternal some primary schooling (1-6 years)
0.117***
Paternal some secondary schooling (7-9 years)
0.164***
Paternal some tertiary schooling (10-plus years)
0.215***
Father’s occupation is non-agricultural
-0.037
Child was breastfed immediately
0.077***
Health access
0.281***
Mother can decide on own healthcare
-0.022
6-11 mo.
12-23 mo.
(2)
(3)
0.126***
0.193***
0.106***
0.179***
0.203***
0.400***
0.377***
0.617***
0.086**
0.132***
0.125**
0.183***
0.158***
0.248***
-0.097**
0.03
0.104***
0.288***
0.219***
0.011
-0.04
12-23 mo.
(4)
0.169***
0.208***
0.360***
0.504***
0.203***
0.205**
0.316***
-0.022
0.133***
0.162**
-0.170***
0.047**
Household wealth is important, about same effect size as maternal primary but smaller than maternal secondary schooling.
Both maternal & paternal schooling effects larger at than 6-11 mo.
Nonfarm occupation is negatively linked to CDD but only at 6-11 mo.
Immediate breastfeeding is positively linked to CDD later, at mo.
Health access has a large effect size
Mother’s decision-making does not have additional significance beyond schooling and other variables
except at higher wealth levels
(4th and 5th quintiles of asset index)
Wealth * Mother can decide on own healthcare
Observations
67,241
24,182
43,059
43,059
Notes: All results include controls for agroclimatic and demographic variables (not shown).
Column (4) results include non-significant interaction terms for wealth with each schooling variable, occupation, health access, breastfeeding and remote location

16. Correlates of child diet diversity across 47 DHS surveys in 39 countries, for 67,241 children from 2006 to 2013
  Child age:
6-23 mo.
(1)
Household wealth (quintile)
0.167***
Night lights intensity index by cluster, upper tercile
0.141***
Night lights intensity index by cluster, med. tercile
0.074**
Remote location (>1 hour to towns of >20k pop)
-0.047*
Population density, upper tercile
0.110***
Population density, middle tercile
0.066**
Distance to coastline, middle tercile
0.079**
Distance to coastline, upper tercile
-0.015
Mean rainfall, middle tercile
0.157***
Mean rainfall, upper tercile
0.238***
Mean temperature, middle tercile
-0.195***
Mean temperature, upper tercile
-0.194***
Altitude by cluster, middle tercile
0.072**
Altitude by cluster, upper tercile
0.052
Growing period in days, medium length
0.078
Growing period in days, high length
0.019
6-11 mo.
12-23 mo.
(2)
(3)
0.126***
0.193***
0.102*
0.162***
0.065
0.077*
-0.004
-0.069**
0.072
0.131***
0.058
0.071**
-0.04
0.144***
0.014
-0.027
0.117*
0.174***
0.253***
0.232***
-0.186***
-0.206***
-0.264***
-0.166***
0.084*
0.063
0.033
0.047
0.029
0.109*
-0.003
“Market access” is linked to CDD only >12 mo.
Climatic conditions are linked to CDD at both 6-11 months and also months
Observations
67,241
24,182
43,059

17. Conclusions: What can we learn when we zoom out to all available DHS data?
Robust wealth effect indicates consistent demand for CDD throughout the range of DHS populations
We find heterogeneity only for women’s say over health care, which raises CDD only for the top two quintiles of wealth
Health access (antenatal care, facility birth, vaccinations) is linked to higher CDD for both 6-11 and months
Higher temperatures also linked to lower CDD for both age groups
Market access (travel time to town, night lights, population density) is linked to higher CDD only at older ages (12-23 months)
Next steps to be discussed today!
Look at heterogeneity over regions? Decompositions/predictions?