Food Policy Child dietary quality


[PDF]Food Policy Child dietary quality - Rackcdn.comhttps://158fc6497e5a64559e1f-d14ef12e680aa00597bdffb57368cf92.ssl.cf2.rackcdn...

0 downloads 179 Views 14MB Size

Food Policy 61( 2016) 185- 197

Contents lists available at ScienceDirect

s ".,'°' ' __

FOOD POLICY

Food ELSEVIER

Child

journal

quality in

dietary

homepage:

Policy

www. elsevier.

Nepal:

rural

mom

com/ locate/ foodpol

Effectiveness

community- level

of a

®

CrossMark

development intervention F. Darrouzet- Nardi ' ,

Amelia

C. Miller , Neena Joshi ,

Laurie

Mahendra Lohani d,

Shubh Mahato ,

Beatrice L. Rogers b Allegheny College, Meadville, PA, USA

b Tufts University, Boston, MA, USA Heifer International Nepal in Katmandu, Nepal

dHeifer International in Little Rock, AR, USA

ARTICLE

Article

INFO

ABSTRACT

history:

Received

18 June

Received

in

form 22

revised

Accepted 31

Nutrition- sensitive agricultural interventions have the potential to improve child dietary quality in rural households, as evidenced by a growing body of work which concurrently measures agricultural and nutrition indicators. Our objective was to investigate whether children in rural farming communities of Nepal participating in a community- level, nutrition- sensitive development intervention had improved dietary quality compared with children living in non- participating matched rural communities. Six rural communities of Nepal where livelihoods were predominantly agricultural were selected to participate in the phased implementation of a community- level development intervention by Heifer International. Households and children in each community were surveyed at baseline, and follow- up surveys were implemented every six months for twenty- four months. Detailed data on food consumption for children older than 6 months of age were collected using a 24 h recall for 17 foods and food groups; parents

2015

2016

March

2016

March

Keywords:

Child

dietary diversity

Smallholder farms Impact

evaluation

Nepal Animal

responded

foods

source

Community based interventions

for children.

of covariance

model

A difference-

in- differences

were used to analyze

model

the resulting

with

household

fixed- effects

panel data. Results

indicated

and

an analysis

that the impacts

of the intervention varied by agro- ecological region and by season. In the Hills region, which is poorer on average and more conducive to livestock production, children who had received the intervention for two years were 2. 2 times as likely to have consumed as likely to have achieved source

foods

minimum

as children

dietary diversity

who received

food from an additional

food group, 1. 27 times

and 1. 38 times as likely to have consumed

the intervention

animal

for only one year. In the Terai region, which is

more conducive to crop production, there was no significant change in dietary quality attributable to the intervention. of communityactivities opment

These results provide level

development

were disparate activities

evidence

activities.

across communities

for particular

locations

that particularly

Given

vulnerable

that the effects

families

within the same country, we conclude

is necessary

and

ment,

nutrition

labor productivity,

Thomas,

2007;

Black

is

and

et

essential

and

al.,

development devel-

Ltd. All rights

reserved.

quality- often measured by dietary diversity- is a direct determi-

motivation

nant of nutritional Adequate

level

that tailoring

for success. 2016 Elsevier

Introduction

can take advantage

of community-

for

child

growth,

health throughout

2013;

Alderman

et

develop-

life ( Strauss

al.,

2006).

and

Dietary

status, especially for key micronutrients

vitamin A and iron, ( Arimond

et al., 2010; Arimond

such as

and Ruel,

2004), which are found predominantly in animal sourced foods. Although somewhat lost in myriad other goals, ending hunger

and improving agricultural productivity are targets of the newly launched This is

not

paper

represents

currently

publication

at

original

work

being considered by Policy, it will not be

Food

that

has

another published

not

been

journal,

published

that

and

elsewhere

in

the

previously, if

that

accepted

for

form.

Each

same

author has seen and approved of the contents of the manuscript.

Corresponding Meadville, E-

mail

author

at:

Allegheny

College,

520

N.

Main

Street,

PA, USA. address:

adarrouzetnardi@allegheny. edu(

http:// dx. doi. org/ 10. 1016/ j.foodpo1. 2016. 03. 007 0306- 9192/© 2016

Elsevier

Ltd.

All

rights

reserved.

A. F. Darrouzet-

Nardi).

Box

E,

Sustainable

Development

Goals ( SDGs),

listed

together

under Goal 2: Zero Hunger( UN, 2015). Ensuring high- quality diets for rural children number

in low- income countries

of strategies,

including

can be addressed

by stimulating

economic

by a

growth,

implementing nutrition- specific programming, or implementing nutrition- sensitive programming ( Headey, 2013; Black et al., 2013;

Ruel and Alderman, 2013). The intervention

evaluated

here

186

is

A.F. Darrouzet- Nardi et al./ Food Policy 61( 2016) 185- 197

a

nutrition- sensitive

program,

determinants

immediate

the

does

which of

directly

not

nutritional

address

Ruel

status (

and

and economic development. Heifer activities focus on the distribu-

tion of livestock

and training

to rural women' s groups

with an

Alderman, 2013). Being forced to pick between any one of these three major strategies is— at best— a false choice, and the effective-

emphasis on income generation. These activities

occur within the

context

of social

strategy will likely depend on specific local context. Regardless, it is important to assess whether and how different

specifically citizen empowerment, values training, social mobilization, group savings and micro- lending, and enterprise.

ness of

types

one

any

of

programs

vulnerable

areas

in April

earthquake

Aguayo

people(

2015

dietary

child

this

level,

and

paper

5. 6

million

the

over

effectiveness

development

improving long- term. To

and

recovery

status

the

dietary

child

a suitable

nutritional

assesses

nutrition- sensitive

improving

on

devastating

with

quality

this

end,

a

about

severely affecting

be important for supporting

will

diets, especially in

child

experienced

2015). Leveraging existing non- governmental diverse programming experience in Nepal

al.,

et

as

of

NGOs)

organizations(

for improving Nepal which

work

such

of

a

intervention

community-

in

a using quality study design which

community- level randomized average treatment effect on the treated( ATET).

Nepal

rural

matched-

pair

estimates

the

of a strong

focus

on the development

capital,

The nutritional effects of Heifer activities in agriculture and in

women' s empowerment

have been formally

evaluated

recently;

the promotion of livestock ownership by Heifer programs was an important mechanism leading to improvements in anthropometric measurements Nepal (

Miller

of children

in Rwanda ( Rawlins

et al., 2014)

and

2014).

However, impacts on child dietary quality have not been yet established, and nutrition is not directly et

al.,

a focus of Heifer programming. To this end, we evaluated the effect of a holistic community- level nutrition- sensitive intervention child dietary diversity and animal source food consumption rural

on in

Nepal.

Potential pathways of impact Child diets in Nepal and nutrition- sensitive

interventions There are four broad potential pathways through which agricul-

Inadequate

DHS)

Survey (

6- 24

months

dren

were

a problem

health

of public

Findings from the 2011

Demographic

in Nepal

for

in Nepal.

significance

Health

dietary quality is

child

indicated

that,

children

and aged

in the 24- h preceding the survey, only 28% of chilgiven foods from at least 4 different food groups, 46%

of children

consumed

consumed

food

a

a

rich

food

in

rich

A,

24% of

children

ICF International,

2012).

vitamin

in iron( MOHP

and

and

concerns

for

exist

predominant

the

population

older

is

82%),

rural (

age

groups

fluctuations

so

across

dietary

and

ICF

seasons may exacerbate development ( MOHP and

Bank,

2012;

Maleta

et al.,

Nutrition- sensitive may

the

ameliorate

tions

on

nutrition

improvements

noted

in

2015) found that

et

child

the

impact

World

diversity

smallholder

2013).

fluctua-

However,

of

for

evidence

animal

a

interven-

of agriculture

and

focus

agricultural of seasonal

low quality

overall

2012),

al.,

an

effects

Alderman,

and

of

an

with

food

source

recently Smale farmer adoption of hybrid though

et

al.

maize

positively associated with dietary diversity in Zambia. The impacts of community- level, nutrition- sensitive activities in farming regions on improving child diets are difficult to measure and was

inconsistently a

specific

Bezner

nutrition

Kerr

A

2009).

et

maternal

some

empowerment,

uses

to

Bhutta

agriculture

were

though

of

and

infant

training

as

300

be

production,

aspect

of the

improved

diets

were

with

chil-

young

et

on

the

or

nutrition

by

Malapit

al. (

2015),

child

indicators

feeding of

rela-

and who and

women' s

or

through

Second,

increased

increased

incomes

could dietary

consumption

incomes.

Third,

of the

child

nutrition

outcomes.

The

latter

mechanism

relates

directly to the design of the Heifer intervention, which focused on developing and facilitating women' s self- help groups. Therefore, for those women

who had been exposed

to the intervention

that they would be better

able— whether through increased knowledge, increased household resources,

improve

or

increased

power

over

household

their children' s diets, when compared

had not been exposed

to the intervention

resources —

to

with women who

for as long a period of

time. There is empirical evidence in Nepal that this is a plausible

mechanism,

as evidenced

by Malapit et al. ( 2015), who estimated

that women' s empowerment

was strongly

associated

with mater-

nal and child nutrition there.

Under this theoretical framework displayed below in Fig. 1, increasing women' s empowerment with a community- level development

intervention

could

result

in

an

outward

shift

of the

demand curve for children' s food, perhaps especially during the seasons.

For our current

empirical

purposes,

we used a

model to estimate

these

effects. Women' s empowerment was not measured for the present

study, and so it is not possible to test this mechanism

directly. We

focused on dietary quality as opposed to nutritional status for four reasons. First, the intervention address

child nutritional

was not specifically

status. The animal

designed to

husbandry

training

and goat donation aspects of the intervention may have direct relationships with animal source food consumption,

but not necessar-

ily nutritional status. Second, the short- term effects of intervention

activities may be more visible in diets than in anthropometrics. Third, there are other key factors such as sanitation which may

not.

interfere with the diet- nutrition

henceforth Heifer)

alleviation,

gender- related

due to the animal

intervention.

through

reduced- form difference- in- differences

intervention

for poverty

own

training

could

production

hungry

al.,

projects

tools

intakes

study. First, household

agricultural

2004;

is a globally active NGO in thirty- two countries. The organization the introduction of livestock and related animal husbandry International(

over

for the present

with improved

et

2015). The mediating

Malapit

and

al.,

improved

allocation

children' s

by

associated

others

et al.,

time

or

et

require

Satzinger

recently in Ghana

in Nepal

and

2008;

al.,

agriculture

and

Berti

succeed (

et

Carletto

empowerment

International

Heifer with

goal

or

2011;

indicators

nutrition

The Heifer

interventions typically

demonstrated

as

Quisumbing ( 2015) found that

focus

such

women' s status(

between

complex,

and

key link between

of women' s

tionships are

al.,

potential

dren' s diets is effects

observed,

and

able

growth

2012;

important

for longer periods of time, we expected dietary

and

dietary

child

Girard

consumption (

insults to

production,

improvement of women' s social standing could lead to more desir-

in food availability

International,

interventions

health

review of studies

systematic

majority

of own

is

2003).

in food availability ( Ruel

tions

the

and

consumption

issues. Three of these potential mechanisms stand out as especially

of

Agriculture

as well.

in Nepal,

76%)

occupation (

agriculture,

husbandry

study focuses on a wider age range of children than typically measured for the DHS( 6 months to 8 years at baseline), but similar

the

as synthesized by Carletto et al. ( 2015): food prices, income from

increase

This

dietary

tural activities can improve child diets and nutritional outcomes,

citizen

empowerment,

relationship.

Further work would

be needed to make connections between dietary quality and nutritional status in this setting. Lastly, dietary diversity was found to be associated with child nutritional

status, independent

of associ-

ations with income and socioeconomic status in a wide range of settings(

Arimond

and

Ruel, 2004).

Darrouzet- Nardi

A. F.

Household

in

participation

community-

Policy

2016) 185-

61(

197

187

Increased consumption of

intervention \>

level

Food

et al./

household

production

for children

1

V Improved

status

of women

in

Child dietary quality

households

participating

A A

V

Increased income from farm ire Child

activities.

nutritional

status

Fig. 1. Conceptual framework.

At the start of the intervention for each community, local lead-

Methods

ers were invited to serve on an advisory panel and as liaisons to the Study design

population

and participants

about

the

project

activities.

Data

collection

was

per-

formed by a local field research NGO, the Nepal Technical Assistance As previously described in Miller conducted of

according

Helsinki,

and

all

Parents

Protections,

study. was

gave

to

NHRC.

At

daily reviews of the data collection, which allowed rapid identification and correction of errors. Enumerators were trained at the

Reference

board in

review

Health

of

a

informed

respondents,

third

visit,

For

visit.

consent

by

each

was

beginning of the project with 6 days of orientation

non-

Human

and

obtained

participants

were

answer

their

children.

ment

consent.

any or all questions Coded, confidential files

were

control

respon-

attempts

guidelines

the

that

measurement

of

households (

docu-

to

maintained

monitor

Heifer International

for participation

in the inter-

were identified

as participating

in

or

not before the baseline survey was implemented. All participating

they or

to

rates exceeded 90% of households

each study site. Households

of

visit,

study

activities

In each community,

to include all households

vention, and participation

house-

each

refresher

Households were identified in cluster samples within each VDC,

parent

to

and

inter- observer reliability.

and invited to participate.

reminded

decline

or

quality

consent

prior

discontinue

ongoing

and maintain

written

literate

to the project,

field pilot testing in three villages not included in the study sites,

for Human and

party, according to the

from the study,

withdraw

decline to

NHRC

Office

Protection

of

Department

each

verbal

visit, witnessed

could

S.

U.

parent

prior

dents, informed

the

Council(

formance and activities of the Field Enumerators, and conducted

patients

the

as

literate

obtained

hold

Research

subjects

Tufts University Institutional Review Board. permission for the inclusion of their children in the

as well

For

Office

the

by

endorsed

Services,

human

1496), the human investigation

845, Renewal #

Research

involving

procedures

Group, not connected to Heifer. Field Supervisors monitored the per-

this study was al. ( 2014), laid down in the Declaration

et

guidelines

the Nepal Health

by

were approved

Nepal

to the

N=

431

included

were included

in the sample

for the baseline

survey

households).

Non- participating households were not in the baseline or follow- up surveys. Each participating

NHRC.

household was visited five times over 24 months at 6 month inter-

participating household were enrolled in the study. The study was conducted in three study sites of Nepal, and Nuwakot. Chitwan in the districts of Chitwan, Nawalparasi,

vals, with some minor adjustments due to monsoon or other diffi-

All

members

Nuwakot

to

procedure

was

approved

both located in the Terai

are

is located in the Hills. These

low- income

vided

consent

by

the

of each

Nawalparasi

and

by

The

subsistence

specific

areas

at

areas

are

farmers. Heifer field the

request

of

region

largely

of

Enumerators,

are

within

in pairs to conduct

ties

is

designee.

defined geopolitically in Nepal as " Village DevelopCommittees" ( VDCs). A waiting list of interested communi-

For

purposes

developed

ties

and

income

was

assigned

to

for

receive

design,

each

Heifer

local

size,

identified, based

were

natural

levels. A

interested

resources,

matched

into Group 2

months.

which

first ( Group

Group

reduces

communities.

the

head

based

on

specifically

of household

or her

standardized

the version

tools

used in

1)

or

second (

sumed

activities

1

and

risk

Group

2

communities

of program

effect

were

spill- over

Although

occasionally,

we expect

and other simple indicators

a 24 h recall is not a perfect sur-

that this

measurement

error was

perform relatively

well ( Prentice

et al.,

2011: Arimond et al., 2010). Other large- scale studies in South Asia

begincontin-

was an

random across households.' Recent evidence comparing biomarkers with different dietary recall instruments indicates that 24 h recalls

Group 2),

activities

2012). The dietary intake questionnaire

vey instrument, and might not accurately measure foods con-

predominant

ued through the entire 24 months, while Group 2 participated in Heifer activities beginning after 12 months; these activities continfrom 12 to 24

by ICF International,

of age in the household.

employ-

of agriculture

the three study sites, and randomly development activities in a staggered

later. Thus, Group 1 participated in Heifer after the baseline survey; these

non- adjacent,

was

individual 24 h recall for each child between 6 months and 8 years

pair of communi-

ning immediately

ued

with the female

questionnaire

ICF International,

on

of

either

The

traveled

the Government of Nepal and Measure DHS in 2006 ( MOHP and of comparable

availability of health care, type demographic features such as

and educational

selected

intervention one year

other

altitude,

as to the assignment

the Nepal Demographic and Health Survey ( DHS) conducted by

VDC.

study, three pairs of the three districts

each

opportunities,

practiced,

each

this

location,

geographic

castes,

of

in

communities

ment

for

maintained

were visited by Field

the visits during which a 116 item question-

naire was completed

prorural

communities, ment

with travel. Households

who were blinded at baseline

of the VDC as Group 1 or Group 2. The Field Enumerators

Nepal;

populated

activities

local NGOs

culties interfering

have (

used

24 h recall

data

to

estimate

dietary

diversity

data

Arimond et al., 2009). A 24 h recall is often used to estimate protein

If there was systematic

measurement

error in the 24- h recall, for example if

families with higher educational attainment were systematically better across all survey rounds at recalling their food intakes, this would be accounted for in the the family fixed- effects analysis with as a time- invariant unobservable characteristic.

188

A.F. Darrouzet- Nardi et al./ Food Policy 61( 2016) 185- 197

or

intake

energy

or

calibrate

misreporting is typically Freedman to

attempt and

estimate

issue

this

so

Consumption ASF)

can

animal

are

due

and

of seventeen amounts

All the who

concern such

as

quality

birth

were

recall

vented

inter-

ingestion

age

Children

strengthening

study. foods

create

house-

or

breastfed (

the time

at

of

knowledge -

associated

to

another

bonds

means

over time, as well a means

to

spread

the

donated

to

livestock

issues.

Further details

on the intervention

in Nepal can

be found in Miller et al.( 2014).

Household

visit

age

8

to

characteristics

years

MOHP

collected

on each

household,

lines( MOHP and ICF International, 2012). The number of animals

ICF

and

was

based on household possessions and quality of housing( e. g. toilet and water facilities), and were calculated using DHS- Nepal guide-

as

by inspection

information

including socio- economic status ( SES), animal ownership, annual income, and amount of land owned. Scores for SES were largely

in detail

excluded

owned was converted

of

to a standardized

score using FAO Global

criteria

Livestock Units( FAO, 2013). At baseline, households owned a vari-

which they were handicaps that pre-

ety of animals, including cattle, oxen, buffalo, sheep, goats, pigs, and poultry. At baseline, 14 households had no livestock. This

at

or children

survey

in

were

with

number

severe

diminished

12 month

excluded.

point,

to

and

2 0

households households

with with

no livestock no

livestock

at the at

the

24 month point. Poultry was the type of livestock most commonly owned, followed by goats. Livestock holdings were not measured

of the intervention

Details

community

sustainable

economic

each

enrolled

in" to the entry

neurologic

diet for

of a normal

were

determined

first

the

as

Allen,

foods from

assessed

of age

who " aged at

a

well

throughout the community. " Cornerstones" training ensures selfhelp practice in community members on a broad range of socio-

of

especially

and

as

Passing on the Gift" thus forms a basis for further

but did not ask about

were

months

was

physical

illnesses

current

if

goats -

Demographic

completely

Child

with

donated

group member."

methods.

in the study

Children

recipients to agree to provide the first female offspring of the

occasional

Murphy

asked

were consumed,

be

to

certificate.

enrolled

eligible.

that the

meaningful,

at baseline (

Children< 6

2012).

the Gift" to their neighbors. Acceptance of this gift requires the

consumption

participating household were health of children age 6 months

likely

International,

sourced

of each

the study.

were

they

present

can signal

study. Diets and resided in participating households of

for the animal

healthy foods is important,

2014). The 24 h

or preparation

mass

This

While frequent

them.

other

food groups

members

part

the

protein

setting is nutritionally

et al.,

body-

study did not intake from the 24 h recall,

in the diet

rare

and

index

questionnaire,

with

2003).

al.,

et

foods

afford

in this

Rawlins

correlated

of major

to the low dietary

2003;

as

not

typically

foods

source

or

additional

occasionally

consumption

Kipnis

energy

was

of

which

hold

2014;

al.,

et

food frequency

a

positively

at the 6- and 18- month observation points.

As previously consisted

of

described

in Miller et al.( 2014), the intervention program of 12- month long participatory

a

Dietary quality indicators

community- level development activities led by Heifer field staff, specifically

tailored

for

International

Nepal,

2014).

the

in

situation

The Heifer

citizen empowerment, and poverty alleviation, development, with a strong emphasis on optimization

based in

Self-Help Groups

women' s

household

members,

facilitator. These

course

of

designed

and

are

workshops,

promote

sometimes

biweekly

meet

the

cohesion

activities

involve a

Heifer and

dietary

(

including animal

reforestation

use

of

addressing livestock into the

of

the

environment),

der

awareness (

household), Nepal,

and

promoted

questionnaire were aggregated into the commonly used WHO diet-

The

ary diversity indicator(

participation

saving biogas), improving health and husbandry, and

improvement

was established

pest

not

Kitchen

gardens

the Heifer

child

diets,

training

nutrition,

or

curriculum

does

health. Baseline

not

staff suggest

that

members

community

are

dominant

were

of

dietary

and also because most children

patterns

groups, the MDD indicator

at baseline is applicable

in the

and

across

various

quality arose with the addition(

age

in this case, even for chil-

dren older than 24 months of age. The widest variation

specifi-

aware

equal to one if the child

study consumed foods from at least three groups: typically a starchy staple, pulses( dal), and then oil( WHO, 2010). Given these pre-

knowledge

generally

milk and dairy

of interest, MDD,

sen as a cut- off for the number of food groups, in accordance with

in dietary

or not) of the fourth food group,

that is, if a fourth food group was consumed at all by the child. The third outcome of interest, ASF, was constructed as a binary

of the nutritional benefits of animal source among participants foods was not measured. Anecdotal qualitative reports from Heifer

field

as a binary variable,

WHO recommendations,

way to generally improve household diet; however, a specific focus on nutrition improvement for the

Notably,

address

techniques.

nuts, and seeds;

second outcome

in the previous 24 h period) at time t and zero if not. Four was cho-

of

household income, encouraging gendecision- making on benefits for the

management

legumes,

achieved MDD( defined as having consumed any four food groups

increasing focusing

eggs;

products ( WHO, 2010). The

practices

animal

ecosystem

WHO, 2010). Thus, seven food groups were

starchy staples ( grains and white potatoes); vitamin- A rich fruits and vegetables; other fruits and vegetables; organ meat,

planning for the future ( Heifer International Participating households also learned about com-

as a

was

program.

tree-

and

agricultural

had

is

program

and

2014).

posting

cally

and

sustainable

management (

integration

there

using

to indicate

Neumann et al., 2003; Ruel, 2003; Steyn et al., 2005). To calculate

meat, and fish;

community accountability

resources,

calculated

DDS and then MDD, the seventeen foods and food groups in the

devised to

encourage

was

the

created:

sharing

MDD)

consumed any animal sourced foods ( ASF) was constructed

for holistic community training is based on the" 12 Cornerstones" modules development. Each " Cornerstone" has specific training

via

into seven groups( WHO, 2010). Then,

diversity" (

food groups. Finally, a binary indicator of whether children

other

self- reliance.

foods, aggregated

minimum

whether or not the child had consumed foods from at least four

are

interac-

specific

seventeen "

trained

training. Over

and

months,

community

community of livestock

with

by

supplemented guidance,

12

approximately

to

These

generation. which

which

meetings

instruction,

tive

income

to

as a means

Dietary quality was represented as three outcome variables of interest. First, a dietary diversity score( DDS) was calculated from

focuses

curriculum

on

management

Heifer

Nepal (

rural

training

variable indicating whether the child had consumed any ASF (

meat, fish, eggs, or dairy) in the previous 24 h.

the nutritional benefits of animal milk, frequently offering it to the elderly, sick, and children when it was available. At the conclusion of the first year of involvement activities,

two

households,

meat-

with

the

type

goats

proviso

were

that

these

donated first

to

Statistical

participating

recipients "

methods

in these

Pass

on

To assess program effects over the two- year study period, we estimated

a

difference- in- differences

model

which

incorporated

household fixed- effects'

key

Nepal.

as

results

between

by

region,

Hills

the

due to the in

Terai

the

and

et al./

model to estimating the difference- in- differences impact of the intervention, we first performed paired sam-

t- tests to

how dietary quality

assess

sex, and

geographic

exploratory

regression

age,

an

stratified

Darrouzet- Nardi

Food

Policy

region

of

varied

with

child.

to

estimate

analysis

key factors

Then,

the

the

such

conducted

we

associations

2016) 185-

61(

197

189

the difference- in- differences test of program

model, which served as the formal

effects.

Difference- in- differences estimation was selected for this study

Before the

assess ple

and

differences

agroecological

A. F.

because we had panel data to take advantage of, and difference-

in- differences

can address endogeneity

sent whenever an intervention

issues, which may be pre-

is not randomized

at the individual

level.

between the treatment variable and dietary quality indicators. Using obtained from this exploratory analysis, we information designed the difference- in- differences model to estimate the average

the

treatment

effect

compared

the

An

the

is

a

in- differences

useful

is directly

rounds

an

more

quality. difference-

with

rounds

outcomes

of

differences

data

of

McKenzie,

endline(

we

analysis

dietary

child

compare

when

between

autocorrelation

with

significant

on

tool to

analytical

especially besides just baseline and

low

tively

results

treatment

of

estimates,

ANCOVA, the survey

levels

various

ANCOVA

collected

intervention. Finally,

the

of

ANCOVA), to test for statistically

covariance(

between

the treated( ATET)

on

difference- in- differences

2012).

are

In

interest

of

an

across

incorporated. Child

diets may have relaespecially across different seasons, and food groups such as animal sourced foods.

autocorrelation,

especially for

particular

variability and low autocorrelation of these dietary outcomes be exploited with ANCOVA to improve statistical power over a

The can

difference- in- differences the

remove

to the

2012).

An ANCOVA

can

that

due

treatment

average

outcomes

effects

was

across

survey rounds. the dependent variables

In the exploratory regression, were the DDS, MDD and ASF indicators of dietary quality. The independent variables were child age, child age squared, child sex, household size,

women' s

Villages were randomly assigned to either Group 1 or Group 2, but within villages, households

pation

in the

level,

education

income, and geographic

season

of

measurement,

region. Notably, although

annual

intervention,

intervention

significant

predictor

regression.

Income

was

24-

and

at

months,

the

hungry hungry

the

the 6-

the

study

be

to

likely

subsistence-

be

should

on

starting after the baseline survey was implemented.

Statistical

the

29%

receiving foreign

for

for Nepal

income

identification

incorporating

fixed-

effects

scores

the

and

estimates

included

season

was

or

Bank,

best

the

the

design and the near universal participation within

each community,

and thus reduced concerns for selection bias. This

strategy is especially useful given that we are concerned with the broader effects of the community- level intervention on individual effects

diets. Concern

with

of the intensity

the

but still important-

separate-

of household- level participation

in the

program may have merited a different analytical strategy, such as propensity score matching, as utilized in Rawlins et al. ( 2014) who explored the Heifer program effects in Rwanda.

Dietary Quality

it=

of+/

f

lit+ B2Pt+

Q6 Cit+

P3( lit* Pt)+

fl4St+

135( lit* St) 1)

µt

Eq. ( 1) above was the difference- in- differences

model used for

was

an indicator

of how long ( in years)

received the intervention at time t( Intervention=

a child

had

0, 1, or 2 years).

It was included in the model to capture any differences between Group

1

Group

and

2

at

baseline. The P("

Phase")

variable

was

a

binary indicator of Phase 1 ( P= 0) and Phase 2( P= 1) of the intervention. The Phase variable was included

to capture the time-

variant factors which may affect the dietary quality outcomes of interest, irrespective sible exogenous

of being in Group 1 or Group 2, such as pos-

variation

in agricultural

productivity

year to year,

were

were pooled into two groups from the original five time points to obtain consistent standard error estimates( Bertrand et al., 2004).

or unmea-

Therefore, to

control

regression

varied

by

effects

into

binary indicator

and randomized

implementation

incomes,

option

as

the

for example from a drought or a pest infestation. The time phases

of program

as a

in

con-

households, instead

incomes

from the phased

received

many

2012)their

part of

households.

across

in

that

unobservable

among

from the exploratory

we stratified

nal groupings

reflect

differences

socioeconomic

Results

therefore,

as a substantial

rea-

variable

given

who

variable

during

diversity

came

households

surwith

two

are

in

impact evaluation of the Heifer intervention. The I(" Intervention")

is that income

reason

World

as a whole(

heterogeneity

unobservable

there

as a control

dietary

with

might

household

incorporating

used

The first

model.

remittances

invariant

time-

not

study setting. Moreover,

of remittances

receipt

ables.

The

points.

children

and

months,

coincident

annual

However,

valid.

was

codetermined

agriculture

households-

trol

month

exactly

Although measuring

season.

why household income the difference- in- differences

sured

therefore

12-

season

of

years

and

18-

month or

was

sons

is

baseline,

for selection

intervention

in the exploratory

ASF

or

at

potential

intervention

annual income

may lead to systematically lower estimates of due to the recency effect, income measurement across

income the

not

for income

schedule

vey

MDD

of

measured

the

sites ( approximately 90%). Using a difference- inmodel allowed us to estimate the ATET of the Heifer

differences

improved over the course of the 24- month study, it was not a statistically

could choose their level of partici-

creating

bias, despite the near- universal participation of households in the

children' s

dietary

antecedent

McKenzie,

model (

in the

variation

Identification strategy and selection bias

of

vari-

region; regio-

in

variable

For Group the

1, the baseline

6- month,

constituted 12- month

12- month,

the

Phase

observations

observation 18- month

2.

For

constituted

and

Group

Phase

24 month

2,

the

constituted Phase

baseline

1

1, and

observations

and

through

the

18-

and

24- month observations constituted Phase 2. Thus, Phase 1 indicates those who had been exposed

to the intervention

less at the time of measurement,

and Phase

for 6 months 2 indicates

or

those

who had been exposed to the intervention for 12 months or longer

at the time of measurement. Along with the benefit for econometric analysis of obtaining consistent standard error estimates, pool2

able

this

fixed

Household

characteristics

effects

of

incorporated in

were

households

study, household data

were

that

remained

collected

order

to

constant

from

for

control

the

across

study

with

along

respondents,

unobserv-

any

For

period.

child- specific

variables. But, some key variables cannot be measured, such as each household' s joint the motivation in providing healthy diets for their kids, or their excitement about Heifer

intervention.

International

regression,

each

categorized

in

unit

non-

Just

of observation(

overlapping

like

with

children,

groups(

such

country-

in the as

case

level fixed of

the

households, in

present

the

in

effects

a

study)

case of

the

panel can

be

present

then enter as a constant term ( over time) in the differencestudy), which in- differences estimates. One child per household need not be picked out to use for measurement

household

over

of

household

the

study

fixed

period.

effects,

because

each

child

belongs

to

only

one

ing the five time points

of observations

Phases explicitly

recognizes

the intervention

to be immediate(

The

I*

P ("

Intervention*

into these

longer

two

that we did not expect the effects of Phase")

Bertrand et al., 2004). variable

was

an interaction

term of the I and P variables. The coefficient on this term was the estimated ATET of the intervention on those children who received

the intervention starting immediately after the baseline survey. The ATET was the difference in the average outcome across Phases

1 and 2 for Group 1, minus the difference across

phases

in the average outcomes

for Group 2. The difference for Group 2

was

subtracted

190

A.F. Darrouzet- Nardi et al./ Food Policy 61( 2016) 185- 197

from the difference impact also

1 to

with

any

have influenced dietary

could

the I*

cient on

for Group

intervention

the

of

P

ferences." Table

conflation

time-

was

Groups.

factors that

estimated

coeffi-

Group

difference in the dif-

the"

below describes the three different

1

Table 1

the true

of

variant

The

outcomes.

therefore,

variable,

avoid

other

intervention levels, by the length

groups

of

harvest

were

the

during

made

season ( S= 0). Season was

hungry

season (

included

S=

1)

2

2

2

2

6

12

18

24

0

6

12

18

1

1

1

2

2

0

0

0

1

1

0

1

1

1

1

0

0

0

1

1

0

1

1

2

2

0

0

0

1

1

ATET,

and

years)

P)

I* P

of

The

I*

P variable

was

used

to

estimate

the

also

24

served

as

the

different groupings for ANCOVA.

in order to control

for the impact of seasons on the dietary quality outcomes,

which

whole, effect sizes are smaller in magnitude, but statistically signif-

in preliminary analysis. The I* S (" Intervention* Season") variable was an interaction term of the I and S indicators. was

2

0

Note:

the

or

1

1)(

Value

observations

1

months)

has been taking place, and the phase for each group. The S(" Season") variable was a binary indicator of whether the

vention

1

Intervention( Phase(

time the

of

1

Observation(

inter-

to the

exposure

1

observed

icant and potentially

useful from a policy perspective.

This variable was included to allow for the fact that the intervenimpact may vary by season, due to nonlinearities in the impact of the intervention such as seasonal fluctuations in food availabiltion

The of term represented the household fixedhousehold is unique, and may have unobservable

ity.

which

correlate

of child

with

child

characteristics,

and

this

Estimating

erties.

allowed

us

to

is

model

term

an error

using

impact

the

assess

dietary quality. Cit u; t

as

represents

the

with

each

a vector

prop-

household- level fixed-

effects

intervention

on

child

household, by effectively quality within each specific removing the impacts of unobservable, time- invariant household because

Standard

not

Wooldridge, 12. 0(

be

to

expected

errors

the

of

observations

within

were

household

a

Bertrand

analyzed

household

by

clustered

child

independent (

Data

2002).

were

same

StataiMP

using

were

2004;

al.,

et

version

(

characteristics.

statistics

Descriptive

and summaries

statistics

of

across groups

at baseline indicate that randomization was relatively successful Table 2). Mean household characteristics, including annual household income, and socioeconomic

ownership

did not differ

status, household

between

Groups

size, and land

1 and 2 within

each

study sites, as expected since these groups were assigned randomly ( Miller et al., 2014). However, animal scores at baseline

were different between Group 1 and Group 2. Further details on the randomization can be found in Miller et al.( 2014). The analyt-

ical design allowed us to account for this difference by controlling for baseline household characteristics.

2011).

StataCorp,

1- 3 present the descriptive

key household

dietary

characteristics.

Tables

characteristics

usual

the Heifer

of

effects,

Household characteristics

There

were

significant

differences

in baseline

characteristics

between study sites in the Terai and the Hills( Table 3). At baseline, the

Exploiting We

multiple

utilized

survey

difference- in- differences McKenzie (

2012).

dependent

dietary

levels

the

of

as

a

estimates,

This

model

outcome

tests

while

for

comparison

the

following

same

controlling

for

are

the

pooled

in

specification

the

whether

the

variables

treatment,

in the Terai had higher mean annual income per house-

hold( one- sided two- sample t-test, p= 0.000), larger mean house-

ANCOVA

an

households

rounds

across

the

of

means

different

lagged

dietary

outcomes.

hold

size (

mean

two-

one- sided

women' s

p= 0. 000) than the Hills ( Table 3). At baseline,

p= 0. 000) than the Terai ( Table 3), a finding consistent ecological

literature(

3i DietaryQuality

t_

3214+ EAt

1+/

t=

to 271

1

t

Eq.(

DietaryQuality;, t 2), and

patterns rent and lagged dietary quality child

for

i

at times

t

by as described and outlined in Table 1 binary indicators mation

and

These

for

to

each prevent

the usual

across

ical

properties.

model is ( 32.

survey

perfect were

represent

collinearity;

is a

constant

which is dropped

andµ;,

coefficient

tis an

error

of interest in

analyzed using

term;

over the study period. The number

76

at

baseline

to

134

at

(

cur Child dietary is Dietary esti- child

age,

quality-

region, and

measured by

DDS,

MDD,

season of

term

version

with Tables 5- 7 below present analyt- same

12.

0

and

measurement.

ASF- varied Figs. 2

age

base-

dietary groups

quality

or

patterns.

indicators did not vary

appreciably

sex ( Tables

child

6). Children in the

and

5

line ( mean

DDS= 3. 05,

SD= 1. 19, one- sided two- sample t-

p=0. 000);however, mean DDS, MDD, and ASF did not differ

Results other age groups nor by sex(Tables 5 and

among

6). Over the study

period, the lower DDS for the 6-12 month age group section

of baseline

null

ysis

cannot

the

not

key

the

groups,

intervention'

s

the

effects,

then

and finally

tion using dif erencethese results

detailing

comple- improved key

results

of

and

demonstrate that the indicators of

exploring

the

differences in-

mean

is

most

pressing.

dietary

the

ANCOVA

When

the 12-month

patterns

across

be

rejected ( one- sided two- sample ttest,

evalua-

shown). The

results for the impact

methods.

Overall, inclusion or exclusion of the youngest age significantly dietary diversity at baseline was the

and mentary foods were just being introduced(

season, when the

results for

the

sample

persisted across

observation, where

the

p=0. 29, ttests

impact

Heifer intervention in the Hills,

estimating

points except

mediated hypothesis that the DDS across all age groups was the same

child dietary quality

baseline). this effect was greater during the hungry need

anal- time

results in detail, starting with an household characteristics that may have

presents

and 3

findings for the dietary

2011). 12 month age group had the lowest mean dietary diversity at test:

This

with > 10 24- months

for

this

StataiMP

from

It,

StataCorp, 6-

all

with the

the

above; At is a set of five

round, one of

Our main

Data

t_ 1

increased

2) Table

indicators( DDS, MDD, and ASF)

respectively;/ 30 1,

t-

and

DietaryQuality,

changed

at 24- months, and the number of households

chickens

µ;

4). In

in the

Metz, 1989).

Livestock holdings 30+/

households

Hills had a higher mean animal score( one- sided two- sample t-test,

of households with at least one goat increased from 197 at baseline

5

DietaryQualityi =/

t- test, p= 0. 015), and higher level ( one- sided two sample t-test,

sample

education

as

a

3 Except

for

the

youngest

age group, 6-12

month

evaluation

were robust to

group, for which

lowest, likely because

n= 87 at

A. F.

Darrouzet- Nardi

et al./

Food

61(

Policy

2016) 185- 197

191

Table 2

Characteristics at baseline across districts between Group 1 and Group 2. Household

Units

of

N

characteristics

measurement

income

Annual

Households

Nepali Rupees

59

SES

Animal

Household

score

Global Livestock

87, 610

1. 38

0. 79

5. 38

1602

117, 581

0. 91

0. 86

2.

08

3205

units

owned

m2

Wealth index

year

per

Number

Land

size

of members

Group 1 in Chitwan Mean SD

Group 2 in Chitwan Mean

44

SD t- test p-

value

65, 434

1. 70

1. 61

5. 41

2741

46, 225

0. 99

1. 86

1. 87

2583

0. 235

0. 096`

0. 004-

0. 962

0. 056'

71, 799

1. 48

2. 80

7. 16

3184

59, 414

1.

10

1. 62

3. 02

2443

Group 1 in Nuwakot Mean

59

SD

Group 2 in Nuwakot Mean

72

SD t- test p-

63, 930

1. 19

2. 95

5. 93

3348

64, 301

1. 16

1. 67

2. 19

2234

0. 161

0. 558

0. 007-

0. 688

0.

value

473

Group 1 in Nawalparasi Mean

62

SD

92, 258

2. 02

2. 80

6. 69

3779

94,711

1. 28

1. 62

2. 84

4481

85, 191

2. 22

2.41

7. 57

4163

1. 11

1. 84

3. 69

5495

0. 669

0. 354

0. 005"

0. 113

0. 666

78, 019

1.

66

2. 07

6. 41

3189

84, 014

1. 16

1. 77

2. 84

3789

0. 178

0.617

0. 976

0. 103

Group 2 in Nawalparasi Mean

68

SD

92,

t- test

p- value

875

All study sites Mean

364

SD t- test

p- value

0. 000"

Notes: Chitwan and Nawalparasi are districts located in the Terai region. Nuwakot is located in the Hills region. T-tests were performed to test the differences in means

between each group, within each of the three districts, and then for the sample as a whole at baseline. Two- sample t- test with equal variances;" p<. 10. p<. 01.

Table

3

p<. 05.

Child dietary patterns across age groups and sex at each obser-

Difference

in

for household

t- tests)

means (

characteristics

across

agroecological

vation point were analyzed using the 24 h recall data( Tables 5 and

regions.

Annual

Household

Women'

income

size

education

Nepali

Number

per

rupees

year

Animal

s

Global

Categorical

of

6). At baseline, almost all children consumed a starchy staple( 96%) and oil( 94%) in the previous day, while many children consumed

score

people

dal( lentils)( 43%),

livestock

and other

fruits

and vegetables(

38%),

and smal-

e

ler

units

proportions

rich

fruits

and

of children vegetables(

consumed

meat (

34%) and Vitamin- A

9%). Mean DDS at baseline ranged from

Region

Hills N=

65, 071

71, 753

5. 00

2. 21

2. 87

2. 78 food groups ( 6- 12 month age group) to 3. 81 food groups

6. 27

3. 21

1. 66

MDD (

19- 24 month

225)

Terai

N= 364) Difference Pr( T<

t)

-

6682

1. 27

0. 048"

or

-

1.

0. 200

00

age 5)

group).

ranged

Across

age groups,

from 22% of children (

achievement 6- 12 month

of age

group) to 58. 6% of children ( 37- 60 month age group). Consump-

1. 21

o. 000"'

Table

tion of ASF ranged from 44% of children ( 6- 12 month age group)

o. 000"'

65% of children ( 37- 60 month age g group) ( Table 6). Dietary quality indicators differed between the Terai and the Hills( Table 5). to

Two-

t- test

sample

with

equal

some or

school;

primary

beyond."

p<.

05.

p<.

01.

p<.

4:

The

variances;

categorical, where 1: no education; finished

women'

s

education

is

variable

2: literacy classes, non- formal education; 3:

primary

school;

5:

some

secondary

school;

6:

SLC

In general, DDS was higher in the Terai than in the Hills( one- sided

two sample t-test: p= 0.05). Mean MDD was also higher in the Terai( one- sided two- sample t- test: p= 0. 00), but ASF consumption

10.

was higher

x.,

in the Hills ( one- sided

two- sample

t- test: p= 0. 00)

Table 5).

Dietary quality indicators varied across seasons( Fig. 1). Seasonal distribution of DDS differed, as more children achieved higher dietTable

ary diversity

4

Change in livestock Number

of

ownership

households

across

with

goats

all

study

sites

between baseline

Baseline

12

377

421

months

24- months.

and

24

months

the

hungry

during

scores

54. 5%)(

seasons(

households

None 1 2

2 Number 0- 5

of

households

with

harvest

seasons(

73. 7%) than during

Fig. 2). DDS were systematically higher

during the harvest seasons compared to the hungry seasons( mean plus SD= 4. 13 plus

Total

the

1. 07 vs. 3. 67 plus 1. 11,

two- sample t-test,

415

p< 0. 001) (

Table

4).

This

pattern

was

seen

in

both

regions

166

36

49

34

14

16

p= 0.006 in the Terai, p= 0.000 in the Hills). There was an apparent disappearance of the ' hungry season' for the Terai Group 2 at the 12- month observation point, indicating that something was con-

37

37

38

136

238

217

206

tributing to better dietary quality in that region( Fig. 2). Thus, there

poultry 133

103

6- 10

89

86

73

10

76

113

134

were seasonal patterns in dietary quality across both regions( Terai and Hills). Overall, the children March

and

April than in October

achieved and

better dietary quality in

November( Fig. 1).

192

A.F. Darrouzet- Nardi et al./ Food Policy 61( 2016) 185- 197

100. 0

4

90. 0

O

80. 0

8

r

70. 0

D

60. 0

d50.0

40. 04

C4 C4

30. 0 o

oip

20. 0

0

10. 0

y

0. 0

u•

6- Months

Baseline Terai

1 ——

Group

12- Months

Terai

Group

2 —

18- Months

Hills

1 ——

Group

24- Months

Hills Group 2

Fig. 2. Percentage of children achieving minimum dietary diversity, over the study period. Note: N= 377 households at baseline, N= 377 at 6- months, N= 421 at 12- months, N= 423 at 18- months, N= 415 at 24- months.

45

4

40-

E 35u

30-

0 on

C" 0 2015u

10

d 5 a

J

0 DDS= 1

DDS= 2

DDS= 3

DDS= 4

DDS= 6

DDS= 5

DDS= 7

DDS= 8

Fig. 3. Dietary diversity scores( DDS) by season. Note: DDS is the dietary diversity score. Black bars indicate the distribution of dietary diversity scores in the' harvest season', and gray bars indicate the distribution of dietary diversity scores during the' hungry season.'.

Impact

The

evaluation

reduced Exploratory ables

are

levels,

and

that

was

Intervention*

the

differences This

coefficient estimated

ratios,

that

so

indicate

no

variable.

Results

the

24 h

show

that

have

to

1.

27

times

1.

38

times

children

as as

likely likely

in the Hills

year.

to to

who

coefficient

of

the

children

0. 981

coefficient

achieving ASF

equal

to

control

or

in the Hills

who

is

of a child

consumption

one

after odds

would

explanatory

had

received

consumed

received

the

ASF ( p= 0. 001) as intervention for only

model

season,

and

the

in

shows

both

that

the

Hills ( Odds

MDD

Terai (

was

Odds

Ratio= 0. 069,

for DDS and MDD, respectively), but it did not mediate the impact intervention

were

have

p= 0. 007)

Ratios=

the

hungry

p= 0. 000). Season also mediated the impact of the intervention in the Hills ( Odds Ratios= 1. 728 and 1. 333, p= 0. 000 and p= 0.001 of the

the

approximately 2. 2 times as an additional food group ( p< 0. 001), have achieved MDD ( p= 0. 015), and had

0. 503,

the

separately( Table 9).

words,

given

of on

Ratio=

during

difference- in-

of

coefficient

of

years

consumed

8). The impact the

among the treated children, in Table 9 are presented as

estimated effect

term

other

likelihood

Coefficients an

Table

the Terai

and

period,

estimated

regions(

vari-

outcome

women' s education

in the DDS, the likelihood

change

the

dietary and

by calculating

ATET. In

intervention for two

likely

one

or

the

size

interaction

the

average

intervention.

across

for the Hills

was

achieving MDD, in the previous the

household

estimated

Phase

model

that

revealed

with

they differ

intervention

the

the

analysis

associated

difference- in- differences

and

in

the

Terai

for

1. 049, p= 0. 826

DDS

or

for

and p= 0. 259

MDD ( Odds

for DDS

and

MDD, respectively). This indicated that the intervention was more effective at improving child dietary quality during the hungry season than in the harvest season in the Hills, but not the Terai. Differ-

ences in program effects across seasons may have occurred because it may be easier to see an effect during the hungry season, since the baseline status quo is worse. In other words, it' s easier to see program benefits where conditions were worse to begin with,

because there are likely to be diminishing

marginal returns to com-

munity development activities. In general, child- level variables were not associated with diet-

ary outcomes, as perhaps foreshadowed by the t- tests in preliminary analysis. Results did not change when including or

excluding the youngest children( aged 6- 12 months) whose diets differed the

most

from the

rest of

the group,

from the

analysis.

A. F.

Darrouzet- Nardi

et al./

Food

Table 5

61(

Policy

2016) 185-

197

193

across the three intervention in

Difference

for

t- tests)

means(

dietary

outcomes

across

age,

gender,

season,

and

levels ( zero exposure, one year of

exposure, two years of exposure). The results of the ANCOVA indi-

region.

cated that there was a significant effect of the amount of the interDDS

MDD

ASF

vention

Girls

3. 86

0. 79

0. 62

Boys

3. 84

0. 78

0. 61

0. 02

0. 01

0. 01

0. 31

0. 11

0. 23

Difference Pr( T>

t)

Pr( T<

or

t)

the intervention

Age 6

12

mos.-

12

Age>

mos.

mos.

Difference Pr( T>

t)

Pr( T<

t)

0. 000), 1. 12 times as likelyto have achieved MDD

p

0. 72

0. 47

0. 97

0. 63

tion for only one year. There were no measurable

0. 15

intervention ANCOVA

0. 25

0. 00"'

0. 000'

0. 00"'

on dietary

in Table

quality

3. 67

0. 55

0. 59

Harvest

season

4. 13

0. 74

0. 61

Difference Pr( T>

t)

Pr( T<

or

0.00-

t)

region

Hills

the intervention

region

t)

Pr( T>

0. 00-

t)

Pr( T<

Two-

DDS=

Dietary Diversity Score. Minimum dietary diversity.

t-test

sample

Animal

Source

with

Foods

equal

to

zero.

These

0. 82

0. 59

3. 81

0. 80

0. 65

The

0. 06

estimated

0. 02

0. 05*

Notes:

equal

0. 05*

0. 00"'

magnitudes

of these

results

analysis odds

lend

further

for all the

ratios

are

and statistically

support

dietary

smaller

than

to

Just

what

as

with

the

statistical

difference- in- differences

model

power.

presented

above, we can aggregate the ANCOVA for the whole sample to

show

generalizability ( Table 12). In this specification, that children who had received the intervention

years were approximately an additional that

appears

boy

children

compared on

1980s

and

South

Asia

1990s

while

others

did

is bias

If there

allocation

marriage

displacing

nization

1997;

against

evidence

Messer,

girl

1997;

Miller,

out

Gittelsohn

Gupta, 1987),

1997),

due to

agricultural

mecha-

results

Gittelsohn

et

al.,

potential

The literature

1997; Messer, 1997).

any disparities in how girls and boys were affected Heifer intervention over time longer periods of time. below

10

this

specification,

p= 0. 001), p= 0. 001),

have

to

likely

as

the

shows

1. 15

and

results

years

were

an

for the

that

show

consumed

times

1. 18

pooled

results

intervention for two

the

received

of

it is difficult to quantify cultural beliefs 1997), and further work would be needed to

Messer,

traditions(

et

by

times

had received significance

the intervention of the model

for only one year. F- tests for joint

indicate

that we can reject the null

hypothesis that the coefficients on all of the variables in the model are equal to zero.

In summary, there are four main results to emphasize. The first seasons,

for children aged 6 months to 8 years, and this variation

this

even using just dietary diversity

as an indicator.

The second result is that a community- level, nutrition- sensitive intervention

without an explicit nutrition

focus had a significant

impact on child dietary quality in rural Nepal. The third result is that this intervention was effective at improving DDS, MDD, and ASF in the Hills, which is arguably the more agro- ecologically vulnerable region of Nepal when compared with the Terai. The

sample

children

as

who

a

had

approximately 1. 52 food group

additional

to have likely as likely to have

as

have achieved MDD ( p= 0.072), and 1. 09 times as likely to have consumed an animal sourced food ( p= 0.001) as children in who

is measurable

assess

In

1997),

that

exemplifies

1. 12 times as likely to have consumed

food group ( p= 0. 034), and 1. 04 times as likely to

pregnancy4

foods during menstruation and 1997), or lower perceived economic

Gittelsohn

children ( area

al.,

for two

result is that child dietary quality in rural Nepal varies with the

of certain

restrictions

Table

bias, et al.,

labor force ( Miller,

the

of

foods in

of

of gender

in the intrahousehold

children

subconscious (

practices ( women

distributions

found

studies

foods, it may be

of

traditional

some

Miller,

not(

may have slightly improved dietary girl children. Evidence from the

with

intrahousehold

the

was mixed:

was

significant effect of the intervention

address

quality indicators

the

outcomes.

model. Still, there is

ing the multiple survey rounds while optimizing

consumption.

01.

times

F- tests for

in the Hills. The comparison of the two models is useful for exploit-

p<. 10.

whole.

of interest.

with the difference- in- differences

a measureable

variances.

p<. 05.

and

outcomes

h Yp othesis that the coefficients on all of the variables in the model

0.0. 111111

3. 88

0. 07

or

on the dietary

difference- in- differences

Difference

in this

from

11 show that we can reject the null hypothesis

joint significance of the model indicate that we can reject the null

are

Terai

girl

effects of the

The results

0. 19

0. 46

Region

1997).

in the Terai.

that there is no statistically significant effect of different levels of season

It

p= 0.002),

3. 88

0.82

or

in the Hills for two years were approximately

3. 05

Hungry

p<.

In

and 1. 20 times as likely to have consumed an animal sourced food p< 0.001) as children in the Hills who had received the interven-

Season

ASF=

on child dietary quality( Table 11).

1. 36 times as likely to have consumed an additional food group

Age

MDD=

had been received

this specification, results show that children who had received

Gender

MDD

achieved

consumed

fourth result is that the intervention

had a stronger

effect during

the hungry season, arguably when it was needed most. The use of an alternate model for comparison, ANCOVA, demonstrates

that

effect sizes may in fact be much smaller than those estimated

by

difference- in- differences.

ASF

p= 0.006) as children who had received the intervention for only The

one year.

magnitude

was estimated to

of

be dismissed

from

To

the

address

we

ANCOVA). effects

of

compared

The the

odds

ratios are smaller

than

what

into

estimate results

purpose

of

intervention

using

this

on

was

DDS,

an

to

MDD

Discussion

but are not small enough

a country- level policy- making standpoint. loss in power that comes with pooling

difference- in- differences groups,

these

for the regions separately,

baseline analysis

determine and

ASF

and of

Children in the Hills experienced a

endline

covariance

whether were

the

different

odds of consuming

This

causal

mechanism

would

not

fit for

our

study

population

per

se

due

to the

ages of the children included in the study( not old enough to be menstruating yet). However, girls

and

the

presence

women

in

of

general.

this

cultural

practice

may

be indicative

of views

towards

DDS and improved

This was

a community- level, nutrition- sensitive program without a specific focus on child dietary improvement. The main mechanism through which the intervention

functioned

is still unclear, but could relate

to the improvement to livestock holdings or women' s empower-

ment. The increase 4

improved

ASF and MDD with the intervention.

in the number of households

owning

goats

and poultry over the study period, as well as the promotion of kitchen gardens, may have provided increased income or direct nutritional benefit through home consumption. The intervention' s emphasis

on

social

capital

and

community development

may

also

194

A.F. Darrouzet- Nardi et al./ Food Policy 61( 2016) 185- 197

Table

6

Dietary quality indicators for children at baseline, by child age and sex groupings. Sub-

sample

N

of children

score(

6-

12

boys

Minimum

Dietary diversity mean)

of

78

with

MDD)

0

2.

girls

and

boys

45

3. 62

53. 3

73. 3

19- 24

month-

old

girls

and

boys

26

3. 81

57. 7

43. 8

25- 36

month-

old

girls

and

boys

26

3. 49

48.8

48. 8

37- 60

month-

old

girls

and

boys

84

3. 72

58. 6

64. 8

162

3. 59

49.8

56. 3

60

old

old

month-

girls

girls

and

and

boys

All

girls

291

3. 86

62.

8

59. 7

All

boys

298

3. 84

61. 6

58. 9

589

3. 57

50. 7

57. 9

Total

foods

43. 9

41

old

month-

18

source

of children with ASF)

month-

13-

22.

Animal

dietary diversity

children

Table 7

Percentage of children consuming particular foods, at each survey point. Observation

N

Starchy

Oil

staple

Dal

Milk

Meat

Eggs

Vit.

A

rich

foods

Other fruits and vegetables

Baseline

589

95. 8

93. 5

43. 0

18. 8

41. 4

93

8. 8

6-

524

99. 0

94. 3

53. 4

20. 4

54. 4

8. 6

13. 2

72. 9

months

445

98. 7

85. 6

51. 5

14.

2

40. 7

13. 5

6. 7

33. 5

months

523

97. 3

94.1

61. 8

15. 1

40.2

8. 0

8. 2

74.4

months

533

98. 5

96. 8

55. 5

143

503

9. 2

11. 8

54.4

months

12-

1824

Table 8

37. 9

advantage of the Heifer intervention more effectively. This may

Exploratory

regression.

Variable

have occurred because the Hills has more land available for grazing Units

1)

2)

(

DDS Exposure

Years

to

Terai

-

0. 560) 0. 408"

Binary

region

(

income

Annual

Nepali

rupees

per

(

0. 000

-

ASF

these rainfed grazing lands( Metz, 1989; LRMP, 1986). In contrast,

0. 085

-

0. 074

lower elevations such as the Terai are characterized by fewer live-

0. 195)

(

0.287)

stock which live and graze closer people' s homes, due to higher

0. 232-

0. 036)

than the Terai, and livestock are more mobile to take advantage of

3)

MDD

0. 100

intervention

(

human population pressure and the larger quantity of irrigated

0. 085

0.002) 0. 000

( -

0.276) 0. 000

0. 547) livestock

FAO

score

Number

size

0. 086

units

of people

s

education

-

Years

Months

age

age2

Child

constant

Number

population pressure in the Terai, or larger mean household size,

0. 014

which might make it more challenging for community- level devel-

(

0. 308)

0. 198***

0.044'

-

(

(

0.051)

(

0. 006

-

0.899)

(

0.300)

0. 000

-

0. 000

(

0. 201 0. 163)

0.625) 0. 000

0.249)

(

0.668) 0. 002

(

0. 000)

464

throughout the year. Wheat, which accounts for just 19% of Nepal' s

cereal production, is harvested in Nepal approximately from March to June( FAO, 2007). Maize, millet, and rice, which together account

from August to November. Other studies have found variation

dietary income

least

squares.

p-

values

results of this study indicate that a community- level, nutrition-

463

in

in

patterns and nutritional status across seasons in lowcountries ( Savy et al., 2006; Mitchikpe et al., 2009). The

sensitive

estimated

members.

Season was an important predictor in both regions, presumably because cereal harvests in Nepal are not uniformly distributed

for 80% of Nepal' s cereal production, are harvested approximately

0.498) 0. 545-

(

opment activities to succeed there, or to allocate program benefits

among more household

0.263) 0. 020

( -

0. 623)

(

0. 257)

0.026

0. 004

461

by ordinary Standard errors are clustered by family. regression

analytical models with the household fixed- effects. The differential

impacts across regions may also be due to the higher relative

observations

A linear

These

into the

-

2. 873'#

of

the Terai.

at baseline are incorporated

0. 012

0. 000)

N

in mind, the Hills

0. 047' -

0. 607) constant

than

0.240)

0. 004

Months2

0. 025

livestock

(

0. 498) Child

to raising

0.237)

0. 072

Binary

0. 717)

(

0. 024

0. 001) ( Sex

conducive

(

0. 061) women'

0. 984)

(

0. 138) Household

are more

time- invariant differences

year

Animal

cropland ( Metz, 1989). With these differences

intervention

such as Heifer may ameliorate

some of the

seasonal fluctuations in food availability. As described by Ruel parentheses;

and Alderman(

2013), nutrition- sensitive programs can be helpful

p<.

10.

templates

for scale- up at the regional

p<.

05.

adapted to be nutrition- specific programs.

or national

level, or be

p<. 0. Strengths and limitations

have improved child dietary quality, and the pathway through this

which

from

may have occurred is through women' in Heifer' s Self- Help Groups.

The

impact in the Hills

positive

baseline dietary

was

achieve

a

short

period

of

animal

also

initially

quality great improvement

could

mean

s empowerment

time. It

scores

in

is

also

the

may have occurred because low that the intervention

so

in dietary quality in a relatively that the higher baseline

possible

Hills

allowed

households

The strengths of this study include the random assignment across villages which allows for valid comparison of the program' s

participation

to

take

effects between implementation

Group 1 and Group 2, the semi- annual

survey

which allows for the analysis of seasonal patterns

in dietary quality, the inclusion of older children up to age 8, which is a rare occurrence in dietary quality studies, and the individuallevel 24 h diet diversity recall data. Household fixed- effects can

A. F.

Darrouzet- Nardi

et al./

Food

Policy

2016) 185- 197

61(

195

Table 9

Difference- in- differences, by region. Outcome

Units

region

Intervention

1)

Years

Phase

2)

3)

4)

5)

6)

DDS Terai

DDS Hills

MDD Terai

MDD Hills

ASF Terai

ASF Hills

1. 171

0. 381***

1. 048

0. 859

0. 958

0. 706-

0. 333)

0. 000)

043

0. 429"""

Binary

1.

Interaction

1. 190

Binary

0.836

Interaction

0. 981

0. 750)

0. 550)

035

1.

0. 000)

0. 128)

686"""

0.

0. 602)

0. 592)

0. 001)

0.810-

0. 994

0. 000)

0. 924)

0. 018)

2700"

Intervention*

phase

2. 197"""

0. 211)

1. 065

0. 000)

1.

0380)

1. 062

0. 015)

1. 380"""

0. 415)

0. 001)

8790"

Hungry

season

0. 291***

0. 118) Intervention*

season

0. 000)

child

Binary

1. 060"

Months

1. 001

Months2

1.

0. 974

age

0. 787) 2

Child

age)

N

0. 504)

0. 690) 0. 992

0. 030) 1.

0. 687)

000

1. 001

0. 792)

000

1.

0. 402)

1632

1. 029

1. 042""

0. 480) 1.

0. 639)

0. 771)

0. 604)

000

0.968

1. 014

1. 001

0. 668)

966

0. 963

0. 990

0374) 1.

0. 391)

1618

0. 001)

1. 001

1. 000

0. 621)

049

0. 259)

0. 075)

003

0. 400)

000

0. 000) 1. 222"""

1. 027"

0. 524) 1.

0. 613'""

0. 025) 1.

0. 000)

0. 068)

Child

728"""

1.

0. 826)

Boy

0.

0. 776)

000

1. 000

0. 625)

972

0. 659)

1619

969

The estimated coefficients on Intervention* Phase is the average treatment effect on the treated of the intervention. Group 1, having received the intervention starting one year earlier than Group 2, had longer durations for the" Intervention" variable. Coefficients are reported as odds ratios; p- values in parentheses; Standard errors are clustered by household; household fixed- effects not shown.

p<. 10. p<. 05. p<. 01.

Table 10

villages self- selected into assessment in- differences for

Difference-

the

sample

as

a whole.

Units

1)

Years

MDD

0. 757'

0. 960

0. 058) Phase

0. 736"

Binary

Interaction

phase

(

1. 524"""

Interaction

season

1.

Boy

1.

Binary

child

013

Child

Child

2

Months2

1.

age)

(

(

0. 632)

(

(

015

Generalizability

indica-

et al.,

of results

As with any impact

(

evaluation,

external

validity

of results

is

important to consider. For comparison and to inform the potential

generalizability of this study, the median household size in Nepal is

0. 883) 1.

0. 657)

in nutritional

animals ( Rawlins

programming activities.

0. 255)

000

1.

from donated

2014). This study focused on quantifying the other aspects of Heifer

1. 000

0. 508)

000

tors which come directly

0. 706)

1. 001

0. 993)

of Heifer projects have found improvements

18o""

1. 017

0. 508)

1. 000

Months

age

(

(

008

1.

0. 622)

1.

0. 001)

resources

0. 508)

1. 117""" (

over

observation point, to elucidate causal mechanisms. Other studies

0. 006)

(

or control

measurement of socioeconomic status and animal holdings at each

0. 972

0.000)

(

229"""

0. 008)

(

0. 765***

0. 000) Intervention"

0. 021)

0. 558"""

Binary

season

(

was not measured. Further work is needed, including the detailed

1.

use, empowerment,

data

0. 852'"

0. 090)

1. 146-

0. 001)

Hungry

(

for impact, since detailed

on women' s time

0. 912"

0. 034)

to assess mechanisms

ASF

0. 012)

(

0. 889"'

0. 014) Intervention*

0. 523)

(

not possible

3)

2)

DDS Intervention

for eligibility for Heifer' s pro-

gramming, thus potentially limiting external validity. Lastly, it was

4. 5 members, and the median years of women' s educational attain-

000

ment

0. 544)

is

1. 15 years

International,

for

2012).

Terai

and

Hill

Other variables

regions

could

combined (

not be directly

ICF

com-

pared to national averages because the DHS reports income and N

2584

2604

2588

socioeconomic scores as quintiles. The FAO livestock index not is The

estimated

the treated

on

coefficients

Intervention*

Phase

is

the

average

treatment

effect

on

Group 1, having received the intervention starting one year earlier than Group 2, had longer durations for the" Intervention" variable. Coefficients are

clustered

p<.

10.

p<.

05.

measured

the intervention.

of

reported

are

as

ratios;

odds

p-

values

in

by household; household fixed- effects

Standard

parentheses; not

errors

stock (

but 78% of household

International,

Average

2012).

in Nepal

gross

own live-

national

income

per capita in Nepal was$ 2410( PPP) in 2011- 2015, which converts to about

shown.

for the DHS,

ICF

256, 336

Nepali

rupees

per year(

World

Bank,

2015).

The

GNI for Nepal as a whole is likely much larger than the income estimated in this study because the study population

p<. 01.

is predominantly

agricultural and consumes much of the own- production of food.

This intervention was administered

at the community- level, to

rural farming communities which met criteria for participation by control

for

inequalities present main

heterogeneity

unobservable within

and

affect

limitations

of

households results (

this

ipation intensity

was

measured

time

at

all

such

as

across

birth

Jayachandran

and

households,

order

bias

Pande,

could

2013).

study include that household- level

not

measured,

points

of

that

key

observation,

variables and

that

but

Heifer. Therefore, communities which would not have met their

be

criteria may not achieve similar results. The six communities were

The

diverse as a group( Tables 2- 4), and arguably large enough to rep-

partic-

were

not

included

resent rural farming communities Nepal.

Moreover,

across

the

study

in the Hills and Terai regions of

only participating period,

which

households is

a

significant

were

measured

limitation

to

196

A.F. Darrouzet- Nardi et al./ Food Policy 61( 2016) 185- 197

Table 11

Analysis of covariance( ANCOVA), stratified by region. Variable

Lagged

Units

1)

Lag

outcome

2)

3)

4)

5)

6)

DDS Terai

DDS Hills

MDD Terai

MDD Hills

ASF Terai

ASF Hills

1. 223'*

1. 052

1. 118-

0. 986

1. 100***

1. 059

0. 000)

0. 194)

0. 000)

0. 712)

0. 002)

0. 157)

0*

Intervention

Years

0.963

Binary

0. 940

Binary

0.639***

Binary

1. 357*

1. 360*

0. 550)

0. 987

0. 000)

1.

0. 638)

119*'

1. 196* 0*

0. 998

0. 002)

0. 941)

0. 000)

0*

Survey

round

2

Survey

round

3

Survey

round

4

3. 960*

0. 561)

0. 973

0. 000)

0. 565)

1. 003

0. 000)

0. 837`*

1. 064

0. 000)

1. 527*'*

0. 953)

0. 922

0. 689)

0. 000)

0. 208)

1. 354*

1. 116***

1. 142**

0. 016)

0. 008)

0. 012)

1. 3820. 000)

0.890*-

1. 169*

0. 022)

0. 026)

0*

0. 001)

0. 957

0. 930

0. 305)

0. 191)

N

1057

614

1071

622

1057

620

F- statistic

20.22

29. 35

10.

54

24. 15

3. 62

7.91

Prob>

0. 000

0. 000

0. 000

0. 000

0.003

0. 000

F

Note: Three survey round binary indicators instead of four( five minus one to prevent perfect collinearity) because the lagged dependent variable is perfectly collinear with the survey round binary indicators as well. The estimated coefficients on Intervention are the estimates of treatment impact. Coefficients are reported as odds ratios; p- values in parentheses.* p<. 10. p<. 05. p<. 01.

Table

12

randomized

Analysis

Covariance( ANCOVA) for

of

Variable

Lagged

Units

outcome

the

whole

(

sample.

1)

results, the analytical methods utilized here increase the external

1. o75***

1. o86***

validity of the study. Many impact evaluations are highly focused

Years

1. 117**

0. 002)

(

(

Survey

round

3

Binary

0. 738***

Survey

round

4

Binary

1. 317#**

(

near

with

the survey

Intervention ratios;

p-

are

the

in

values

the

1693 21.

0. 000

0. 000

as

well.

The

to the impairment

of external

of impact evaluations.

Conclusion

0. 585)

The results

0. 937*

(

minus

estimated are

suggest that the impact of interventions

to improve

status of children depends on the initial agroecological

conditions

in

the

geographic

region

concern,

child dietary quality, even when this outcome is not a specific goal

0. 000

of the program.

to

precolli-

coefficients

reported

as

on

odds

parentheses.

Our results demonstrate

through which community

tions can improve the collection

through

nutrition.

of longitudinal

affect

that there are multiple

and agricultural

interven-

Further work is needed, especially data, to investigate

which community- level development

as the one studied here operate. A potential

p<. 10.

may

that

7. 66

one

interventions

and

1677

pathways

nutrition- sensitive

of

community- level,

is perfectly

variable

nutritional

0.057)

78

impact. Coefficients

treatment

of

sometimes

between these two key characteristics

0.004)

(

four( five

of

lagged dependent

validity,

0. 978

0. 001)

1671

binary indicators

round

estimates

(

0. 000)

28. 14

binary indicators instead

collinearity) because

0.004)

1. 120-

(

on internal

validity. With this study design, we aimed to improve the balance

1. 124`*

0. 001)

(

0.000)

F

(

145-

0. 858***

0.001)

statistic

0. 001) 1. 070-

0. 072) 1.

572-*

0.000)

N

(

1. 041*

0.034)

round

participa-

1.

1.

perfect

to maximize

Lag

Binary

vent

attempt

endogenous

ASF

2

Note: Three survey

trials which

the potentially

MDD

165***

round

Prob>

field

DDS

Survey

F-

controlled

validity. Although

tion levels are a threat to internal validity and may have inflated

3)

2)

0. 000) Intervention

internal

the mechanisms

interventions key extension

such of this

study would be to collect data for and estimate the Women' s

p< os p<. 01.

Empowerment

in

Agriculture

Index (

Alkire

et

al.,

2013).

More

detailed measurement of micronutrient intake for children would

also be helpful to inform nutrition- sensitive programming. generalizability involvement

within in

communities,

community- level

if

there

intervention

is

endogenous If

activities.

every

household in all participating communities were measured regardless

of participation

attenuate.

up- take

levels in the Heifer programming,

However,

non- universal

of programming

participation

activities

more

and

could

results

This study was supported

endogenous

accurately

Sources of funding

reflects

and funded by Heifer International.

the Conflicts of interest

real- world.

We would expect better external validity for this interventionrandomized

and administered

at

the

community- level-

than

for

a

The authors do not have any conflict of interest to declare.

intervention which was randomized and administered at the individual or

household- level-

because

typically target communities the

expense

eligible

interventions ferent

and

logistical

participants such as

contexts,

for this

is less

a

national

policies

and programs

or groups and not individuals,

challenge

of means-

program.

Evaluation

one, which

common

than

could

be

testing and finding of broad, holistic

adapted

evaluations

of

to

many dif-

highly-

Authors' contributions

due to

specific

L.M., M. L., and N.J. designed the study. L.M., N.J., M. L., P. S., and S. M. supervised

tion activities. data

analysis,

the field work and data collection, and the interven-

L. M. managed A.D. N.

and

data cleaning;

L. M. drafted the

A. D. N. conducted

manuscript;

L.M.

the

and

B.

Darrouzet- Nardi

A. F.

R.

consulted

to the

on statistical

manuscript

and

development

model

all

authors

and

it for

approved

made

et al./

revisions

Food

Policy

2016) 185-

61(

Jayachandran,

Approving

ethics

Rev. 1, 77- 100. R., 2013.

Why

are Indian

Children

Shorter

than African

Children? Department of Economics, Northwestern University, Mimeo. Kipnis, Victor, Subar, Amy F., Midthune, Douglas, Freedman, Laurence S., Ballard-

committee

was

study

Develop.

S., Pande,

Barbash,

Rachel,

Schatzkin,

This

197

Gupta, M. D., 1987. Selective discrimination against female children in rural Punjab, India. Popul.

submission.

197

by

approved

the Nepal

Health

Council

Research

Carroll,

P.,

Bingham, Sheila,

Raymond

Schoeller, Dale A.,

2003. Structure of dietary measurement error: results of the OPEN biomarker study. Am. J. Epidemiol. 158(

NHRC).

Troiano, Richard

Arthur,

1),

J.,

14- 21.

LRMP, 1986. Land Utilization Report. His Majesty' s Government of Nepal/ Government of Canada. Kenting Earth Sciences Limited. Malapit, Hazel Jean L, Quisumbing, Agnes R., 2015. What dimensions of women' s empowerment in agriculture matter for nutrition in Ghana? Food Policy 52,

Acknowledgements

54- 63.

The

authors

acknowledge

referees

for their valuable

families

who

Heifer field

Lab

comments

A. Masters

Malapit, Hazel Jean L, Kadiyala, Suneetha, Quisumbing, Agnes R., Cunningham,

anonymous

and suggestions,

to all the

intervention in Nepal,

all

field

enumerators

Tufts

University,

Profes-

and office staff,

Kenaa,

the Nutrition

Innovation

of Tufts University, and Rupa Koirala Health Sciences in, Dharan, Nepal. All remaining errors

Singh

the B. P.

Insti-

of

are

Tyagi,

Parul,

2015.

Women' s

mitigates

the

negative

Develop.

Stud. 51(

8), 1097-

1123.

Maleta, K., Virtanen, S. M., Espo, M., et al., 2003. Seasonality of growth and the relationship between weight and height gain in children under three years of in rural

Malawi.

Acta

Paediatr.

92(

4), 491-

497.

McKenzie, D., 2012. Beyond baseline and follow- up: the case for more T in experiments. J. Dev. Econ. 99( 2), 210- 221. Messer, E.,

1997. Intra- household

findings and understandings-

our own.

empowerment

effects of low production diversity on maternal and child nutrition in Nepal. J.

age

of

Dr.

at

tute

to two

grateful

Preeti Subbha, Deepak Thapa, Nepal Technical

staff,

William

are

in the Heifer

participated

Assistance Group( NTAG) sor

and

allocation of food and health care:

current

introduction. Soc. Sci. Med. 44( 11), 1675- 1684.

Metz, John J., 1989. A framework for classifying subsistence production types of Nepal.

Human

Ecol.

17(

2),

147-

176.

Miller, B. D., 1997. Social class, gender and intrahousehold

References

food allocations to

children in South Asia. Soc. Sci. Med. 44( 11), 1685- 1695. Miller, L. C., Joshi, N., Lohani, M., Rogers, B., Loraditch, M., Houser, R., Singh, P.,

Aguayo,

Victor

nutrition

11),

M., Sharma,

e665-

Alderman,

Hoddinott, J.,

H.,

women'

Arimond

Women'

s

Nutrition

in

Delivering Lancet

essential

Glob. Health

3

A.,

of

consequences

early

474.

Quisumbing, A., Seymour, index. World Dev.

G., Vaz, A., 2013. 31(

agriculture

diverse, Arimond,

Assistance Doris,

10),

poor

Ruel,

52),

71- 91.

et

2010.

al.,

food

E.,

Mullainathan, S.,

Krasevec,

FitzGerald,

J.,

interventions

in

2004.

S.,

2004.

improving

How

Econ.

Ql

estimates?

119(

A

much

1),

in 5

child

Nutr.

trust

we

249- 275.

review

nutrition

should

the

of

Health

of

Nutr.

05), 599- 609.

7( Bezner

R.,

Kerr,

and

Berti,

nutrition

Health Bhutta,

Nutr.

14(

maternal

Parul,

and

De

Lancet Carletto,

child

Onis,

nutritional

Assessment ah869e00.

Population

HTM(

(

of a

agriculture

participatory

in

growth

Malawi.

northern

Public

et

2008.

al., and

Susan

Majid,

371(

P., Bhutta,

et

and

9610),

417-

Maternal

and

income

middle-

440.

child

countries.

427- 451.

introduction Organization, to

A., 2015. special

2007.

Special

Nepal

Organization

Dataset).

Report:

United

http:// data. fao. org/

<

html? version=

1. 0>(

Stud.

FAO/ WFP org/

improved 7, 8- 9.

Food

docrep/

ah869e/

Latest

Nations. GLIPHA. Livestock

ref/

21

4813cbb2-

21

update:

Nov

S., Commins,

Laurence

Kipnis,

Victor,

validation

studies and

energy

Headey,

Derek

D.,

John

Midthune, of

dietary

protein

2013.

M., Moler, James

Douglas, self- report

intake.

Am.

J.

Developmental

et

f0- 4f6c-

2013,

aebe-

Accessed

instruments

21

Pooled

using

of

Girard,

production

young

J. L.,

McAuliffe,

strategies

children:

5

a

on

C.,

the

systematic

et

health review.

al.,

2012.

and

The

change:

nutrition

Paedia.

Perinat.

31

accessed

effects

a

cross-

of

May 2014).

household food

outcomes

Epidemiol.

of

women

26(

sl),

and

205-

222. Gittelsohn,

35(

3), 312-

326.

pattern but not in energy and nutrient intakes of rural Beninese school- aged

Prentice,

11),

3941S-

Ross

3932S-

3935S.

3949S.

L,

Mossavar- Rahmani,

Yasmin,

Huang,

Ying,

Van

Horn,

Linda,

Beresford, Shirley A.A., Caan, Bette, Tinker, Lesley, et al., 2011. Evaluation and comparison of food records, recalls, and frequencies for energy and protein assessment by using recovery biomarkers. Am. J. Epidemiol.( kwr140) International' s

livestock

donation

programs

in

Rwanda

on

nutritional

outcomes. Food Policy 44, 202- 213. Ruel, M. T., 2003. Operationalizing dietary diversity: a review of measurement issues and research priorities. J. Nutr. 133( 11), 39115- 39265. Ruel, Marie T., Alderman,

Harold,

2013.

Maternal

and child nutrition

study group.

Nutrition- sensitive interventions and programmes: how can they help to progress and child nutrition?". Lancet 382 in improving maternal

accelerate

Satzinger, F., Bezner Kerr, R., Shumba, L., 2009. Intergenerational participatory discussion

groups

foster

knowledge

exchange

to

improve

child

nutrition

and

food security in northern Malawi. Ecol. Food Nutr. 48( 5), 369- 382. Savy, M., Martin- Prevel, Y., Traissac, P., et al., 2006. Dietary diversity scores and nutritional status of women change during the seasonal food shortage in rural Faso. J. Nutr. 136(

10), 2625- 2632.

Smale, Melinda, Moursi, Mourad, Birol, Ekin, 2015. How does adopting hybrid maize affect dietary diversity on family farms? Micro- evidence from Zambia. Food Policy 52, 44- 53. StataCorp, 2011. Stata Statistical Software: Release 12. StataCorp LP, College Station, TX.

Steyn, N., Nel, J., Nantel, G., et al., 2005. Food variety and dietary diversity scores in children: are they good indicators of dietary adequacy? Public Health Nutr. 9 5), 644- 650.

Strauss, J., Thomas, D., 2007. Chapter 54 health over the life course. In: Paul Schultz, T., Strauss

John,

A.( Eds.),

Handbook

of Development

Economics,

vol. 4. Elsevier,

United

Nations,

2015.

Sustainable

Development

Knowledge

Platform.

https:// sustainabledevelopment.un.org/ topics>( Accessed 27 October 2015). Wooldridge, J. M., 2002. Econometric Analysis of Cross Section and Panel Data. The MIT

Press.

World Bank Group, 2012. World Development Indicators. World Bank Publications. World Bank Group, 2015. Data Catalog. < http:// data. worldbank. org/ indicator/ NY. GDP. PCAP. CD? page= 3>( last accessed 5 April 2016).

J.,

Thapa,

micronutrient

1739-

bull.

pp. 3375- 3474.

World

analysis.

A. W., Self,

from

biomarkers

recovery

nutritional

Develop. 42, 76- 88. Heifer International Nepal.< http:// www. heifernepal. org/>( country

Baer, David

results

kwu116)

Epidemiol.(

drivers

E., Arab, Lenore,

2014.

al.,

Nutr.

children. Public Health Nutr. 12( 03), 414- 422.

Burkina

Security

010/

2013).

the

of

to

pathways

issue. J. Develop.

http:// www. fao.

22 November

accessed

Farm- level

to the

Food

9891), 536- 551.

Christian,

A.,

Zulfiqar

2013.

al.,

in low- income

Interventions for

works?

Lancet

survival.

Walker,

Ezzati,

What

2013).

Freedman,

for

Cesar G.,

Mission

8855b340dadd.

J.,

R. E.,

overweight

Agriculture

and

Black,

M., Winters, P., Zezza,

status:

Effects

child

1472.

Mercedes, and

Agriculture

and

on

undernutrition

9890),

382(

G., Ruel,

Nov

T.,

L, 2011.

project

1466-

E., Victora,

undernutrition

Food

8),

Z. A., Ahmed,

Black, Robert

Shumba,

P. R.,

education

health.

Rawlins, R., Pimkina, S., Barrett, C. B., et al., 2014. Got milk? The impact of Heifer

effectiveness

Public

outcomes.

and

Maryland.

11),

2004.

evidence

growth

Neumann, C., Bwibo, N., Murphy, S., et al., 2003. Animal source foods improve dietary quality, micronutrient status, growth and cognitive function in Kenyan school children: background, study design, and baseline findings. J. Nutr. 133

20695.

20595-

child

Murphy, S. P., Allen, L.H., 2003. Nutritional importance of animal source foods. J.

diets

s

on

Daniels,

Simple

of women'

adequacy

11),

Effects

Food

2585.

Duflo,

agriculture

140(

Nepal:

Ministry of Health and Population( MOHP)[ Nepal], New ERA, and ICF International Inc., 2012. Nepal Demographic and Health Survey 2011. Kathmandu, Nepal: Ministry of Health and Population, New ERA, and ICF International, Calverton,

Nutr. 133( Alicia,

Carriquiry,

Nadia,

Fogny,

Nutr.

DC:

Mahato, S., 2014. Community development and livestock promotion in rural

Mitchikpe, C. E., Dossa, R.A., Ategbo, E. A., et al., 2009. Seasonal variation in food

FHI 360.

Elodie,

micronutrient

J.

Washington,

Dietary diversity is associated with from 11 demographic and health surveys. J.

T.,

differences- in- differences P. R.,

II Project,

Fanou-

settings.

Site.

Bangladesh

Becquey,

predict

Marie

status:

2579-

M.,

Bertrand,

Megan,

indicators

resource-

nutritional

Rural

Technical

Deitchler,

Mary,

from

Results

Wiesmann,

diversity

group

Food

Econ.

Dick, R., Peterman,

Diets:

s

Mary, Melissa C.,

Berti,

2015.

Raj,

earthquake.

2006. Long term Pap. 58( 3), 450-

B.,

Kinsey,

empowerment

Arimond,

134(

Giri

the Nepal

Mary, Liv Elin Torheim, Doris Wiesmann, Maria Joseph, Alicia Carriquiry, Dietary Diversity as a Measure of the Micronutrient Adequacy of

2009. and

Subedi,

after

Oxford

malnutrition.

S., Meinzen-

The

Anirudra,

children

e666.

childhood

Alkire,

for

services

1749.

M., Landman,

sufficiency

in

L. T.,

rural

1997.

Nepali

Cultural

factors,

households.

Soc.

caloric

Sci.

intake

Med.

and

44 ( 11),

World Health Organization, Feeding

Practices.

World

2010. Indicators Health

for Assessing Infant and Young Child

Organization,

Geneva.