52
Promoting Improved Infant and Young Child Feeding: Evidence from a Field Experiment in Ethiopia * Seollee Park , Yaeeun Han , and Hyuncheol Bryant Kim § September 2018 [PRELIMINARY AND INCOMPLETE: PLEASE DO NOT CITE OR CIRCULATE] Please find the latest version HERE. Abstract Unhealthy diet with very low dietary diversity is prevalent among young children in the developing world, which increases the risk of chronic undernutrition. Such suboptimal child-feeding behaviors are partly due to mothers’ lack of nutritional knowledge, income, or both. This study implemented a clustered randomized experiment in Ethiopia to examine the effects of nutrition behavior change communication (BCC) and food vouchers on child-feeding behaviors and child growth. We find that BCC increases knowledge on infant and young child feeding and improves children’s dietary quality significantly, while food vouchers alone do not. The impacts on improved child-feeding behaviors are largest for those treated by both BCC and food vouchers, and we find evidence for stunting reduction in this group. Keywords: infant and child nutrition, nutrition education, behavior change communication, food vouchers, cluster randomized control trial, Ethiopia JEL classification: I12, I26, J22, O12, O15 * We wish to thank John Hoddinott, as well as seminar participants at Cornell University for their in- valuable feedback. We also thank Jieun Kim, Yong Hyun Nam, Minah Kim, Hyolim Kang, Jiwon Baek, Tembi Williams, Soo Sun You, and Jeong Hyun Oh at Africa Future Foundation (AFF) for their excellent fieldwork, and Rahel Getachew, Chulsoo Kim, Hongryang Moon at Myungsung Christian Medical Center and director Hyun Jin Song for their support. This project was supported by AFF, Korea Foundation for International Healthcare (KOFIH), Seoul Women’s Hospital Bucheon Branch, and Dr. Taehoon Kim. All views expressed are ours, and all errors are our own. The study was approved by ethical review committees at the Oromia Health Bureau (Ethiopia, BEIO/AHBHN/1-8/2670), Myungsung Medical College (Ethiopia), and Cornell University (USA, 1612006823). Dyson School of Applied Economics and Management, Cornell University Division of Nutritional Sciences, Cornell University § Department of Policy Analysis and Management, Cornell University 1

Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

  • Upload
    others

  • View
    5

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Promoting Improved Infant and Young Child Feeding:Evidence from a Field Experiment in Ethiopia∗

Seollee Park†, Yaeeun Han‡, and Hyuncheol Bryant Kim§

September 2018

[PRELIMINARY AND INCOMPLETE: PLEASE DO NOT CITE OR CIRCULATE]

Please find the latest version HERE.

Abstract

Unhealthy diet with very low dietary diversity is prevalent among young children in the developingworld, which increases the risk of chronic undernutrition. Such suboptimal child-feeding behaviorsare partly due to mothers’ lack of nutritional knowledge, income, or both. This study implementeda clustered randomized experiment in Ethiopia to examine the effects of nutrition behavior changecommunication (BCC) and food vouchers on child-feeding behaviors and child growth. We find thatBCC increases knowledge on infant and young child feeding and improves children’s dietary qualitysignificantly, while food vouchers alone do not. The impacts on improved child-feeding behaviorsare largest for those treated by both BCC and food vouchers, and we find evidence for stuntingreduction in this group.

Keywords: infant and child nutrition, nutrition education, behavior change communication, foodvouchers, cluster randomized control trial, Ethiopia

JEL classification: I12, I26, J22, O12, O15

∗We wish to thank John Hoddinott, as well as seminar participants at Cornell University for their in-valuable feedback. We also thank Jieun Kim, Yong Hyun Nam, Minah Kim, Hyolim Kang, Jiwon Baek,Tembi Williams, Soo Sun You, and Jeong Hyun Oh at Africa Future Foundation (AFF) for their excellentfieldwork, and Rahel Getachew, Chulsoo Kim, Hongryang Moon at Myungsung Christian Medical Centerand director Hyun Jin Song for their support. This project was supported by AFF, Korea Foundation forInternational Healthcare (KOFIH), Seoul Women’s Hospital Bucheon Branch, and Dr. Taehoon Kim. Allviews expressed are ours, and all errors are our own. The study was approved by ethical review committeesat the Oromia Health Bureau (Ethiopia, BEIO/AHBHN/1-8/2670), Myungsung Medical College (Ethiopia),and Cornell University (USA, 1612006823).†Dyson School of Applied Economics and Management, Cornell University‡Division of Nutritional Sciences, Cornell University§Department of Policy Analysis and Management, Cornell University

1

Page 2: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

1. Introduction

Economists are often concerned about the causes and consequences of health and nutrition.

In developing countries, nutritional status is a critical component of health, especially for

children under 2. A particular focus is on chronic child undernutrition, as it is linked to

nearly half of all deaths in children under 5 and affects more than 150 million young children

(World Bank 2017). Chronic undernutrition during infancy and young childhood leads to

poorer health, education, and labor outcomes in adulthood (T. Ahmed et al. 2012; Avula

et al. 2013; Barrett 2010; Black et al. 2008; Hoddinott, Behrman, et al. 2013; UNFAO 2006).

The high prevalence of chronic child undernutrition in many developing countries, char-

acterized by high rates of stunted linear growth, has spawned a large body of research into

possible explanations. This includes genetic predispositions (Nube 2009), poor quality di-

ets and food systems (Headey et al. 2012), intrahousehold biases (Jayachandran and Pande

2017; Pande 2003), low status of women (Schroff et al. 2009; Menon 2012), and the inefficacy

of nutritional programs and strategies (Gupta et al. 2005, World Bank 2006).

Substantial research has sought to address these causes, mostly focusing on a single

aspect and on immediate causes of undernutrition. Interventions that address a single aspect

of undernutrition such as micronutrient deficiencies (Muller et al. 2003; Newton et al. 2007;

Merwe et al. 2013), lack of knowledge (Prina and Royer 2014), and lack of income (Manley et

al. 2013) have often shown limited success. Moreover, it is estimated that the interventions

addressing the immediate, one-dimensional causes, at nearly full coverage, would reduce

stunting by only 20% (Bhutta et al. 2013; Ruel et al. 2013). Hence, while there is a good

understanding of the causes of chronic child undernutrition, less is known about what will

accelerate its reduction. This generates the question as to whether interventions that address

multiple constraints simultaneously would be more effective in reducing undernutrition.

In this paper, we study ways to improve infant and young child nutrition by investigat-

2

Page 3: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

ing the roles of knowledge and income in changing mothers’ child-feeding practices. To do so,

we implement a community-based cluster randomized experiment in Ethiopia that provides

nutrition education in the form of behavioral change communication (BCC) and food vouch-

ers. Specifically, we explore the effects of releasing the knowledge constraint, the income

constraint, or both constraints by randomly providing four-month-long BCC only (BCC),

voucher only (V oucher), and both BCC and voucher (BCC + V oucher) interventions for

mothers with one or more children between 4 and 20 months of age. We target this age

range because stunting prevalence increases rapidly after the first 6 months because exclu-

sive breastfeeding no longer meets the energy and nutrients needed for rapid child growth,

while relatively low in the first 6 months due to breastfeeding (Cunningham et al. 1991;

Dewey et al. 1995; Beaudry et al. 1995). Adopting healthy child-feeding practices during the

transitional period from exclusive breastfeeding to complementary feeding1—i.e., 6 months

after birth—is particularly crucial for preventing undernutrition (Black et al. 2008; Jones

et al. 2003; Ruel et al. 2013).

This research contributes to several strands of literature. First, it contributes to the

empirical literature on the effects of nutrition education or BCC for caregivers on caregivers’

nutritional knowledge, child-feeding behaviors, and child growth. Secondly, this research

adds to the literature on the role of vouchers in improving child nutrition in developing

countries. Third, we contribute to the growing literature on the effectiveness of multifaceted

programs on poverty reduction. To our knowledge, this study is the first study to examine

the combined effects of providing both BCC and vouchers on child-feeding behaviors and

child nutrition.1Appropriate complementary feeding means feeding children a diverse diet that meets the nutritional

requirements. This entails feeding vitamin A-rich fruits and vegetables daily, in addition to a range of otherfruits and vegetables. Meat, poultry, fish, or eggs also need to be consumed daily to ensure the intake ofcertain micronutrients critical for growth found only in animal source foods. In this regard, healthy food inthis paper refers to these food groups (WHO 2010a).

3

Page 4: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Our study finds that BCC significantly improves mothers’ nutritional knowledge while

voucher alone does not. This is consistent with the growing literature that examines the

causal effect of BCC on mothers’ nutritional knowledge. A recent experimental study con-

ducted in Bangladesh shows that a 2-year-long BCC resulted in answering about four more

questions correctly at endline, out of 18 infant and young child feeding (IYCF) questions,

than the control group (Hoddinott, I. Ahmed, et al. 2017). A similar 2-year-long experiment

in Burkina Faso finds that, out of three IYCF questions, BCC significantly increases the

likelihood of answering two questions correctly by more than 15 percentage points (Olney

et al. 2015). Another experiment in Malawi, which examines the effect of a nutrition BCC

that visits mothers five times over the period of just before giving birth and 5 months after,

shows a somewhat weaker effect on knowledge, with only one out of seven IYCF questions

more likely to be answered correctly by 7 percentage points (Fitzsimons et al. 2016). Our

study finds that BCC causes mothers to answer approximately two more questions correctly

for both BCC and BCC + V oucher groups out of 34 questions. In terms of proportion of

mothers answering correctly, we find that 17 questions are significantly more likely to be

answered correctly by the BCC and the BCC + V oucher groups. Hence, we show that

a similar or greater impact on mothers’ knowledge can be attained with a relatively short

treatment length.

On child-feeding behaviors, BCC has positive impacts on a number of child diet quality

measures. The impact is greatest when BCC is combined with vouchers. However, voucher

alone does not increase child diet quality on any standard child diet quality measure we

examine. As our primary measure of diet quality, we find that child dietary diversity score

(CDDS) increases by 0.3 and 0.6 food group for BCC and BCC+V oucher, respectively. The

impacts on CDDS are driven by significant increases in the proportion of children consuming

animal source foods among the BCC group, and animal source foods, vitamin A-rich fruits

4

Page 5: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

and vegetables, and nuts and legumes for the BCC+V oucher group. Furthermore, children

in the BCC + V oucher and the BCC groups were 15.2 and 7.7 percentage points more

likely to meet the World Health Organization (WHO) guidelines for minimum acceptable

diet, respectively. The BCC + V oucher group was also 17.6 percentage points more likely

to meet the minimum dietary diversity. The limited existing literature on the effects of BCC

or nutrition education on child-feeding practices find more moderate effects. For example,

Fitzsimons et al. (2016) find that only one out of eight food groups had a significant increase

in the proportion of children consuming the food group due to BCC. Olney et al. (2015) show

that BCC combined with agriculture input support and training increases the proportion of

children meeting minimum dietary diversity by 12.6 percentage points, but do not report

results on other measures. A similar study that evaluates the impact of a nutrition education

program coupled with agricultural intervention finds a comparable effect, with an increase

in CDDS by 0.52 food group (Reinbott et al. 2016). In addition, we find improvements in

diet quality at the overall household level, which is in line with the findings of Fizsimons et

al. (2016).

In terms of physical growth of children, we find evidence for chronic undernutrition

reduction by 9.5 percentage points in the BCC + V oucher group, as measured by stunting

prevalence. We do not find significant effects for other groups. This is a striking result

given a 6-month lapse between baseline and endline, on average. Other studies have yet

to find evidence on the causal effect of BCC or BCC combined with other interventions on

stunting reduction. Fitzsimons et al. (2016) find that BCC decreases wasting prevalence

by 4.2 percentage points but is significant at the 10% level, and do not show evidence for

stunting reduction. Other studies that coupled BCC with agricultural interventions do not

find evidence for stunting reduction (ibid.; Olney et al. 2015). We further show that there are

varying distributional effects by treatment arm, which confer important policy implications.

5

Page 6: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

While BCC + V oucher is effective for improving growth outcomes among undernourished

children, BCC works favorably for those who are relatively well-nourished. This suggests

that social protection programs that target undernutrition reduction may be more effective

when providing both knowledge and resources simultaneously.

Lastly, we add to the growing literature on the effectiveness of multifaceted programs

on poverty reduction, shifting the focus from poverty to undernutrition reduction. This lit-

erature suggests that a multipronged approach, often combining skills training or education

with income support and thereby releasing multiple constraints simultaneously, brings sus-

tainable and cost-effective results in reducing poverty (Banerjee et al. 2015; Bandiera et al.

2017). Consistent with this literature, our results on child-feeding behaviors and growth out-

comes collectively suggest that the impacts are greatest when both the knowledge and the

income constraints are addressed simultaneously through the BCC + V oucher treatment.

This is particularly evident in stunting reduction results, where the BCC and the V oucher

treatments, individually, show no impact but the combined BCC+V oucher treatment shows

a sizable effect.

The remainder of this paper is organized as follows: Section 2 presents the study design

and the intervention programs; Section 3 describes the data and sample characteristics;

Section 4 sets out the methods; and Section 5 presents the results. We discuss the results

and conclude in Section 6.

2. Study Design and the Interventions

2.1. Study Context

Ethiopia is an appropriate setting for this study because it faces daunting chronic child under-

nutrition challenges which coexist with knowledge and income constraints. The prevalence of

6

Page 7: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

stunting in Ethiopia, an indicator for chronic undernutrition, was 38% among children under

five (Ethiopia DHS 2016). The prevalence of stunting rapidly increases after 6 months of age,

largely due to poor infant and young child-feeding practices in Ethiopia. As shown in Figure

1, at the age of six months, 16% of children are stunted in Ethiopia but the corresponding

number increases to a staggering 47% by the age of 24 months (ibid.). Low dietary diversity

is particularly striking among young children in Ethiopia, with only 7% of children aged

6-23 months meeting the minimum acceptable dietary standard which is a WHO standard

for adequate infant and young child nutrition.

As the second-most populous country in Africa, Ethiopia has a GDP per capita of

707USD per year in 2016, which is less than two dollars per person per day (World Bank

2017). There are also many misconceptions about child nutrition among mothers in Ethiopia.

For example, many Ethiopian mothers believe that babies under 12 months should not be

fed animal source foods. Also, it is common to give infants as old as 9 months only thin

gruel, with the misbelief that they are not able to digest solid or semi-solid food. The widely

available and inexpensive healthy food items in the area, such as mangos rich in vitamin

A, are not well recognized. This contextual evidence suggests there are considerable income

and knowledge constraints for child nutrition in Ethiopia.

2.2. Study Sample Description

Our study area is Ejere district (woreda) located in the Oromia region of central Ethiopia,

approximately 50 km west of the capital, Addis Ababa. Ejere is primarily a rural district

which is further subdivided into three urban and 27 rural wards (kebeles). Ejere has a

population of around 112,000 spread over these 30 wards, who are predominantly engaged in

mixed crop-livestock farming at a small scale. Most farmers engage in traditional practices

of rain-fed subsistence agriculture.

7

Page 8: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

The study sample is mothers with at lease one child aged between 4 and 20 months

(‘eligible group’ hereafter).2 Pregnant women and mothers with a child with less than 4

months at the time of baseline survey were additionally included in the surveys to measure

potential spillover of the knowledge (‘spillover group’ hereafter).

2.3. Experimental Design

This study is based on a cluster randomized control trial. The clusters were randomly selected

from three urban and three randomly selected rural wards in Ejere (Figure 2). A total of 79

villages (garees) from these six wards in Ejere entered a lottery, and were randomly selected

into one of four arms: BCC only (BCC), vouchers only (V oucher), BCC and vouchers

(BCC + V oucher), and the control group. Randomization was stratified by wards. There

are 101 (15), 96 (14), 154 (13), and 289 (37) mother and child pairs (villages) randomly

assigned to the BCC, V oucher, BCC + V oucher, and control study groups, respectively.

We also include pregnant women and women with children under 4 months in the same

villages to study IYCF knowledge spillovers (spillover group). The corresponding numbers

for the spillover group are 86, 54, 97, and 107 mother and child pairs, respectively. Figure 3

summarizes the study design.

2.4. Interventions

There are two interventions in this study: nutrition BCC and food vouchers. The inter-

ventions were designed by the study team through a series of focus group discussions and

pilot-testing as shown in study timeline (Figure 4), and implemented in collaboration with2 We selected the age range between 4 and 20 months as the treatment eligibility criteria in order to target

the age range that is most susceptible to undernutrition due to malpractices in child-feeding. In particular,we seek to address chronic undernutrition caused by suboptimal practices in complementary feeding, whichstarts at 6 months of age. We do not include children under 4 months because the BCC intervention doesnot address breastfeeding practices.

8

Page 9: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Africa Future Foundation (AFF), a local non-governmental organization.

BCC.3 The BCC treatment offered weekly IYCF information sessions for 16 weeks to a

group of seven to fourteen mothers who has a child from 4 to 20 months of age at baseline.4

All eligible mothers living in the same village formed a group to receive BCC education. The

BCC treatment was a multifaceted and interactive information intervention complemented

by various participatory learning methods including weekly sharing of mothers’ experiences

applying new IYCF activities, videos and visual images, role-plays, and cooking sessions.

The focus of the BCC sessions and supporting activities was on the need to increase dietary

diversity of children aged 6-23 months, with an emphasis on animal source foods and vitamin

A-rich fruits and vegetables, appropriate feeding amounts and frequency, and feeding and

caregiving practices. Each session ended with an action plan the mothers agreed upon, and

the proceeding session reviewed and discussed past week’s action plans. In addition, the

BCC participants also received a small handbook containing a summary of IYCF contents

and weekly action plans based on contents learned each week, and a self-check diary. An

overview of the BCC curriculum is provided in Appendix B.

The BCC facilitators consisted of local female community workers who had been work-

ing in the community as AFF social workers for at least six months up to five years. The

selected BCC facilitators went through three rounds of training, passed regular knowledge3BCC is the strategic use of communication to promote positive health outcomes, based on proven

theories and models of behavior change. BCC employs a systematic process beginning with formativeresearch and behavior analysis, followed by communication planning, implementation, and monitoring andevaluation. Audiences are carefully segmented, messages and materials are pre-tested, and mass media(which include radio, television, billboards, print material, internet), interpersonal channels (such as client-provider interaction, group presentations) and community mobilisation are used to achieve defined behavioralobjectives (MEASURE Evaluation 2018).

4 The BCC program curriculum is developed based on the Alive & Thrive’s BCC program implementedin Ethiopia. Alive & Thrive is an initiative to save lives, prevent illness, and ensure healthy growth anddevelopment through the promotion and support of optimal maternal nutrition, breastfeeding, and comple-mentary feeding practices. Alive & Thrive has worked in Ethiopia since late 2009 to address widespreadand limited recognition of the long-term consequences of stunting and find ways to reach mothers (Alive &Thrive, 2018).

9

Page 10: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

tests, and implemented two pilot programs in both urban and rural areas. Each group had

two designated facilitators—one leader and one helper. The lead facilitator taught the ses-

sions and led discussions and role-plays, while the other facilitator helped by encouraging

discussion and assisting illiterate mothers. The sessions were conducted at the ward office

or health posts. Throughout the study, two supervisors randomly visited the BCC ses-

sions for quality control. The BCC facilitators, supervisors, and the study team had weekly

group meetings to discuss progress and challenges. The supervisors also made home visits

to mothers who missed more than two consecutive sessions to encourage attendance.

Food vouchers. The voucher treatment provided food vouchers of 200 ETB (approximately

10 USD) per month for four months to either mother or father of the household, which could

be used at nearby markets.5 This amount is similar to the cash or food transfers amount

of Ethiopia’s Productive Safety Net Program which was set to be about 8.5 USD at the

time of the program design (MOA 2014). Also, we provide food vouchers instead of cash or

food because food vouchers are proven to be most effective in improving dietary diversity

(Hidrobo et al. 2014).

Food vouchers were redeemable for any kind of food items sold at the market including

cereals, roots and tubers, fruits, vegetables, legumes, meat and fish, milk products, eggs,

and spices. Food vouchers were distributed every four weeks at the nearest market or at the

participant’s household if not picked up from the market. Vouchers were given in denomi-

nations of 5, 10, and 20 ETB to facilitate small transactions, and could be used over a series

of visits per month. Vouchers were required to be redeemed within expiration date noted

in the voucher (four weeks). At the first disbursement, voucher recipients were provided5 The voucher constituted approximately 10% of household food expenditure. As there is one public

market per each of the six kebeles in our study area, the vouchers could be used at any of these six publicmarkets.

10

Page 11: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

detailed instructions on how to use the vouchers. Unique household identification numbers

and names were printed on the voucher and the study participants were required to present

household photo IDs, provided by the study team, to redeem the vouchers. On all market

days of the study period, our voucher staff visited the market to facilitate transactions and

recorded voucher-based transactions.6

3. Data

3.1. Data Sources

The primary data sources are (1) census data including household demographic and socioe-

conomic information, (2) the baseline and follow-up surveys, and (3) administrative data

collected during the intervention including BCC attendance rates and voucher usage records.

The timeline of the data collection and interventions is summarized in Figure 4.

AFF conducted a census of all households in 22 wards of Ejere in May-September 2016,

covering approximately 22,000 households.7 The census collected a variety of demographic,

socioeconomic, and health variables such as age of mother and children, marital status,

education and employment, household asset, and birth history of mother. Using the census

data, we randomly selected three rural wards and selected all three urban wards, which

consisted of 108 villages in total. From these villages, we randomly selected 79 villages to be

included in this study. These villages had a total of 640 eligible mother and child pairs, all

of which were included in the study for the treatment and control groups, and 344 mother

and child pairs for the spillover group.6 The vouchers were redeemable in all major markets in the study area. The voucher staff were stationed

at the market on all market days during the study period, and followed the voucher-holders to record eachtransaction whenever they visited the market.

7 Out of 30 wards in Ejere, 8 wards in the southern part of the district were excluded from the census dueto security reasons. There were strong anti-government sentiments in this region which spread to hostilitytoward NGOs and surveyors.

11

Page 12: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

The baseline survey was conducted in April-August 2017 before the intervention pro-

gram began. The follow-up survey was conducted upon program completion in December-

March 2018, about 6 months after the baseline survey. Both the baseline and the follow-up

questionnaires include detailed information on IYCF knowledge and practices, child food

consumption, household food and non-food expenditures, health, gender, social networks,

anthropometry, demographics, and socioeconomic information. The follow-up survey also

has a section on the mothers’ experience with the program.

During the baseline and follow-up surveys, we asked study participants to list up to ten

closest friends (including relatives) within the ward (kebele). Using this social network data,

we construct BCC peer variables including whether the mother has any BCC-participating

friend and the number of BCC-participating friends by cross-referencing the networks with

BCC participants and vice versa.8 At baseline, 55% of BCC treatment mothers and 36%

of spillover group mothers had at least one BCC-participating friend, defined either by own

network or the other person’s network.

In addition, our study team collected administrative data on BCC attendance and

voucher usage during the intervention from August 2017 to February 2018. On voucher

usage, the voucher staff collected information on the type of food item, the quantity bought,

and the amount spent using the vouchers. It confirms that about most (94%) of the voucher

participants utilized the vouchers to buy food, and 77% of face value of the voucher had been

redeemed (on average 153 birr out of 200 birr). AFF staff also collected GIS information on

village and markets in the catchment area. Administrative data confirm that the mothers8 The social network was created from the network module in the baseline and follow-up surveys asking

the respondent to list the top 10 closest friends living in the same ward. Matching was done initially bymatching phone numbers, then by matching the friends’ names with survey respondent and spouse namesusing the similarity score generated by the ‘matchit’ command in Stata. Name matches with similarity scoreabove 0.6, out of a range from 0 to 1, were manually compared across name, spouse name, sex, ward, andphone number to confirm the match. Manual confirmation was necessary due to inconsistent spelling ofAmharic and Oromo names.

12

Page 13: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

attended the weekly BCC sessions regularly (74% attendance rate).

3.2. Outcome Variables

The primary outcomes for this study are IYCF knowledge scores among mothers and child

dietary diversity score (CDDS). We also constructed other measures of children’s diet quality,

household food and non-food consumption and expenditures, measures of household diet

quality, and anthropometric measures of child development.

We use standardized mothers’ IYCF knowledge scores measured in the baseline and

follow-up surveys. The knowledge score is the percentage of questions answered correctly

out of 34 questions. We construct CDDS as a measure of children’s diet quality. The CDDS

sums the number of distinct food groups consumed by the child in the past 24 hours among

the following seven different food groups: cereals, roots and tubers, legumes, nuts and seeds,

dairy products, meat/poultry and fish, eggs, vitamin A-rich fruits and vegetables, other

fruits and vegetables (WHO 2010a; WHO 2010b).

We also look at other measures of child diet quality to capture the proportion of chil-

dren given appropriate diet quality and quantity. These indicators include the minimum

acceptable diet, minimum dietary diversity, and minimum meal frequency standards. The

minimum acceptable diet indicator is created assessing two different IYCF components com-

piled into one index, adjusted for child’s age: minimum dietary diversity and minimum meal

frequency. Minimum dietary diversity is proportion of children who receive food from 4 or

more food group, and minimum meal frequency is the proportion of children who consumed

minimum number of meals appropriate for the age (WHO 2010a; WHO 2010b).9 Minimum9 Minimum dietary diversity is a proxy for adequate micronutrient density of foods. The cut-off of four

food groups is associated with better-quality diets for both breastfed and non-breastfed children. The fourfood groups should come from a list of seven food groups: grains, roots, and tubers; legumes and nuts; dairyproducts (milk yogurt, cheese); flesh foods (meat, fish, poultry, and liver/organ meat); eggs; vitamin A-richfruits and vegetables; and other fruits and vegetables. Minimum meal frequency, a proxy for a child’s energyrequirements, examines the number of times children received foods other than breastmilk. The minimum

13

Page 14: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

acceptable diet differs from CDDS in that it accounts for feeding frequency in addition to

diversity.

To gauge how knowledge changes perception, we also measured mothers’ perception of

relative child growth by asking how the child fares compared to other children of the same

age in terms of diet quantity and quality on a five-tiered scale ranging from very well to very

poor. Additionally, we constructed a variable for timely introduction of complementary food

using a total standardized score aggregating indicator variables for whether the child started

eating a certain food after six months but before 12 months of age across eight different

complementary food items. This outcome measures how well mothers are doing in terms of

introducing various complementary food to their children at appropriate ages—not too early

as to incur digestive problems but not too late so that children are not undernourished. 10

In order to examine household-level expenditure responses to nutritional knowledge

and food vouchers, we calculate household food expenditure by summing the value of food

items purchased in the past seven days by food group or in total. Food items include cereals,

roots and tubers, nuts and legumes, fruits and vegetables, meat and poultry, eggs, milk and

milk products, and spices and condiments. Household non-food expenditure is calculated

from the value of durable items purchased in the last six months and non-durable items

purchased in the last month. All values are converted to weekly per capita values. Non-food

items include clothing, household items, medical costs, educational costs, energy costs, repair

costs, and wedding/funeral costs. All values are converted to weekly per capita values.

To assess how household-level diet quality changes as a result of knowledge and vouch-

ers, we also construct a food consumption score (FCS) which measures household diet quality

in terms of both energy and diversity Weismann et al. 2009.11 FCS less than or equal to 35

number is specific to the age and breastfeeding status of the child (WHO, 2010).10 The complementary food items asked are water or other non-breastmilk liquids, solid or semi-solid food,

meat, eggs, legumes, green vegetables, fruits, and snacks.11 The FCS is calculated by summing the number of days that the household consumed each of the eight

14

Page 15: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

is considered having poor to borderline consumption (WFP 2008).

Lastly, we examine the treatment effect on children’s physical growth using child an-

thropometry which was measured three times during each survey to minimize error. Child

growth outcomes include height-for-age Z scores (HAZ) and stunting, weight-for-height Z

scores (WHZ) and wasting. These were constructed using a mean of the three measurements

for height and weight. HAZ and WHZ are standardized Z scores relative to the WHO refer-

ence population. Stunting and wasting are dummy variables equal to 1 if a child’s HAZ and

WHZ, respectively, are 2 standard deviations (SD) below the WHO reference population.

3.2. Sample Characteristics and Randomization Balance

Table 1 presents the summary statistics for the whole sample (Column 1), the control

group (Column 2), and the difference between each treatment groups and the control group

(Columns 3 to 5) and between treatment groups (Columns 8-10). Panels A, B, and C present

mother, child, and household characteristics at baseline, respectively. Mothers in our sample

are, on average, 28 years old, 14 percent are household heads, 77 percent are Oromos, 84

percent are Orthodox Christians, have approximately 2 children, 77 percent are married, 56

percent have work, 50 percent are able to read, 49 percent are able to write, have about 4

years of schooling, and 45 percent are from rural areas. The mean mother IYCF knowledge

score is 21.5 out of 32 (67 percent) and the mean CDDS is 2.4. Mean age of the eligible

child is approximately 12 months and average household size is 4.5. Only 13 percent of the

sample met the minimum acceptable diet at baseline. The mean HAZ is -1.1 with a 27

percent stunting prevalence. The mean FCS is 43. Average total weekly food and non-food

expenditure per capita are approximately 132 ETB and 43 ETB, respectively.

food groups (staples, pulses, vegetables, fruit, meat and fish, milk and dairy, sugar and honey, oils andfats), multiplying the summed number of days by the food group’s weighted frequencies, and summing theseweighted scores across food groups.

15

Page 16: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Columns 3 to 10 confirm that the randomization was successful, with the sample well

balanced across intervention and control clusters at baseline. Across 150 (25 x 6) difference-

in-means tests, only four differences are statistically significant at the 5% level, suggesting

that the baseline characteristics are balanced overall. With 150 tests being considered, the

probability of rejecting a true null hypothesis for at least one outcome is nearly 100%. As a

robustness check, we also control for these baseline covariates in the analysis.

As shown in Panel D, eligible mothers’ attrition rate at the follow-up survey is 8.4%.

Table 1 shows no significant difference in attrition rates across intervention groups. The

attrition rate of follow-up child anthropometry is 16.6 percent. It is significantly different

between the BCC and the V oucher groups and between the V oucher and the BCC +

V oucher groups at the 10 percent level (Columns 7 and 9), but these comparisons are not

the main focus of our analysis on anthropometry.

4. Methods

Our estimation strategy relies on the randomized design of the program. Our basic treatment

effects specification estimates the following equation:

yij1 = β0 + β1BCCij + β2V oucherij + β3BCC&V oucherij + β4yij0 + β5Xij + εij

where yij is the outcome of interest for household i from village j at follow-up including

mother’s nutritional knowledge score, household food and non-food expenditures, nutrition

indicators including CDDS, HDDS, and FCS, and child’s HAZ and WHZ scores. BCCij ,

V oucherij , and BCC&V oucherij are dummy variables equal to one if the participant was

randomly assigned to the BCC, V oucher, or the BCC + V oucher group, respectively, and

16

Page 17: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

zero otherwise. β1 , β2 , and β3 represent the intent-to-treat estimators. Yij0 is the outcome

of interest at baseline. Xij is a control vector of household i ’s characteristics including

demographic variables (mother’s age, eligible child’s age, marital status, household size,

number of children, ethnicity, religion) and socioeconomic status (mother’s literacy, years

of schooling, employment status, and household assets). εij is an error term and errors are

clustered at the village level. We present results using the specification that includes the

control vector, but the outcomes are almost the same when different specifications are used.

To address the issue of small number of clusters, we use the wild-cluster bootstrap method

for inference (Cameron et al. 2008; Rosenbaum 2002; Fisher 1935). In each results table, we

report clustered standard errors as well as the p-value computed using the wild-bootstrap

cluster-t procedure.

Next, we use the network data to estimate whether the treatment also influenced the

outcomes of peers of the participants. The extent of such spillover effects or information

spillovers can be estimated with the following specification:

yij1 = α0 + α1Peerij + α2yij0 + α3Xi + εij

where yij1 and yij0 are nutritional knowledge scores for household i from village j at follow-up

and baseline, respectively . Peerij is the number of BCC-participating friends the spillover

group respondent has. We use three different definitions of the Peerij variable: 1) the number

of BCC participants who listed the spillover group mother as a friend, and the spillover group

mother also listed the BCC participants as a friend; 2) the number of BCC participants who

listed the spillover group mother as a friend; and 3) the number of BCC participants the

spillover group mother listed as friends.

17

Page 18: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

5. Results

5.1. First-stage Results: BCC Attendance and Knowledge

Table 2 presents our first-stage results on BCC attendance and mothers’ IYCF knowledge.

Using the BCC administrative data, Column 1 of Table 2 compares the overall BCC atten-

dance rates across treatment. Note that attendance rate for the V oucher group and the

control group are set to zero. On average, the BCC and the BCC + V oucher group have

73% and 76% attendance rates, respectively, and they are not statistically different from

each other.

In Column 2, we find that attendance in the BCC sessions led to significant knowledge

gains: 0.47SD and 0.42SD for the BCC and the BCC + V oucher groups, respectively. In

other words, mothers in the BCC and the BCC + V oucher groups answered about 2 more

questions correctly compared to the control group. These effects are statistically significant

at the 1% level. However, voucher itself does not have such effect. The size of the impacts are

not significantly different between the BCC and the BCC+V oucher groups, suggesting that

receiving vouchers in addition to the BCC intervention does not further increase knowledge

gains. For reference, attendance rates and knowledge scores by IYCF topic are presented in

Table 11.

5.2. First-stage Results: Voucher Redemption

Using voucher purchase record data, Table 3 presents the next set of first-stage results on

households voucher usage. Note that voucher redemption amount in the control group are

set to zero and the BCC group is not included in this analysis. Column 1 shows that the

V oucher and the BCC+V oucher groups spent, on average, 151 and 153 ETB worth of food

vouchers per month, respectively, redeeming about 76% of the disbursed voucher amount.

18

Page 19: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

The amount redeemed per month is not statistically different between the V oucher and the

BCC + V oucher groups.

From Columns 2-11, we find that the food vouchers are spent on most food groups in

similar amounts between V oucher and BCC + V oucher. While large amounts are spent on

starchy staples and oils and fats, it is interesting that households diversity a third of their

voucher spending on non-staple food including eggs, fruits and vegetables, and nuts and

legumes. Meat and milk are not usually bought with vouchers, as they are usually not sold

in the market but obtained from their own or neighbor’s livestock. The BCC + V oucher

group spends more voucher on other fruits and vegetables and sugar, drinks, and spices

than the V oucher group. While these differences are statistically significant, they are not

economically significant, differing by less than 2 ETB per week. On other food groups, we

do not find statistically significant differences in the amount of vouchers spent on other food

groups.

5.3. Child Diet Quality

Having established that the BCC intervention resulted in knowledge gain among BCC-

participating mothers and that voucher recipients spent the food vouchers, we now look

at how this changed mothers’ child-feeding behaviors, reflected in the quality of children’s

diets. The first outcome we examine is CDDS which is our pre-specified primary outcome

(Column 1 of Table 4). Among children in the BCC + V oucher group, CDDS increased by

0.62 food group, an impact statistically significant at the 1% level. The BCC treatment by

itself had a smaller impact, 0.35, and was less precisely measured. The V oucher treatment

had no effect on CDDS. The result for the BCC group suggests that mothers are able to feed

more diverse food to their children when provided appropriate education to some extent, even

without financial support. However, we find that the increase in the BCC +V oucher group

19

Page 20: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

is almost twice as big as that of the BCC group, although the difference is not statistically

significant.

Results on minimum acceptable diet (Column 2 of Table 4) confirm that mothers in

the BCC + V oucher group show the greatest changes in child-feeding behaviors, which is

consistent with the results on CDDS. The proportion of children who met the minimum

acceptable diet standard increases for both the BCC and the BCC + V oucher groups by 8

and 15 percentage points, respectively. The magnitude of the increase for theBCC+V oucher

group is double that of the BCC group and statistically significant at the 10% level. This

measure is different from CDDS in that it accounts for not only diet quality but also quantity,

by accounting for feeding frequency.12 Also, by using a minimum cutoff, this measure focuses

on improvements in the lower tail of the distribution.

As shown in Column 3 of Table 4, timely introduction of complementary food score

increased for children in the BCC and the BCC +V oucher groups, but not for those in the

V oucher group. Unlike the previous two measures, it is interesting that the coefficient on

BCC has nearly the same size as the coefficient on BCC+V oucher for this outcome, 0.15SD

and 0.13SD, respectively. As there is relatively little or no cost to adjusting the timing of

introducing various foods, compared to increasing food quantity or diversity, both BCC and

BCC + V oucher households similarly improved their child-feeding behavior in this regard.

The proportion of mothers who perceive that their children consume relatively better

diets in terms of quantity and quality increases for the BCC + V oucher group by 5 and 8

percentage points, respectively, but not for those who receive BCC only (Columns 4 and 5

of Table 4). Interestingly, mothers in the voucher group perceive that their children have

better diet quality when, in fact, they do not. This affirms that, without a proper education

program, mothers may have a misconception of what constitutes a good quality diet for their12Results on minimum dietary diversity and minimum feeding frequency, which are the components of

minimum acceptable diet, are presented in Table 12.

20

Page 21: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

children.

5.4. Child Consumption by Food Group

To explain what is driving the improvements in child diet quality, we examine child food

consumption by food groups. Columns 1 to 4 present results on animal products, including

meat and fish, milk and milk products, eggs, and Column 5 on vitamin A-rich fruits and

vegetables, which were highlighted during the BCC program. Food groups in Columns 6 to

8 were not emphasized in the BCC program.

Table 5 shows that the BCC treatment increases child’s consumption on meat and milk

products. The BCC + V oucher treatment increases child’s consumption on a wider range

of food groups including meat and fish, milk and milk products, eggs, and vitamin A-rich

fruits and vegetables, all of which were highlighted as important sources of micronutrients

needed for healthy child growth in the BCC sessions. However, among children in the

V oucher group, meat is the only food group on which they increase consumption. While

more children in both the BCC and the BCC +V oucher groups ate animal source foods, it

is worth noting that the magnitude of the increase in the BCC+V oucher group is more than

twice as large as that of the BCC group—larger by more than 10 percentage points (Column

4). This difference in point estimates are statistically significant at the 5% level. This greater

impact of BCC + V oucher on children’s animal source food consumption shows that it is

the greater consumption of animal source foods that is driving the greater improvements in

diet quality in the BCC + V oucher group.

5.5. Household Expenditures

We now turn to changes in household expenditures in response to the treatments to explain

the child diet results. Table 6 presents results household food expenditure per week per

21

Page 22: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

capita in total and by food groups. Note that household expenditure data was collected

during the follow-up survey which was conducted after the completion of the interventions.

Hence, household expenditure data do not include voucher expenditures. Column 1 presents

weekly total household expenditure which should equal the sum of all expenditures by food

group in Columns 3-12.

We first look at expenditures at the aggregate level, including food expenditures per

week per capita (Column 1) and non-food expenditures per week per capita (Column 2). We

find positive coefficients on total food expenditures, but they are not statistically significant

(Column 1). The size of the coefficients on V oucher and BCC+V oucher are similar to that

of the voucher transfer in per week per capita terms (200 ETB ÷ average household size of

4.5 ÷ 4 weeks ≈ 11 ETB). This suggests that the households at least did not drop their food

expenditures immediately after intervention completion, which is a common phenomenon

among voucher schemes in developed countries, but rather maintained comparable food

spending amounts. The BCC group spent similar amounts on food to other treatment

groups.

As shown in Column 2, we do not find any impact on non-food expenditures for the

BCC and the V oucher groups. This is particularly interesting for the V oucher group,

as it suggests that there is no evidence for crowding-out of food expenditures into non-

food expenditures due to the food vouchers. We find increases in non-food expenditures in

the BCC + V oucher group, but is imprecisely measured. While this may be a suggestive

evidence for a partial crowding-out of food expenditures among the BCC +V oucher group,

food expenditure results suggest otherwise. While an increase in non-food expenditures

should lead to a decrease in food expenditures–especially given that the expenditure data

was collected after the voucher scheme was completed—we do not find this among the BCC+

V oucher group.

22

Page 23: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

We investigate further in to household expenditures by looking at food group expen-

ditures. We find that households in the V oucher and BCC + V oucher groups continue to

spend more on non-staple food groups. This result is comparable to existing studies that find

a positive relationship between income and non-staple nutrients (Dereje 2015; Bilal et al.

2013; Skoufias et al. 2011). There is a statistically significant difference in impacts on vitamin

A-rich fruits and vegetables expenditures between the V oucher and BCC+V oucher groups,

suggesting that BCC expanded the set of food expenditures to this food group. However,

the magnitude of this effect is 1 ETB per week per capita which is not large.

As for the BCC group, we find that they also increased non-staple expenditures to

some extent, particularly on animal source foods, although it is less precisely measured. This

suggests that nutritional knowledge gained through BCC influences households to diversify

food expenditure to some extent, even without additional income.

Taking together the results on voucher redemption from Section 5.2 and food expendi-

tures, we find that increasing income through vouchers causes non-staple expenditures to rise

in a similar way for both V oucher and BCC+V oucher groups. Comparing this finding with

the diverging results on child diet quality between these two groups, we can conclude that,

while the food items bought at the household level are similar, it is the nutritional knowledge

gained through BCC that causes mothers in the BCC + V oucher group to allocate more

non-staple food to young children.

5.6. Child Physical Growth

Given the improvements in child diet quality, we further investigate how the changes in

children’s diets affected their physical growth and stunting, in particular. We find that the

BCC + V oucher treatment improves nutritional status at least in the short run. Table 7

presents the results on HAZ and stunting, measures of chronic nutritional status among

23

Page 24: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

infants and young children (Columns 1 and 2), as well as on WHZ and wasting which are

related to acute nutritional status (Columns 3 and 4).

We do not find impact on HAZ. However, stunting prevalence significantly decreases

by 9.5 percentage points among children in the BCC + V oucher group, but not in other

groups. In the overall sample, stunting prevalence increased from 27% at baseline to 42%

at follow-up. This pattern is similar to the increasing trend of stunting rates with age

among children over 6 months in developing countries. Amid this rapidly increasing trend,

our stunting results show that the BCC + V oucher treatment prevented stunting from

occurring for about 10% of its participants. Likewise, as Panel A of Figure 5 illustrates, the

BCC + V oucher group shows greater improvements among children in the lower tail of the

HAZ distribution, representing that chronic nutritional status improves for those who are

otherwise more prone to be chronically undernourished.

We find an increase in WHZ scores by 0.3SD and a decrease in wasting prevalence for

the BCC group but the latter is not statistically significant. Graphical evidence shows that

WHZ scores improve for the upper tail of the distribution for the BCC group, which explains

why there is no effect on wasting (Panel B of Figure 5). This implies that weight-for-height

increases among children who are relatively well-nourished for the BCC group. One possible

explanation could be that children in the upper tail of the WHZ distribution are most likely

from households that were able to adjust their spending to improve child diet quality. On

the other hand, for the BCC + V oucher group, the additional financial support may have

empowered households with children in the lower tail of the HAZ distribution to afford IYCF

improvements. To summarize, our child growth results suggest that the BCC + V oucher

treatment improves the nutritional status of chronically undernourished children, while the

BCC treatment helps improve the nutritional status of relatively well-nourished children.

24

Page 25: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

5.7. Household Food Consumption

Although our interventions were focused on improving children’s diets, we find positive

impact on household diet quality. As shown in Column 1 of Table 8, diet quality, as measured

by FCS, also improves at the household level among the BCC and the BCC + V oucher

groups but not the V oucher group. While the BCC program focused on healthy child-

feeding practices, it is possible that some nutritional information with general application was

applied to the overall household diet—e.g., the emphasis on dietary diversity and essential

micronutrients. This is also supported by results on household food consumption by food

group (Columns 2-11 of Table 8), which shows that improvements in household diet quality

is driven by the consumption of food groups highlighted in the BCC sessions, notably animal

source foods and vitamin A-rich fruits and vegetables.

5.8. Spillover Effects

Lastly, we examine whether a one-time intervention could be sustained in the community

through peer networks (Table 9). It is plausible that mothers primarily seek IYCF advice

from their peers who gave birth just a few months ahead of them, in which case we expect

the spillover group mothers who have a BCC-participating friend to be better-informed than

those who do not. To assess this, we take advantage of the data on the spillover group which

consists of mothers with children under 4 months and pregnant women at baseline. Column

1 of Table 9 presents the effect of having BCC friends defined by the networks of both the

spillover group and BCC participants. Columns 2 and 3 show the effect of having BCC

friends defined by the spillover group respondent’s network and BCC participants’ network,

respectively.

We find suggestive evidence that knowledge on IYCF can be transferred to mother’s

peer group. Those in the spillover group who have a friend who received the BCC treatment

25

Page 26: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

have higher IYCF knowledge score compared to those who do not. The coefficients are

all positive and economically large even though the size and significance of the coefficients

are different across the definition of the peer. For an additional BCC-participating peer,

overall knowledge scores shown in Columns 1 to 3 increase by 0.11SD, 0.18SD, and 0.07SD,

respectively, although statistically significant only in Column 2. It is also worth noting

that the magnitude of this increase is approximately one-third of the increase among BCC

participants.

6. Extensions and Robustness Checks

6.1. Heterogeneity Analysis

We conduct heterogeneity analysis to assess whether treatment impacts differ by various

household characteristics. We examine differences in the following characteristics: urban/rural,

whether mother had no formal education, whether eligible child is the first child, age of the

mother and of the eligible child, and whether single mother. We analyze this by including the

heterogeneous variable of interest as control and interacting this variable with the treatment

group dummy variables.

Selected results are reported in Table 10. We find heterogeneous impacts by area,

mother’s schooling, eligible child age, whether single, and wealth. For households in rural

areas (Panel A), the effect of BCC + V oucher on CDDS is smaller. Mothers in rural

areas are more likely to increase self-employed farming labor supply in response to BCC.

Stunting prevalence is higher for rural population that received BCC + V oucher, although

not significant using bootstrap -t p-value. For mothers with no formal schooling (Panel B),

the impact of BCC and BCC + V oucher on IYCF knowledge is greater by 0.4SD, farm

labor supply is lower for both fathers and mothers, and wasting prevalence is lower by 10

26

Page 27: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

percentage points. For households with 12 months or younger child at baseline (Panel C),

the impact of BCC + V oucher on CDDS is greater and mothers among the BCC group

increase farm labor supply, but the impact of BCC + V oucher on stunting prevalence is

higher for this group. The impact of BCC and BCC + V oucher on is smaller for single

mothers and Vouchers increase stunting prevalence in this group (Panel D). As for other

household characteristics we examined, we do not find significant differences in impact.

6.2. Robustness Checks

We perform several robustness checks. First, to address the issue of small number of clusters,

we use the wild-cluster bootstrap method for inference (Cameron et al. 2008; Rosenbaum

2002; Fisher 1935). Our results show that the degrees of statistical significance do not differ

for the most part when using the wild-cluster bootstrap p-values. As a second robustness

check, we estimated all of the regressions in Tables 3 to 10 without control variables. We

find that the results are robust to the exclusion of control variables, and the point estimates

and their degree of statistical significance remain similar.

7. Discussion and Conclusion

High rates of stunting in many developing countries pose important health threats to young

children. Many interventions that target a single dimension of causes of child undernutrition

have often found limited effect. Interventions that address multidimensional and interrelated

causes of undernutrition, such as lack of awareness and affordability, may be more effective in

bringing about healthy child development. We test this by implementing a community-based

cluster randomized experiment in Ethiopia that randomly provide IYCF education through

a nutrition BCC and increase affordability through food vouchers to mothers of children

27

Page 28: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

aged between 4 and 20 months.

We find that receiving child nutrition education through the BCC program improves

mothers’ child-feeding knowledge, which in turn translates into better child-feeding practices,

while food vouchers alone do not. When only provided with knowledge and not financial

support, mothers purchase more diverse food to some extent and allocates them to their chil-

dren, evidenced by household food group expenditures and child food consumption results.

Also, to procure additional food or income to support improved child-feeding practices, these

mothers may increase their self-employed farming labor supply. With these new purchases

or production, children are fed more diverse food items that they would not have eaten

otherwise. Moreover, BCC improves the diet quality of the household at large regardless

of whether with or without vouchers.Informed mothers also start introducing these healthy

food items at an earlier age so that complementary feeding is not delayed, either with or

without financial support.

Receiving food vouchers in addition to the BCC program considerably augments the

positive impacts. While the magnitude of the knowledge gain is similar between the BCC

and the BCC + V oucher groups, the increase in child dietary diversity doubles for the

BCC + V oucher group. This suggests that the lack of resources could be a barrier to

improving child-feeding practices for those with proper child-feeding knowledge. The increase

in dietary diversity is driven by eating more diverse animal source foods and vitamin A-rich

fruits and vegetables. Diet quantity also increases doubly, evidenced by increased feeding

frequency reflected in the minimal acceptable diet standard. Mothers also become more

confident in how their children fare in terms of diet quantity and quality compared to other

children of the same age. Consequently, improved diet quality leads to stunting reduction. In

terms of introducing complementary foods at appropriate ages, BCC and BCC + V oucher

improve similarly. From this, we can infer that when there is little or no cost associated

28

Page 29: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

with a behavior change—i.e., there is no income constraint to behavior change—providing

knowledge alone brings about similar improvements to providing both knowledge and income.

This has important implications for improving other suboptimal health behaviors caused by

misinformation but not cost, such as breastfeeding practices.

Comparing the BCC + V oucher group with the V oucher group, we find that income

through vouchers increases the overall non-staple food stock, whereas nutritional knowledge

through BCC alters mothers’ intrahousehold food allocation decisions. When provided both

education and income, mothers allocate the healthy non-staple food purchased with the

vouchers to infant and young children. Voucher recipients who did not receive BCC still

purchase more non-staple food, but they simply do not allocate it to their children. Food

vouchers alone play a very limited role in changing mothers’ feeding behaviors, and does not

increase child nor household dietary diversity.

This study has some limitations. First, our study looks at relatively short-term results

measured during the period between completion of the intervention and three months there-

after. Thus, we are not able to examine whether improved knowledge and IYCF practices are

sustained in the long-term, and whether or not chronic nutritional status improves further

with time. Secondly, there are only 79 villages (clusters) in our study sample, which calls

for small-sample correction of standard errors.

Our results show that both awareness and affordability are challenges for improved

IYCF which is critical for preventing stunted growth. We also demonstrate that an intensive

nutrition education program (BCC) successfully improves feeding practices, while financial

support (voucher) itself does not effectively improve IYCF. The impacts on IYCF are greatest

when education and financial programs are combined, leading to stunting reduction.

29

Page 30: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Figures and Tables

Figure 1: Predicted Height-for-age Z Scores by Child Age in Months

Source: Local polynomial smoothing predictions with 95% confidence intervals estimated

using the DHS data (Ethiopia DHS, 2000, 2011).

30

Page 31: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Figure 2: Map of Ejere District

31

Page 32: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Figure 3: Study Design

32

Page 33: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Figure4:

Stud

yTim

eline

33

Page 34: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Figure5:

Heigh

t-for-ag

e(H

AZ)

andweigh

t-for-height

(WHZ)

ZScoreKernelD

ensity

Graph

34

Page 35: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Table1:

BaselineMeanCha

racteristics

byIntervention

Group

s

Mean

Difference

betw

een

treatm

entan

dcontrol

P-

value:

NDifference

betw

een

treatm

ents

All

Con

trol

B-C

B-V

BV-C

B=V=BV

B-V

B-B

VV-B

V(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

Pan

elA

.M

othe

rch

arac

teri

stic

sMotherage(years)

28.29

28.21

-0.681

-0.063

0.696

0.318

632

0.618

1.377

0.759

Motheris

Oromo

0.766

0.765

-0.016

0.000

0.001

0.988

634

0.016

0.017

0.001

Motheris

Ortho

doxChristian

0.844

0.851

0.018

0.010

-0.038

0.776

634

-0.008

-0.057

-0.048

Motheris

married

0.768

0.778

-0.049

-0.055

0.028

0.403

633

-0.007

0.077

0.083

Motherha

swork

0.564

0.543

0.045

0.040

0.042

0.998

634

-0.005

-0.003

0.002

Motherab

leto

read

0.495

0.471

0.081

0.052

0.008

0.797

633

-0.029

-0.074

-0.045

Motherab

leto

write

0.485

0.457

0.085

0.056

0.021

0.846

633

-0.029

-0.064

-0.034

Motheryearsof

scho

oling

4.258

3.958

0.595

-0.221

0.835

0.611

634

-0.817

0.240

1.057

MotherIY

CFkn

owledg

escore

21.49

21.46

0.005

-0.168

0.195

0.808

634

-0.173

0.190

0.363

Pan

elB.C

hild

char

acte

rist

ics

Elig

ible

child

age(m

onths)

12.48

12.29

1.100∗∗

-0.183

0.016

0.057

634

-1.282∗∗

-1.083∗

0.199

Child

dietarydiversityscore

2.359

2.433

0.022

-0.236

-0.185

0.456

634

-0.258

-0.207

0.051

Minim

umacceptab

lediet

0.128

0.116

0.074

0.004

0.004

0.361

627

-0.070

-0.070

0.000

Heigh

t-for-ageZscore

-1.060

-1.038

-0.043

-0.111

-0.128

0.930

613

-0.068

-0.084

-0.016

Stun

ting

0.272

0.266

0.005

-0.015

0.045

0.549

613

-0.020

0.040

0.060

Pan

elC

.H

ouse

hold

char

acte

rist

ics

Femaleho

useholdhead

0.139

0.135

0.026

0.004

-0.008

0.772

634

-0.022

-0.033

-0.012

Hou

seho

ldsize

4.539

4.495

-0.100

0.160

0.140

0.500

634

0.260

0.240

-0.020

Num

berof

child

ren

2.348

2.315

-0.101

0.126

0.146

0.572

634

0.227

0.247

0.020

Asset

index

-0.015

-0.057

0.162

-0.086

0.113

0.807

634

-0.249

-0.050

0.199

Rural

0.445

0.505

-0.017

-0.060

-0.194

0.673

634

-0.043

-0.177

-0.133

Total

weeklyfood

expe

nditure,

percapita

131.7

129.9

11.26

-11.92

2.275

0.818

634

-23.18

-8.984

14.20

Total

weeklyno

n-food

expe

nditure,

percapita

17.05

16.82

3.522

-5.266∗∗

1.302

0.076

634

-8.788∗

-2.220

6.568

Hou

seho

ldfood

consum

ptionscore

43.19

43.28

-1.003

-0.016

0.333

0.830

634

0.987

1.337

0.349

Distanceto

thenearestmarket(km)

3.593

4.978

-1.924

-2.327

-3.013

0.521

626

-0.403

-1.089

-0.686

Pan

elD

.A

ttri

tion

Follo

w-upSu

rvey

Attrition

Rates

0.084

0.093

0.007

-0.040

-0.013

0.478

634

-0.047

-0.020

0.027

Anthrop

ometry

Attrition

Rates

0.166

0.156

0.045

-0.049

0.038

0.049

634

-0.094∗

-0.007

0.087∗∗

Tab

le1repo

rtsmeanof

selected

baselin

evariab

les.

Brepresents

thosewho

wereoff

ered

theBCC

program

only.V

represents

thosewho

wereoff

ered

thevouche

rprogram

only.BV

represents

thosewho

wereoff

ered

both

theBCC

andthevouche

rprograms.

Colum

ns1-2show

asummaryof

thewho

lesamplean

dthecontrol

grou

p.Colum

ns3-5repo

rtmeandiffe

rences

andsign

ificancelevels

from

test

ofmeandiffe

renc

esbe

tweeneach

treatm

entgrou

pan

dcontrol.

Colum

n6show

sp-values

from

thejointtest

ofequa

lityof

parametersrepo

rted

incolumns

3-5,

column7thenu

mbe

rof

observations,an

dcolumns

8-10

test

ofmeandiffe

rences

betw

eentreatm

entgrou

ps.

∗,∗

∗,a

nd∗∗

∗de

note

sign

ificanceat

10%

,5%

,and

1%,r

espe

ctively.

35

Page 36: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Table 2: Effects on BCC Attendance and Mother IYCF Knowledge

BCCAttendance rate

Mother IYCFknowledge score(standardized)

(1) (2)

BCC (B) 0.730∗∗∗ 0.468∗∗∗

(0.020) (0.100)[0.000]

Voucher (V) -0.004 0.060(0.007) (0.137)

[0.669]

BCC & Voucher (BV) 0.762∗∗∗ 0.415∗∗∗

(0.011) (0.103)[0.001]

Observations 630 577R-squared 0.898 0.127Control group mean 0.000 -0.166P-value: B=V 0.007P-value: B=BV 0.182 0.657P-value: V=BV 0.028P-value: B+V=BV 0.553This table reports results on overall BCC attendance rate and mothers’ IYCFknowledge score (standardized). Column 1 uses administrative data and com-pares BCC attendance rates with the control group where the control andthe voucher group’s attendance rates are set to zero. Column 2 uses surveydata on mothers’ IYCF knowledge. All estimations include area dummies,mother’s age, whether mother is married, working, able to read, and able towrite, mother’s years of schooling, eligible child age, household size, number ofchildren, whether female-headed household, household asset index, ethnicity,and religion. Column 2 additionally controls for the baseline outcome. Ro-bust standard errors clustered at the unit of randomization, the village level,in parentheses. Wild-cluster bootstrap p-values in square brackets. The p-value for difference in coefficients is a F-test for whether the coefficient differsbetween treatment groups. The p-value for difference between B+V and BVtests whether there is any complementarity between BCC and vouchers. ∗∗∗

p<0.01, ∗∗ p<0.05, ∗ p<0.1.

36

Page 37: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Table3:

Effe

ctson

Vou

cher

Redem

ption(D

uringIntervention

)

Average

total

vouche

rexp./m

onthAverage

voucherredemptionpe

rweekby

food

grou

p

Meat

andfish

Milk

and

milk

prod

-ucts

Eggs

Animal

prod

-ucts

total

Vitam

inA-rich

fruits

&veg.

Other

fruits

and

veg.

Nuts

and

legu

mes

Starchy

stap

les

Oilan

dfats

Sugar,

drinks,

and

spices

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

Vou

cher

(V)

151.7∗∗∗

0.145

0.045

1.394∗∗∗

1.584∗∗∗

1.232∗∗∗

5.463∗∗∗

1.870∗∗∗

14.90∗∗∗

10.10∗∗∗

2.742∗∗∗

(4.073)

(0.093)

(0.040)

(0.282)

(0.332)

(0.144)

(0.396)

(0.435)

(0.900)

(0.646)

(0.281)

[0.000]

[0.011]

[0.276]

[0.000]

[0.000]

[0.000]

[0.000]

[0.000]

[0.000]

[0.000]

[0.000]

BCC

&Vou

cher

(BV)

153.2∗∗∗

0.031

0.100∗

1.119∗∗∗

1.250∗∗∗

1.566∗∗∗

7.375∗∗∗

1.527∗∗∗

13.50∗∗∗

9.672∗∗∗

3.316∗∗∗

(3.176)

(0.028)

(0.053)

(0.167)

(0.166)

(0.158)

(0.394)

(0.455)

(0.755)

(0.505)

(0.195)

[0.000]

[0.301]

[0.172]

[0.000]

[0.000]

[0.000]

[0.000]

[0.000]

[0.000]

[0.000]

[0.000]

Observation

s516

516

516

516

516

516

516

516

516

516

516

R-squ

ared

0.902

0.071

0.030

0.261

0.248

0.398

0.623

0.317

0.622

0.611

0.403

P-value:V=BV

0.785

0.217

0.350

0.348

0.324

0.129

0.002

0.547

0.239

0.589

0.081

Thistablerepo

rtsresultson

averagevo

uche

rrede

mptionin

totalpe

rmon

than

dby

food

grou

pspe

rweekusingthevouche

rpu

rcha

serecord

data.

The

’Animal

prod

ucts’food

grou

pis

anaggregationof

meatan

dfish,

milk

andmilk

prod

ucts,an

deggs.Using

thevo

uche

rusageda

ta,theresults

compa

retheBCC+Vou

cher

grou

pan

dtheVou

cher

grou

pwiththecontrolgrou

pwhich

haszero

vouche

rspen

ding

.Allestimations

controlforarea

dummies,

mothe

r’sage,

whe

ther

mothe

ris

married

,working

,ab

leto

read

,an

dab

leto

write,mothe

r’syearsof

scho

oling,

eligible

child

age,

househ

old

size,n

umbe

rof

child

ren,

whe

ther

female-he

aded

househ

old,

househ

oldassetinde

x,ethn

icity,

andrelig

ion.

Rob

uststan

dard

errors

clusteredat

theun

itof

rand

omization,

thevilla

gelevel,in

parenthe

ses.

Boo

tstrap

pedclusteredstan

dard

errors

insqua

rebrackets.W

ild-cluster

bootstrapp-values

insqua

rebrackets.The

p-valuefordiffe

rencein

coeffi

cients

isaF-testforwhe

ther

thecoeffi

cientdiffe

rsbe

tweentreatm

entgrou

ps.

∗∗∗p<

0.01,

∗∗p<

0.05,

p<0.1.

37

Page 38: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Table 4: Effects on Child Diet Quality

CDDSMinimumacceptable

diet

Timely intro.of comple-mentaryfood score

(standardized)

Perceivedrelative

child dietaryquantity

Perceivedrelative

child dietaryquality

(1) (2) (3) (4) (5)

BCC (B) 0.331∗ 0.078∗∗ 0.146∗∗ 0.004 0.037(0.174) (0.034) (0.065) (0.041) (0.027)[0.086] [0.045] [0.049] [0.916] [0.197]

Voucher (V) 0.032 0.006 -0.046 0.0474 0.0534∗∗

(0.184) (0.030) (0.085) (0.032) (0.025)[0.883] [0.824] [0.646] [0.144] [0.044]

BCC & Voucher 0.604∗∗∗ 0.152∗∗∗ 0.130∗∗ 0.047∗ 0.076∗∗∗

(BV) (0.172) (0.031) (0.063) (0.026) (0.024)[0.014] [0.002] [0.081] [0.102] [0.004]

Observations 576 531 565 577 577R-squared 0.125 0.126 0.045 0.067 0.053Control group mean 3.073 0.124 0.153 0.893 0.905P-value: B=V 0.152 0.059 0.050 0.361 0.528P-value: B=BV 0.177 0.070 0.826 0.332 0.140P-value: V=BV 0.007 0.0001 0.073 1.000 0.317P-value: B+V=BV 0.381 0.193 0.796 0.941 0.687This table reports results on child dietary diversity score (CDDS), minimum acceptable diet standard, stan-dardized score on timely introduction of complementary foods, and mothers’ perception of their child’s relativedietary quantity and quality. All estimations include the baseline outcome, area dummies, mother’s age, whethermother is married, working, able to read, and able to write, mother’s years of schooling, eligible child age, house-hold size, number of children, whether female-headed household, household asset index, ethnicity, and religion.Robust standard errors clustered at the unit of randomization, the village level, in parentheses. Wild-clusterbootstrap p-values in square brackets. The p-value for difference in coefficients is a F-test for whether thecoefficient differs between treatment groups. The p-value for difference between B+V and BV tests whetherthere is any complementarity between BCC and vouchers. ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1.

38

Page 39: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Table5:

Effe

ctson

Child

Food

Con

sumption

Whether

child

atein

thelast

24ho

urs:

Meat

Milk

Eggs

Aminal

prod

ucts

total

Vitam

inA-rich

fruits

&veg.

Other

fruits

&veg.

Nuts&

legu

mes

Starchy

stap

les

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

BCC

(B)

0.131∗∗

0.091∗

0.083

0.094∗

0.003

-0.022

0.067

-0.024

(0.050)

(0.050)

(0.066)

(0.050)

(0.055)

(0.062)

(0.057)

(0.017)

[0.029]

[0.098]

[0.299]

[0.087]

[0.957]

[0.779]

[0.219]

[0.224]

Vou

cher

(V)

0.136∗∗∗

-0.023

-0.010

0.139∗∗

-0.055

-0.043

0.031

-0.004

(0.037)

(0.045)

(0.070)

(0.056)

(0.041)

(0.053)

(0.071)

(0.013)

[0.003]

[0.628]

[0.886]

[0.030]

[0.197]

[0.478]

[0.696]

[0.860]

BCC

&Vou

cher

(BV)

0.119∗∗∗

0.102∗∗

0.181∗∗∗

0.200∗∗∗

0.098∗

0.004

0.087∗

0.005

(0.024)

(0.044)

(0.051)

(0.038)

(0.050)

(0.046)

(0.047)

(0.010)

[0.001]

[0.039]

[0.008]

[0.000]

[0.113]

[0.933]

[0.107]

[0.682]

Observation

s576

576

576

576

576

576

576

576

R-squ

ared

0.099

0.099

0.100

0.173

0.060

0.042

0.047

0.038

Con

trol

grou

pmean

0.119

0.275

0.286

0.437

0.226

0.805

0.368

0.992

P-value:B=V

0.932

0.036

0.273

0.490

0.323

0.753

0.624

0.267

P-value:B=BV

0.817

0.832

0.186

0.049

0.149

0.709

0.722

0.125

P-value:V=BV

0.677

0.015

0.017

0.297

0.005

0.413

0.418

0.529

P-value:B+V=BV

0.023

0.631

0.307

0.677

0.055

0.442

0.912

0.176

Thistablerepo

rtsresultson

child

food

consum

ptionby

food

grou

p.Eachou

tcom

eindicateswhe

ther

thechild

atean

yfood

item

from

the

food

grou

pin

thelast

24ho

urs.

The

’Animal

prod

ucts’foo

dgrou

pisan

aggregationof

meatan

dfish,

milk

andmilk

prod

ucts,a

ndeggs.

Allestimations

controlfor

theba

selin

eou

tcom

e,area

dummies,mothe

r’sage,

whe

ther

mothe

rismarried

,working

,ableto

read

,and

able

towrite,m

othe

r’syearsof

scho

oling,

eligible

child

age,

househ

oldsize,n

umbe

rof

child

ren,

whe

ther

female-he

aded

househ

old,

househ

old

assetinde

x,ethn

icity,

andrelig

ion.

Rob

uststan

dard

errors

clusteredat

theun

itof

rand

omization,

thevilla

gelevel,in

parenthe

ses.

Wild

-cluster

bootstrapp-values

insqua

rebrackets.The

p-valuefordiffe

rencein

coeffi

cients

isaF-testforwhe

ther

thecoeffi

cientdiffe

rsbe

tweentreatm

entgrou

ps.The

p-valuefordiffe

rencebe

tweenB+V

andBV

testswhe

ther

thereis

anycomplem

entarity

betw

eenBCC

andvouche

rs.

∗∗∗p<

0.01,∗

∗p<

0.05,∗

p<0.1.

39

Page 40: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Table6:

Effe

ctson

Hou

seho

ldExp

enditures(A

fter

Intervention

)

Total

food

expe

n-diture

Total

non-

food

expe

n-diture

Amou

ntspentpe

rcapita

inthelast

week:

Meat

and

fish

Milk

and

milk

prod

-ucts

Eggs

Animal

prod

-ucts

total

Vitam

inA-rich

fruits

&veg.

Other

fruits

&veg.

Nuts

and

legu

mes

Starchy

stap

les

Oils

and

fats

Sugars,

drinks,

spices

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

(12)

BCC

(B)

12.80

-3.927

9.858∗∗

-1.048

-0.136

8.742∗

0.209

1.915∗

3.613∗∗∗

0.558

-0.326

-1.726

(10.47)

(3.161)

(4.276)

(1.248)

(0.250)

(4.775)

(0.257)

(1.043)

(1.315)

(6.000)

(1.140)

(2.628)

[0.293]

[0.236]

[0.042]

[0.460]

[0.609]

[0.108]

[0.479]

[0.111]

[0.026]

[0.946]

[0.825]

[0.521]

Vou

cher

(V)

13.23

4.602

11.70∗∗

-0.381

0.058

11.51∗∗

-0.185

0.322

0.273

-2.112

1.015

2.078

(12.02)

(3.664)

(5.152)

(1.513)

(0.293)

(5.326)

(0.222)

(0.658)

(1.119)

(6.813)

(1.330)

(2.306)

[0.311]

[0.217]

[0.049]

[0.805]

[0.861]

[0.0591]

[0.433]

[0.636]

[0.832]

[0.784]

[0.468]

[0.365]

BCC

&Vou

cher

(BV)

14.40

9.752∗

5.732

1.662

0.650∗

8.058∗

1.004∗∗∗

0.973

0.429

-1.541

1.127

3.747∗∗

(10.27)

(5.252)

(3.716)

(1.338)

(0.331)

(4.179)

(0.278)

(0.610)

(1.162)

(6.710)

(0.946)

(1.679)

[0.206]

[0.219]

[0.172]

[0.248]

[0.073]

[0.086]

[0.005]

[0.121]

[0.728]

[0.844]

[0.311]

[0.024]

Observation

s576

576

576

576

576

576

576

576

576

576

576

576

R-squ

ared

0.231

0.167

0.097

0.112

0.104

0.125

0.154

0.256

0.104

0.173

0.062

0.124

Con

trol

grou

pmean

81.312

31.451

13.053

4.320

0.964

18.336

0.772

6.611

3.012

30.942

5.819

15.820

P-value:B=V

0.974

0.035

0.763

0.695

0.544

0.670

0.159

0.132

0.036

0.693

0.402

0.188

P-value:B=BV

0.894

0.012

0.392

0.074

0.027

0.897

0.012

0.408

0.047

0.751

0.289

0.040

P-value:V=BV

0.928

0.266

0.286

0.252

0.099

0.563

0.0001

0.423

0.914

0.931

0.940

0.480

P-value:B+V=BV

0.493

0.098

0.025

0.151

0.094

0.104

0.012

0.373

0.078

0.999

0.821

0.360

Thistablerepo

rtsresultson

weeklyho

useh

oldfood

andno

n-food

expe

nditurein

totalan

dby

food

grou

p.Eachou

tcom

eindicatestheam

ount

spentby

househ

oldin

thelast

weekpe

rcapita

inEthiopian

Birr.

The

’Animal

prod

ucts’food

grou

pis

anaggregationof

meatan

dfish,

milk

andmilk

prod

ucts,an

deggs.Allestimations

controlfortheba

selin

eou

tcom

e,area

dummies,

mothe

r’sage,

whe

ther

mothe

ris

married

,working

,ab

leto

read

,an

dab

leto

write,

mothe

r’syearsof

scho

oling,

eligible

child

age,

househ

oldsize,nu

mbe

rof

child

ren,

whe

ther

female-he

aded

househ

old,

househ

oldassetinde

x,ethn

icity,

and

relig

ion.

Rob

uststan

dard

errors

clusteredat

theun

itof

rand

omization,

thevilla

gelevel,in

parenthe

ses.

Wild

-cluster

bootstrapp-values

insqua

rebrackets.

The

p-valuefordiffe

rencein

coeffi

cients

isaF-testforwhe

ther

thecoeffi

cientdiffe

rsbe

tweentreatm

entgrou

ps.The

p-valuefordiffe

rencebe

tweenB+V

and

BV

testswhe

ther

thereis

anycomplem

entarity

betw

eenBCC

andvouche

rs.

∗∗∗p<

0.01,∗

∗p<

0.05,∗

p<0.1.

40

Page 41: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Table 7: Effects on Child Physical Growth

HAZ Stunted WHZ Wasted(1) (2) (3) (4)

BCC (B) -0.048 0.080 0.328∗∗ -0.043(0.173) (0.073) (0.160) (0.034)[0.789] [0.312] [0.041] [0.254]

Voucher (V) -0.178 0.073 -0.024 -0.002(0.155) (0.059) (0.156) (0.035)[0.304] [0.280] [0.863] [0.964]

BCC & Voucher (BV) 0.181 -0.095∗∗ -0.0003 0.009(0.159) (0.046) (0.190) (0.034)[0.386] [0.094] [0.999] [0.783]

Observations 486 486 494 494R-squared 0.325 0.232 0.126 0.070Control group mean -1.543 0.416 0.026 0.082P-value: B=V 0.529 0.940 0.056 0.364P-value: B=BV 0.221 0.021 0.125 0.227P-value: V=BV 0.071 0.016 0.914 0.806P-value: B+V=BV 0.099 0.011 0.268 0.337This table reports results on height-for-age Z scores (HAZ), stunting prevalence,weight-for-height Z scores (WHZ), and wasting prevalence. All estimations includebaseline outcome, area dummies, mother’s age, whether mother is married, work-ing, able to read, and able to write, mother’s years of schooling, eligible child age,household size, number of children, whether female-headed household, householdasset index, ethnicity, and religion. Robust standard errors clustered at the unit ofrandomization, the village level, in parentheses. Wild-cluster bootstrap p-valuesin square brackets. The p-value for difference in coefficients is a F-test for whetherthe coefficient differs between treatment groups. The p-value for difference be-tween B+V andBV tests whether there is any complementarity between BCC andvouchers. ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1.

41

Page 42: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Table8:

Effe

ctson

Hou

seho

ldFo

odCon

sumption

FCS

Whether

househ

oldatein

thelast

week:

Meat&

poultry

Milk

&milk

prod

.Eggs

Animal

prod

-ucts

total

Vitam

inA-rich

fruits

&veg.

Nuts&

legu

mes

Other

fruits

&veg.

Stap

les

Oils

&fats

Sugar&

spices

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

BCC

(B)

5.804∗∗∗

0.153∗∗

-0.042

0.0133

0.149∗∗

-0.042

-0.035∗

-0.014

0.014∗

0.016∗

-0.004

(1.707)

(0.070)

(0.050)

(0.065)

(0.061)

(0.050)

(0.019)

(0.016)

(0.008)

(0.009)

(0.012)

[0.003]

[0.061]

[0.458]

[0.866]

[0.035]

[0.458]

[0.084]

[0.450]

[0.088]

[0.082]

[0.731]

Vou

cher

(V)

1.566

0.069

0.023

0.002

0.086∗

0.023

-0.002

-0.019

-0.006

0.003

-0.007

(2.172)

(0.065)

(0.057)

(0.066)

(0.048)

(0.057)

(0.028)

(0.012)

(0.016)

(0.010)

(0.011)

[0.535]

[0.340]

[0.684]

[0.981]

[0.100]

[0.684]

[0.961]

[0.174]

[0.763]

[0.835]

[0.726]

BCC

&Vou

cher

(BV)

5.748∗∗∗

0.137∗∗

0.171∗∗∗

0.224∗∗∗

0.213∗∗∗

0.171∗∗∗

0.014

-0.014

0.005

0.003

0.004

(1.658)

(0.060)

(0.058)

(0.052)

(0.042)

(0.058)

(0.027)

(0.011)

(0.011)

(0.014)

(0.004)

[0.008]

[0.047]

[0.009]

[0.002]

[0.000]

[0.009]

[0.635]

[0.227]

[0.711]

[0.825]

[0.355]

Observation

s576

576

576

576

576

576

576

576

576

576

576

R-squ

ared

0.226

0.203

0.246

0.207

0.289

0.246

0.063

0.065

0.055

0.047

0.052

Con

trol

grou

pmean

53.425

0.360

0.234

0.345

0.536

0.402

0.061

0.992

0.985

0.985

1.000

P-value:B=V

0.067

0.323

0.281

0.890

0.339

0.281

0.229

0.733

0.201

0.159

0.852

P-value:B=BV

0.976

0.834

0.001

0.002

0.256

0.001

0.085

0.962

0.390

0.271

0.426

P-value:V=BV

0.084

0.370

0.017

0.001

0.007

0.017

0.663

0.728

0.549

0.957

0.346

P-value:B+V=BV

0.588

0.393

0.0185

0.025

0.769

0.0185

0.226

0.408

0.863

0.280

0.372

Thistablerepo

rtsresultson

househ

oldfood

consum

ptionscore(F

CS)

andho

useh

oldfood

consum

ptionby

food

grou

p.Outcomes

inColum

ns2-11

indicate

whe

ther

theho

useh

oldatean

yfood

item

from

thefood

grou

p.The

’Animal

prod

ucts’food

grou

pis

anaggregationof

meatan

dfish,

milk

andmilk

prod

ucts,

andeggs.Allestimations

controlfortheba

selin

eou

tcom

e,area

dummies,

mothe

r’sage,

whe

ther

mothe

ris

married

,working

,ab

leto

read

,an

dab

leto

write,

mothe

r’syearsof

scho

oling,

eligible

child

age,

househ

oldsize,nu

mbe

rof

child

ren,

whe

ther

female-he

aded

househ

old,

househ

oldassetinde

x,ethn

icity,

and

relig

ion.

Rob

uststan

dard

errors

clusteredat

theun

itof

rand

omization,

thevilla

gelevel,in

parenthe

ses.

Wild

-cluster

bootstrapp-values

insqua

rebrackets.

The

p-valuefordiffe

rencein

coeffi

cients

isaF-testforwhe

ther

thecoeffi

cientdiffe

rsbe

tweentreatm

entgrou

ps.The

p-valuefordiffe

rencebe

tweenB+V

and

BV

testswhe

ther

thereis

anycomplem

entarity

betw

eenBCC

andvouche

rs.

∗∗∗p<

0.01,∗

∗p<

0.05,∗

p<0.1.

42

Page 43: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Table 9: Spillover Effects

(1) (2) (3)

Peer definitions:Mutually listed

as friends

BCC partic-ipant listed

spillover groupmother as friend

Spillover groupmother listedBCC partici-pant as friend

Dependent variable: Mother IYCF knowledge score (standardized)

# of BCC peers 0.111 0.182∗ 0.071(0.305) (0.103) (0.104)

Observations 275 275 275R-squared 0.112 0.120 0.113This table reports results on the number of BCC-participating peers spillover group mother has defined by both the

spillover group and BCC-participating mother networks (Column 1), the BCC-participating mothers’ networks

(Column 2), and the spillover group mothers’ networks (Column 3). All estimations control for the baseline

outcome, area dummies, mother’s age, whether mother is married, pregnant, working, able to read, and able to

write, mother’s years of schooling, total number of friends listed as network, eligible child age, household size,

number of children, whether female-headed household, household asset index, ethnicity, and religion. Robust

standard errors clustered at the unit of randomization, the village level, in parentheses. ∗∗∗ p<0.01, ∗∗ p<0.05, ∗

p<0.1.

43

Page 44: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Table 10: Heterogeneity Analysis

(1) (2) (3) (4) (5)Mother

knowledgescore (stan-dardized)

CDDS (0-7)Mother self-employedfarming

Stunting Wasting

Panel A. Rural

BCC x Rural (B) 0.388∗ 0.204 0.128∗∗ -0.169 -0.0703(0.198) (0.341) (0.062) (0.122) (0.056)[0.090] [0.556] [0.064] [0.231] [0.254]

Voucher x Rural (V) -0.174 0.0259 -0.117 0.0847 0.005(0.282) (0.369) (0.079) (0.103) (0.068)[0.604] [0.964] [0.197] [0.459] [0.961]

BCC & Voucher 0.083 -0.599∗∗ 0.001 0.156∗ -0.035x Rural (BV) (0.226) (0.290) (0.096) (0.084) (0.057)

[0.754] [0.082] [0.994] [0.123] [0.549]

Observations 580 579 580 513 526R-squared 0.086 0.113 0.360 0.177 0.015P-value: B=V 0.0575 0.670 0.003 0.066 0.321P-value: B=BV 0.222 0.024 0.212 0.012 0.597P-value: V=BV 0.449 0.105 0.296 0.507 0.584Panel B. Mothers with no formal schooling

BCC x No schooling 0.423∗ 0.221 -0.080 -0.012 -0.102∗

(B) (0.239) (0.428) (0.107) (0.120) (0.056)[0.114] [0.651] [0.472] [0.901] [0.118]

Voucher -0.157 -0.170 -0.201∗∗ -0.009 -0.001x No schooling (V) (0.274) (0.368) (0.089) (0.155) (0.068)

[0.570] [0.650] [0.045] [0.954] [0.988]

BCC & Voucher 0.337∗ -0.224 -0.092 0.004 -0.105∗

x No schooling (BV) (0.171) (0.300) (0.085) (0.091) (0.054)[0.092] [0.488] [0.317] [0.97] [0.106]

Observations 580 579 580 513 526R-squared 0.100 0.110 0.247 0.168 0.021P-value: B=V 0.081 0.448 0.320 0.985 0.183P-value: B=BV 0.726 0.332 0.918 0.894 0.968P-value: V=BV 0.087 0.896 0.288 0.932 0.166

continued on next page

44

Page 45: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

continued from previous page

(1) (2) (3) (4) (5)Mother

knowledgescore (stan-dardized)

CDDS (0-7)Mother self-employedfarming

Stunting Wasting

Panel C. Eligible child 12 months or younger at baseline

BCC x Under 12m -0.254 0.0875 0.108∗ 0.057 -0.031(B) (0.229) (0.296) (0.057) (0.129) (0.064)

[0.310] [0.800] [0.060] [0.674] [0.658]

Voucher 0.089 -0.025 0.004 -0.051 -0.018x Under 12m (V) (0.247) (0.255) (0.125) (0.097) (0.065)

[0.723] [0.920] [0.983] [0.603] [0.798]

BCC & Voucher 0.254 0.438∗∗ 0.0477 0.141∗ -0.0447x Under 12m (BV) (0.177) (0.212) (0.065) (0.075) (0.056)

[0.209] [0.049] [0.481] [0.065] [0.426]

Observations 580 579 580 513 526R-squared 0.094 0.109 0.364 0.172 0.014P-value: B=V 0.237 0.755 0.398 0.438 0.876P-value: B=BV 0.029 0.276 0.321 0.498 0.858P-value: V=BV 0.504 0.112 0.728 0.041 0.728Panel D. Single mothersBCC x Single mother 0.370 -0.755∗ -0.199 0.210 0.133(B) (0.256) (0.387) (0.143) (0.199) (0.132)

[0.170] [0.082] [0.199] [0.348] [0.365]

Voucher -0.122 -0.401 -0.196 0.328∗∗ 0.176x Single mother (V) (0.357) (0.467) (0.122) (0.153) (0.110)

[0.763] [0.439] [0.145] [0.0480] [0.157]

BCC & Voucher 0.177 -0.939∗ -0.206 0.102 0.0324x Single mother (BV) (0.510) (0.487) (0.142) (0.127) (0.088)

[0.901] [0.190] [0.158] [0.449] [0.716]

Observations 580 579 580 513 526R-squared 0.088 0.113 0.360 0.174 0.024P-value: B=V 0.137 0.518 0.983 0.550 0.782P-value: B=BV 0.695 0.744 0.956 0.547 0.473P-value: V=BV 0.586 0.385 0.931 0.067 0.219This table reports heterogeneous impacts of intervention by various household characteristics (whether reside in ruralarea, whether mother has no formal schooling, whether eligible child is 12 months or younger at baseline, and whethersingle mother). The point estimates are from interaction terms between treatment and the household characteristic ofinterest. All estimations include baseline outcome and area dummies. Robust standard errors clustered at the unit ofrandomization, the village level, in parentheses. Wild-cluster bootstrap p-values in square brackets. The p-value fordifference in coefficients is a F-test for whether the coefficient differs between treatment groups. ∗∗∗ p<0.01, ∗∗ p<0.05,∗ p<0.1.

45

Page 46: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

ReferencesAhmed, T, M Mahfuz, S Ireen, A Ahmed, S Rahman, M Islam, and F Choudhury (2012).

“Nutrition of children and women in Bangladesh: Trends and directions for the future”.In: Journal of Health Population and Nutrition, pp. 1–11.

Avula, R, P Menon, K Saha, M Bhuiyan, A Chowdhury, S Siraj, and E Frongillo (2013).“A program impact pathway analysis identifies critical steps in the implementation andutilization of a behavior change communication intervention promoting infant and childfeeding practices in Bangladesh”. In: The Journal of Nutrition 143.12, pp. 2029–2037.

Bandiera, O, R Burgess, N Das, S Gulesci, I Rasul, and M Sulaiman (2017). “Labor Mar-kets and Poverty in Village Economies”. In: The Quarterly Journal of Economics 132.2,pp. 811–870.

Banerjee, A, E Duflo, N Goldberg, D Karlan, R Osei, W Pariente, J Shapiro, B Thuysbaert,and C Udry (2015). “A multifaceted program causes lasting progress for the very poor:Evidence from six countries”. In: Science 348.6236.

Barrett, Chris (2010). “Measuring Food Insecurity”. In: Science 327, pp. 825–828.Beaudry, M, R Dufour, and S Marcoux (1995). “Relation between infant feeding and infec-

tions during the first 6 months of life”. In: Journal of Pediatrics 126, pp. 191–97.Bhutta, Z, J Das, A Rizvi, M Gaffey, N Walker, S Horton, P Webb, A Lartey, and R Black

(2013). “Evidence-based interventions for improvement of maternal and child nutrition:what can be done and at what cost?” In: The Lancet 382.9890, pp. 452–477.

Bilal, S, G Dinant, R Blanco, R Crutzen, A Mulugeta, and M Spigt (2013). “The influenceof father’s child feeding knowledge and practices on children’s dietary diversity: a studyin urban and rural districts of Northern Ethiopia”. In: Maternal & Child Nutrition 12,pp. 473–83.

Black, E, H Allen, A Bhutta, E Caulfield, M de Onis, M Ezzati, C Mathers, J Rivera, andMaternal and Child Undernutrition Study Group (2008). “Maternal and child undernu-trition: global and regional exposures and health consequences”. In: The Lancet 371.9608,pp. 243–260.

Cameron, C, J Gelbach, and D Miller (2008). “Bootstrap-based improvements for inferencewith clustered errors”. In: Review of Ecnomics and Statistics 90.3, pp. 414–427.

Cunningham, A, D Jelliffe, and E Jelliffe (1991). “Breastfeeding and health in the 1980s: aglobal epidmiologic review”. In: Journal of Pediatrics 118, pp. 659–66.

Dereje, M (2015). Food Consumption Changes in Ethiopia. Addis Ababa, Ethiopia.Dewey, K, M Heinig, and L Nommsen-Rivers (1995). “Differences in morbidity between

breastfed and formula fed infants”. In: Journal of Pediatrics 126, pp. 696–702.Ethiopia DHS (2016). Ethiopia DHS. url: https://dhsprogram.com/what-we-do/survey/

survey-display-478.cfm.Fisher, R (1935). The Design of Experiments. MacMillan.Fitzsimons, E, B Malde, A Mesnard, and M Vera-Hernandez (2016). “Nutrition, information

and household behavior: Experimental evidence from Malawi”. In: Journal of Develop-ment Economics 122, pp. 113–126.

Gupta, M Das, M Lokshin, M Gragnolati, and O Ivaschenko (2005). Improving Child Nutri-tion Outcomes in India: Can the Integrated Child Development Services Be More Effec-tive? Washington DC.

46

Page 47: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Headey, D, A Chiu, and S Kadiyala (2012). “Agriculture’s role in the Indian enigma: Helpor hindrance to the malnutrition crisis?” In: Food Security 4, pp. 87–102.

Hidrobo, M, J Hoddinott, A Peterman, A Margolies, and V Moreira (2014). “Cash, food, orvouchers? Evidence from a randomized experiment in northern Ecuador”. In: Journal ofDevelopment Economics 107, pp. 144–156.

Hoddinott, J, I Ahmed, A Ahmed, and S Roy (2017). “Behavior change communicationactivities improve infant and young child nutrition knowledge and practice of neighboringnon-participants in a cluster-randomized trial in rural Bangladesh”. In: PLoS One 12.6,e0179866.

Hoddinott, J, J Behrman, J Maluccio, P Melgar, A Quisumbing, M Ramirez-Zea, and RMartorell (2013). “Adult consequences of growth failure in early childhood”. In: TheAmerican Journal of Clinical Nutrition, ajcn–064584.

Jayachandran, S and R Pande (2017). “Why Are Indian Children So Short? The Role ofBirth Order and Son Preference”. In: American Economic Review 107.9, pp. 2600–2629.

Jones, G, W Steketee, E Black, A Bhutta, S Morris, and Bellagio Child Survival Study Group(2003). “How many child deaths can we prevent this year?” In: The Lancet 362.9377,pp. 65–71.

Manley, J, S Gitter, and V Slavchevska (2013). “How effective are cash transfers at improvingnutritional status?” In: World Development 48, pp. 133–155.

MEASURE Evaluation (2018). Behavior Change Communication–MEASURE Evaluation.url: https://www.measureevaluation.org.

Menon, P (2012). “Childhood undernutrition in south Asia: Perspectives from the field ofnutrition”. In: CESinfo Economic Studies 58.2, pp. 274–295.

Merwe, L Van der, S Moore, A Fulford, K Halliday, S Saikou Drammeh, and S Young (2013).“Long-chain PUFA supplementation in rural African infants: a randomized controlled rialof effects on gut interity, growth, and cognitive development”. In: Amreican Journal ofClinical Nutrition 97, pp. 45–57.

MOA (2014). Productive Safety Net Programme phase IV programme implementation man-ual. Addis Ababa.

Muller, O, M Garenne, P Reitmaier, A Baltussen Van Zweeden, B Kouyate, and H Becher(2003). “Effect of zinc supplementation on growth in West African children: a randomizeddouble-blind placebo-controlled trial in rural Burkina Faso”. In: International Journal ofEpidemiology 32, pp. 1098–102.

Newton, S, S Owusu-Agyei, and B Kirkwood (2007). “Is there any monitoring of the quality ofvitamin A capsules used in supplementation programs?” In: American Journal of ClinicalNutrition 86, p. 1254.

Nube, M (2009). “The Asian enigma: Predisposition for low adult BMI among people ofSouth Asian descent”. In: Public Health Nutrition 12, pp. 507–516.

Olney, K, A Pedehombga, M Ruel, and A Dillon (2015). “A 2-Year Integrated Agricultureand Nutrition and Health Behavior Change Communication Program Targeted to Womenin Burkina Faso Reduces Anemia, Wasting, Diarrhea in Children 3-12.9 Months of Ageat Baseline: A Cluster-Randomized Controlled Trial”. In: The Journal of Nutrition 145.6,pp. 1317–1324.

Pande, R (2003). “Selective Gender Differences in Childhood Nutrition and Immunizationin Rural India: The Role of Siblings”. In: Demography 40, pp. 395–418.

47

Page 48: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Prina, S and H Royer (2014). “The importance of parental knowledge: Evidence from weightreport cards in Mexico”. In: Journal of Health Economics 37, pp. 232–247.

Reinbott, A, A Schelling, J Kuchenbecker, T Jeremias, I Russell, O Kevanna, M Krawinkel,and I Jordan (2016). “Nutrition education linked to agricultural interventions improvedchild dietary diversity in rural Cambodia”. In: British Journal of Nutrition 116, pp. 1457–1468.

Rosenbaum, P (2002). Observational Studies. Springer.Ruel, M, H Alderman, and Maternal and Child Nutrition Study Group (2013). “Nutrition-

sensitive interventions and programmes: how can they help to accelerate progress inimproving maternal and child nutrition?” In: The Lancet 382.9891, pp. 536–551.

Schroff, M, P Griffiths, L Adair, C Suchindran, and M Bentley (2009). “Maternal autonomyis inversely related to child stunting in Andhra Pradesh, India”. In: Maternal and ChildNutrition 5.1, pp. 64–74.

Skoufias, E, V di Maro, T Gonzales-Cossio, and S Ramirez (2011). “Food quality, caloriesand household income”. In: Applied Economics 43, pp. 4331–4342.

UNFAO (2006). World Agriculture Toward 2030/2050, Interim Report. Rome.Weismann, D, L Bassett, and John Hoddinott (2009). “Validation of the World Food Pro-

gramme’s Food Consumption Score and Alternative Indicators for Household Food Se-curity”. In: IFPRI Discussion Paper 00870.

WFP (2008). Food Consumption Analysis: Calculation and Use of the Food ConsumptionScore in Food Security Analysis. Rome.

WHO (2010a). Assessing Infant and Young Child Feeding Practices. Part 2: Measurement.Geneva.

— (2010b). Indicators for assessing infant and young child feeding practices.World Bank (2006). Repositioning Nutrition as Central to Development: A Strategy for Large-

scale Action. Washington DC.— (2017). World Development Indicators. url: http://data.worldbank.org.

48

Page 49: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Appendix A. Figures and Tables

Table A1: Effects on BCC Attendance and Mother IYCF Knowledge by Topic

IYCF Topics:

Animalsourcefoods

VitaminA-richfruits& veg.

Malnutrition& care

Feedingquantity,frequency,thickness

Age ofintro-duction

Hygiene

(1) (2) (3) (4) (5) (6)Panel A. Attendance rate by topic

BCC (B) 0.690∗∗∗ 0.607∗∗∗ 0.732∗∗∗ 0.764∗∗∗ 0.655∗∗∗ 0.960∗∗∗

(0.028) (0.040) (0.052) (0.027) (0.024) (0.022)Voucher (V) -0.009 -0.013 -0.006 -0.0001 -0.0004 -0.001

(0.011) (0.021) (0.017) (0.010) (0.005) (0.012)BCC & Voucher (BV) 0.721∗∗∗ 0.664∗∗∗ 0.718∗∗∗ 0.782∗∗∗ 0.789∗∗∗ 0.808∗∗∗

(0.024) (0.043) (0.025) (0.016) (0.018) (0.033)

Observations 630 630 630 630 630 630R-squared 0.837 0.741 0.757 0.878 0.832 0.842P-value: B=BV 0.389 0.294 0.819 0.572 0.000 0.000Panel B. Knowledge score by topic

BCC (B) 0.352∗∗ 0.392∗∗∗ 0.334∗∗ 0.203∗∗ 0.308∗∗∗ 0.035(0.134) (0.085) (0.127) (0.092) (0.097) (0.160)[0.034] [0.000] [0.024] [0.035] [0.004] [0.832]

Voucher (V) 0.008 0.098 0.096 0.002 0.015 0.005(0.122) (0.105) (0.167) (0.114) (0.102) (0.118)[0.944] [0.394] [0.608] [0.992] [0.887] [0.971]

BCC & Voucher (BV) 0.282∗∗ 0.302∗∗∗ 0.346∗∗∗ 0.242∗∗∗ 0.209∗∗ 0.043(0.113) (0.094) (0.086) (0.088) (0.093) (0.103)[0.026] [0.004] [0.000] [0.006] [0.044] [0.695]

Observations 577 577 577 577 577 577R-squared 0.079 0.075 0.081 0.072 0.105 0.073P-value: B=V 0.025 0.010 0.226 0.088 0.004 0.866P-value: B=BV 0.638 0.371 0.931 0.708 0.297 0.966P-value: V=BV 0.056 0.067 0.158 0.050 0.065 0.791P-value: B+V=BV 0.695 0.177 0.706 0.814 0.437 0.992This table reports results on BCC attendance rate and mothers’ IYCF knowledge score (standardized) by IYCFtopic. Panel A uses administrative data and compares BCC attendance rates with the control group wherethe control and the voucher group’s attendance rates are set to zero. Panel B uses survey data on mothers’IYCF knowledge. All estimations include area dummies, mother’s age, whether mother is married, working, ableto read, and able to write, mother’s years of schooling, eligible child age, household size, number of children,whether female-headed household, household asset index, ethnicity, and religion. Panel B additionally controlsfor the baseline outcome. Robust standard errors clustered at the unit of randomization, the village level, inparentheses. Wild-cluster bootstrap p-values in square brackets. The p-value for difference in coefficients is aF-test for whether the coefficient differs between treatment groups. The p-value for difference between B+VandBV tests whether there is any complementarity between BCC and vouchers. ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1.

49

Page 50: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Table A2: Effects on Other Child Diet Quality Measures

Minimumdietarydiversity

Minimummeal

frequency

Numberof timesbreastfedyesterday

Number oftimes ate solidor semi-solidfood yesterday

(1) (2) (3) (4)

BCC (B) 0.044 0.070 0.347∗∗ 0.105(0.052) (0.076) (0.171) (0.228)[0.443] [0.414] [0.067] 0.637]

Voucher (V) -0.021 0.038 0.259 0.281∗

(0.050) (0.061) (0.179) (0.145)[0.648] [0.548] [0.185] 0.062]

BCC & Voucher (BV) 0.176∗∗∗ 0.135∗ -0.024 0.457∗∗

(0.050) (0.070) (0.178) (0.195)[0.021] [0.085] [0.913] 0.038]

Observations 577 434 484 573R-squared 0.129 0.064 0.147 0.057Control group mean 0.328 0.565 5.272 2.678P-value: B=V 0.227 0.684 0.672 0.401P-value: B=BV 0.030 0.476 0.085 0.188P-value: V=BV 0.001 0.186 0.147 0.360P-value: B+V=BV 0.060 0.802 0.019 0.822This table reports results on minimum dietary diversity, minimum meal frequency, number of timesbreastfed yesterday, and number of times ate solid or semi-solid food yesterday. All estimations includethe baseline outcome, area dummies, mother’s age, whether mother is married, working, able to read,and able to write, mother’s years of schooling, eligible child age, household size, number of children,whether female-headed household, household asset index, ethnicity, and religion. Robust standard errorsclustered at the unit of randomization, the village level, in parentheses. Wild-cluster bootstrap p-valuesin square brackets. The p-value for difference in coefficients is a F-test for whether the coefficient differsbetween treatment groups. The p-value for difference between B+V and BV tests whether there is anycomplementarity between BCC and vouchers. ∗∗∗ p<0.01, ∗∗ p<0.05, ∗ p<0.1.

50

Page 51: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Table A3: Effects on Hygiene Behavior

Hygiene score (standardized)(1)

BCC (B) 0.055(0.051)[0.337]

Voucher (V) -0.111∗∗

(0.051)[0.051]

BCC & Voucher (BV) 0.038(0.040)[0.363]

Observations 518R-squared 0.131Control group mean -0.071P-value: B=V 0.013P-value: B=BV 0.765P-value: V=BV 0.008P-value: B+V=BV 0.220This table reports results on hygiene score which takes a mean of howoften the mother washes hands before cooking, washes hands before feed-ing, washes food before preparing child meal, washes dishes before usingto feed child, washes dish with clean water, washes bottle before feeding,and sterilizes bottle (on a scale of 1=Never to 5=Always). All esti-mations control for the baseline outcome, area dummies, mother’s age,whether mother is married, working, able to read, and able to write,mother’s years of schooling, eligible child age, household size, numberof children, whether female-headed household, household asset index,ethnicity, and religion. Robust standard errors clustered at the unit ofrandomization, the village level, in parentheses. Wild-cluster bootstrapp-values in square brackets. The p-value for difference in coefficients is aF-test for whether the coefficient differs between treatment groups. Thep-value for difference between B+V and BV tests whether there is anycomplementarity between BCC and vouchers. ∗∗∗ p<0.01, ∗∗ p<0.05, ∗

p<0.1.

51

Page 52: Promoting Improved Infant and Young Child Feeding ...barrett.dyson.cornell.edu/NEUDC/paper_524.pdf · significantly, while food vouchers alone do not. The impacts on improved child-feeding

Appendix B. Mother IYCF BCC Curriculum

Week Contents Week Contents

1 Introduction 9A: Frequency & amount of complementary food

B: Eating schedule & discussion

2 Dietary diversity and weekly diet schedule 10 Recipe and cooking demonstration

3 When to start complementary feeding 11 Responsive feeding

4 Thickness & consistency of complementary food 12 Feeding during illness

5 Role play & discussion 13 Role play & discussion

6 Food variety-iron, proteins from meat 14 Hygienic preparation & storage of food

7A: Enrichment of complementary food

15 Group discussion & reviewB: Household food processing strategy

8 Role play & discussion 16 Testimonials & ceremony

52