Preventive Medicine 100 (2017) 152–158

Contents lists available at ScienceDirect

Preventive Medicine

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

The impact of household and community cash transfers on children's

food consumption in Indonesia
Dian Kusuma a,b,⁎, Margaret McConnell a, Peter Berman a, Jessica Cohen a
a
Harvard T.H. Chan School of Public Health, Department of Global Health and Population, 655 Huntington Avenue, Boston, MA 02115, USA
b
Center for Health Economics and Policy Studies, Faculty of Public Health, Universitas Indonesia, Kampus Baru UI, Depok 16424, Indonesia

a r t i c l e i n f o a b s t r a c t

Article history: The current state of child nutrition is critical. About 5.9 million children under the age of five still died worldwide
Received 24 August 2016 with nearly half are attributable to undernutrition. One explanation is inequality in children's food consumption.
Received in revised form 12 April 2017 One strategy to address inequality among the poor is conditional cash transfers (CCTs). Taking advantage of the
Accepted 22 April 2017
two large clustered-randomized trials in Indonesia from 2007 to 2009, this paper provides evidence on the im-
Available online 24 April 2017
pact of household cash transfer (PKH) and community cash transfer (Generasi) on child's food consumption.
Keywords:
The sample sizes are 14,000 households for PKH and 12,000 households for Generasi. After two years of imple-
Indonesia mentation, difference-in-differences (DID) analyses show that both cash transfers lead to significant increases
Conditional cash transfers in food consumption particularly for protein-rich items. The programs significantly increase the consumption
Household of milk and fish by up to 19% and 14% for PKH and Generasi, respectively. Both programs significantly reduce
Community some measures of severe malnutrition. PKH significantly reduces the probability of wasting and severe wasting
Children's food consumption by 33% and 41% and Generasi significantly reduces the probability of being severely underweight by 47%. This un-
Clustered-randomized trials derscores the potential of household and community cash transfers to fight undernutrition among the poor.
© 2017 Published by Elsevier Inc.

1. Introduction consumption of fruit and vegetables, and of animal products by 4.6%,

21.9%, and 17.4% in Mexico; increased food expenditure (largely on
The current state of child nutrition is critical. About 5.9 million chil- meat, fruits and vegetables) by 24.8 percentage points in Nicaragua
dren under the age of five died in 2015 worldwide and nearly half of all (Hoddinott and Skoufias, 2004; Maluccio and Flores, 2005).
child deaths are attributable to undernutrition (UNICEF et al., 2015a; However, the literature is limited in three ways. First, existing stud-
Black et al., 2013). One explanation is inequality in children's food con- ies use household-level but lack evidence using community-level CCTs.
sumption since poor children are more likely to have less diversified and The former is relatively more expensive given the need to monitor indi-
nutritious diets and to live in food-insecure households (Bhutta et al., viduals for conditionality. In Indonesia, the average administrative cost
2013; Black et al., 2008; Darnton-Hill and Coyne, 1998). About 159 mil- of a CCT (PKH) is almost three times as high as that of unconditional
lion children under-five were stunted and 50 million were wasted glob- cash transfer (UCT/BLT) (Alatas et al., 2011). Also, household-level
ally in 2014, of whom 96% live in poorer countries in Asia and Africa CCTs potentially lack community involvement such as supplementary
(UNICEF et al., 2015b). In Indonesia, the proportion of children under- feeding program. Secondly, evidence from large-scale RCT evaluation
weight is two times higher among the poorest compared to the wealth- is limited. The Mexican PROGRESA used a longitudinal sample of ap-
iest (Utomo et al., 2011). Since child nutrition contributes to many proximately 24,000 households while the Nicaraguan Red de Protección
health and productivity outcomes, addressing child undernutrition Social (RPS) used only 1742 households (Hoddinott and Skoufias, 2004;
among the poorest is needed (Victora et al., 2008). Maluccio and Flores, 2005). A large-scale randomized evaluation helps
One strategy is conditional cash transfers (CCTs), which provide cash establish evidence of whether the CCT impact is possible when taking
payments to poor households in exchange for compliance with health/ into account the complexity of implementation namely issues with
education-related conditionalities (Fiszbein et al., 2009). Evidence cash distribution, monitoring conditionality, and supply-side improve-
from randomized controlled trials (RCT) shows that CCTs improve ments (Madon et al., 2007). Thirdly, despite some evidence that CCTs
children's food consumption: increased total caloric availability, improve children's food consumption and expenditure the existing ev-
idence shows that significant impacts on nutritional outcomes are not
⁎ Corresponding author at: Harvard T.H. Chan School of Public Health, Department of
assured (Fiszbein et al., 2009; Leroy et al., 2009).
Global Health and Population, 655 Huntington Avenue, Boston, MA 02115, USA. Taking advantage of two large clustered-randomized trials in
E-mail address: [email protected] (D. Kusuma). Indonesia, we provide evidence on the impact of household cash

http://dx.doi.org/10.1016/j.ypmed.2017.04.020
0091-7435/© 2017 Published by Elsevier Inc.
 D. Kusuma et al. / Preventive Medicine 100 (2017) 152–158 153

transfers (PKH) and community cash transfers (Generasi) on children's Table 2

food consumption. The context and features of these programs may ex- Comparing program features of PKH and Generasi.

plain why we may not actually expect a cash transfer to change food Features PKH Generasi
consumption so the test that cash transfers significantly increase food Cash Quarterly cash transfer to Block grant to villages each
consumption at all is actually a novel one. In PKH, there were no specific mothers, through nearest post year; $8500 (2007) and
rules on how the cash must be spent by households and there were office; $60–220 per household $18,200 (2009) per village on
many implementation issues including delay in cash payments. In per year (approximately average; only for health and
15–20% of income); no specific education use.
Generasi, block grants were allocated to village management team
rules about how the transfer MIS data: health allocation is
who gets to decide on which activities to be funded toward health and must be used. 44% of total, which consists of
education. Furthermore, a comparison of the effectiveness of household 42% on SFPs, 25% on financial
and community cash transfers is important for policy options to over- assistance for mothers, 17% on
come the aforementioned limitations of household CCTs. One might ex- infrastructure, 9% on
facilities/equipment, 4% on
pect both programs to have differential results due to different financial incentive health
approaches (household v. community) and different characteristics of workers, and 2% on
population (PKH subdistricts are 75% urban and in Java while Generasi training/BCC
subdistricts are 90% rural), which could influence the outcomes differ- Conditionality Health and education Health and education
cash penalty indicators; cash penalty design indicators; no cash penalty.
ently. Also, community cash transfers are potentially less expensive
including first breach is
since monitoring is done at the village level rather than at the household warning; second breach is 10%
level. Previous evaluation of PKH found no significant effects on aggre- discount; third breach is
gate household consumption and that of Generasi provided no evalua- expulsion
tion of consumption (Alatas et al., 2011; Olken et al., 2011). Field Trained facilitators to advise Trained facilitators to advise
facilitators beneficiaries on conditionality village team on allocation of
and cash penalty. funds through social mapping
and others.
2. Methods Supply Mostly urban areas; readiness is Mostly rural areas
readiness precondition based on existing
health and education facilities;
2.1. Cash transfers: PKH and Generasi
threshold for readiness was set
lower for off-Java
The government of Indonesia piloted in 2007 two large-scale pilots: Target Very poor households (UCT Villages
(1) Program Keluarga Harapan (PKH), a CCT to household; and (2) beneficiaries database) with
Generasi, an incentivized community block grant program. The goals pregnant/lactating women and
children 0–15 years.
are to reduce poverty, maternal and child mortality, and to ensure uni-
Provinces West Java, East Java, North West Java, East Java, North
versal coverage of basic education. Details of the programs are provided Sulawesi, Gorontalo, East Nusa Sulawesi, Gorontalo, and East
elsewhere (Kusuma et al., 2016; Alatas et al., 2011; Olken et al., 2011) Tenggara, and Jakarta Nusa Tenggara
and in the Appendix. In brief, PKH is a traditional CCT program to very Note: MIS = management information systems; SFP = supplementary feeding program;
poor households (e.g. PROGRESA) while Generasi adopts that idea and BCC = behavioral change and communication; UCT = unconditional cash transfers. The
applies it in a way that allows communities the flexibility to address UCT database is of poor households, based on economic and asset-based poverty measure-
supply or demand constraint. PKH provided quarterly cash transfers di- ments by the Department of Statistics, who received the unconditional cash transfers in
2005 to mitigate the inflationary impact caused by fuel price adjustments. In terms of lo-
rectly to mothers in the amount of approximately 15–20% of income
cation, there were no overlapping districts in the pilot of PKH and Generasi even though
with no specific rules on how the cash must be used while Generasi pro- the provinces are similar.
vided annual block grant to villages to be used only for health and edu-
cation. Both programs were designed to achieve the same target
indicators or conditionality (Table 1). Like most CCTs, the conditions
are not specifically on nutrition but rather on growth monitoring and 2.2. Causal mechanisms of cash transfers and children's food consumption
nutritional supplements, which may affect child nutritional outcomes.
PKH was mostly in urban areas where area readiness is based on There are at least four pathways by which PKH and Generasi could
existing health and education facilities while Generasi was mostly in improve children's food consumption. The first pathway, for both PKH
rural areas. The summary of program features is provided in Table 2. and Generasi, is through the conditionalities. While households in
PKH areas are required to take children to health facilities for monthly
growth monitoring, those in Generasi areas are conditioned to use the
Table 1 block grants on health and education. This should encourage improve-
Conditionality and target indicators for PKH and Generasi. ment in children's food consumption in both programs. The second
pathway, also for both PKH and Generasi, is through improved knowl-
Health indicators
edge/information from increased interaction with health systems.
1. Four prenatal care visits
Both programs have the element of supply-side improvement, but the
2. Taking iron tablets during pregnancy
3. Delivery assisted by a trained professional
timing is different. PKH required areas to improve health facilities be-
4. Two postnatal care visits fore participating while Generasi targeted the less supply-ready areas
5. Complete childhood immunizations but allowed improvement activities. The management information sys-
6. Adequate monthly weight increases for infants tem (MIS) data show that 30% of Generasi health grants were for im-
7. Monthly weighing for children under three and biannually for children under five
proving infrastructure, facilities, equipment, and incentives for health
8. Vitamin A twice a year for children under five
workers (Olken et al., 2011). The third pathway is through the child
Education indicators cash element, which mainly applies for PKH. The cash element in PKH
9. Primary school enrollment of children 6-to-12 years old is an additional $80 per year only for households with a child under-
10. Minimum attendance rate of 85% for primary school-aged children five and/or a pregnant woman, which could serve as a nutrition cash el-
11. Junior secondary school enrollment of children 13-to-15 years old ement for these households. However, the MIS data for Generasi shows
12. Minimum attendance rate of 85% for junior secondary school-aged children only 9% of villages provided financial assistance for child health. The
Source: MOSA (2007); MHA (2008). fourth pathway, which applies only for Generasi, is supplementary
154 D. Kusuma et al. / Preventive Medicine 100 (2017) 152–158

feeding program (SFP). The MIS data indicate that 97% of the villages roots, tubers; (2) milk; (3) eggs; (4) meat; (5) fish; and (6) fruit and
implemented SFPs and took up 42% of the total health grants during vegetables. While these variables are closely related to the conditional-
2007/2008, which should help improve children's food consumption ity of growth monitoring in PKH and Generasi, eating milk and fish for
(see Table A.2 in the Appendix). instance is not a condition of the cash transfer programs. Since the
SFPs have been part of national and global nutrition policy for a long unit of analysis is at child level, any number of children in the reported
time (MOH, 2011; Horton et al., 2010; Bailey and Raba, 1976). The idea age range qualifies. The respondent was the mother or caregiver living
of SFP is to provide foods (free of charge or at low cost) to cover diet de- in the house. We did not use the variable of number of days due to lim-
ficiencies among vulnerable preschool children, school children, and itation in grouping the days. See the Appendix for more details.
pregnant/lactating women. Depending on the type of food SFPs can be In addition to the dependent variables, we use data on household
delivered on a communal basis or distributed periodically to homes food spending (the monetary value of the previous week's food con-
(Bailey and Raba, 1976). In Generasi, this includes: (a) monthly SFP sumption) and on posyandu activities (the number of children weighed
for children and mothers is given at posyandu events; (b) recovery and given supplementary foods) as additional analysis to suggest poten-
SFP for malnourished children is administered daily for three consecu- tial causal pathways. The household food spending was based food re-
tive months for severe malnutrition and one month for less severe call on monetary food consumption bought or obtained from other
cases; and (c) monthly SFP for school children. In some Generasi vil- sources in the last week reported by household head, spouse, or any fe-
lages, SFP was only provided to malnourished children in the forms of male adult in the house. Finally, we provide analysis on whether
biscuits and milk in West Java and rice and side dishes in NTT children's food consumption actually translates into nutrition outcomes
(Febriany et al., 2011). In Generasi, most SFP activities (79%) and including underweight (weight-for-age), wasting (weight-for-height),
funds (63%) were allocated to SFP Monthly for children and mothers and stunting (height-for-age). While stunting is an indicator of chronic
(see Table A.3 in the Appendix). malnutrition, both underweight and wasting reflect recent and severe
processes that lead to substantial weight loss. Wasting is calculated by
2.3. Evaluation design and data comparing weight-for-height of a child with a reference population
and stunting is calculated by comparing the height-for-age (Caulfield
We analyze the data from baseline (2007) and follow-up (2009) sur- et al., 2006; WFP, 2014). In the analyses, we use the proportion of un-
veys conducted by the National Planning Agency and World Bank for derweight and severe underweight and the proportion of wasting and
both PKH and Generasi. The pilots were large-scale with the overall severe wasting. Underweight and wasting use b− 2SD, while severe
samples of approximately 14,000 and 12,000 households for PKH and cases use b− 3SD of Z-score (WHO, 2006). In Stata, we use the
Generasi, respectively. The baseline surveys took place from June to Au- -zscore06- command to calculate anthropometric z-scores using the
gust 2007 for both programs. The follow-up surveys were implemented 2006 WHO child growth standards (Leroy, 2011).
from October to December 2009 for PKH and from October 2009 to
January 2010 for Generasi. Both were piloted in five provinces: West 2.5. Empirical specifications
Java, East Java, North Sulawesi, Gorontalo, and East Nusa Tenggara.
Additionally, PKH was also piloted in Jakarta. PKH covered mostly We use a difference-in-differences (DID) estimator to determine the
urban areas while Generasi did mostly rural areas. Ethical approval is average effect of the programs, controlling for differences at baseline.
not required because of using public data without identifiable private For PKH, we employ panel-DID because there is almost no attrition in
information. Random allocation was incorporated and there were no the sample of 24–36 month olds (N = 1394). For Generasi, we employ
overlapping districts. Detailed sampling is provided elsewhere repeated-cross-section-DID because of its follow-up survey design that
(Kusuma et al., 2016; Alatas et al., 2011; Olken et al., 2011; Sparrow et took half of the total sample from the baseline sample and took the
al., 2008) and in the Appendix. other half from new respondents (in the same area). This design is fa-
We use datasets for children's food consumption, household food vorable since Generasi targets the community as a whole as beneficia-
spending, and health provider. Children dataset is based on food con- ries. Below are the main specifications in our analyses.
sumption in the previous week for children 24–36 months old (at base-



line), which is the age limit in the survey questionnaire. For PKH, those Y ihjt ¼ α10 þ α11 PKH j þ α12 ðPostt Þ þ β11 PKH j  Postt þ X ihj γ1
children were followed up for two years. The PKH children sample can þ ϵihjt ð1Þ
be considered without attrition because it was 1394 at follow-up out



of 1395 at baseline. For Generasi, we use samples of children 24– Y ihjt ¼ α20 þ α
21 Generasi j þ α22 ðPostt Þ þ β21 Generasi j Postt
36 months in both survey periods – a repeated cross-section sample of þ X ihj γ2 þ ϵihjt ð2Þ
1481 children. For household food spending in the previous week, the
datasets have 1376 and 1472 households for PKH and Generasi, respec- where Yihjt is the outcome variable for child i in household h from sub-
tively. These are households with children 24–36 months old. For health district j in time period t; Post = 1 if in the follow-up period (2009)
provider posyandu SFPs data, we have 2831 and 2103 posyandus for and 0 otherwise; X is a vector of characteristics for child, household,
PKH and Generasi, respectively. The posyandu data are only at follow- and subdistrict; and ε is unobserved idiosyncratic error (assumed to
up since those at baseline are unavailable. be uncorrelated with all other variables). Eq. (1) is for the PKH program
where PKH = 1 if the child lives in a PKH treatment area and zero oth-
2.4. Dependent variables erwise. Eq. (2) is for the Generasi program where Generasi = 1 if the
child lives in a Generasi treatment area (both versions) and zero
In the datasets, there are only two variables on children's consump- otherwise.
tions based on food recall in the last week: whether the child consumed We estimate all equations using ordinary least squares (OLS) using
each type of food (binary variable) and how many days a child con- Stata 14 (StataCorp, 2014). The characteristics of children, households
sumed the food. The original types of food include milk, egg, meat, and subdistricts in Table 3 panel (a) are included as control variables
pork, chicken, fish, rice, vegetables, fruits, and snacks. We adapt food to improve the precision of the estimation. All standard errors are calcu-
grouping from nutrition literature (WHO, 2010) and cash transfers liter- lated allowing for heteroscedasticity and for clustering at the level of
ature on child nutrition (Macours et al., 2012; Attanasio and Mesnard, randomization (subdistrict). In Eqs. (1) and (2), the parameters of inter-
2006; Maluccio and Flores, 2005; Hoddinott and Skoufias, 2004). est are the β estimates, which are the DID estimates of the programs' ef-
Based on this, we have six main dependent variables of whether or fects for 2009 (relative to 2007). The programs' effects are identified by
not a child ate any of the following over the previous week: (1) grain, the randomized design. Because we do not condition on actual program
 D. Kusuma et al. / Preventive Medicine 100 (2017) 152–158 155

Table 3
Baseline characteristics and tests for balance in PKH and Generasi.

PKH Generasi PKH v. Generasi

Control Treatment Control Treatment Control mean

Mean Difference Mean Difference Difference
[1] [2] [3] [4] [5] = [1] − [3]

(a)
Child characteristics N = 1394 N = 627 N = 900
Female (=1) 0.45 0.06⁎⁎ 0.44 0.03 0.00
Age in months 30.11 0.09 30.06 0.26 0.05
Household characteristics N = 1375 N = 623 N = 889
Block wall (=1) 0.43 −0.02 0.57 −0.05 −0.14⁎⁎
Dirt floor (=1) 0.32 0.01 0.13 0.06 0.19⁎⁎
Zinc roof (=1) 0.27 −0.02 0.32 −0.03 −0.04
Tile roof (=1) 0.55 0.03 0.48 0.09 0.07
Latrine in house (=1) 0.48 −0.04 0.61 0.06 −0.13⁎⁎
House has electricity (=1) 0.77 −0.02 0.64 0.05 0.13⁎
Piped water into house (=1) 0.12 0.00 0.11 0.01 0.01
Wood & coal cooking fuel (=1) 0.77 0.03 0.74 0.02 0.03
Land owned (square meters) 2543.7 −495.5 4999.9 −381.54 −2456.28⁎⁎
At least one animal (=1) 0.54 −0.01 0.64 0.04 −0.10⁎⁎
Per capita expenditure (Ln) 5.36 −0.02 7.19 −0.01 −0.12⁎⁎
Mother's years of education 0.71 −0.01 0.81 0.05 −0.10⁎⁎
Mother's age in years 35.77 −0.33 33.47 −0.4 2.30⁎⁎
Household size 5.80 −0.13 5.10 −0.07 0.70⁎⁎
Number of under-five 1.46 −0.02 1.33 0.05 0.12⁎⁎
Sub-district characteristics N = 340 N = 233 N = 242
Doctor (number of, #) 6.21 0.02 4.15 0.08 2.06⁎⁎
Nurse (#) 9.93 1.12 9.57 −0.21 0.35
Midwife (#) 11.43 −0.06 9.80 −0.39 1.63⁎⁎
Pharmacy (#) 2.55 −0.10 1.69 −0.08 0.86⁎⁎
Health facility (#) 49.83 0.87 43.71 1.46 6.13⁎⁎

(b)
Dependent variables: children's food consumption in previous week (yes/no) N = 1394 N = 627 N = 900
Grain, roots, tubers 0.97 0.014⁎ 0.97 0.011 0.00
Milk 0.40 −0.015 0.49 −0.024 −0.09⁎
Meat 0.38 −0.057⁎ 0.44 0.037 −0.06
Fish 0.76 −0.026 0.80 −0.074 −0.03
Eggs 0.66 −0.005 0.69 0.016 −0.03
Fruit, vegetables 0.95 0.006 0.94 0.026 0.00

Note: Data comprise of children 24–36 months old and their households and subdistricts. Analyses use OLS regression comparing the means between treatments and control groups at
baseline. The estimates of PKH v. Generasi (column 6) are comparing the control means between PKH (column 1) and Generasi (column 3). Standard errors are not shown to save
space and are all clustered at the subdistrict level. The bold and italics show statistically significant estimates.
⁎ p b 0.1.
⁎⁎ p b 0.05.

participation when using the survey data, but only on whether the households are on average poorer but PKH subdistricts have more
household resides in a treatment locality, the estimates reflect the in- health workers and facilities (column 5).
tent-to-treat average effect of the program. In Table 4, we show the main results that both programs lead to sig-
We provide postestimation tests for comparisons between PKH and nificant increases in children's food consumption, particularly for milk
Generasi only in terms of their impact on nutrition. We start with test- and fish. PKH increases milk and fish consumption by 6.7 and 5.5 per-
ing the equality of treatment effects (β11 = β21), which would be valid if centage points (column 2), which translate into increases of 19% and
recipients of both programs were from similar samples. The balance 7% over the average level in control areas (column 1) – 19% is 0.067 di-
test, however, shows that the two programs have different starting vided by 0.36 and 7% is 0.055 divided by 0.81. Similarly, Generasi in-
levels of dependent variables and different sample characteristics. creases the consumption of milk and fish by 9.6 and 10.6 percentage
Thus, we also look at whether the overall treatment group means of de- points (column 5), which translate into 19% and 14% over the control
pendent variables at follow-up are similar (α10 + β11 = α20 + β21), level, respectively (column 4). When comparing between PKH and
where we use the constant term α without covariates. We include the Generasi, the postestimation tests show that we reject the equality of
characteristics of children, households and subdistricts as control vari- treatment group means at follow-up for those consumptions (p-values
ables in regressions to account for some of the baseline differences in of 0.031 and 0.088), indicating the impact is significantly larger in
the samples. More details are provided in the Appendix. Generasi (column 8).
Since both programs improve food consumption but have dissimilar
3. Results delivery platforms (i.e., household and community-based), we provide
evidence for potential causal mechanisms namely food spending and
In Table 3, we examine the differences in means at baseline between posyandu activities in Table 5. In terms of spending, results show that
treatment and control groups for child, household and subdistrict char- PKH significantly increases household spending on milk/eggs by
acteristics (panel a) and for dependent variables (panel b) in PKH and 1537.5 points, which translates to an 18% increase compared to the con-
Generasi. The balance test shows that the differences are mostly not sig- trol group mean (panel a column 2). Results also show that Generasi
nificant in PKH and Generasi (columns 2 and 4), as expected from ran- significantly increases household spending on fish by 3495.6 points or
dom allocation. We also compare between the two programs, which a 21% increase relative to the control group mean (panel a column 5).
show that household and subdistrict characteristics are different. PKH The postestimation tests show that we can reject the equality of
156 D. Kusuma et al. / Preventive Medicine 100 (2017) 152–158

Table 4
Difference-in-differences estimates of the impact of PKH and Generasi on children's food consumption.

Whether or not a child ate any of the PKH Generasi Postestimation tests (p-value)
following in the previous week (yes/no)
Control Treatment effects Control Treatment effects Ho: equality of treatment effects Ho: equality of treatment
group means at follow-up
Mean β11 SE Mean β21 SE β11 = β21 α10 + β11 = α20 + β21
[1] [2] [3] [4] [5] [6] [7] [8]
N = 1394 N = 1481

Grain, roots, tubers 0.99 −0.010 (0.010) 0.98 −0.003 (0.017) 0.690 0.905
Milk 0.36 0.067⁎ (0.035) 0.51 0.096⁎ (0.049) 0.634 0.031
Meat 0.36 0.041 (0.036) 0.43 0.037 (0.057) 0.945 0.201
Fish 0.81 0.055⁎ (0.033) 0.76 0.106⁎⁎ (0.050) 0.390 0.088
Eggs 0.68 0.033 (0.031) 0.68 0.014 (0.052) 0.750 0.855
Fruit, vegetables 0.97 −0.002 (0.015) 0.96 −0.010 (0.020) 0.770 0.489

Note: Data comprise children 24–36 months old (at baseline). Analyses use OLS regression using Eqs. (1) and (2) for PKH and Generasi for panel (a). Control mean is at the follow-up
period (2009). Control variables include characteristics of children, households and subdistricts (Table 3 panel a). For postestimation, we test the equality of treatment effects (β11 =
β21) and the equality of treatment group means at the follow-up period (α10 + β11 = α20 + β21) where we use the constant term without covariates. Standard errors (SE) are in paren-
theses and all clustered at the subdistrict level. The bold and italics show statistically significant estimates.
⁎ p b 0.1.
⁎⁎ p b 0.05.

treatment group means at follow-up for milk and fish spending (panel a 4. Discussion
columns 7 and 8), which indicate that those numbers are significantly
different between PKH and Generasi. In terms of posyandu activities, re- Our results show empirical evidence that PKH, a large-scale house-
sults show that the number of children weighed and given supplemen- hold CCT program, improves children's food consumption. While
tary foods significantly increases in Generasi but not in PKH areas. PROGRESA, a large-scale household CCT program in Mexico, increased
Generasi increases the number of children given supplementary foods children's consumptions of fruits, vegetables, and animal products by
by 16.38 units, which translate into a 51% increase compared to the con- up to 21.9%, PKH increases milk and fish consumption among children
trol means (panel b column 5). The postestimation tests show that we by up to 19%. Our results add new evidence on the external validity
can reject both equalities, indicating larger effects in Generasi (panel b of household CCT impact on children's food consumption despite
columns 7 and 8). potentially different demographics, food availability, and health sys-
Finally, we show whether the improvement of children's food tems between Latin America and South East Asia. Both are evidence of
consumption translates into improved child nutritional outcomes in effectiveness at scale taking into account various implementation issues
Table 6. PKH significantly reduces wasting and severe wasting by 6.3 such as monitoring conditions and supply-side improvement. Our re-
and 3.7 percentage points, which translate into reductions of 33% and sults also provide new evidence that Generasi, a large-scale community
41% compared to the control group means, respectively. Similarly, CCT program, improves children's food consumption. There is currently
Generasi reduces the prevalence of severely underweight children no evidence from large-scale and RCT evaluation in the world. Similar to
by 5.6 percentage points, which translates into a reduction of 47% PKH, Generasi are shown to increase milk and fish consumption by up to
compared to control areas (panel a columns 2 and 5). However, we can- 19%. And given the large scale, our results are already taking into ac-
not reject the equality of treatment effects and of treatment group count various implementation issues.
means at follow-up. Results show no impact of the programs on While the evidence from both PKH and Generasi provides policy op-
stunting. tions to children's food consumption and undernutrition, there are a

Table 5
Difference-in-differences estimates of the impact of PKH and Generasi on household food spending and posyandu activities.

PKH Generasi Postestimation tests (p-value)

Control Treatment effect Control Treatment effect Ho: equality of treatment effects Ho: equality of treatment
group means at follow-up
Mean β11 SE Mean β21 SE β11 = β21 α10 + β11 = α20 + β21
[1] [2] [3] [4] [5] [6] [7] [8]

(a) Household food spending in previous week N = 1376 N = 1472

Grain and roots 59,598.4 914.6 (1817.2) 52,983.7 −2859.7 (3118.6) 0.292 0.090
Milk and eggs 8441.8 1537.5⁎ (834.3) 14,665.3 1333.9 (1845.2) 0.919 0.037
Meat 8556.7 478.3 (1234.3) 12,919.7 −629.5 (2124.1) 0.650 0.300
Fish 15,659.8 −259.2 (957.8) 16,719.5 3495.6⁎⁎ (1630.7) 0.045 0.001
Fruit and vegetables 23,830.8 654.1 (1322.1) 23,220.8 1032.1 (1823.7) 0.866 0.122
(b) Posyandu activities in previous week N = 2831 N = 2103
Number of kids weighed 41.22 1.97 (1.44) 39.67 11.82⁎⁎⁎ (1.85) 0.001 0.001
Number of kids given supplementary feeding 36.18 0.42 (1.52) 32.30 16.38⁎⁎⁎ (1.94) 0.001 0.001

Note: Data comprise households with children 24–36 months old (at baseline) and posyandus for panel (a) and (b), respectively. Analyses use OLS regression using Eqs. (1) and (2) for
PKH and Generasi, respectively. Posyandu estimates in panel (b) are single-difference since baseline data are unavailable. Control means are at follow-up period (2009). Control variables
include characteristics of children, households (excluding per capita expenditure because of endogeneity), and subdistricts for panel (a) and include only characteristics of subdistrict for
panel (b). For postestimation, we test the equality of treatment effects (β11 = β21) and the equality of treatment group means (α10 + β11 = α20 + β21) where we use the constant term
without covariates. Standard errors (SE) are in parentheses and all clustered at the subdistrict level. The bold and italics show statistically significant estimates.
⁎ p b 0.1.
⁎⁎ p b 0.05.
⁎⁎⁎ p b 0.01.
 D. Kusuma et al. / Preventive Medicine 100 (2017) 152–158 157

Table 6
Difference-in-differences estimates of the impact of PKH and Generasi on nutrition outcomes.

PKH Generasi Postestimation tests (p-value)

Control Treatment effect Control Treatment effect Ho: equality of treatment effects Ho: equality of treatment group means at follow-up

Mean β11 SE Mean β21 SE β11 = β21 α10 + β11 = α20 + β21
[1] [2] [3] [4] [5] [6] [7] [8]
N = 1394 N = 1481

Underweight 0.38 −0.040 (0.036) 0.34 −0.020 (0.051) 0.739 0.337

Severe underweight 0.10 −0.025 (0.024) 0.12 −0.056⁎ (0.034) 0.453 0.320
Wasting 0.19 −0.063⁎⁎ (0.032) 0.19 −0.010 (0.035) 0.265 0.635
Severe wasting 0.09 −0.037⁎ (0.022) 0.10 −0.021 (0.025) 0.640 0.869
Stunting 0.55 0.035 (0.046) 0.48 0.034 (0.055) 0.992 0.477
Severe stunting 0.29 0.047 (0.053) 0.29 −0.06 (0.053) 0.149 0.003

Note: Data comprise children 24–36 months old (at baseline). Analyses use OLS regression using Eqs. (1) and (2) for PKH and Generasi for panel (a). Outcome variables are proportions of
underweight/severe, wasting/severe, and stunting/severe. Underweight, wasting, and stunting use b−2SD, while severe cases use b−3SD of Z-score. Control means are at the follow-up
period (2009). Control variables include characteristics of children, households, and subdistricts (Table 3 panel a). For postestimation, we test the equality of treatment effects (β11 = β21)
and the equality of treatment group means at follow-up (α10 + β11 = α20 + β21), where we use the constant term without covariates. Standard errors (SE) are in parentheses and all
clustered at the subdistrict level. The bold and italics show statistically significant estimates.
⁎ p b 0.1.
⁎⁎ p b 0.05.

few considerations in favor of Generasi, particularly in developing coun- to fight undernutrition among the poor and to reach their goals toward
try settings. First, because activities in Generasi also include supply-side poverty reduction and human capital improvement.
improvement (e.g. infrastructure renovation and equipment procure-
ment), Generasi shows impact in resource-limited settings. Secondly,
Conflict of interest statement
because there is no individual monitoring for conditionality in Generasi,
None declared.
it is likely to be less expensive than household CCTs. While we do not
have information on the comparison of overall investment between
Appendix A. Supplementary data
PKH and Generasi, Alatas et al. (2011) mentioned that the average ad-
ministrative cost of PKH is almost three times as high as that of uncon-
Supplementary data to this article can be found online at http://dx.
ditional cash transfer. Third, instead of heavily relying on each
doi.org/10.1016/j.ypmed.2017.04.020.
household as in PKH, Generasi relies on each community. Generasi com-
munity was involved in planning activities such as identifying target
groups and prioritizing village health problems, guided by trained facil- References
itators and targets indicators. The community also involved in simple
health systems interventions most notably SFPs (e.g. preparing for Alatas, V., et al., 2011. Program Keluarga Harapan: Main Findings From the Impact
Evaluation of Indonesia's Pilot Household Conditional Cash Transfer Program.
rice, dishes, biscuits and milk) (Febriany et al., 2011). SFPs were a World Bank.
major part of Generasi during the pilot and implemented in nearly all Attanasio, O., Mesnard, A., 2006. The impact of a conditional cash transfer programme on
villages (97%) and took up almost half (42%) of total health grants. In ad- consumption in Colombia. Fisc. Stud. 27 (4), 421–442.
Bailey, K.V., Raba, A., 1976. Supplementary feeding programmes. Chapter 21. In: Beaton,
dition, Generasi also provides financial assistance for mothers with chil- G.H., Bengoa, J.M. (Eds.), Nutrition in Preventive Medicine. World Health Organiza-
dren, which was implemented in nearly all villages (96%) and took up a tion, Geneva (1976).
quarter of total health grants. Bhutta, Z.A., Das, J.K., Rizvi, A., Gaffey, M.F., Walker, N., Horton, S., ... Black, R.E., 2013. Ev-
idence-based interventions for improvement of maternal and child nutrition: what
While comparison between household and community cash
can be done and at what cost? Lancet 382 (9890), 452–477.
transfers is important for policy, direct comparison is limited in our Black, R.E., Allen, L.H., Bhutta, Z.A., Caulfield, L.E., De Onis, M., Ezzati, M., ... Rivera, J., 2008.
analysis due to different program objectives and sample characteristics. Maternal and child undernutrition: global and regional exposures and health conse-
quences. Lancet 371 (9608), 243–260.
Generasi aims toward improvement in less supply-ready rural areas,
Black, R.E., Victora, C.G., Walker, S.P., Bhutta, Z.A., Christian, P., De Onis, M., ... Uauy, R.,
while PKH aims toward poorer households in supply-ready urban 2013. Maternal and child undernutrition and overweight in low-income and mid-
areas. Nevertheless, postestimation tests were used to help with the dle-income countries. Lancet 382 (9890), 427–451.
comparisons. Future research should provide direct comparison: house- Caulfield, L., Richard, S., Rivera, J., Musgrove, P., Black, R., 2006. Stunting, wasting, and mi-
cronutrient deficiency disorders, chapter 28. In: Jamison, D.T., Breman, J.G., Measham,
hold v. community cash transfers. A.R., Alleyne, G., Claeson, M., Evans, D.B., …, Musgrove, P. (Eds.), Disease Control Pri-
orities in Developing Countries. World Bank Publications (2006).
Darnton-Hill, I., Coyne, E.T., 1998. Feast and famine: socioeconomic disparities in global
nutrition and health. Public Health Nutr. 1 (01), 23–31.
5. Conclusion Febriany, V., Toyamah, N., Sodo, J., Budiyati, S., 2011. Qualitative Impact Study for PNPM-
Generasi and PKH on the Provision and the Utilization of the Maternal and Child
This paper provides causal impact of two large-scale household and Health Services and Basic Education Services in the Provinces of West Java and East
Nusa Tenggara. The SMERU Research Institute, Jakarta, Indonesia.
community cash transfers on children's food consumptions. After two Fiszbein, A., Schady, N.R., Ferreira, F.H., 2009. Conditional Cash Transfers: Reducing Pres-
years of implementation of PKH (a large-scale household CCT) and ent and Future Poverty. World Bank Publications.
Generasi (a large-scale incentivized community block grant), analyses Hoddinott, J., Skoufias, E., 2004. The impact of PROGRESA on food consumption. Econ.
Dev. Cult. Chang. 53 (1), 37–61.
show that both programs significant increase children's food consump-
Horton, S., Shekar, M., McDonald, C., Mahal, A., Krystene Brooks, J., 2010. Scaling Up Nu-
tion particularly for protein-rich items. They increase milk and fish con- trition: What Will It Cost? World Bank, Washington, DC.
sumption by up to 19% and 14%. There is evidence that the effects of Kusuma, D., Cohen, J., McConnell, M., Berman, P., 2016. Can cash transfers improve deter-
minants of maternal mortality? Evidence from the household and community pro-
Generasi are attributable to supplementary feeding. It is likely that the
grams in Indonesia. Soc. Sci. Med. 163, 10–20.
improved food consumptions, additional cash, and improve health Leroy, J.L., 2011. zscore06: Stata Command for the Calculation of Anthropometric z-scores
knowledge from increased interaction with health facilities contribute Using the 2006 WHO Child Growth Standards. http://www.ifpri.org/staffprofile/jef-
to improved nutrition outcomes. Our analyses show that PKH reduces leroy.
Leroy, J.L., Ruel, M., Verhofstadt, E., 2009. The impact of conditional cash transfer
wasting and severe wasting up to 41% and Generasi reduces severe un- programmes on child nutrition: a review of evidence using a programme theory
derweight up to 47%. All this underscores the potential of both programs framework. J. Dev. Eff. 1 (2), 103–129.
158 D. Kusuma et al. / Preventive Medicine 100 (2017) 152–158

Macours, K., Schady, N., Vakis, R., 2012. Cash transfers, behavioral changes, and cognitive UNICEF, WHO, WB, UN, 2015a. Levels and Trends in Child Mortality. UN Inter-agency
development in early childhood: evidence from a randomized experiment. Am. Econ. Group for Child Mortality Estimation.
J. Appl. Econ. 4 (2), 247–273. UNICEF, WHO, WB, 2015b. Levels and trends in child malnutrition. UNICEF – WHO –
Madon, T., Hofman, K.J., Kupfer, L., Glass, R.I., 2007. Implementation science. December 20 World Bank Group Joint Child Malnutrition Estimates. Key Findings of the 2015
2007. Science Magazine. Edition.
Maluccio, J., Flores, R., 2005. Impact Evaluation of a Conditional Cash Transfer Program: Utomo, B., Sucahya, P.K., Utami, F.R., 2011. Priorities and realities: addressing the rich-
The Nicaraguan Red de Protección Social. Intl Food Policy Res Inst. poor gaps in health status and service access in Indonesia. Int. J. Equity Health 10
MHA, 2008. PNPM Generasi Operational Guideline (in Bahasa). Ministry of Home Affairs, (47), 1–14.
Jakarta. Victora, C.G., Adair, L., Fall, C., Hallal, P.C., Martorell, R., Richter, L., Sachdev, H.S., 2008. Ma-
MOH, 2011. Guideline for Supplementary Feeding Program for Malnourished Children (in ternal and child undernutrition: consequences for adult health and human capital.
Bahasa). Ministry of Health, Jakarta. Lancet 371 (9609), 340–357.
MOSA, 2007. PKH General Guideline (in Bahasa). Ministry of Social Affairs, Jakarta. WFP, 2014. Hunger Glossary. Retrieved from. www.wfp.org/hunger/glossary.
Olken, B.A., Onishi, J., Wong, S., 2011. Indonesia's PNPM Generasi Program: Final Impact WHO, 2006. WHO child growth standards: length/height-for-age, weight-for-age,
Evaluation Report. World Bank. weight-for-length, weight-for-height and body mass index-for-age: methods and de-
Sparrow, R., Moeis, J.P., Damayanti, A., Herawati, Y., 2008. Conditional Cash Transfers in velopment. World Health Organization.
Indonesia: Baseline Survey Report. Program Keluarga Harapan and PNPM-Generasi. WHO, 2010. Indicators for Assessing Infant and Young Child Feeding Practices: Part 2:
World Bank. Measurement. World Health Organization, Geneva.
StataCorp, 2014. Stata Statistical Software: Release 14.1. StataCorp LP, College Station, TX.