Agricultural & Applied Economics Association

Child Growth, Shocks, and Food Aid in Rural Ethiopia Author(s): Takashi Yamano, Harold Alderman, Luc Christiaensen Source: American Journal of Agricultural Economics, Vol. 87, No. 2 (May, 2005), pp. 273-288 Published by: Oxford University Press on behalf of the Agricultural & Applied Economics Association Stable URL: http://www.jstor.org/stable/3697844 . Accessed: 11/04/2011 17:35
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at . http://www.jstor.org/action/showPublisher?publisherCode=aaea. . Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Agricultural & Applied Economics Association and Oxford University Press are collaborating with JSTOR to digitize, preserve and extend access to American Journal of Agricultural Economics.

http://www.jstor.org

CHILD GROWTH, SHOCKS, AND FOOD AID IN RURAL ETHIOPIA

TAKASHIYAMANO,HAROLD ALDERMAN,AND LUC CHRISTIAENSEN
Child stuntingin Ethiopiahas persistedat alarmingrates,despite enormousamountsof food aid, often procured responseto shocks.Using nationally in data,the studyfindsthat while representative harvestfailureleadsto childgrowthfaltering, food aid affectedchildgrowthpositivelyandoffsetthe

shocksdid not receivefood aid.In sum,whilefood aid has helpedreducechildmalnutriexperienced tion, inflexiblefood aid targeting, education,has togetherwithendemicpovertyandlimitedmaternal left the prevalenceof childstuntingat alarming levels. childgrowth,childmalnutrition, Keywords: Ethiopia,food aid,shocks.

of in effects shocks communities received aid.However, communities that that food negative many

Children that grow slowly experience poorer psychomotor development and interact less frequently in their environment (GranthamMcGregor et al.). They tend to delay school enrolment, and score less well on cognitive tests (Martorell, 1997). Moreover, the detrimental effects of slow height growth during early childhood may be long lasting. For example, Alderman, Hoddinott, and Kinsey find that in Zimbabwe, lowered stature as a pre-schooler following exposure to the 1982-84 drought resulted in a permanent loss of stature of 2.3 cm, a delay in starting school of 3.7 months, and 0.4 grades less of completed schooling. The combined effect of these factors was estimated to reduce lifetime earnings by 7%. Rural households in developing countries often live in risky environments, unable to fully protect their consumption against temporary income shocks such as droughts (Dercon). The available empirical evidence to date on the effect of such income shocks on child growth suggests pervasive growth retardation (Martorell, 1999; Hoddinott and Kinsey). As such temporary income shocks may cause permanent damage to children's future welfare and cogTakashi Yamano is Fellow at the Foundation for Advanced Studies on International Development. Harold Alderman is Lead Human Development Economist. Luc Christiaensen is Economist in the Africa Region of the World Bank. The authors would like to thank John Hoddinott, Martin Ravallion, Norbert Schady, and John Strauss as well as seminar participants at the Chronic Poverty Conference at Manchester University for useful comments. Nonetheless, the findings,interpretations, and conclusions expressed are entirely those of the authors, and they do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent.

nitive abilities (World Bank), further empirical investigation to quantify the magnitude of the effect of such shocks on early child growth is called for. A common intervention to alleviate the effects of drought shocks is food aid, often motivated by explicit reference to its beneficial effect on child malnutrition. Ironically, however, there is limited research on the effect of food aid on child growth (Barrett). The literature has so far mainly focused on food aid targeting, i.e., whether the poor are reached or not (von Braun; Sharp;Clay, Molla, and Habtewold; Jayne et al.), without examining the actual welfare effects of food aid for its beneficiaries. One notable exception is Quisumbing who finds positive effects of food aid programs on weight-for-height z-scores of children using panel data from Ethiopia.' Examining the effect of shocks and food aid on child growth is often complicated by the lack of sufficiently integrated data sets as well as the methodological difficulties in separating the causal effects of food aid on children's nutritional status from the reverse causality. Food aid programs are generally
'Other related studies include Dercon and Krishnan who examine the extent to which food aid helps households smooth their consumption (as opposed to nutritional outcomes) in the face of negative income shocks while taking into account the existing informal risk sharing arrangements. Their results, based on panel data from Ethiopia, indicate positive effects of food aid on consumption smoothing, though largely via intra-village risk sharing and not through direct targeting. Brown, Yohannes, and Webb, and Webb and Kumar look at the relation between child malnutrition and participation in food for work programs. They find positive relationships but were unable to establish causality.

2005AmericanAgricultural EconomicsAssociation Copyright

Amer. J. Agr. Econ. 87(2) (May 2005): 273-288

274

May2005

Amer.J. Agr Econ.

targeted to poorer areas and neglecting the endogeneity of program placement may lead to substantial underestimates of their effect (Rosenzweig and Wolpin; Pitt, Rosenzweig, and Gibbons). This study addresses these challenges by integrating three different nationally representative surveys from Ethiopia conducted over the period 1995-96. In doing so, it complements the findings of Quisumbing which are based on a sample from 15 purposively selected villages. Moreover, the study differs from others in the literature in that it focuses on child growth rather than achieved status, thereby matching the flow of food aid with the outcome, and by explicitly disentangling the effect of the shock per se on nutrition from the effect of the assistance. To control for program placement effects, food aid allocations have been instrumented with past food aid needs assessments and longterm rainfall patterns as captured by average rainfall and the coefficient of variation of the rainfall. Our focus on Ethiopia is motivated by the alarmingly high pre-school child stunting rates that have persisted at around 60% since the early 1980s and are among the highest in the world (Christiaensen and Alderman). Yet, Ethiopia has received massive amounts of food aid over the past decades often in response to severe droughts, which are a frequently recurring phenomenon.2 These facts have led some to question the effectiveness of food aid in reducing child malnutrition. Our results indicate that crop damage has a large detrimental effect on early child growth (measured in height) with children aged six to twenty-four months experiencing about a 0.9 cm growth loss over a six-month period compared to communities whose percentage of damaged crop area was 50% points lower. We also find that food aid affects child growth positively, especially among the six to twentyfour months old who grew on average 1.8 cm faster in the food aid receiving communities than if no food aid would have been available. The empirical analysis further suggests that the total amount of food aid distributed offsets the growth damage from the income shock in food aid receiving communities. However, many communities who experienced shocks,
2 About one-fifth to a quarter of all food aid deliveries to Africa over the past decades has gone to Ethiopia, with food aid attaining up to 20% of domestic production in drought years (Jayne et al.). According to World Food Programme estimates, Ethiopia has been the second largest recipient of food aid in the world for 1994-98 (after Bangladesh).

did not receive food aid, due to inflexible targeting rules, leaving many children's growth unprotected from drought shocks, consistent with the high child stunting rates observed. In the next section, we lay out the conceptual framework and our estimation strategy. Subsequently, the data are described. We then discuss determinants of food aid allocation and the effect of food aid and income shocks on child growth, before making concluding remarks. Conceptual Framework As outlined in Foster, Deolalikar, and Dercon and Hoddinott, we first consider a general intertemporal household utility model defined over household consumption and child health including preference shifters (A), such as gender-based parental attitudes towards their children. This maximization problem is further subjected to an intertemporal budget constraint and a health production function. A child i's height at t + 1, hit+l, can then be derived as a function of its initial height, hit, its household income, yit, observed characteristics at the individual, household, and community level, Xit, as well as unobserved individual (eit), household (uit), and community (vjt) characteristics: = f (hit, yit, Xi,, eit, uiv, t, A). Household income, Yit, is determined by household characteristics including household assets, and community characteristics. Drought or insect related crop failure also affects income, especially among subsistence farmers who form the large majority in rural Ethiopia. Income can be defined as inclusive of transfers, including food aid. Household income, yit, can thus be written as: (1) hit+l
(2) Yit = y(Sit, Fit, Xit, uit, Vjt).

By substituting equation (2) into equation (1), a child i's height becomes
(3) hit+1 =

A) (hit, Sit, Fit, Xit, eit, uit, Vjtr, .

Plot damage is assumed to affect household consumption negatively, and thus child growth, especially when households are unable to insure their consumption from income shocks. Food aid is expected to have a positive effect on child growth by supplementing household income and increasing food consumption. Its

Yamano,Alderman, and Christiaensen

Child Growth and Food Aid

275

effect on child growth will further depend on erations; it is also consistent with the policy the intra-household allocation and thus age question we investigate. The evaluation literand gender preferences of the parents (A) re- ature makes the distinction between the imgarding their children as well as the modal- pact of a program on those who receive the ity under which it is distributed-free distribu- treatment and the impact of the "intention to tion (FD) or food for work (FFW). When food treat." While both convey useful information, aid is freely distributed, as is mostly the case Heckman, Lalonde, and Smith observe in their in our sample, it also frees up time for child review of econometric methodologies for evalcare, another important input in child growth uation that often it is the latter that is of policy (Engle, Menon, and Haddad). Moreover, to relevance. the degree that the labor requirement in FFW Thus, we evaluate based on the intention to is more strenuous than it is in alternative ac- treat, in this case, child growth conditional on tivities, food aid increases energy outlay. This the allocation of food aid to the community, suggests that free food aid may have a larger ef- and not the household choice to take up this fect on child growth than an equivalent amount opportunity. If we were to model the impact of food for work. In addition, depending on on self-selecting participants or those chosen the degree of market integration, food aid may by a local administration, it would be neceshave an indirect positive effect on child growth sary to make a set of additional assumptions.3 While we do not address the marginal impact by lowering food prices. The effect of food aid on child growth is, of of additional food (or income) on households, course, contingent on the amount a household we note that the impact that we do measure receives, i.e., the targeting rules of the food aid allows one to assess the degree to which inprogram as well as individual choices regarding creases in the amount of food aid allocated to labor on public works. For example, if allocated the community prevent child malnutrition. in response to household income shocks (Fit = From equation (3), we derive an estimable F(Sit)), food aid may also mitigate the nega- growth equation tive effect of income shocks. While the theoret- hit =hhit ical literature on targeting has devised optimal (4) + PFFit + = hit+1 FSjt F allocation rules given information constraints + Ix Xit + eit + uit + vjt. (Besley and Kanbur;Besley), the actual allocation of food aid is often the combined result of However, food aid may be directed to those a host of factors including considerations of opareas where child malnutrition is high, potentimal targeting, but also spatial inertia in protially leading us to underestimate its effect gram operations due to fixed operational costs : 0. To overcome the food aid pro(Jayne et al.), and the political economy of re- (E(Fjtit)) we use the average source allocations at the national and regional gram placement problem, in food-aid-need assessments 1984-88, up to level. Yet, different political economy theoeleven years before the period being studied, ries predict quite different allocation rules. Acand its squared term along with rainfall recording to altruism theories of public transfers, lated variables capturing chronic needs, as inthe least endowed ought to receive the highest struments to predict the quantity of food aid transfer (Roberts), while pressure group thereceived. Given the inertia involved in the loories predict that groups small in number and cation of food aid programs as a result of high with considerable resources for lobbying take fixed start up costs, earlier needs assessments the highest share of public transfers (Becker). during the second half of the 1980s have been Thus, in practice, whether and how much food observed to be good predictors of future food aid a household is likely to receive is mostly a aid in Ethiopia (Jayne et al.). The selection of context-specific matter that needs to be deter- the instrumental variables is discussed in more mined empirically. detail below. We are not able to match our informaTo examine the potential differential eftion on food aid and plot-related shocks for fect of FD and FFW, the predicted amounts each household with their children's growth. received of each kind are also separately Instead, we use information on the average included in the child growth regression amount of food aid received (Fi,) and average at the community level. As a plot damage 3 While the element of self-selection in free distribution is less (Sit) result, the take up of food aid by a household than with food for work, the perspective of evaluation as the inis not modeled in this study. Yet, this choice tention to treat is also relevant to avoid assumptions regarding the does not follow only from practical consid- mechanisms of the intra-community allocation process.

276

May 2005

Amer. J. Agr. Econ.

1FFjt = IFDFFDjt + FFWFFFWjt. The presence of intra-household gender differentiation is explored through the inclusion of interaction terms of the sex of the child with the shock variables.

Child Growth, Food Aid, and Crop Failure in Rural Ethiopia

Over the period 1995-96, a series of three nationally representativehousehold surveys were conductedin Ethiopiaas part of the Rural Integrated Household Survey Program. We ther at the household level or the enumeration area (EA)/community level. Anthropometric

integrate informationfrom these surveys eion information pre-schoolchildren provided is by the 1995/96 Welfare Monitoring Survey
(WMS), which covered twelve randomly se-

this survey, 2,414 rural children under five years old were measured twice and matched successfully across two rounds with a six

lected householdsin each EA/community.In

months interval.4After excludingcases with more than twelve-month age difference reported, negative height growth, more than
25 cm growth in six months, and a HAZ-score mained, spread over 469 EAs out of a total of 531 EAs.

beyond the [-6, +6] range,2,089 childrenreThe Food SecuritySurvey (FSS) was conducted on a sub-sampleof the WMS (7 out

mation on the amounts of food aid received by each householdin the past twelve months
4 While 5,012 children aged six to fifty-four months old were measured in the first round and 5,121 children aged twelve to sixty months old were measured in the second round, only 2,414 were measured twice and matched successfully across two rounds. Yet, even though only 48% of original children were measured twice, 86% of the households in the original survey were in the second round. Thus, it is likely that much of the difference in children in survey rounds reflects coding errors-a common problem of panel surveys-and not selective attrition. Indeed, no large mortality rates among children or migration were reported during the few months between surveys, and further investigations indicated miscoding child IDs across rounds and a lower rate of follow up among children near the age cut-off as important reasons for the attrition. We also tested for attrition biases following Fitzgerald, Gottschalk, and Moffitt, and Alderman et al. Since all children have an initial height we were able to test for the influence of unobservable differences in the sample in the base period. This test confirmed that the group with subsequent attrition is on the same nutrition production function as the rest of the sample. Therefore, it seems that the reasons for attrition are not self-selection in nature and that attrition bias is not a serious concern in our data. 5 Those excluded are less likely to be in peri-urban areas, less likely to have an educated father, less likely to come from a household that owns land, and more likely to be older. Other variables, including the EA-level food aid and crop damage variables, are not significantly correlated to the probability of being excluded due to unreasonable measurements.

of 12 households) and collected recall infor-

(June 1995-May1996),whichcovers the 1995 Meher(mainrain)season.The 1995/96 Annual AgriculturalSample Survey(ASS) covered a (25 largerset of householdsin eachcommunity in total, includingthose coveredin the WMS) and collected informationon crop damageon each plot duringthe 1995 Meher season. As the plot size of each plot was physicallymeaof sured,we couldcalculatethe proportion plot area damagedfor each household.In the abcommonhouseholdidensence of appropriate tifiers,we merged the ASS and FSS with the we WMS at the EA level. In particular, proxied per capitahousehold food aid availability and the proportionof total plot areadamaged per householdby their respectivecommunity averages. Food aid in Ethiopia is allocated in two stages. In a first stage, food aid is assignedto basedon theirneeds assessments the Woredas, receivedfolandthe overallfood aidquantities lowing the annualappeal to the international communityfor food aid.6Needs assessments are estimates of the proportionof people in need of food aid,carriedout annuallyfor each Woredaby the Disaster Preventionand Preparedness Commission (formerly the Relief and RehabilitationCommission)in consultaOn tion withNGOs andthe donorcommunity. eachyeara food the basisof these assessments, aid appeal goes out to the internationalcommunity.The Woredassubsequentlydistribute the food aid in a second step to the households using a federalguidelinethat most food should be distributedthroughFFW arrangements and that only householdswith no able shouldreceivefree food. In bodied individuals in practice,however,as illustrated table 1, free is food distribution more commonthan FFW, with the level of effort for publicworks often being nominal. About one in five communities(116 out of 531 communities)received food aid between 1996.Of those commuJune 1995andJanuary nities receivingfood aid, 53% reportedusing this aid exclusivelyfor FD, 21%only for FFW, Comand 27%had both types of distribution. munitiesreceivingaid experiencedlowerrainfall on average as well as higher variationin
their rainfall patterns as captured by the coefficient of variation. These findings suggest that food aid allocations are somewhat targeted

6 A Woreda is the second lowest administrative unit in Ethiopia and corresponds to what is commonly known as a district in other countries. There are about 560 Woredas in the country.

and Yamano, Alderman, Christiaensen

and ChildGrowth FoodAid

277

Table 1. Food Aid Distribution and Plot Damage at EA Level EAs with Food Aid EAs withoutFood Aid Numberof EAs Numberof EAs withFD only Numberof EAs withFFW only Numberof EAs with FD and FFW Annualexpenditures per capita(Birr)(Net of food aid) Per capitafood aid received (Birr) (June1995-January 1996)
Inertia/chronicpoverty measures

Total 531

116 61 24 31 828.6(518.1) 22.5 (32.7)

415

1,111*(1,172) 0.0**(0.0)

1,049(1,070) 5.06 (18.3)

Needs assessments 1984-88 in Long-runaverage(LA) rainfall(mm) (1967-2001) Coefficientof variationof rainfall(1967-2001) Shocks Ratio of damagedplot area withinEA Breakdownby causesof damage Too little rain Too muchrain Cropdisease/insect problem

0.288(0.248) 891.8(332.4) 0.286 (0.092) 0.323 (0.262) 0.164 (0.262) 0.072 (0.12) 0.087(0.116)

0.089** (0.160) 1,135.9** (358.9) 0.227** (0.094) 0.176** (0.176) 0.061** (0.130) 0.036** (0.072) 0.079(0.116)

0.133(0.200) 1,082.6(367.1) 0.240(0.096) 0.208(0.206) 0.084 (0.173) 0.044 (0.087) 0.081(0.112)

Note: Needs assessments, long-run average rainfall and coefficient of variation of rainfall are measured at the Woreda level. Numbers in parentheses are standard deviations. In 1996, 1US$ equals about 6.5 Ethiopian Birr. * and ** indicate a statistically significant difference at the 5% and 1% level, respectively, on a particular attribute between communities with and without food aid.

to chronically poor communities although the allocations may also suffer from inertia. Finally, communities that received food aid were also observed to be poorer in 1995/96 as reflected by their lower average household expenditure per capita. Food aid programs also appear targeted to communities that experienced crop damage in the 1995/96 Meher season. The average percentage of damaged crop areas was about 32% in communities with food aid, while it was only 17.6% in communities without food aid. Most of the damage was caused by rainfall shocks (mostly droughts), though a non-negligible proportion (about 40%) of the damage was related to insect attacks and crop diseases. Comparing shock incidence in communities with and without food aid, it appears that food aid was especially responsive to droughts (and flooding) though not to (idiosyncratic) insect attacks or crop diseases. To explore the relationship between child growth and food aid, we plot child growth (in cm) over the six-month interval against child age at the first measurement (figure 1), both for all children (426) in the food aid receiving communi-

ties as well as for those (1,663 in total) in the non-food aid receiving communities. Figure 1 is created using locally weighted smoothed scatter plots, LOWESS (Cleveland). The observed pattern reflects normal growth curves, i.e., growth velocity declining by age. More strikingly,we also find that, in general, children in food aid receiving communities grow faster than children in communities without food aid, especially those younger than two years old. Consistent with figure 1, table 2 shows that six to twenty-four month old children grew about 0.41 cm faster in communities with food aid compared to those without, although the difference is not statistically significant (t-stat = 1.28). We do not find any difference in growth among children aged twenty-five to sixty months old. Following common child growth specifications (Deolalikar; Hoddinott, and Kinsey), other variables in our regressions include the individual child, household, and community characteristics.The descriptive statistics are reported in table 3. We control for individual child characteristics by including initial height, gender of the child, and child age.

278

May 2005

Amer. J. Agr. Econ.

7
With Food Aid

E
S.

O0

6r

I
C

o
Without Food Aid

412 24 36 Initial Age in Month 48 60

Figure 1. Child growth in height (cm) in a six-month period and food aid Table 2. Child Growth in Height by Food Aid in Rural Ethiopia ChildrenAged 6-24 months A
Number of children Food aid (F = 1)a Growth in height (cm)

ChildrenAged 25-60 months B 1,006 801

205

All No food aid (F = All

0)a

1,083 862
221

No food aid (F = 0)a Food aid (F = 1)a

6.68 (4.27) 7.19 (4.33) +0.41 [1.28]

6.78 (4.25)

5.48 (4.04) 5.44 (3.57) -0.05 [0.16]

5.49 (4.15)

Difference(Yes-No)[t-statistics]

Note: Numbers in parentheses are standard deviations, and numbers in brackets are absolute t-values. aNo Food Aid includes children who live in enumeration areas (EAs) where no food aid program was available between June 1995 and January 1996 (the period between the first and second measurement of child height in the Welfare Monitoring Surveys). Food Aid includes children who live in EAs where at least one food aid program (free distribution, food for work, or both) was available between June 1995 and January 1996.

Household characteristics included in the model are mother'sage, educationalinformation on household members, gender of the householdhead,the compositionof the household,householdassets,andthe sourceof drinking water.We proxy the educationalstatus of the household by using the highest grade attained by the most educatedmale and female adult in the household,as opposed to education of the parents,to capturepotentialintrahouseholdexternalities fromeducation.These areespeciallyimportant wheneducationlevels are low (Basu and Foster;Gibson). While the highest grade attainedby the most educated male adult in the household is twice as large as the highestgradeattainedby the most educated female adult,at an averageof 2.3 grades

for the most educated male adult per household, educationalattainmentsin Ethiopiaare As individualexpendituresreflectfood aid received,whichwe cannotnet out in the data we have, the estimatingequationsdo not include household expenditures.However, we control for wealth at the household level using assets. Nearly all householdsreport owning land, and about 60% possessed a plough. Yet, less than20% of the householdspossessa radio,a sign of widespreadpoverty;and only
7 Because of a very low education level in Ethiopia, the correlation between mother's and the most educated female member's education level is quite high, 0.924. Thus, we did not find any significant differences in estimation results using one or the other.

extremely low.7

and Yamano, Alderman, Christiaensen

and ChildGrowth FoodAid

279

Table 3. Means of Socioeconomic Determinants of Child Growth in Rural Ethiopia ChildrenAged 6-24 months ChildrenAged 25-60 months Mean S.D. Mean S.D. Growthin heightin six months(cm)
EA -level food aid Child growth

6.68 3.7 2.8 0.9 0.21 1,104 6.79 67.2 0.47 13.5 28.60 0.03 2.16 1.13 0.11 1.24 1.30 0.93 0.60 0.19 0.17 0.71 0.04 0.05 0.15
1,989

4.27 11.5 10.4 4.0 0.20 1,159 0.57 7.7 0.50 6.6 6.94 0.17 3.48 2.68 0.31 0.76 0.61 0.25 0.49 0.40 0.37 0.45 0.20 0.22 0.36
467

5.48 3.1 2.2 0.9 0.21 1,081 6.80 85.2 0.54 38.7 30.68 0.05 2.45 1.24 0.11 1.32 1.32 0.91 0.59 0.22 0.20 0.71 0.05 0.06 0.16
1,987

4.04 10.3 9.3 3.7 0.20 1,034 0.53 7.6 0.5 8.9 7.28 0.22 3.75 2.95 0.32 0.82 0.62 0.28 0.49 0.41 0.40 0.45 0.22 0.24 0.37
449

Per capitafood aid received(Birr) PC food aid fromfree distribution (Birr) PC food aid fromfood for work (Birr)
EA-level variables

Damagedplot areas(ratio) Per capitaexpenditure(net of food aid) In(percapitaexpenditure)

Child characteristics

Initialheight(cm) Gender (boy = 1) Age (months)

Household characteristics

Mothers'age (years) No motherinfo (No info = 1) Maximum male education(years) Maximum female education(years) Femaleheadedhouseholds(= 1) Numberof men Numberof women Ownership(= 1): land Ownership(= 1): plough Ownership(= 1): animals Ownership(= 1): radio (= Ownership 1): sickle Ownership(= 1): stove Watersource:protectedwell (= 1) Watersource:tap (= 1) Pop.density(per arableland km2)/1,000 Peri-urban 1) (= Tarmac roadavailablein zone (= 1) Numberof children
Community characteristics Elevation (m)

0.327 0.17 0.52 1,083

0.399 0.38 0.50

0.336 0.18 0.54 1,006

0.427 0.39 0.50

one in five households reports ownership of animals. About 15% of the households have access to a tap for drinking water. Dummy variables for peri-urban areas, availability of a tarmac road in the zone as well as elevation, a proxy for malaria infestation, capture some important location characteristics for child growth.8,9 About 17% of the children in our sample live in peri-urban areas and slightly more than half of the children live

8 There are about fifty-five Zones in Ethiopia which is the administrative unit between the Woreda (district) and the Killil (the largest administrative unit). 9 Malaria is usually absent in most of the Ethiopian highlands, though its overall incidence in the country has been increasing steadily over the years in Ethiopia. In 1995, the incidence was 1.1 million cases, while it increased to 1.5 million in 2000 (World Health Organization).

in zones with a tarmac road. We also include population density, measured as the number of people per kilometer squared of arable land. This is a possible proxy for both unobserved infrastructure and land quality. Finally, we include nine Killil dummies to control for other spatially correlated characteristics such as food prices, the presence of development programs, and quality of service delivery. From the descriptive discussion (figure 1 and table 2) as well as from other studies in the literature, it appears that our key variables of interest, shocks and food aid, may have differential effects according to child age. Consequently, we estimate separate child growth regressions both for children younger and older than two years old. We begin however by examining the determinants of food aid reception.

280

May 2005

Amer. J. Agr. Econ.

Food Aid Allocationin RuralEthiopia Ourmaininteresthereis in determining rethe of food aid allocationsto income sponsiveness shocks,whichcombinedwiththe effect of food aid on child growth (discussedin the section below), permits us to analyze how effective food aidis in mitigating effect of shockson the child growth.More detailed empiricaldiscussions of the food aid allocationrules in rural Ethiopia have been provided by Jayne et al. and Dercon and Krishnan.10 dependent The variable is the community'stotal per capita value of food aid received (whetherused for FD or FFW) between the first and the second surveyround,whichcomprisesup to eight months.Becauseonly one-fifthof the sampled communitiesreceived food aid, we use Tobit models. The results in column A, table 4, indicate that the EA-level per capitaexpenditure(excludingaveragefood aid) is negativelycorrelated with food aid. However, the resultsalso indicatethat,even aftercontrolling current the expenditure level, proxies of inertia/chronic poverty (Z) are important determinantsof food aid distribution.In particular,the food aid needs assessmentsduring1984-88, which encompassedthe major1984/85famine,have a significantpositive effect on the amount of food aidreceivedby communities. mayinThis dicate that communitiesthat were considered vulnerablein 1984-88 are still vulnerableand in need of food aid in 1996.Alternatively, this may also reflect inertia in food aid programs due to high fixed costs related to programestablishmentleading to a high degree of spatial continuityin food aid allocationswith the currentspatialpatternof food allocationsstill allocationset up in reflectingthe geographical responseto the 1984/85famine.The empirical evidence presentedby Jayneet al. favors the latter interpretation. returnto this in the We analysisbelow. An importantproxy for chronicpoverty is the coefficient of variation for rainfall, especially in rural Ethiopia which largely deNot only may pends on rain-fed agriculture.
high rainfall variability force farmers to adopt low-risk-low-return production technologies, trapping them into chronic poverty, but rainfall variability is also negatively correlated to

long run average rainfall."' In other words, the depressing effects of low average rainfall on living standards are exacerbated by increased uncertainty. The larger the coefficient of variation, the larger the amount of food aid received in the eight-month period. Since the coefficient of variation is distinct from the current shock, the significance of this variable should be interpreted in terms of long run conditions in the communities. We also find that food aid programs are responsive to crop damage, represented by the ratio of damaged plots in the community. However, the amount of food aid delivered in response to shocks seems small compared with the amount of food aid determined by the inertia/chronic poverty measurements. When decomposing the average predicted value of total food aid per capita (5.95 Ethiopian Birr over eight months) into food aid allocated in response to inertia and chronic poverty and food aid allocated in response to shocks, we find that the lion's share of all food aid (84% = (4.98/5.95) x 100) has been allocated in response to inertia and chronic poverty (as well as the other community characteristics). We also find that only a small part (16% = (0.97/5.95) x 100) has been allocated in response to shocks. While these results may partly follow from the fact that only 20% of the crop area was damaged, they are in keeping with Dercon and Krishnan who also report a limited response of food aid to shocks in their purposively selected sample of fifteen villages in Ethiopia surveyed three times between 1994 and 1995.12,13 Nonetheless, to judge how effective food aid is in mitigating the effect of shocks on child growth, we also need to know how food aid reception and plot damage affect child growth. We revisit this issue later, when we discuss the empirical results on the effect of the different child growth determinants on child growth. When looking at the determinants of free food aid and food for work allocations separately (table 4, columns B and C,
" Pearson correlation coefficient = -0.4. This is statistically significant at the 1% level. 12 The amount of food aid determined by inertia and chronic poverty has been estimated by setting the shock variable (S) and the interaction term with the shock variables (S x P) equal to zero. The amount of food aid responding to the shocks has been predicted based on the shock variables and its interaction terms. The latter were included to examine if the responsiveness of the food aid distribution system to shocks depends on the inertia of the system, which was not supported by the data. 13Doubling the plot damage ratio to 40% increases the percentage food aid allocated in response to shocks to 27.4%.

10To examine the food aid allocation rules we augmented the data set used by Jayne et al. with more disaggregated rainfall data covering a much longer time period than used in the earlier study.

and Yamano, Alderman, Christiaensen

and ChildGrowth FoodAid

281

Table 4. EA-Level Food Aid Received (Per Capita) in Birr (EA-Level Analysis: Tobit) Food Aid A
Inertia/chronic poverty measures (Z)

FD B 332.4(3.30)** -211.6 (2.37)* -0.139 (1.87) 0.003(0.18) 59.26(1.21) 51.13(2.46)* -86.11 (1.53) -19.39 (2.83)** 9.416(1.29) -0.033 (2.04)* -85.69 (3.00)** -11.15 (1.45) 50.89(0.95) 5.80 [0.00]** 3.10[0.05]* 4.89 4.19 0.70 0.129 92 531

FFW C -29.50 (0.68) -34.97 (0.85) 0.102(3.21)** -0.010 (1.45) 24.80(1.11) 6.450(0.73) 9.088(0.34) -0.809 (0.29) 8.482(2.57)* -0.005 (1.22) -26.35 (2.35)* -6.928 (2.04)* -30.75 (1.29) 4.31 [0.00]** 0.99 [0.37] 1.00 0.74 0.26 0.127 55 531

Needs assessmentin 1984-88 (P) Needs assessmentsquared(p2) Assessment(P) x long-runrainfall Long-runaveragerainfall1967-2001 C.V.of rainfall1967-2001
Shocks (S)

235.6(2.78)** -191.6 (2.48)* -0.019 (0.32) -0.002 (0.13) 74.83(1.81) 40.94(2.36)* -63.40(1.30) -18.75 (3.22)** 12.24(1.95) -0.021 (1.98)* -72.52 (3.52)** -13.18 (2.03)* 47.34(1.03) 7.49 [0.00]** 2.92[0.06] 5.95 5.04 0.91 0.123 116 531

Damagedplot areas(ratio) Damagedplot areas x assessment(P)

EA-level variables (X)

In(EA-levelper capitaexpenditure) Elevation Pop.density(per arableland km2) Peri-urban Good roadavailable(=1) Constant
Joint significance tests and predictions

On inertia/chronic povertymeasures(Z) On shocks(S) and S x P Predicted: total food aid Predicted: permanenttransfer Predicted: responseto shocks PseudoR-squared Numberof EAs with food aid Numberof EAs

Note:Nine Killildummies alsoincluded not reported. on standard are but Numbers parentheses absolute in t-values calculated heteroskedasticity-robust are errors withcluster(EA) effects. *indicates significance 5% 1% level;and**indicates significance.

respectively), we find that while free food aid has been allocated both in response to the current expenditure level, chronic needs, and shocks, food for work allocations seem largely unaffected by shocks.14 This would suggest that, in practice, food for work programs have been largely set up to address chronic food insecurity, while free food aid may serve a limited insurance function. Yet, further investigation is needed, as experience in the sample studied by Quisumbing and Dercon and Krishnan seems to suggest the opposite. Estimated Effects of Shocks and Food Aid on Pre-school Child Growth The results on child growth in table 5 show that children aged six to twenty-four months old are quite vulnerable to shocks, consistent
14The sum of the number of EAs that received food aid in both regressions exceeds the total number of EAs with food aid in our sample because some EAs used food aid both for free food distribution and food for work. In these EAs, we calculated the average per capita value of each type of food aid.

with findings by Hoddinott and Kinsey. A 10% point increase in the proportion of damaged plot areas within a community corresponds to a reduction in child growth by 0.12 cm over a six-month period (column A). Due to the fact that the average growth rate among this age group is 6.68 cm, a 0.12 cm decline represents a 1.8% reduction in growth. When we add food aid variables to the growth models in columns B and C, the coefficient on plot damage increases in absolute value (from -1.174 to -1.886) as does the precision of the estimate." This suggests that food aid mitigates the negative effect of plot damage on child growth. When the food aid variable is excluded (column A), the estimated coefficient on plot damage not only picks up the (negative) effect of plot damage but also the (positive) effect of food aid on child growth because
15 While we also control for food aid in model D, the latter model is applied to a restricted sample, i.e., excluding those communities that distribute food aid both through FD and FFW. As a result, the coefficient on damaged plot areas is not strictly comparable with those in models B and C, even though the size, sign, and statistical significance are very similar.

282

May2005

Amer. Agr.Econ. J.

Table5. ChildGrowthin Height (cm):ChildrenAged 6-24 Months-IV Models

InitialHeight Is Endogenous A B Per capitafood aid received (Birr)a PC food aid fromFDa PC food aid fromFFWa
EA-level variables EA-level food aid

Plus,Food Aid Is Endogenous D C 0.090 (2.58)** 0.068 (0.85) 0.332 (0.68)

0.042 (2.62)**

Damagedplot areas(ratio)
Child characteristics

-1.174 (1.92)

-1.505 (2.42)* 0.099 (1.33) -0.483 (1.56) -0.250 (1.66) 0.003 (0.80) 0.001 (0.04) 0.555 (0.55) -0.038 (0.77) 0.074 (0.89) 0.024 (0.05) 0.001(0.00) -0.219 (0.97) 0.798 (1.32) -0.272 (0.76) -0.144 (0.40) -0.042 (0.10) 0.135 (0.41) -0.257 (0.35) -0.783 (1.39) 0.030(0.06) -0.264 (0.69) 0.683 (1.84) 0.192 (0.32) 0.259 (0.81) 2.661 (0.66) 0.53 [0.78] 58.5
3.36 [q = 2]

-1.886 (2.80)** -2.187 (2.02)* 0.104(1.40) -0.487 (1.56) -0.255 (1.68) 0.003(0.78) 0.005(0.22) 0.620 (0.61) -0.046 (0.92) 0.073 (0.87) 0.097(0.19) 0.018(0.09) -0.243 (1.07) 0.745(1.23) -0.273 (0.77) -0.154 (0.42) 0.018(0.04) 0.140(0.42) -0.296 (0.40) -0.773 (1.34) -0.003 (0.01) -0.334 (0.86) 0.701(1.88) 0.358 (0.58) 0.303(0.95) 1.579(0.38) 0.50 [0.81] 24.6 11.9
8.23 [q = 5]

Initialheight (instrumented)b 0.091 (1.23) Gender(boy = 1) -0.473 (1.54) -0.242 (1.61) Age (month) 0.003 (0.80) Age squared
Household characteristics

0.114 (1.49) -0.531 (1.57) -0.236 (1.52) 0.002(0.59) 0.018 (0.78) 0.197 (0.17) -0.047 (0.76) 0.077(0.85) 0.035 (0.06) -0.067 (0.32) -0.317 (1.21) 0.861 (1.39) -0.142 (0.38) -0.124 (0.32) -0.039 (0.09) 0.164 (0.42) -0.023 (0.03) -0.841 (1.27) -0.095 (0.19) -0.238 (0.54) 0.688(1.62) 0.557(0.90) 0.255 (0.77) 0.795(0.19) 2.86 [0.06] 0.46 [0.89] 23.1 8.1/2.0
6.03 [q = 5]

Max.male education(years) Max.female education(years) Mother'sage (years) No motherinfo (no info = 1) Femaleheadedhouseholds Numberof men Numberof women land Ownership: Ownership: plough animals Ownership: radio Ownership: sickle Ownership: stove Ownership: Watersource:protectedwell Watersource:tap Elevation Pop. density(per arableland km2)/1000 Peri-urban Good road available(= 1) Constant
Joint significance tests

-0.003 (0.13) 0.500 (0.50) -0.031 (0.63) 0.075 (0.91) -0.039 (0.08) -0.014 (0.07) -0.199 (0.87) 0.844 (1.39) -0.270 (0.75) -0.134 (0.38) -0.095 (0.22) 0.130 (0.39) -0.223 (0.31) -0.794 (1.43) 0.056 (0.11) -0.204 (0.54) 0.665 (1.80) 0.050 (0.08) 0.222 (0.69) 3.745 (0.94) 0.56 [0.76] 60.3
3.36 [q = 2]

On FD and FFW On assets

Joint significance tests on instruments

F-stat of IVs on initialheight F-stat of IVs on food aid (FD/FFW)

Over-identification tests: Chi-squared

R-squared Numberof children

0.03 1,083

0.03 1,083

0.01 1,083

0.01 1,005

Note: Killil dummies (n = 9) are also included but not reported. Numbers in parentheses are absolute t-values calculated on heteroskedasticity-robust standard errors. * indicates 5% ** significance level; and indicates 1% significance. aEndogenous variables in columns C and D. bEndogenous variables in all models.

areas with plot damage are more likely to receive food aid, as indicated in table 4. Thus, by controlling for food aid programs in columns B and C, we are able to get a more accurate estimate of plot damage per se.

Food aid has a positive effect on the growth of children between six and twentyfour months old. Moreover, the positive coefficient on food aid on child growth increases from 0.042 to 0.090 when we

Yamano, Alderman, and Christiaensen

Child Growth and Food Aid

283

control for program placement effects through the use of instrumental variables (column C). As discussed previously, the instrumental variables are Woreda-level variables on the inertia/chronic poverty measurements (Z). These variables are included in the instrumental variables model as instruments, together with the other variables in the child growth model. The F-tests on the instruments in the first stage regression presented at the bottom of table 5 clearly show high predictive power. The instruments also pass an over-identification test (Wooldridge), presented at the bottom of table 5 for each IV model providing additional confidence in the validity of our identifying variables. Two key messages emerge from these results. First, the substantial change in the size of the coefficient on food aid-the coefficient more than doubles, when instrumenting the food aid allocations-underscores the importance of controlling for program placement in examining the effect of food aid on individual welfare. This is consistent with our expectations since food aid programs appeared to be located in communities with poor child nutrition and growth (table 1). It is also in line with Pitt, Rosenzweig, and Gibbons, who reported increases of up to 100% in the estimated effect of public programs on human development outcomes when accounting for program placement. Second, the effect of food aid on child growth among the six to twenty-four month old children in our sample is considerable. Children in communities who received food aid grew on average 2.0 cm (0.090 times 22.5 Ethiopian Birr) faster in a six-month period than if no food aid would have been available.16 This would help compensate poor child nutrition and growth in communities that are targeted for food aid. We further investigate if the effects of food aid programs differ by the modality of food aid utilization (table 5, column D). To do so, we predict both types of food aid separately with the same set of instrumental variables, proxies for inertia and chronic poverty measurements (Z), using the instrumental variable procedure. We also restrict the sample to communities that do not have dual use. Given that our food aid variables are matched with individual child growth at the community level, inclusion of communities that use food aid for both purposes may confound the estimation of
16 Recall from table 1 that the average value of food aid received among the food aid receiving communities was 22.5 Ethiopian Birr.

the differential effect of FD and FFW. Application of model C to the restricted sample yields very similar results as those obtained from the full sample, apart from the coefficient on total food aid, permitting us to use the restricted sample for examining the differential effect of both uses of food aid. The results indicate that both uses have positive coefficients. While they are both imprecisely estimated, the F-test indicates that they are jointly significant. It seems that we are unable to separate the impact of one program from the other. The F-test on instrumental variables on FD and FFW indicate that instrumental variables are weakly correlated with FFW (F-stat = 2.0). In order to separate one program's impact from the other, we need to have instruments that are closely correlated with one program but not with the other. As Gebremedhin and Swinton observe, FFW is partially determined by individual choice as well as community targeting. This, in turn, is partially influenced by the efficiency of factor markets (Barrett and Clay). Furthermore, FFW is influenced in some areas by the availability of appropriate projects. Such information is currently not available for this study to use as instruments. The estimated coefficient of the initial height is about 0.10, with a t-statistic around 1.4, suggesting no catch-up growth in this short period. We have treated initial height as an endogenous variable, using weight and its squared term as instrumental variables. Although this does not solve a potential omitted variables problem between the initial height and unobserved child or household characteristics, it helps reduce the measurement error problem in the initial height.17 To further examine the robustness of the estimated coefficients of food aid and crop damage, we also excluded initial height (results not reported here), as it is very difficult to find plausible instruments that are (a) of sufficient magnitude and persistence to affect a child's height in the initial period, (b) sufficiently variable across households, and (c) sufficiently transitory not to affect the child's stature in the subsequent period. The estimated coefficients of interest are similar to the ones reported in table 5.

17 When we do not instrument initial height, the estimated coefficient of initial height is about -0.25, with a very high t-statistic. While nominally this implies that shorter children grow faster, this result should not be interpreted as proof of catch-up growth since measurement error in lagged height would be negatively correlated with the dependent variable.

284

May 2005

Amer. J. Agr. Econ.

We also ascertainwhether the results are robust to the additionof the currentrainfall shock to the instrumentalvariables.This potential instrumentis not included in the results reportedin tables 4 and 5 because there is a possibility that rainfall affects nutrition However,if the throughits effectson mobility. currentrainfallshock is included, the coefficient of food aid corresponding the one reto ported in table 5, column C, is 1.823 and the joint significanceof the instrumentsis 10.7. These are very close to the ones reportedin table 5. While it might be somewhat of a surprise that the asset variables are jointly insignificant, this is less of a contrastto the existing literatureon wealth and nutrition,including studies from Ethiopia, than it first appears. Most studies regress achieved status (height or weight), which is a stock variable,against the stock of assets or expenditureswhichmay be interpretedas a measureof permanentincome and, hence, also a stock indicator. In contrast,this model has a flow variableas the dependent variable.This, as well as the fact that many assets includingland had comparatively little variance, reduces the precision of the estimates.Nevertheless,their inclusion providessome protectionagainstmissingvariable bias. In keeping with this logic, it is also less of a surprisenot to find statisticallysignificantresultsfor the coefficientson most of the otherstockvariables, whilethe coefficients on the flow variables(food aid and shocks), which are also our variablesof primaryinterest, are estimatedwithmuchgreaterprecision. We do note a positiveeffect of populationdensity on child growth that may be related to the availability more fertile lands and pubof lic infrastructure more densely populated in areas. In additionto the resultsreportedin table 5, we havealso triedan interaction termbetween the ratio of damagedplot area and the gender of the child (boy = 1). The inclusion of the interactiontermindicatesthat on average the growthof girlsundertwo suffersless from income shocks than the growth of boys under two, either because of greater biological resilience or due to intra-household dynamics. The coefficients on the interaction term and the damage variable are -2.06 and -0.81, respectively, and they are jointly significant (F-test = 4.38). This result-which is also borne out by other evidence from Ethiopia (Christiaensen and Alderman)-is in keeping with Svedberg who finds that boys are often more malnourished in Sub-Saharan Africa.

Our findingsregardinggrowth of children and agedtwenty-five sixtymonthsold aresimilarto the resultson childrenagedsixto twentyfour months, but less precisely estimated (table 6). The coefficienton the plot damage variable has a negative sign and its size infor creases as we controland instrument food aid-as in the case of younger childrenthoughin none of the models(A-D) is it statisThe ticallysignificant. point estimateof the estimatedcoefficientis alsomuchsmaller.Thisis consistentwith other studies(Martorell,1997; Jensen;Hoddinottand Kinsey) that find children between twelve and twenty-four months to be especially vulnerablein the face of income shocks. Similarly,the estimated coefficient of the food aidvariablehas a positivesign when food aid is instrumented, though,again, The it is not statistically significant. point estimate of the coefficient(0.040) is also smaller than the one found among younger children (0.090). We can use the resultsin table 5, columnC, to examinehow effectivefood aidis in protecting child growthfrom plot damageshocks.To obtain a marginalresponseof food aid to plot response damage,we decomposethe marginal into two components:the marginaleffect on the probability of receiving food aid times the average amount of food aid usually received and the marginaleffect on the average amountof food aid receivedtimesthe average of probability receivingfood aid(Wooldridge). we Accordingly, re-estimatethe food aid allocation model in table 4 in a two-stepmethod, and find that a Probit and TruncatedOLS,18 10%point increasein plot damagearea is expected to increasethe amountof food aid by 0.96 Birr.19 Using the results in column C in
18 Because Tobit is a nonlinear model, we cannot directly interpret Tobit coefficients as marginal effects. Although it is possible to calculate the marginal effects from Tobit coefficients, we prefer to use a two-step method discussed in text because Tobit constrains coefficients to be the same in affecting the probability of receiving food aid and the amount of food aid received among recipients. The simulation results, however, are similar in both Tobit and two-step methods. 19Denote by y the dependent variable (food aid in our example) and xi independent variable i (e.g., plot damage). The Tobit coefficient on xi can then be decomposed into:

aE[y Ixi] axi

- Pr ob[yi > 0]

aE[ylxi, y > 0] axi a Pr ob[y > 0] + E[y Ixi, y > 0]

axi

By using the results from Probit and Truncated OLS and average values, the total marginal response of a 10% increase in plot damage on the amount of food aid can then be calculated as the average probability of receiving some food aid times the marginal impact of plot damage on the amount of food aid from the truncated OLS (23.44, t = 1.16) plus the average amount of food aid conditional aid being non-zero times the marginal effect of plot damage on the

and Yamano, Alderman, Christiaensen

and ChildGrowth FoodAid

285

Table 6. Child Growth in Height (cm): Children Aged 25-60 Months-IV Models InitialHeightIs Endogenous Plus,FoodAid Is Endogenous A B C D Per capitafood aid received(Birr)a PC food aid fromFDa PC food aid fromFFWa
EA-level variables EA-level food aid

-0.001 (0.12)

0.040(0.88)

0.059(0.76) 0.069(0.22) -1.414 (0.86) 0.003(0.09) 0.069(0.25) -0.119 (0.75) 0.001(0.39) 0.032(1.40) 0.160(0.22) 0.040(0.73) -0.019 (0.29) -0.256 (0.47) -0.160 (0.79) 0.365(1.48) -0.484 (0.98) 0.338(0.96) 0.280(0.75) -0.204 (0.48) -0.207 (0.62) 0.220(0.40) -0.800 (1.50) 0.232(0.47) 0.001(0.00) 0.479(1.38) -0.735 (1.18) 0.069(0.20) 7.042(1.83) 0.58 [0.56] 0.52 [0.79] 68.9 7.1/1.3
1.69 [q = 5]

Damagedplot areas(ratio)
Child characteristics

-0.082 (0.11) -0.071 (0.09) -0.411 (0.49) 0.005(0.14) 0.004(0.13) 0.011(0.33) 0.179(0.71) 0.179(0.70) 0.187(0.73) -0.048 (0.36) -0.047 (0.35) -0.073 (0.52) -0.000 (0.09) -0.000 (0.10) 0.000(0.05) 0.025(1.21) 0.246(0.40) 0.017(0.33) -0.004 (0.06) -0.132 (0.28) -0.130 (0.69) 0.308(1.37) -0.498 (1.07) 0.158(0.51) 0.339(1.01) -0.160 (0.42) -0.159 (0.52) 0.239(0.45) -0.904 (1.80) 0.230(0.47) 0.061(0.17) 0.534(1.59) -0.893 (1.52) 0.173(0.55) 5.439(1.64) 0.55 [0.77] 190.8
0.70 [q = 2]

Initialheight(instrumented)b Gender(boy = 1) Age (months) Age squared

Household characteristics

Max.male education(years) Max.femaleeducation(years) Mother'sage (years) No motherinfo (no info = 1) Femaleheadedhouseholds Numberof men Numberof women land Ownership: Ownership: plough animals Ownership: radio Ownership: sickle Ownership: stove Ownership: Watersource:protectedwell Watersource:tap Elevation Pop.density(per arablelandkm2)/1,000 Peri-urban Good roadavailable(= 1) Constant
Joint significance tests

0.025(1.20) 0.247(0.40) 0.017(0.33) -0.004 (0.06) -0.133 (0.28) -0.130 (0.69) 0.308(1.37) -0.499 (1.07) 0.158(0.51) 0.338(1.00) -0.160 (0.42) -0.159 (0.52) 0.239(0.45) -0.906 (1.80) 0.231(0.47) 0.062(0.18) 0.533(1.59) -0.897 (1.53) 0.173(0.55) 5.452(1.65) 0.55 [0.77] 189.6
0.70 [q = 2]

0.026(1.28) 0.233(0.38) 0.012(0.23) -0.000 (0.00) -0.100 (0.21) -0.125 (0.65) 0.309(1.37) -0.489 (1.05) 0.154(0.50) 0.373(1.09) -0.162 (0.42) -0.140 (0.45) 0.221(0.42) -0.859 (1.70) 0.206(0.42) 0.024(0.07) 0.563(1.65) -0.779 (1.29) 0.164(0.51) 5.083(1.53) 0.55 [0.77] 74.0 8.7
1.81 [q = 5]

On FD andFFW On assets

Joint significance tests on instruments

F-statof IVs on initialheight F-statof IVs on food aid (FD/FFW)

Over-identificationtests: Chi-squared

R-squared Numberof children

0.05 1,006

0.05 1,006

0.04 1,006

0.01 943

Note: Killil dummies (n = 9) are also included but not reported. None of the asset ownership variables and water source variables has a significant coefficient. Numbers in parentheses are absolute t-values calculated on heteroskedasticity-robust standard errors. * indicates 5% significance level; and ** indicates 1% significance. aEndogenous variables in columns C and D. bEndogenous variables in all models.

table 5, this amount of additional food aid less growth. Food aid compensates on averwill increase child growth by 0.086 cm (0.96 age about 46% of the negative impact on child Birrtimes0.090).Yet, a 10%point increasein growth following from crop damage. This average marginal response represents plot damageis also associatedwith -0.189 cm
probability of receiving food aid from the probit (0.201, t = 2.06 in our re-estimation), or aE[y Ixi] = 0.218 x 23.44 x 0.1 22.5 + axi x 0.201 x 0.1 = 0.96 Birr.

three very different sets of communities. First, consider communities that had not previously received aid (about 78% of our sampled communities). Due to an increase of need resulting from the 10% increase in plot damage the probability of receiving food aid increases by

286

May 2005

Amer. J. Agr. Econ.

2%. Yet, while the marginalincreasein probability is incremental,the actual response is lumpy;a communityis either completely included or it is not. For those that are allowed to enter into the food distribution response in to this particular shock,the averageallocation of food aid (22.5Birr)increaseis farmorethan necessaryto offset the impactof plot damage. Conversely,any communitywith a shock yet not addedto the food aid lists bear the consequences of the shocks.The thirdgroup,communitiesthat alreadyreceive food aid (about 22%of the sampledcommunities)also appear to be protected,thoughthe marginalincrease is impreciselyestimated.That is, a 10% plot damage increasesthe amount of food aid by 2.3 Birr, which is more than enough to compensate the negativeimpacton growth.While it is encouragingto observe that food aid can alleviate the negative effect of frequentlyreshocksin Ethiopiaon childgrowth,the curring resultsare also consistentwith the continuing persistenceof high child malnutritionas well as withresultsthatshowinertiain the targeting of food aid.

erly estimate the effect of food aid on child growth. In addition,basedon the empirical targeting rules derivedfrom the data, the total amount of food aid appearson an averagesufficientto offset the negative effects of plot damage on child growth in food aid receivingcommunities. This result is encouragingas it indicates that food aid has indeed been effectivein protecting early child growthfrom droughtsand otherincomeshocksin food aidreceivingcommunities.Yet, it appearsthat food aid reception has been largely determined by factors other than shocks,and as a result,manycommunitieswhoexperienceshockstendnot to get food aid. Indeed, child stuntinghas persisted at alarminglevels despite massiveamountsof food aid. This points not only to the inflexible but food aid targeting, also to the endemicnature of povertyandthe extremelylow levels of maternaleducationin Ethiopia.Food aid targeting rules more responsiveto shocksas well are as other insurancemechanisms called for. [ReceivedJune2003; June2004.] accepted References Alderman, H., J. Behrman, H. Kohler, J. Maluccio,and S. Watkins."Attritionin Household SurveyData:Some Longitudinal Tests Three for Country Samples." Developing
Alderman, H., J. Hoddinott, and B. Kinsey. of "LongTermConsequences EarlyChildhood DC: Malnutrition." Washington Unpublished, FoodPolicyReWorldBankandInternational 2003. searchInstitute, and C. Barrett, "FoodSecurity FoodAssistancePrograms."In B. Gardner,and G. Rausser,eds.
Handbook of Agricultural Economics, vol. 2B. Demographic Research 5(2001):78-124.

and Summary ConcludingRemarks Using three nationallyrepresentative surveys conductedduring1995-96,we findthatincome shocks, measured by crop damage, reduce child growth substantially,especially among childrenaged six to twenty-four months.Children in this age groupmay lose about 0.9 cm growth over a six-month interval when half of their crop area is damaged.As early child growth falteringmay cause permanentdaminsurance mechanisms help to age,appropriate householdsprotecttheirconsumption fromincome shocks are crucial.This holds especially in Ethiopia,wherestuntingamongpre-school childrenhas persistedat alarminglevels over the past decadesand where droughtsare a recurrentphenomenon. Food aid has often been procured in response to shocks and has been motivatedby
its beneficial effect on child malnutrition. This depends, of course, critically on the allocation rules and the marginal effects of food aid on child growth. Our empirical results indicate that the average value of food aid received in a community has indeed a large positive effect on early child growth. The results further underscore the critical importance of controlling for program placement effects to prop-

NorthHollandPress,2003. Amsterdam: Barrett,C., and D. Clay. "How AccurateIs Foodfor-WorkSelf-Targetingin the Presence of Imperfect Factor Markets? Evidence from
Ethiopia." Journal of Development Studies

39(2003):152-80. Basu, K., and J. Foster. "On MeasuringLiteracy."

Economic Journal 108(1998):1733-49.

Becker,G. "ATheoryof CompetitionamongPressure Groupsfor PoliticalInfluence." Quarterly Besley, T. "Political Economy of Alleviating In TheoryandInstitutions." M. Bruno, Poverty:
and B. Pleskovic, eds. Annual World Bank Journal of Economics 98(1983):371-400.

and Alderman, Christiaensen Yamano, Conference on Development Economics 1996. Washington DC: World Bank, 1997, pp. 117-34. Besley, T., and R. Kanbur. "Food Subsidies and Poverty Alleviation." Economic Journal 98(1988):701-19. Brown, L., Y. Yohannes, and P.Webb. "Rural LaborIntensive Public Works: Impacts of Participation on Pre-schooler Nutrition: Evidence from Niger." American Journal of Agricultural Economics 76(1994):1213-18. Christiaensen, L., and H. Alderman. "Child Malnutrition in Ethiopia: Can Maternal Knowledge Augment the Role of Income?" Economic Development and Cultural Change 52(2004):287312. Clay, D., C. Molla, and D. Habtewold. "Food Aid Targeting in Ethiopia: A Study of Who Needs It and Who Gets It." Food Policy 24(1999):391409. Cleveland, W. "Robust Locally-Weighted Regression and Smoothing Scatterplots." Journal of the American Statistics Association 74(1979):829-36. Deolalikar, A. "Child Nutritional Status and Child Growth in Kenya: Socioeconomic Determinants." Journal of International Development 55(1996):273-308. Dercon, S., ed. Insurance Against Poverty. Helsinki: Oxford University Press and World Institute of Development Economics Research, 2004. Dercon, S., and J. Hoddinott. "Health, Shocks and Poverty Persistence." In S. Dercon, ed. Insurance Against Poverty. Helsinki: Oxford University Press and World Institute of Development Economics Research, 2004. Dercon, S., and P. Krishnan. "Food Aid and Informal Insurance." Center for the Study of African Economies WorkingPaper Series 200301, 2003. Engle, P., P. Menon, and L. Haddad. "Care and Nutrition: Concepts and Measurement." World Development 27(1999):1309-37. Fitzgerald, J., P. Gottschalk, and R. Moffitt. "An Analysis of Sample Attrition in Panel Data." The Journal of Human Resources 33(1998):251-99. Foster, A. "Prices, Credit Markets and Child Growth in Low-Income Rural Areas." Economic Journal 105(1995):551-70. Gebremedhin, B., and S. Swinton. "Reconciling Food-for-Work Project Feasibility with Food Aid Targeting in Tigray,Ethiopia." Food Policy 26(2001):85-95. Gibson, J. "Literacy and Intra Household Externalities." World Development 29(2001):155-66. Grantham-McGregor, S., C. Walker, S. Chang, and C. Powell. "Effects of Early Childhood Sup-

and ChildGrowth FoodAid

287

plementation with and without Stimulation on Later Development in Stunted Jamaican Children." American Journal of Clinical Nutrition 66(2001):247-53. Heckman J., R. Lalonde, and J. Smith. "The Economics and Econometrics of Active Labor Market Programs." In O. Ashenfelter and D. Card, eds. Handbook of Labor Economics, vol. 3A. Amsterdam: Elsevier Science, 1999, pp. 1865-2097. Hoddinott, J., and B. Kinsey. "Child Growth in the Time of Drought." Oxford Bulletin of Economics and Statistics 63(2001):409-36. Jayne, T. S., J. Strauss, T. Yamano, and D. Molla. "Targeting of Food Aid in Rural Ethiopia: Chronic Need or Inertia?" Journal of Development Economics 68(2002):247-88. Jensen, R. "Agricultural Volatility and Investments in Children." American Economic Review 90(2000):399-404. Martorell, R. "Undernutrition during Pregnancy and Early Childhood and Its Consequences for Cognitive and Behavioral Development." In M.E. Young, ed. Early Child Development: Investing in the Future. Amsterdam: Elsevier, 1997. . "The Nature of Child Malnutrition and Its Long-Term Implications." Food and Nutrition Bulletin 20(1999):288-92. Pitt, M., M. Rosenzweig, and D. Gibbons. "The Determinants and Consequences of the Placement of Government Programs in Indonesia." In D. van de Walle, and K. Nead, eds. Public Spending and the Poor: Theory and Evidence. Baltimore: World Bank and John Hopkins University Press, 1995. Quisumbing, A. R. "Food Aid and Child Nutrition in Rural Ethiopia." World Development 31(2003):1309-24. Roberts, R. "A Positive Model of Private Charity and Public Transfers." Journal of Political Economy 92(1984):136-48. Rosenzweig, M., and K. Wolpin. "Evaluating the Effects of Optimally Distributed Public Programs: Child Health and Family Planning Interventions." American Economic Review 76(1986):470-82. Sharp, K. "Targetingof Food Aid in Ethiopia." Unpublished, Save the Children Fund-Ethiopia, 1997. Svedberg, P "Undernutrition in Sub-Saharan Africa: Is There a Gender Bias?" Journal of Development Studies 26(1990):469-86. von Braun, J., ed. Employment for Poverty Reduction and Food Security. Washington DC: International Food Policy Research Institute, 1995.

288

May 2005

Amer. J. Agr. Econ.

Webb, P., and S. Kumar. "Food and Cash for Work in Ethiopia: Experiences during Famine and Macroeconomic Reform." In J. von Braun, ed. Employment for Poverty Reduction and Food Security. Washington DC: International Food Policy Research Institute, 1995. World Health Organization. WHO Cooperation Strategies in Ethiopia: Investing for Healthy

Generation. Addis Ababa: World Health Organization, 2002. Wooldridge, J. M. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: The MIT Press, 2002. World Bank. World Development Report. Investing in Health. New York: Oxford University Press, 1993.