macro shocks and micro outcomes: child nutrition during indonesia’s crisis

Economics and Human Biology 2 (2004) 21–44

Macro shocks and micro outcomes: child nutritionduring Indonesia’s crisis

Steven A. Blocka,∗, Lynnda Kiessb, Patrick Webbc,Soewarta Kosend, Regina Moench-Pfannerb,

Martin W. Bloemb, C. Peter Timmerea Fletcher School of Law and Diplomacy, Tufts University, Medford, MA 02155, USA

b Helen Keller International Asia-Pacific Regional Office, Jakarta, Indonesiac Friedman School of Nutrition Science and Policy, Tufts University, Medford, MA, USA

d National Institute for Health Research and Development, Ministry of Health,Government of Indonesia, Jakarta, Indonesia

e Development Alternatives, Inc., Bethesda, MD, USA

Received 19 December 2003; received in revised form 19 December 2003; accepted 19 December 2003

Abstract

A survey of households in rural Java is used to assess the nutritional impact of Indonesia’s droughtand financial crisis of 1997/1998. A time-age-cohort decomposition reveals significant nutritionalimpacts. However, child weight-for-age (WAZ) remained constant throughout the crisis, despiterapid increases in food prices and the consequent household consumption shock. The evidence isconsistent with the hypothesis that within households, mothers buffered children’s caloric intake,resulting in increased maternal wasting. However, reductions in the consumption of high-qualityfoods further resulted in increased prevalence of anemia for both mothers and children. The com-bined effects were particularly severe for cohorts conceived and weaned during the crisis.© 2004 Elsevier B.V. All rights reserved.

JEL classification: I12; I31; D12; O12; O15

Keywords: Nutrition; Micronutrients; Crisis; Indonesia; Consumption shocks; Malnutrition; Health

1. Introduction

Indonesia’s dramatic economic crisis beginning in 1997, is generally referred to as a“financial” crisis, but is more accurately portrayed as an interaction among three separate

∗ Corresponding author.E-mail address: [email protected] (S.A. Block).

1570-677X/$ – see front matter © 2004 Elsevier B.V. All rights reserved.doi:10.1016/j.ehb.2003.12.007

22 S.A. Block et al. / Economics and Human Biology 2 (2004) 21–44

Table 1Price changes for selected foods, January 1997 to October 1998 (%)

Mean price increase Standard deviation

Rice 195.2 29.2Other cereals and tubers 137.5 101.8Fish 89.1 67.4Meat 97.0 49.3Dairy and eggs 117.1 31.9Vegetables 200.3 129.5Pulses, tofu, and tempeh 95.2 76.0Fruit 103.7 61.3Oils 122.0 74.8Sugar, coffee, and tea 142.9 28.3Prepared food and beverages 81.4 51.7

Source: Friedman and Levinsohn (2002)based on their analysis of SUSENAS and BPS surveys of urban marketsin 27 provinces.

processes: a financial crisis with regional ramifications, a national political upheaval leadingto a change of government, and a series of agricultural shocks linked to global climaticevents. The Indonesian economy shrank by 14% in 1998 alone as the meteoric depreciationof the Rupiah beginning in January interacted with a banking crisis and loss of investorconfidence, to generate severe inflation (Radelet, 1999; Radelet and Sachs, 1998). Thissituation undermined the regime of President Suharto, who resigned in May 1998 in thewake of serious street rioting, leaving a political vacuum in his wake. In addition, most ruralareas were also suffering a severe drought (and wildfires) linked to theEl Niño phenomenon,which preceded and exacerbated the financial crisis.

For central Java—the focus of this study—the period of lowest rainfall extended fromFebruary 1997 to January 1998 causing a serious reduction in the subsequent harvest(Gilligan et al., 2000). The production shortfall generated food shortages, which droveup food prices, thereby contributing further to aggregate inflation (Table 1) (Friedman andLevinsohn, 2002). Indeed, January 1998 saw the highest monthly inflation in 24 years(6.9%), caused in large part by a greater than 10% per month increase in food prices (withrice alone accounting for 15% oftotal inflation) in that month (Government of Indonesia,1998).

Although the macroeconomics of Indonesia’s “implosion” have been widely documentedand analyzed, its household-level consequences have received relatively less attention. Sev-eral recent studies have used household data to assess real price effects of the crisis toderive improved deflators for measuring poverty (Levinsohn et al., 2003; Suryahadi et al.,2000; Friedman and Levinsohn, 2002). One study examines the effects of the crisis on agri-cultural households, emphasizing income, production, and input demand effects (Gilliganet al., 2000). Micronutrient intake is of particular concern in this context, as food priceshocks may lead poor consumers to buffer their caloric intake at the expense of thequalityof their diets in terms of micronutrient content. Micronutrient deficiencies can cause learn-ing disabilities, impair work capacity, and have been associated with heightened morbidityand mortality—particularly among pre-school children and pregnant women (World Bank,1994; Commission on Macroeconomics and Health, 2002).

S.A. Block et al. / Economics and Human Biology 2 (2004) 21–44 23

Indeed, the households in our sample substantially reduced their consumption of micronu-trient-rich foods during the crisis. As eggs are a relatively affordable and important source ofmicronutrients, as well as protein and calories, they likely to be a good proxy for high qualityfoods in the diet.1 Household egg consumption declined steeply from December 1996 toOctober 1998, falling at an average rate of 2.5% per month. The level of egg consumptionduring that period fell from 0.54 to 0.24 eggs per person per week. During 1999 and 2000we observe a moderate recovery and stabilization of egg consumption, though at a levelduring 2000 that was only half the level during 1996.Friedman and Levinsohn (2002)report a 117% increase in dairy and egg prices in national markets over the same period(Table 1).

Similarly, dark green leafy vegetables are an important and relatively inexpensive sourceof iron, Vitamin A, calcium, folate, and other trace minerals. Between December 1996and July 1998, per capita consumption of dark green leafy vegetables fell by nearly 6% (astatistically significant difference). Though a relatively small decline in percentage terms,its nutritional impact could still be substantial for poor households. For instance, vegetablesaccounted for two-thirds of child Vitamin A intake (adjusted retinol equivalent) in thesample.2 Household consumption of dark green leafy vegetables plummeted more rapidlyfollowing the height of the crisis, falling 30% between July and December 1998. Indeed,average per capita consumption of dark green leafy vegetables during 2000 was 20% lowerthan the average level for the pre-crisis year of 1996 (a statistically significant difference;t = 24.4).

Rice consumption per capita also fell during this period. Using SUSENAS3 data forrural central Java for 1996 and 1999,Skoufias (2003)reports declines in both total calo-ries and total calories from cereals of 7.7 and 7.1%, respectively. Per capita caloric intakeamong those in the bottom income quartile fell by 10.3% over this period. Consistentwith the NSS data, Skoufias also reports significant declines in the share of total calo-ries derived from fish, meat, and dairy products between 1996 and 1999, with increasesin the calorie share from tubers. Households, particularly in the bottom income quartile,thus sacrificed both micronutrient-rich foods and preferred grains (rice) in an effort tominimize their decline in total caloric intake. The per capita changes, however, do notaddress potential changes in the intra-household allocation of calories in response to thecrisis.

The impact of the crisis on nutritional outcomes has also been addressed in several stud-ies, albeit with sometimes conflicting results (Saadah et al., 1999; Bloem and Darnton-Hill,2000; Atmarita et al., 2000; de Pee et al., 2000). The most comprehensive work on In-donesian households during the crisis has been based on the Indonesia Family Life Survey(IFLS), summarized inFrankenberg et al. (1999)andStrauss et al. (2002). The IFLS studiesaddress a broad range of topics, including nutritional outcomes. In those instances where

1 Eggs are also a proxy in the sense that egg consumption is correlated with consumption of more expensivemicronutrient-rich foods, including meat and fish (Skoufias, 2003).

2 In Bangladesh,Bouis and Novenario-Reese (1997)found vegetables to account for nearly 95% of Vitamin A,75% of vitamin C, and 25% of iron intake.

3 SUSENAS is the National Socio-Economic Household Survey conducted the National Bureau of Statisticsevery 3 years.


our results are directly comparable to the IFLS nutrition results, we find general agree-ment. In other instances, the two results are not comparable, as they pertain to differentsub-groups of the population. For instance,Frankenberg et al. (1999)report increases inmean hemoglobin concentration while we find declines. Yet, our result pertains specificallyto children under 5 years old, while the 1999 IFLS result aggregates adults and childrenof all ages. Similarly,Strauss et al. (2002)report no change in the body mass index (BMI)of adult women, while we report declines. Yet, our data are limited to mothers of youngchildren, while the IFLS result pertains to all women. Indeed, the IFLS result reported inFrankenberg et al. (1999)that adult BMI declined while child weight-for-height increasedis consistent with our hypothesis that households buffered the caloric intake of children inresponse to the crisis.

The only direct contradiction between our results and those reported from IFLS data relateto changes in child hemoglobin concentration, whereStrauss et al. (2002)find declines forboth boys and girls to be statistically significant only for boys, while we report statisticallysignificant declines for both boys and girls. Indeed, the differences that remain between ourresults and the IFLS results may also be resolved by greater regional disaggregation of theIFLS results, which have been reported only at the national level, while our study pertainsonly to rural central Java.

The present paper, to our knowledge, is the first to use high frequency nutritional datawith an econometric technique that decomposes trends into the specific impacts of time,age, and cohort effects (Browning et al., 1985; Deaton, 1997). This model has two benefitsin the present context. The first is that it enables us to isolate the time path of particularnutritional indicators, disentangling the time effect from the potentially confounding effectsof age and cohort. This technique has the added benefit of creating a framework that permitsus to link maternal nutrition experience during shocks with the subsequent nutritional andhealth outcomes of particular cohorts of offspring.

2. Data

Nutrition surveillance activities were started in rural central Java in December 1995 aspart of a monitoring and evaluation system for a social marketing campaign focused onVitamin A. Five rounds of data collection were completed through January 1997. Nutritionsurveillance system (NSS) data collection was reinitiated in central Java in June 1998, andcontinued at approximately 3-month intervals.4 The present analysis uses all 14 rounds ofdata, covering the period from December 1995 to January 2001.5

For each round, a random sample of 7200 households was selected using a multi-stagecluster sampling design. A total of 30 villages were selected from each of rural central

4 The NSS data did not begin to record expenditure data until well after the peak crisis period.5 Two earlier papers (de Pee et al., 2000; Bloem and Darnton-Hill, 2000), used the same nutrition surveillance

data (repeated cross-sectional surveys for central Java), focusing on up to six waves of data collection. By contrast,the present analysis is able to depict a more detailed picture of the dynamics of nutrition change during Indonesia’scrises by using all 14 rounds of the data. Dates of survey rounds: 1, January 1996; 2, April 1996; 3, July 1996; 4,October 1996; 5, December 1996; 6, July 1998; 7, October 1998; 8, December 1998; 9, April 1999; 10, October1999; 11, May 2000; 12, August 2000; 13, October 2000; 14, January 2001.


Table 2Descriptive statistics

Pre-crisis(December 1996)

Peak crisis (July1998)

Post-crisis(January 2001)

Egg consumption per adultequivalent per week

0.54a (0.42)b [7125]c 0.29 (0.25) [5422] 0.26 (0.12) [6761]

Dark green leafy vegetableconsumption per a.e. per weekd

0.38 (0.25) [7125] 0.36 (0.28) [5422] 0.29 (0.24) [6761]

Child weight-for-ageZ-score −1.15 (1.18) [7120] −1.16 (1.10) [5430] −1.45 (0.99) [6759]Child hemoglobin concentration 11.00 (1.28) [1115] 10.45 (1.27) [1753] 10.92 (1.23) [1279]Maternal body mass index 21.42 (3.04) [7130] 21.10 (2.99) [5344] 21.70 (3.29) [6742]Maternal hemoglobin

concentration12.77 (1.31) [1172] 12.60 (1.29) [1712] 12.84 (1.32) [1331]

a Sample mean.b Standard deviation.c Number of observations.d a.e. indicates adult equivalent, calculated here as the number of adults plus one-half the number of children

under 6 living within the household.

Java’s six ecological zones by probability proportional to size sampling techniques. Eachvillage provided a list of households containing at least one child under 36 months of age(the age eligibility criterion was expanded to 59 months in round 7 (August 1998)). Fromthis list, 40 households were selected by fixed interval systematic sampling using a randomstart. The total sample size for the 14 rounds is 33,600 households, providing observationson 107,753 children. The number of children observed across the 14 rounds varied from5450 to 10,553.6

The population represented in this sample is rural, yet only 20% own their own ricefields. The plurality of males (32%) work as daily laborers, and an additional 15% areemployees or civil servants. Among women, 76% describe themselves as “housewivesor unemployed”. Both men and women complete on average approximately 7 years ofschooling, and the average family size in our sample was 5.3 when inclusion in the sam-ple required one child under 36 months and 5.2 when this ceiling was raised to 59months.

Table 2provides descriptive statistics for variables included in the analysis at three timesrepresenting pre-crisis, peak crisis, and post-crisis survey rounds. Household intake of darkgreen leafy vegetables was obtained by asking the respondent if vegetables had been pre-pared in the last 3 days and if so, how much (in kg) from different sources (e.g., marketpurchase versus own production or gathering). This information was used to calculate theamount prepared in the household per day. Egg intake was estimated by recording the num-ber of eggs consumed from own production as well as those purchased in the market. Weightfor children under 59 months was measured to the nearest 0.1 kg, length and height mea-surements to the nearest 0.1 cm. Blood samples were collected from a random sub-sample(approximately 18%) of children and mothers by finger-prick to measure hemoglobin

6 Greater detail on nutritional surveillance methods is available inde Pee et al. (2000), and in annual reports ofHelen Keller International.


concentration. The figures inAppendix A illustrate trends in child weight, child height,and maternal body mass index over the survey period.

3. Methodology

The methodological challenge in assessing the effects of Indonesia’s drought and financialcrisis on child nutrition lies in the need to trace an inherently dynamic process in theabsence of panel data in which individuals are followed over time. Rather, the NSS datapresent successive cross-sections of clusters that are randomly re-sampled in each surveyround.7 While this data structure precludes tracing individuals’ nutritional status over time,it does enable us to divide the sample population into relatively homogeneous groups andto trace the average status of those groups over time. In the present setting, date of birthprovides a natural grouping for individuals in the sample. Tracking the experience of theresulting cohorts thus provides a means of approximating the dynamics of the phenomenonof interest—in this case child nutrition. Hence, we adopt (with modifications describedbelow) the average-cohort techniques proposed inBrowning et al. (1985), Deaton andPaxson (1994), Attanasio (1998), and reviewed inDeaton (1997).

The underlying motivation for this approach is that the “snapshot” of a single cross-sectionmay distort the dynamics of interest. Our primary interest lies in the combined effects ofdrought and financial crisis on child nutrition in Indonesia (the political crisis being moreremote to child outcomes in rural central Java). Yet, if child nutrition is also a function ofa child’s age, variation in age can confound the interpretation of cross-sectional evidence.8

If the sample’s age composition changes over time (as does ours), the potential dynamicdistortions become even greater in considering the sequence of “snapshots” provided bymultiple survey rounds.9 Moreover, if there are secular changes over time in nutritionalstatus,cohort effects provide an additional confounding variable. For instance, a post-crisis‘rebound’ in economic growth may be relatively more beneficial to children born morerecently (into the recovery period). The techniques developed by Deaton and others addressthese potentially confounding effects.

As noted, prior applications of this decomposition methodology have been limited tolower frequency data.10 In such applications, cohorts are typically identified by year of

7 Cluster sampling is the accepted method of rapidly assessing large population groups, particularly in thecontext of emergencies. The aim is to secure sufficient information that is representative of the total population aswell as for any subgroup that may be distinguished with the total (WHO, 2000; WFP, 2001).

8 For example,Sahn and Alderman (1997)showed that for Mozambique increases in household incomes onlyaffects the nutritional status of children older than 2 years, while increases in mother’s education only affectsthe nutritional status of children under 2 years of age. In other words, since the determinants of anthropometrycan be different for different ages, age differences within a cross-section need to be controlled for. Indeed, theinterpretation of anthropometry differs by age group.

9 Growth faltering starts as early as 3–4 months and plateaus by 3 years in most countries, while stuntingcompounds itself over time.10 Previous applications of this approach includeDeaton and Paxson (1994)andAttanasio (1998), who examine

life cycle models of savings and consumption,Hall (1971)who examines technical change for vintages of pickuptrucks, andWeiss and Lillard (1978), who examine age, cohort, and time effects on earnings of scientists.


birth, and observed annually. As a result, tracking a given cohort involves observing thoseage 25 in the first survey, age 26 in the second survey, and so on. The subsequent cohortwould begin with those aged 26 in the first survey, aged 27 in the second survey, and soon. By contrast, the present study, based on observations separated by as few as 4 months,requires a definition of cohorts based on the month of birth relative to the final survey round,with subsequent age effects also measured in months.

Our decomposition model of child nutrition outcomes builds from several related hy-potheses. The first is that the real-time income and consumption shock imposed byIndonesia’s crises affected child nutrition. The second hypothesis, supported by exten-sive biological evidence, is that child anthropometry and blood hemoglobin concentrationstend to follow known patterns as a function of child age. Finally, it follows from the firsttwo hypotheses that the time shock may have shifted downward the age profile of partic-ular cohorts that were at especially vulnerable ages at the time of the crises. Thus, for agiven nutrition indicator,N, we begin with the model:N = f(t, a, c), wheret indicatestime,a indicates age, andc, indicates cohort. A cohort is defined here as all children bornin a particular month. The numbering of these cohorts is arbitrary, and is defined heresuch that the youngest cohort born during the last survey round is designated 1. For localchanges in this function, we can express this model as an additive function in logarithms:lnN = ln f(t) + ln g(a) + ln h(c), where the specific functional forms off, g, andh areunknown a priori.11

Cohorts are numbered to indicate when they were born relative to the NSS surveys. The14 survey rounds used here were collected over a period spanning 61 months (1996–2001).The cohorts (c) are numbered according to their age (a) at the time of the final survey. Thus,the youngest cohort observed (born during round 14, or survey month 61 to January 2001) isdesignated as cohort 1; that is,c = a− survey month+ 61. Given the age range of childrenin the sample (0–35 months in survey rounds 1 through 5, 0–59 months beginning in round6) and 14 rounds of surveys, we have at least one observation of 95 monthly cohorts.12

The least restrictive (parametric) approach to estimating the above equation is to allowthe data to define the specific functions by representing each of the functions with dummyvariables. FollowingDeaton and Paxson (1994)andDeaton (1997), we write C, T, andA as matrices of dummy variables for each cohort, survey month (time), and age. Thenumbers of columns, respectively, in these matrices are 81, 14, and 60. Rewriting the aboveequation in terms of these dummy variable matrices and adding an error term yields ourestimating equation: lnN = ιβ+ Tψ+Aα+Cγ + ε, whereι is a vector of ones, andψ, α,andγ are parameters to be estimated for time, age, and cohort effects. From each dummyvariable matrix (as with any dummy variable specification, in which estimation requiresthe exclusion of one category), we must eliminate one column. This specification, however,presents additional problems for identification arising from the complete determination

11 Note that whereN indicates aZ-score (such as in the case of weight-for-height, etc.), we use theZ-scoredirectly in place of a logarithm.12 In implementing this approach cohorts numbered 87 and above are omitted since they “graduated” from the

sample at age 36 months and were thus not observed in any post-crisis survey rounds. The first five cohorts arealso omitted from this specification since they did not reach 6 months of age before the final round (and so wereobserved only 1–3 times).


of cohorts by age and time. That is, as cohort= age− time + a constant, knowing anytwo dimensions automatically determines the third and they are thus linearly dependent.Overcoming this problem requires dropping one more column from one of the dummyvariable matrices to ensure that the third dimension is not a strict linear combination of theother two.

Without imposing further structure, however, the linear relationship between age, co-hort, and time effects cannot be separately identified. AsAttanasio (1998)notes, the dif-ferences between two individuals observed at the same age could be due to time or tocohort effects; the differences between two individuals observed at the same time couldbe due to age or to cohort effects. In either case, we can estimate only two linear com-binations of the three coefficients. Addressing this problem requires a strong identifyingassumption, making one of the three effects orthogonal to the others and zero on average(i.e., the residual).13

The approach taken in previous studies was to construct the time effects to be orthogonal,with mean equal to zero. This is equivalent to assuming that all linear trends in the data canbe interpreted as a combination of age and cohort effects. In the context of those life cycleand technology vintage studies it was appropriate to treat the time effect as a zero-meanbusiness cycle effect. However, this standard identifying assumption isnot appropriate forpresent purposes. In this case, time effects represent the dynamic effects of the crisis, whichcontinued throughout our period of observation. Forcing the time dummies to sum to zerowould thus predetermine our results. Instead, we make the explicit assumption that all lineartrends in these data can be interpreted as a combination of time and age effects, leaving thecohort effect as an orthogonal residual.

This identifying assumption is justified by a combination of biological and economicconsiderations. The methodology rests on the assumption that there is no trend or pre-dictable pattern in the dimension chosen as residual. The present application concentrateson hypothesized negative impacts of a known (and ongoing) economic crisis, making timean inappropriate candidate for residual status. Biological evidence strongly supports theexistence of predictable age patterns in the indicators of interest. It is known, for example,that blood hemoglobin concentration tends to increase with child age after 12 months, fol-lowing a dramatic decline between 0 and 5 months. Thus, age is also not an appropriatecandidate for residual status.

The only remaining candidate is the cohort effect. In the present context this dimensionis the best choice for residual, not only for the economic and biological reasons just noted,but also because in such high frequency data, in which the cohorts are separated in birthby only 1 month and observed only until age 5, it is reasonable to assume that there is noapparent trend in that dimension. If the crisis did have differential impacts across cohortsthat result would not appear as a trend, but would still appear in the estimated cohort effects.

FollowingDeaton (1997), we impose the normalization thats′cγ = 0, wheresc is a (vec-tor) arithmetic sequence (0, 1, 2, 3,. . . ) of the length given by the number of columnsin the cohort dummy variable matrix. Like Deaton, we implement this normalization byconstructing the cohort dummies (dc equal to one for cohortc, zero otherwise) asd∗

c =

13 Note that this approach precludes estimation of interaction effects between time, age, and cohort.


dc − [(c − 1)d2 − (c − 2)d1]. Thus, implementation of the estimating equation becomesa regression of the given health or nutrition indicator on dummies for each time period(excluding the first), for each age in months (excluding the first), and for each cohort (ex-cluding the first two). These regressions are run on cell means for each of the cohort/surveyround combinations in the sample.14

4. Results

4.1. Child weight-for-age (WAZ)

The age effect in weight-for-age (conditional on time and cohort) reflects a steep de-cline by 1.25 standard deviations during the first year of life, after which the sample meanstabilizes at approximately 1.5 below the international reference (Fig. 1). This pattern iscommon in samples from developing countries (WHO, 1995), and thus provides an addi-tional illustration of the justification for treating cohort effects as the trendless residual inour model.15

Consistent with previous findings byAtmarita et al. (2000), Frankenberg et al. (1999),and Strauss et al. (2002)the time effect for WAZ (e.g., crisis impacts, conditional onage and cohort effects) fails to show a substantial decline during Indonesia’s crisis period(Fig. 2).16 While a statistically significant decline is apparent during the pre-crisis year, thedecrease of 0.16 standard deviations relative to international reference standards increasedthe prevalence of underweight from 27 to 30%.17

The key result to be explained, however, is how young children were essentially ableto maintain their weight-for-age during a period of precipitous decline in household-levelrice consumption. One possibility is that households reallocated calories (primarily in theform of rice) from mothers to children, thus buffering them from the caloric shock to thehousehold resulting from the rapid increase in food prices. Evidence consistent with thisinterpretation comes from examining the changes in maternal nutrition outcomes duringthis period.

4.2. Maternal body mass index and hemoglobin concentration

Maternal nutrition deteriorated: mean maternal BMI on the eve of the crisis (Decem-ber 1996) was 21.4 (S.D. = 3.04). By July 1998, mean maternal BMI had decreased to

14 As a precaution against the disproportionate influence of outliers these regressions are run using a robustestimator, though in practice, the results are not sensitive to this precaution. The approach is first to estimate anOLS regression to screen for and eliminate gross outliers, based on a measure of residuals (Cook’s distance >1). With the remaining observations, a weighted least squares regression is estimated in which the weights arecalculated as the inverse of each observation’s absolute residual. In practice, the results are not sensitive to thisprocedure (which is available asrreg in Stata).15 The cohort effect (graph omitted), while jointly different from zero, indicates that the crisis did not differentially

affect different cohorts of children in rural central Java.16 This is true for both boys and girls, when we disaggregate this path.17 As is standard practice, we adopt a cutoff for underweight of−2 S.D. from the international reference mean.

30 S.A. Block et al. / Economics and Human Biology 2 (2004) 21–44-1

.5-1

.25

-1-.

75-.

5-.

250

.25

.5

0 12 24 36 48 60Child Age (months)

Cha

nge

in Z

-Sco

re R

elat

ive

to B

irth

Fig. 1. Conditional age path of child weight-for-age.

21.1 (S.D. = 2.99).18 This reflects a mean decrease in maternal weight of 0.83 kg, con-ditional on height (Fig. 3). This difference is statistically significant at greater than 0.01level (t = 5.93). These changes reflect an increase in the prevalence of maternal wasting(BMI ≤ 18.5) from 14.4 to 17.4% (an increase of 20% from baseline).19 Indeed, maternalBMI returned to its pre-crisis level by October 1999, only to exceed it.20 This is stronglysuggestive of changes in intra-household caloric distribution to buffer children (though welack the individual-specific intakes to substantiate that notion). In addition, mean maternalhemoglobin concentrations declined (statistically significantly) over this period (Fig. 3),sufficiently to increase the prevalence of maternal anemia from approximately 9–12%,lending credence to this interpretation.

A pattern thus emerges in which young children were able to maintain weight for themost part during a period of declining household food consumption, potentially as a resultof caloric transfers from mothers to children. The finding of increased maternal wasting

18 These data thus suggest secular improvements in female BMI when compared withKusin et al. (1979)whoreported a mean female BMI of 18.9 in east Java.19 de Pee et al. (2002)present similar evidence from the same data source.20 The decrease in maternal BMI is consistent with the evidence cited above of declining per capita caloric intake,

though BMI may also decline as a result of increased physical activity and infectious disease for a given quantityof calories.Strauss et al. (2002)report increased rates of female employment over this period (though the preciseimplications of this will depend on type of employment, location, and other factors).


-.5

-.25

0.2

5.5

Jan96 Jul96 Jan97 Jul97crisisJan98 Jul98 Jan99 Jul99 Jan00 Jul00 Jan01Date

Cha

nge

in Z

-Sco

re R

elat

ive

to B

ase

Peri

od

Fig. 2. Conditional time path of child weight-for-age.Note: Base period is first observation.

supports this interpretation.21 Yet, given the dramatic increases in the price of both starchystaples and micronutrient-rich foods, and the consequent declines (at least at the householdlevel) in the consumption of both, it is natural to consider changes in child micronutri-ent status during this period, as caloric buffering may have displaced micronutrient-richfoods from children’s diets. A subtler picture of nutritional impact emerges when we applythe time/age/cohort decomposition to the highly responsive blood chemistry indicator ofhemoglobin concentration. Indeed, this approach reveals a direct link between maternalwasting andsubsequent micronutrient status of those cohorts conceived and weaned duringthe crisis.

4.3. Child micronutrient status

Blood hemoglobin concentration provides the most revealing picture of crisis impacts—one that reveals the effects of poor dietaryquality in addition to quantity. Iron deficiency andiron deficiency anemia have been associated with an increased risk of mortality and reduced

21 This result offers an interesting complement to the IFLS result reported inStrauss et al. (2002)that there wasno decline in BMI forwomen in general. All of the women in the present NSS survey are by definition mothersof young children. Thus, it is consistent with both studies to conclude that there was no increase in wasting forwomen in general, while there was among mothers of young children (who would be the only women engaged inbuffering the caloric intake of young children).


Fig. 3. Maternal hemoglobin concentration and BMI.Note: Base period is first observation.

learning capacity. The peak crisis period in Indonesia was accompanied by substantialdeclines in household consumption of eggs and dark green leafy vegetables—foods thatare important sources of iron and other micronutrients. Decomposing trends in blood ironlevels (measured by blood hemoglobin concentration) in children reflects the expectedconsequence for micronutrient status. Indeed, the share of children in rural Java vulnerableto iron deficiency anemia is substantial. Anemia in pre-school children is defined as ahemoglobin concentration (Hb) of less than 11.0 g/dL of blood (WHO, UNICEF, UNU,1998). Mean Hb in children in the NSS data set is 11.02 g/dl, and the prevalence of anemiaamong children over the entire sample period is 47%.22

Fig. 4 aggregates the time effects of Hb for all children under 5. The decline in meanchild hemoglobin concentration from December 1996 to July 1998 (conditional on age andcohort) was 0.64 g/dL, resulting in a decline in the mean level from 11.0 to 10.36 g/dl. Thetime effects are statistically significant (as indicated by the 95% confidence bands inFig. 4),as are the differences in levels between December 1996 and July 1998. The result of the

22 Anemia is commonly used as an indicator of iron deficiency. Although other causes of anemia (hookworm,malaria) may be present in the study area, the high prevalence of anemia in this population and the high ironrequirements for pre-school children and pregnant women suggest that most of the anemia is caused, in part or intotal, by iron deficiency and other micronutrient deficiencies (Vitamin A, folate, B6, and B12) to a lesser extent.Anemia is a more severe form of iron deficiency. Thus, the prevalence of iron deficiency is estimated to be 2–3times the prevalence of anemia.


Fig. 4. Conditional time path of child hemoglobin concentration.Note: Base period is first observation.

decline between December 1996 and July 1998 (approximately the peak crisis period) wasthus sufficient to boost the prevalence of child anemia from 52 to 68%. Average child Hbtended to stabilize at a post-April 1999 average that was 0.5 g/dl lower than the level in theinitial survey round (jointF-score= 16.89).23

One can also consider gender differences in the time path for hemoglobin concentration.Strauss et al. (2002)report from IFLS data that there were differences between young boysand girls with respect to changes in hemoglobin concentration such that boys suffered somedecline, while the small reduction for girls was statistically insignificant. This is the onerespect in which our results may differ from IFLS.Fig. 5demonstrates that while boys andgirls experienced similar declines in mean Hb during the peak crisis period in central Java,boys recovered more rapidly than girls through 1999 and 2000.24 The time paths for boysand girls are statistically different from each other, and both are statistically different fromzero (Fig. 5) (as are the decreases between December 1996 and July 1998).

Fig. 6 isolates the changes in hemoglobin concentration for boys and girls over theentire probability distributions as of December 1996 and July 1998. The vertical line at 11

23 Note that comparing the conditional time path of Hb concentration inFig. 4with the unconditional time path,we find that ignoring age and cohort effects leads to a substantial underestimation of the decreases in mean childHb.24 Our finding that boys recovered more quickly than girls is similar to findings reported byDangour et al. (2003),

who found, at least with respect to growth performance, that girls (age 4 years) were more severely affected thanboys by economic instability in Kazakhstan during the 1990s.

34S.A

.Block

etal./Econom

icsand

Hum

anB

iology2

(2004)21–44

-1-.

75-.5

-.25

0.2

5.5

Jul96 Jan97 Jul97 crisis Jan98 Jul98 Jan99 Jul99 Jan00 Jul00 Jan01Date

chb_boys chb_girls

Cha

nge

Rel

ativ

e to

Bas

e Pe

riod

(g

/dL

)

Fig. 5. Conditional time path of child hemoglobin concentration by gender.Note: Base period is first observation.


0.1

.2.3

.4Pe

rcen

t

7 9 11 13 15 17child hb

Hemoglobin Concentration -- Girls0

.1

.2

.3.4

Perc

ent

7 9 11 13 15 17child hb

Dec. '96 July '98

Hemoglobin Concentration -- Boys

Fig. 6. Nonparametric kernel density functions.Note: The vertical line at 11.0 g/dl indicates the cutoff for anemia.

indicates the cutoff for anemia. The distribution shifts to the left for both genders, sufficientlyto increase the prevalence of anemia among boys from 56 to 70% and among girls from46 to 66%. The combined findings that child WAZ remained essentially unaffected, whilehemoglobin concentrations declined among both mothers and children, and while mothersbecame increasingly wasted, is consistent with the hypothesis that mothers acted to bufferyoung children’s caloric intake by transferring consumption of staples to children whileboth mothers and children consumed fewer micronutrient-rich foods.

Disaggregating Hb results by cohort reveals a moderate degree of heterogeneity notobserved in WAZ.Fig. 7 traces the mean age paths (from age 0 to 36 months) for distinctgroups of close cohorts. This allows a comparison of mean levels of hemoglobin of theseage paths for different groups of cohorts when each cohort is between 6 and 18 monthsof age (e.g., when deficiencies have potentially the most severe and permanent effects onchild growth and development).25 Note that each cohort is a different age at any given pointin time (thus close groups of cohorts are in a particular age range at any given point intime). Systematic vertical shifts in cohort-specific age profiles would indicate that the crisisaffected differentially various cohorts.

25 Yip and Dallman (1996)show that iron deficiency peaks in this age range as a result of rapid growth, depletediron stores, and low iron content of the diet.

36S.A

.Block

etal./Econom

icsand

Hum

anB

iology2

(2004)21–44

Fig. 7. Age paths of child hemoglobin concentration by cohort groups (averaged over groups of cohorts).


Panels A through D present the age paths for increasingly older groups of cohorts, withpanels summarizing the experience of a group of close cohorts by the mean path withinthat group. Panel D groups together cohorts (cohorts 59 and above) that were already 18months or older by August 1997, when the Rupiah began its rapid decline. The age tracesfor those cohorts lie near to the anemia cutoff of 11.0 g/dL indicated by the horizontal lineduring the ages from 6 to 18 months (framed by the vertical lines in each panel). PanelC presents the average age path for all cohorts that were between the ages of 6 and 18months during the period from August 1997 to July 1998 (cohorts 41–58). In contrastto the older group, those cohorts that entered the most vulnerable ages during the crisisexperienced significantly lower hemoglobin concentrations—often well below the anemiacutoff. Panel B presents the average age path for all cohorts born during the period fromAugust 1997 to July 1998 (cohorts 31–41). These cohorts, too, reflect a downward shiftin their age path relative to the oldest cohorts. Indeed, the most severe impacts are illus-trated in Panel A, which presents the average age path for those cohortsconceived duringthe peak crisis period (cohorts 21–31). This is particularly striking, given that these co-horts did not enter the window of 6–18 months until up to 2 years after the peak of thecrisis.

Fig. 8. Conditional cohort effect of child hemoglobin concentration.Note: Cohorts 21–31 were conceived duringthe crisis; cohorts 32–41 were born during the peak crisis period, cohorts 42–58 were within the age range of 6–18months during the peak crisis period; and, cohorts 59 and above were already 18 months of age at the onset of thecrisis. The “initial cohort” in this analysis (e.g., the youngest cohort) is cohort 8, which was born in May 2000.This is the youngest cohort for which there are four observations.


Table 3Tests for differences in mean hemoglobin concentration during ages 6–18 months across cohort groups

Cohort group Mean Hb for ages6–18 months

Pairwisedifferences

Pairedt-test scores(two-tailed)

A Conceived during peak crisis 10.44 (0.061) A− B = −0.137 t = −2.209,P > |t| = 0.047B Born during peak crisis 10.58 (0.062) B− C = 0.040 t = 0.364,P > |t| = 0.723C Ages 6–18 months during

peak crisis10.53 (0.083) C− D = −0.362 t = −4.103,P > |t| = 0.002

D >18 months at onset of crisis 10.89 (0.037) D− A = 0.429 t = −5.844,P > |t| = 0.0001

Standard errors in parentheses.

Table 3summarizes the mean hemoglobin concentrations for each of these cohort groupsduring ages 6–18 months. These means range from 10.89 g/dL for the oldest cohorts to10.44 g/dL for the youngest. Pairedt-tests confirm that the mean hemoglobin concentra-tions between the cohorts in Panel A differ statistically from those in Panel B, that is,those conceived during the crisis fared worse than those born during the crisis. This re-sult may reflect the effect of increased maternal wasting during that period. There is nostatistical difference in the means for those born during the crisis (Panel B) and those6–18 months of age during the peak crisis period (Panel C). Yet, the mean hemoglobinconcentration for the oldest cohorts (those already 18 months of age at the onset of the

Fig. 9. Conditional age path of child hemoglobin concentration.


crisis (Panel D)) are statistically differently greater than all of the younger cohorts, thoughthe mean for the oldest cohorts still fell below the anemia cutoff (P > |t| = 0.017).26

While it appears that there was some recovery among the latest cohorts, there may alsohave been longer-term developmental consequences. A key question for future explorationis whether micronutrient deficiency in utero during a crisis period (fetal insult) has differ-ent child growth consequences than micronutrient deficiency in the early period of infantdevelopment.

Fig. 8summarizes these cohort effects for hemoglobin concentration, reflecting the dropoff among the younger cohorts (i.e., those with lower cohort numbers) more clearly with athird-order polynomial trend line.Fig. 9similarly summarizes the conditional age path forhemoglobin concentration.

5. Discussion and conclusions

This study decomposes trends into time, age, and cohort effects to analyze high frequencynutritional surveillance data. This approach has two potential advantages over more typi-cal cross-sectional approaches. Most directly, the decomposition applied here allows us todisentangle the potentially confounding effects of time, age, and cohort. Indeed, failing todo so would result in substantial underestimation of the nutritional impact of Indonesia’smultiple crises, and in the overestimation of the nutritional recovery. While we find nomeaningful decline in child weight-for-age, blood hemoglobin concentration—a more re-sponsive indicator, and one that provides insight into the quality, as well as the quantityof the diet—declined sharply during the crisis. Indeed, hemoglobin concentration had notrecovered to its pre-crisis level by January 2001. The crisis thus significantly reversed a20-year period of improving nutritional status in Indonesia, at least at the micronutrientlevel.27

In addition, applying cohort decomposition opens greater possibilities for linking theoutcomes of maternal malnutrition with the subsequent nutrition of identifiable offspring.This important aspect of cohort analysis provides at least suggestive evidence that the cohortsconceived and born to increasingly anemic mothers in central Java also experienced a higherincidence of anemia.

These findings invite further investigation along several dimensions. Having establishedsome of the basic nutritional impacts in central Java of Indonesia’s crises a next step willbe to differentiate these impacts by types of household in policy-relevant ways that canbetter inform the design of interventions to mitigate the most damaging nutritional impactsof such crises. Were some household types more vulnerable to the crisis than others? Moregenerally, what are the roles of maternal education, nutrition knowledge, occupation group,

26 This finding is consistent with the results of a case–control study in Jordan, which linked heightened prevalenceof iron-deficiency anemia in infants to anemia in mothers (Kilbride et al., 1999). See alsode Pee et al. (2002).27 Chernichovsky and Meesook (1984)report, for instance, that the prevalence of iron deficiency in central Java

derived from the 1978 SUSENAS survey was greater than 85% versus the 50% baseline rate found among childrenin the NSS sample used here.


and other household characteristics in determining the demand for child micronutrientstatus?28

Our finding that the nutritional consequences of Indonesia’s crisis were particularly con-centrated at the micronutrient level (as reflected by hemoglobin concentration) invites fur-ther investigation of the determinants of child micronutrient status. In this context, we expectmaternal nutrition knowledge to be critical, given the hidden nature of foods’ micronutrientcontext. Indeed, preliminary results (Block, 2002) suggest that maternal nutrition knowl-edge is critical, more so even than formal schooling, in determining child micronutrientoutcomes. Geographic and environmental distinctions across central Java’s six ecologicalzones may also yield policy-relevant insights into what makes certain population groupsmore or less vulnerable to nutritional shocks.29

More generally, the finding that a macro-shock may have its most long-term effectsvia its micro impacts has immediate policy implications for appropriate government re-sponses. Clearly, the poor are the most vulnerable during times of economic crisis. Yet,these findings indicate that public interventions to protect the poor during crises shouldspecifically include safeguards, not simply for gross caloric intake, but the quality ofchildren’s diets, as well. The nutritional consequences of Indonesia’s crisis demonstratethat economic stress can lead households to substitute into lower quality foods to main-tain gross caloric intake, thus increasing the likelihood of potentially irreparable devel-opmental damage to young children who suffer from micronutrient malnutrition as aresult. Indeed, recent work (Ross and Horton, 1998) finds that the adult labor produc-tivity lost as a result of childhood iron deficiency can lead to substantial reductions in GDP,as well.

Acknowledgements

The authors are grateful to Jack Molyneaux, Julie Schaffner, and John Komlos, as wellas to seminar participants at the Friedman School of Nutrition Science and Policy (TuftsUniversity), Yale University, and the World Bank for helpful comments and input. Remain-ing errors are those of the authors. This paper was produced through the USAID/IndonesiaFood Policy Support Activity.

Appendix A

Absolute levels of child height, child weight, and maternal BMI over time, rural centralJava (Figs. 10–12).

28 Initial related studies using the NSS data to address these issues includeBlock (2002, 2004)andWebb andBlock (2004).29 Block and Webb (2001)examine related questions regarding response to shocks in Ethiopia. Geographical

analyses with current data would also permit comparison with a series of early studies of malnutrition in GunungKidul, Indonesia byBailey (1961, 1962).


Fig. 10. Levels of mean child height, by age (children less than 36 months of age).

Fig. 11. Levels of mean child weight, by age (children less than 36 months of age).


Fig. 12. Levels of mean maternal BMI.Note: Sample restricted to mothers above the age of 21.

References

Atmarita, J.A.B., Latief, D., Soekirman, Tilden, R.L., 2000. The effect of economic crisis on the nutritional statusof Indonesian preschool children. Gizi Indo, 33–41.

Attanasio, O., 1998. Cohort analysis of saving behavior by U.S. households. J. Hum. Resources 33 (3), 575–609.

Bailey, K.V., 1961. Rural nutrition studies in Indonesia. II. Clinical studies of hunger oedema in the cassava areasof Java. Trop. Geogr. Med. 13, 234–254.

Bailey, K.V., 1962. Rural nutrition studies in Indonesia. X. Weight and height of Gunung Kidul adults. Trop. Geogr.Med. 14, 230–237.

Block, S., 2004. Nutrition knowledge and the demand for micronutrient-rich foods: evidence from Indonesia. J.Dev. Stud., in press.

Block, S., 2002. Nutrition Knowledge Versus Schooling in the Demand for Child Micronutrient Status. WorkingPaper No. 93, Harvard University Center for International Development.

Block, S., Webb, P., 2001. The dynamics of livelihood diversification in rural Ethiopia: incomes, assets, andperceptions. Food Policy 26, 333–350.

Bloem, M.W., Darnton-Hill, I., 2000. Micronutrient deficiencies: first link in a chain of nutritional and health eventsin economic crises. In: Bendich, A., Deckelbaum, R.J. (Eds.), Primary and Secondary Preventative Nutrition.Humana Press, Inc., Totowa, NJ.

Bouis, H., Novenario-Reese, M., 1997. The Demand for Micronutrients in Bangladesh. Unpublished WorkingPaper, International Food Policy Research Institute, Washington, DC.

Browning, M., Deaton, A., Irish, M., 1985. A profitable approach to labor supply and commodity demands overthe life-cycle. Econometrica 53 (3), 503–543.

Chernichovsky, D., Meesook, O.A., 1984. Patterns of Food Consumption and Nutrition in Indonesia. World BankStaff Working Paper No. 670.


Dangour, A.D., Farmer, A., Hill, H.L., Ismail, S.J., 2003. Anthropometric status of Kazakh children in the 1990s.Econ. Hum. Biol. 1, 43–53.

de Pee, S., Bloem, M.W., Graciano, F., Sari, M., Halati, S., Tjiong, R., Satoto, 2000. Economic crisis causesconsiderable weight loss among Indonesian women and adolescents. Malaysian J. Nutr. 6 (2), 203–214.

de Pee, S., Bloem, M.W., Sari, M., Kiess, L., Yip, R., Kosen, S., 2002. High prevalence of low hemoglobinconcentration among Indonesian infants aged 3–5 months is related to maternal anemia. J. Nutr. 132 (8(August)), 2215–2221.

Deaton, A., Paxson, C., 1994. Saving, growth, and aging in Taiwan. In: Wise, D. (Ed.), Studies in the Economicsof Aging. University of Chicago Press, Chicago (for the National Bureau of Economic Research).

Deaton, A., 1997. The Analysis of Household Surveys. The Johns Hopkins University Press, Baltimore, MD (forthe World Bank).

Frankenberg, E., Thomas, D., Beegle, K., 1999. The Real Costs of Indonesia’s Economic Crisis: Findings Fromthe Indonesia Family Life Surveys. Unpublished Working Paper, RAND, Santa Monica, CA.

Friedman, J., Levinsohn, J., 2002. The distributional impact of Indonesia’s financial crisis on household welfare:a “rapid response” methodology. World Bank Econ. Rev. 16 (3), 397–423.

Gilligan, D., Jacoby, H., Quizon, J., 2000. The Effects of the Indonesian Economic Crisis on AgriculturalHouseholds: Evidence from the National Farmers Household Panel Survey (PATANAS). Unpublished WorkingPaper, Center for Agro-Socioeconomic Research and the World Bank.

Government of Indonesia, 1998. Quarterly Macroeconomic Reports (various dates). Ministry of Finance, Jakarta.Hall, R.E., 1971. The Measurement of quality change from vintage price data. In: Griliches, Z. (Ed.), Price Indices

and Quality Change. Harvard University Press, Cambridge, MA.Kilbride, J., Baker, T.G., Parapia, L.A., Khoury, S.A., Shuqaidef, S.W., Jerwood, D., 1999. Anaemia during

pregnancy as a risk factor for iron-deficiency anaemia in infancy: a case–control study in Jordan. Int. J.Epidemiol. 28, 461–468.

Kusin, J.A., Kardjati, S., de With, C., Sudibia, I.K., 1979. Nutrition and nutritional status of women in East Java.Trop. Geogr. Med. 31, 571–585.

Levinsohn, J., Berry, S., Friedman, J., 2003. Impacts of the indonesian economic crisis: price changes and thepoor. In: Dooley, M., Frankel, J. (Eds.), Managing Currency Crises in Emerging Markets. University of ChicagoPress (for NBER), Chicago.

Radelet, S., Sachs, J., 1998. The east Asian financial crisis: diagnosis, remedies, prospects. Brookings PapersEcon. Activity 28 (1), 1–74.

Radelet, S., 1999. Indonesia: Long Road to Recovery. Unpublished Working Paper, Harvard Institute forInternational Development, Cambridge, MA.

Ross, J., Horton, S., 1998. The Economic Consequences of Iron Deficiency. The Micronutrient Initiative, Ottowa,CA.

Saadah, F., Waters, H., Heywood, P., 1999. Indonesia: undernutrition in young children. East Asia and pacificregion. World Bank: Watching Brief, Issue 1, January.

Sahn, D., Alderman, H., 1997. On the determinants of nutrition in mozambique: the importance of age-specificeffects. World Dev. 25 (4), 577–588.

Skoufias, E., 2003. Crises, food prices, and the income elasticity of demand for nutrients: some estimates fromIndonesia. Unpublished working paper, Inter-American Development Bank, Washington, DC.

Strauss, J., Beegle, K., Dwiyanto, A., Herawati, Y., Pattinasarany, D., Satriawan, E., Sikoki, B., Sukamdi, Witoelar,F., 2002. Indonesian Living Standards Three Years After the Crisis: Evidence from the Indonesia Family LifeSurvey. Unpublished Working Paper, RAND, Santa Monica, CA.

Suryahadi, A., Sumarto, S., Suharso, S., Pritchett, L., 2000. The Evolution of Poverty During the Crisis in Indonesia,1996 to 1999 (Using Full Susenas Sample). Policy Research Working Paper No. 2435, World Bank, Washington,DC.

Webb, P., Block, S., 2004. Nutrition information and formal schooling as inputs to child nutrition. Econ. Dev.Cultural Change 52 (4), in press.

Weiss, Y., Lillard, L., 1978. Experience, vintage, and time effects in the growth of earnings: American scientists,1960–1970. J. Polit. Econ. 86, 427–447.

WFP (World Food Programme), 2001. Food and Nutrition Handbook. Rome.WHO (World Health Organization), 2000. The Management of Nutrition in Major Emergencies. Geneva.


WHO, 1995. Expert Committee. Physical Status: The Use and Interpretation of Anthropometry. Technical ReportSeries 854, Geneva.

WHO, UNICEF, UNU, 1998. Iron deficiency anemia: prevention, assessment and control. Report of a JointWHO/UNICEF/UNU Consultation. Geneva.

World Bank, 1994. Enriching Lives. Overcoming Vitamin and Mineral Malnutrition in Developing Countries.World Bank, Washington, DC.

Yip, R., Dallman, P.R., 1996. Iron. In: Siegler, E.E., Filer, L.J. (Eds.), Present Knowledge in Nutrition. ILSI Press,Washington, DC, pp. 277–292.

macro shocks and micro outcomes: child nutrition during indonesia’s crisis

Documents