ravallion huppi poverty and undernutrition in indonesia during the 1980s

Public Disclosure Authorized
Public Disclosure Authorized
Public Disclosure Authorized
Public Disclosure Authorized

andResearch
Policy,Planning,

WORKING

1

PAPERS

AgriculturalPolicies

Agricultureand RuralDevelopment
Department
The WorldBank
September1989
WPS286

Poverty
and Undernutrition
in Indonesia
during the 1980s
Martin Ravallion
and
Monika Huppi

Because of sustained growth in average real consumption, a
modest improvement in overall equity, and gains to the rural
sector - particularly the poorest of the poor - poverty and
undernutrition continued to be alleviated during Indonesia's
recent period of macroeconomic adjustment.

The Policy, Planning, and Research Complcx disuibutes PPR Working Papers to disscminate the fuindingsof work in progress and to
encourage the exchange of ideas among Bank staff and aU oLhersintcrested in devclopm.entissues. These papers carry the names of
the authors, reflect only their views, and should be used and cited accordingly. The findings, interpretauons, and conclusions are the
authors'own. They should not be attributed to the World Bank, its Board of DusrciGrs,its management, or any of its member countnes.

Plc,Planning,sndiResearch

AgriculturalPollce

Indonesia adjusted rapidly to sharply falling
external terms of trade during the 1980s using a classic package of currency devaluation,
budgetary and monetary restraint, and regulatory
relaxation.
How did the rountry fare in its efforts to
alleviate poverty and undernutrition during that
period?
It is difficult to measure poverty at one point
in time - much less compare poverty in two
periods - but Ravallion and Huppi answer that
question using two large, comparable sets of
household data for 1984 and 1987. They tested
a wide range of possible poverty lines and
poverty measures - and the sensitivity of key
results to many of the underlying assumptions
about poverty.

They found robust evidence that noverty
continued to decline during the period.

Although caloric intake data are far from
ideal, they found evidence that the extent of
undemutrition also fell significantly. For a
caloric intake level which 37 r -rcent of the
population failed to attain in 1984, they found
that only 27 percent of the population failed to
attain it in 1987.
Why was this so? Gains to the rural sector
contributed greatly to the alleviation of poverty.
Gains to the urban sector and population shifts
from the rural to the urban sector also helped but
were quantitatively less important than dire
gains to the rural poor.
Increases in average real consumption and
an improvement in overall equity both helped to
reduce poverty. In addition, Indonesia's recent
economic history had created conditions favorable to alleviating poverty so long as modest

growth in private consumption per capita could
be maintained during the adjustment period.

This paper is a product of the Agricultural Policies Division, Agriculture and Rural
Development Department. Copies are available free from the World Bank, 1818 H
Street NW, Washington DC 20433. Please contact Cicely Spooner, room N8-039,
extension 30464 (54 pages with figures and tables).

The PPR Working Paper Series disseminates the findings of work under way in the Bank's Policy, Planning, and Research |
Complex. An objective of the series is to get these findings out quickly, even if presentations are less than fully polished.
The findings, interpretations, and conclusions in these papers do not necessarily represent official policy of the Bank.
Produced at the PPR Dissemination Center

Poverty and Undernutrition in Indonesia during the 1980s*
by
Martin Ravallion
and
Monika Huppi

Table of Contents

1.

Introduction

1

2.

Measuring Poverty and Undernutrition

3

3.

The Data and Results

7

4.

Proximate Causes

12

5.

An Alternative Poverty Assessment

18

6.

Implications for Future Poverty Alleviation Prospects

20

7.

Conclusions

22

Appendix 1: The Distributions

25

Appendix 2: Lorenz Curve Parameterizations

33

v!otes

35

References

42

Tables

46

Figures

52

* This Is the first In a series of papers reporting results of an ongoing World Bank
research project, "Policy Analysis and Poverty: Applicable Methods and Case Studies"
(675-04), based in the Agricultural Policies Division, Agriculture and Rural Development
Department. We are grateful for the financial support and encouragement of the Bank's
Research Committee. The paper also provides background support for the 'Indonesia:
Poverty Assessment and Strategy Report" being prepared by Asia-Country Department V.
The paper has benefited in many ways from the assistance and comments of staff of the
Country Department; we are particularly grateful to Kyle Peters and Nicholas Prescott.
We have also had useful comments on the paper from Anne Booth, Francois
Bourguignon, Gaurav Datt, Paul Glewwe, Nanak Kakwanl, Lyn Squire, and Dominique
van de Walle. The household level data tapes used here were provided by the Central
Bureau of Statistics, Jakarta. We are most grateful to the Bureau's staff for their
considerable and able help at various stages of this study.

1. Introduction

The 1970s saw a significantdecline in the proportionof
Indonesia'spopulationwho did not attain minimal nutritionaland other
consumptionneeds (Rao, 1984; CBS, 1984). Concern has been expressedabout
whether this success in poverty alleviationhas been sustainedthrough the
far more difficult1980s.1 The various externalshocks of the 19809
includedsharp falls in the price of the country'smain export and source of
public revenue, oil. During 1986 alone, this amountedto about a one third
drop in the country'sexternalterms of trade. It is now widely agreed that
the government'spolicy rusponsesto these shocks were effective in
stabilizingthe main macroeconomicaggregates. However,we know little
about their effects on poverty.
Rapid macroeconomicadjustmentprograms in LDCs can have adverse
effects on the poor, and there has been some casual speculationthat this
may have happened in Indonesia. The public expenditurecuts were certainly
severe; the govornment'stotal rea±

tpendituresfell by about nine percent

between 1985/86 and 1986/87. However,a good deal of the immediateburden
of adjustmentfell on domestic savingsand investmentrather than private
consumption,which did sustain a modest but still positive (per capita)
growth rate over the period. Povertywill not increasewhen mean
consumptionis maintained,providedthat (and it is an importantproviso)
2
the poor do not lose from changes in the distributionof consumption.

What then happenedto Indonesia'sdistributionof consumptionin
the 1980s? Existing evidenceis inconclusive. It appears that the
governmentdid try to avoid its expenditurecuts falling too heavily on
programs (both current expendituresand investments)in which the poor may

- 2 be expected to participatedisproportionately.Its efficacy in achieving
3
this end is less obvious.

The currencydevaluationsand the boom in non-

4

oil exports may have offer-d some protectionto the rural sector.

However,

the extent to which this includedthe rural poor is unclear;while it is
plausible that many of Indonesia'srural poor tend to be net producers of
tradeable goods (and hence gain from real devaluations),there are also
likely to be many poor hou-eholdsin both urban end rural areas who are not.
There have been reports of a decline in real wage rates in rural Java,
5
though there ic some conflictingevidence.

It is also far from obvious that adverse income effects of shortrun macroeconomicadjustmenton the poor will be reflectedin their
consumption. It is not implausiblethat most of the poor do have strategies
of some sort for coping with short-runincome declines,such as through
adjustmentsin their labor market behavior, intertemporalconsumption
6
behavior and/or their participationin the "moral economy".
What is less

clear, and of greater importance,is how well these coping strategies
perform in practice.
This paper examineswhat happened to aggregatepoverty and
undernutritionin Indonesiaduring the period 1984 to 1987. The twin
objectivesare: i) to describeempiricallyhow poverty and undernutrition
changed over this period, and ii) to examine the proximate causes of those
changes. In pursuing both objectiveswe also hope to illustratethe
usefulnessof a number of recent theoreticaladvancesin poverty analysis.
Section 2 gives an informaldiscussionof the methodologicalissues to do
with how poverty and undernutritionare to be measured, concentrating
particularlyon sensitivityto measurementassumptionsand the relevanceof
recent developmentsin the "dominanceapproach"to poverty analysis. The

paper's main empirical results are presented in Section 3 where we give
poverty assessmentsfor various indicatorsof the standardof living of the
poor. While the primary objectiveof this paper is to measure and describe
how poverty and undernutritionin Indonesiachanged over this period, it is
of interestto go at least slightlyfurther into the causes of those
changes. Section 4 attempts to do this by asking what contributions
sectoralgains and populationshifts (on the one hand), economicgrowth and
changes in inequality(on the other) made to aggregatepoverty alleviation.
The importanceof Indonesia'sfavorabledistributionalparametersat che
beginning of the period is also discussed. Section 5 uses the results of
Section 4 to make an alternativeassessmentof how poverty changed over the
period. The alternativemethod does not assume that the levels of household
consumptionare comparableacross the two data sets. In the light of the
paper's results,Section 6 discussesIndonesia'sprospects for future
poverty alleviation. Section 7 offers some conclusions.

2. MeasurinR Poverty and Undernutrition

A fundamentalquestionwhich arises when assessingpoverty is that
of how an individual's"standardof living" is to be measured. Individuals
and the states of their environmentswill differ in many ways which might be
deemed relevantin principle,but are not quantifiable. A similar comment
applies to measures of undernutrition,where variabilityin nutrient
requirements(both between people and over time) is important,but is
difficult to quantify. Though we shall discuss these problems later, for
the moment we shall assume that a measure is availablefor individualliving
standard (generally)or nutritionalintake (as a particularly important

-4aspect of living standard). The problem we face now is how to compare
distributionsof that measure which, in this instance,are the observed
survey distributionsat two dates.
There is now a large theoreticalliteratureon the measurementof
7
poverty, establishinga number of desirablepropertiesfor such measures.

The popular "headcountindex" satisfiesvery few of those properties. There
has been much recent interest in a class of measuresproposed by Foster,
Greer and Thorbecke (FGT) (1984). This is a single parameter index which
can be made to satisfy the main axioms of poverty measurementthrough a
suitablechoice of that parameter. Each member of the FGT class of poverty
measures is identifiedby the value taken by the parametera and the
correspondingmeasure is denoted Pa.8

Three members of the FGT class are

consideredin this study:
*i) The FGT poverty measure for a

0. This is simply the headcount

index, given by the proportionof the populationwith a standard
of living below the poverty line. Thus, if forty percent of the
populationare deemed to be poor then Po = 0.4.
(ii) The poverty measure for a = 1.

This is the average poverty gap

in the population,expressedas a proportionof the poverty
line. Thus a value of P1 = 0.1 means that, when the aggregate
deficit of the poor relativeto the poverty line is averaged
over all households (whetherpoor or not), it represents10
percent of the poverty line.
(iii) The measure for a = 2. This is a distributionallysensitive
measure, based on the sum of the squaredpoverty gaps of the
poor. This measure satisfiesthe main axioms for a desirable
povertymeasure in the literature,includingSen's (1976) (Weak)

-5'TransferAxiom" which requiresthat when a transferis made
from a poor person to someoneun

is poorer, the measure

indicatesa decrease in aggrega- poverty. In view of the
desirablepropertiesof this meaaure we shall simply refer to it
as our "preferredmeasure".
However,whilc the search for better cardinalmeasures of poverty
and undernutritionhas advancedfar over the last fifteen years or so, there
is still widespreadconcern about arbitrarinessin the choice of a poverty
line, or nutrition cut-off point, and in the choice of a specificfunctional
form for the poverty measure. For example,the popular FGT measure P2
discussed above uses only one of a number of possible functionalforms that
might be suggested,all satisfyingthe main axioms for a desirablepoverty
measure (Atkinson,1987). Fortunately,for many (thoughnot all) of the
purposes of measurement,includingsome policy analyses,all that one is
really concernedabout is the ordinal ranking of distributionsin terms of
poverty or undernutrition. For example,the main questionof interestmay
be: did poverty increaseas a result of (say) structuraladjustment? As a
rule, one will be able to answer this questionwith much less information,
includingfewer arbitraryassumptions,than are needed for making the
cardinalcomparison: how much has poverty changed?
Rather than confine ourselvesto a particularpoverty line and
povertymeasure,we shall draw on recent results on the use of dominance
conditionsin orderingincome or expendituredistributionsin terms of
9
poverty.

For this purpose one can considera very broad class of poverty

measures,which encompassthe main contendersfound in the literature. And,
instead of only allowingone or two specificpoverty lines, one can consider
a range of such lines up to some maximum. As long as the class of poverty

measures satisfiescertain rather mild conditions(notablythat the measures
are continuous,separable,symmetricand weakly monotonic),poverty will
have unambiguouslyfallen between two dates if the cumulativedistribution
of income for the latter date lies nowhere above that for the former date,
over the entire intervalup to the maximum allowablepoverty line (Atkinson,
1987, Condition IA). This is called the first-orderdominancetest. If the
result of that test is ambiguous,then we know that differentpoverty lines
and/or measures will rank the distributionsdifferently;some will indicate
a decrease in poverty while others will not. When such ambiguity exists
according to the first-orderdominancetest, then stronger (higherorder)
dominance conditionsmay prove useful. Sen's TransferAxiom is an important
example of the sorts of further conditionswhich may then prove to be
revealingin decidingwhether poverty has gone up o down. When applied to
the poverty measure, this axiom yields the second-orderdominancetest which
basically requiresthat the area under the cumulativedistributionfunction
up to the maximum poverty line is greater (or no less) over the entire range
of admissiblelines (see Atkinson, 1987, ConditionIIA).
A similar approachcan be used to assess changes in the extent of
undernutrition,when (as is certainly the case) there is uncertaintyabout
nutritionalrequirementsand their interpersonaldistribution.Provided that
the interpersonaldistributionof requirementshas not changed, first-order
dominance of the caloric intake distributionfor one date over another
implies an unambiguouschange in the aggregateheadcount index of
undernutrition(Kakwani,1988). This holds no matter where the caloric norm
is located. Furthermore,second-orderdominanceof one intake distribution
over another implies an unambiguousranking accordingto a broad class of
undernutritionmeasures, includingmeasureswhich attach higher weight to

more undernourishedpersons, such as those discussedby Kakwani (1988) and
Ravallion (1988b). We shall only attempt to make ordinal comparisonsof
undernutrition,followingthe dominanceapproach.

3. The Data and Results

The SUSENAS tapes give data on household consumption(from both
market expendituresand own production)for 50,000 randomly sampled
householdsat each date. Followingpast practice for Indonesia,we shall
mainly base our poverty assessmentson distributionsof household
consumptionper person, adjustedto February 1984 urban prices using the
Consumer Price Index (CPI). The CPI is far from ideal for our purposes.
The index is only constructedfor urban areas and its compositionis
inappropriatefor the poor. We have, however, re-weightedthe CPI so as to
better reflect the conaumptionpattern of the poor, and have also built an
allowance for urban-ruralprice differentialsinto the index.10

Appendix 1

discussesthe data and their limitationsin greater detail. For present
purposes, the most worrying featuresof these data are the possibilitythat
the rate of inflationin rural areas may be underestimatedby the CPI and
that SUSENAS data imply a growth rate of real private consumptionper capita
which is higher than implied by the nationalaccounts. We shall return to
these points later.
As noted above, the aeterminationof a poverty line is always
likely to be contentious,and so it is importantto have some idea of how
sensitivepoverty assessmentsare to this choice. This study considersa
range of poverty lines, embracingmost of the alternativesfound in the
literatureon poverty in Indonesia (see, for example, the survey in CBS,

-81984). Results are presentedhere for the two main poverty lines: a 1984
expenditureof Rp 11,000 per month per person and one of Rp 15,000. The
lower povert) line is chosen to :orrespondto that used in past Bank
studies,after adjusting for inflation (Rao, 1984, 1986). We take this to
be the preferredpoverty line.
Table 1 gives our cardinalestimatesof poverty in Indonesia for
various poverty measures and for both poverty lines. All three measures
(includingthe preferred "distributionally
sensitive"measure)and both
poverty lines indicatea significantdecreasein poverty for both urban and
rural sectors.
For the lower poverty line we find that the proportionof the
populationwho are poor decreasedfrom about one in three at the beginning
of the period, to slightlyover one in five by 1987; this is a substantial
contractionover just three years. The correspondingabsolutenumber of
poor persons declined from 52 million to 36 million. The poverty gap
measure implies that the aggregateconsumptionshortfallof the poor
declined from about Rp 937 per month per head of Indonesia'spopulation
(representingabout 5.5 percent of nationalmean consumption)to Rp 464 in
1987 (about 2.3 percent of the nationalmean). Such calculationsare,
however, contingenton the precise poverty line chosen. The higher poverty
line in Table 1 also ir.dicates
significantdeclines in poverty, though the
numbers are a good deal less dramatic.
Are the qualitativeresults robust to the choice of poverty line
and measure? Figure 1 gives the cumulativefrequencydistributionof
consumptionfor the two dates. The 1984 distributionlies entirelyabove
the 1987 distribution. Thus the first-orderdominanceconditionholds and
so one can conclude that all well-behaved("monotonic")poverty measuresand

- 9 -

al'. possible poverty lines will show an unambiguousdecrease in aggregate
poverty between the two dater

This was found to hold for both urban and

rural areas.
From Figure 1 we can also assess the sensitivityof this conclusion
to possible underestimationof price increasesfacing the poor. The 1987
poverty line (in 1984 prices) which would be needed for the 1987 headcount
index of poverty to equal that in 1a84 is Rp 12,818. Thus, an additional
inflation rate of at least 16.5 percentagepoints (on top of the CPI based
estimateof about 20 percent)would have been needed over the three years to
reverse the conclusionthat poverty has decreasedby this measure.
Similarly,the true inflationrate would need to be about 14.1 points higher
over the three years to equalize the headcount indices for the two dates at
the higher poverty line. Thus the conclusionthat poverty has decreased
would be robust to even quite substantialmeasurementerror in the CPI; the
inflation rate would need to have been underestimatedby at least 50 percent
to reverse our conclusion.
A potentiallyimportantobservationabout the results in Figure 1
is that the poverty lines are found on a steep segment of the consumption
distribution. This is illustratedmore clearly by the density functionof
consumptiongiven in Figure 2.

(This is a non-pararatricestimateusing a

Gaussian kernel). The lower poverty line is very close to the mode, where
the slope of the comsumptiondistributionfunctionreaches its maximum.
This has two implicationsof interesthere:
i) Estimatesof the headcount index of povertywill be
particularlysensitiveto the exact locationof the poverty line, as our
comparisonof the lower and upper poverty lines in Table 1 has suggested.
ii) Poverty levels will be very responsiveto horizontalshifts in
the distributionof consumption;indeed,when the poverty line is at the

-

10

-

mode, the responseof the headcount index to an additive gain or loss at all
consumptionlevels will be at its maximum. As the results of the following
sectionwill demonstrate,the responseof poverty in Indonesiato
multiplicativeshifts (in the form of distributionally
neutral changes in
the mean) was also high in the mid-1980s. This will be seen to have
implicationsfor understandingthe effects of recent economic growth on
poverty.
Is our qualitativeresult on the change in poverty over this period
robust to the choice of an indicatorof the standardof living? Three
alternativeswill be considered: income, food expenditureshare, and
caloric intake.
Figure 3 gives the distributionsof householdincome per person. A
comparisonof the entire frequencydistributionagain reveals that the
first-orderdominanceconditionholds. No matter where one draws the
poverty line, or what poverty measure one uses (withina broad class),
aggregatepoverty in terms of incomesunambigiouslyfell between 1984 and
1987. This conclusionis also robust to even substantialmeasurementerror
in the CPI.
Keeping in mind the problems of comparing surveyedconsumptionand
income levels over time, it is of interestto consideranother alternative
measure of the standardof living. It is well known that a household's
apportionmentof consumptionexpenditurebetween food and non-food goods is
an indicatorof that household'sreal consumptionlevel; the share devoted
to non-food goods is generally found to be a monotonic increasingfunction
of real consumption. However, that functionwill only be the same for
householdswho are homogeneousin relevantrespects;the real consumption
level correspondingto a given food share will generallyvary accordingto

relativeprices, demographicfactors and tastes. Differencesin relative
prices and (possibly)tastes between urban and rural areas could well be the
most important factor mitigatingthe welfare interpretationof interhousehold differencesin food shares in Indonesia. For this reason it is
probably better to considerurban and rural areas separately.
Figure 4 gives the cumulativefrequencydistributionsof the share
of non-food goods in total consumptionfor each of urban and rural areas for
each date. First-orderdominancestill holds up to high non-food shares, so
a wide range of poverty lines and measureswill continueto indicate a
decrease in poverty in both sectors over this period. The proportionof the
rural populationwith a food share in excess of 75 percent (a commonlyused
poverty line) fell from 39.2 percent in 1984 to 35.8 percent in 1987, while
for the urban sector it fell from 10.5 to 8.5 percent. The decline is not
as dramatic as that suggestedby the CPI adjustedconsumptio.and income
data, but it is still evident in the distributionof food shares.
Did undernutritionalso diminish? Figure 5 gives the distributions
of calorie intake per person for the two dates. The 1987 distributionlies
below that for 1984 up to a high intake level (the ranking only reversesfor
the upper nine percent of persons). First-orderdominancethus holds up to
high caloricnorms. The "second-orderdominancecondition"discussed in
Section 2 holds over the entire distribution,so that it can be concluded
that a broad class of undernutritionmeasures show an improvementwhatever
the underlyingdistributionof requirements. These resultswere also found
to hold in both urban and rural areas. The quantitativeshift in the
caloric intake distributionis quite substantialat the lower end; for
example, the estimatedproportionof the populationconsumingless than 1500
caloriesper day decreased from 37 percent in 1984 to 27 percent in 1987.

- 12 -

4. ProzimateCauses

We do not intend to go very deeply here into the reasons for
Indonesia'sevident success at alleviatingpoverty during a period of
macroeconomicadjustment. But some clues are offered by the following
statisticaldecompositionsof the reductionin aggregatepoverty.

SectoralGains and PopulationShifts

The results of Section 3 indicatesiCnificantpoverty alleviation
in both urban and rural sectors. There was also a shift in populationover
the period, with a declining share in the poorer rural sector (the
proportionof the populationliving in rural areas fell from 76.5Z in 1984
to 73.6Z in 1987). 'Whatwas the relativecontributionof these factors to
the reductionin poverty?
Using the Pa class of povertymeasure, the change in aggregate
poverty between the two dates can be decomposedinto sectoralgains,
populationshifts, and interactioneffects,as follows:

87
87

84

Pa

87

84 84

= (Pau - Pau)nu

87

+ (Par

gain to urban
sector at 1984
populationshare
r87

+ E(n8
i-u

84 84

_ ni

)Pai

-

84 84

Par)nr

gain to rural
sector at 1984
populationshare
r 87
+ £(Pai

gain due to
populationshifts
at 1984 sectoral
poverty levels

-

84
87
Pai )(ni

-

84
ni )

(1)

i=u

aggregateinteraction
effects between sectoral
gains and population
shifts

where Pai denotesmeasured poverty in sector i (i=u,r for urban and rural)

-

13 -

t
at date t (t-1984,1987),with correspondingpopulationshare ni.
The
aggregate interactioneffect can be interpretedas a measure of the
correlationbetween populationshifts and changes in poverty.
Using this formula,Table 2 gives the urban/ruraldecompositionof
the aggregatepoverty alleviationgains. For the headcount index, using the
lower poverty line, we find that about 10 percent of the aggregate reduction
in povertywas due to a lower prevalanceof poverty amongst the urban sector
(while accountingfor about one quarter of the population).On the other
hand, 85 percent was due to the lower rural poverty,while about 7 percent
was due to the higher urban populationshare. The net interactioneffect is
-2 percent.
It can be seen from Table 2 that both populationshifts and gains
to the urban sector made positivecontributionsto aggregatepoverty
alleviationduring the period for all measures, and were dampenedonly
slightlyby the negative interactioneffect. The quantitativeimportanceof
the sectoralgains to rural householdsis notable, particularlywhen the
measure focusesmore on the poorest of the poor (a-2). The gains to this
sector accounted for the vast majorityof aggregatepoverty alleviation,and
generallymore than the sector'spopulationshare (there is one exception
for the headcount index at the higher poverty line). Lower poverty lines
and monotonic poverty measures tend to enhance considerablythe importance
of gains to that sector; at the lower poverty line and for the preferred
poverty measure, the gains to the rural sector representedover 90 percent
of the aggregategain.
In the light of this result, it is of interestto take a closer
look at the distributionof gains within that sector. For this purpose, all
householdswere classifiedby their principal source of income, as recorded

- 14 -

in the SUSENAS, giving 21 distinctsources for each of urban and rural areas
(Huppi and Ravallion,1989). Significantreductionsin all poverty measures
(generallyat least as high as the sector means) were experiencedby the
poorest rural subsectors,notably farm laborersand self-employedfarm
households. While 11 and 57 percent of all rural persons in the sample were
principallyemployed in these two subsectorsrespectively,they accounted
for 17 and 61 percent of the aggregatedrop in the preferredpoverty measure
using the lower poverty line (Huppiand Ravallion,1989).

Growth, Inequalityand Poverty

An alternativeway to decomposethe poverty alleviationgain is in
terms of the parametersof the aggregateconsumptiondistribution. Roughly
speaking,one can identifytwo proximatecauses of a change in poverty: a
change in the mean consumptionlevel at given inequality,and a change in
the inequalityof consumptionaround the mean; the former can be thought of
as the 'growtheffect on poverty"while the latter is the "distributional
1l.
effect"

However, the qualitativeeffect on poverty of a reductionin

inequalityat a given mean is not obvious a priori; for example,while a
transferof income from someoneat the poverty line (or only slightlyabove
it) to someonewell below will reduce inequality,it will also increasethe
headcount index of poverty.
To derive a useful decompositionformula for resolvingthis issue
empirically,let P17* denote the poverty level that would have occured in
1987 if the change in mean consumptionsince 1984 had not been associated
with any change in relativeconsumptionlevels; i.e., Pa7* is obtainedby
applying the 1987 mean to the 1984 Lorenz curve. Similarly,let P17**

- 15 -

denote th.,poverty level that one would have found in 1987 if only the
Lorenz curve had shifted since 1984, leaving the mean unchanged. The
observed change in poverty between two dates can then be decomposedinto
"growth"and "distributional"
effects as follows:

87

pa

84
87*
84
87**
84
-Pa
.(Pa _pa ) +(pa
_ Pa) +residual
change in
poverty due
to change in
the mean
holding 1984
Lorenz curve
constant

(2)

change in
poverty due
to shift in
the Lorenz
curve holding
1984 mean
constant

However, one should be cautious in drawing policy implicationsfrom
this decomposition. Distributionally
neutral growth is not the same thing
as growth with distributionallyneutral policies. The laizze-fairegrowth
path of an economy need not be distributionally
neutral, and so policy
interventionsaimed at reducing relevantinequalitiesmay well be essential
to attainingeven distributionallyneutral growth. The "growth-equity"
decompositionis a simple descriptivedevise intendedto throw light on the
proximate causes of poverty alleviation;a deeper analysisof those causes
would be needed to draw sound policy implications.
Turning to the Indonesiandata we find that the period 1984-1987
saw a simultaneousincreasein mean consumptionand a reduction in the
overall inequalityof consumption,in both urban and rural sectors. The
three year growth rate in urban consumptionimplied by the SUSENASwas 12.1
percent,while the rural rate was 14.6 percent. Table 3 gives the Lorenz
curves. The 1987 Lorenz curves unambiguouslydominatethose for 1984 in
both sectors and, of course,nationally. Thus all well-behavedinequality
measures will indicate a reduction in inequalityover the period.1 2' 13

The

- 16 -

aggregateGini coefficient (for example)dropped from 0.331 in 1984 to 0.321
in 1987.14
Table 4 gives our estimatesof the relativecontributionsto
poverty alleviationof growth and greater equity, using the decomposition
formula in equation (2).15 The shifts in the Lorenz curve contributedto
poverty alleviationin almost all cases; the exception for rural areas using
the higher poverty line is probably due to the fact that the rural mean for
1984 was actually slightlybelow that poverty line.1 6

In all cases

consideredin Table 4, the majority of the reductionin poverty can be
attributedto higher mean consumptionat given relativeconsumptions;the
contributionof greater equity (upwardshifts in the Lorenz curve) is small
for the headcount index, though it accountsfor a far higher proportionof
the change in the preferredmeasure of poverty.
It is of interest to also considerthe point elasticityof poverty
in 1984 to distributionally
neutral growth. FollowingKanbur (1987a)and
Kakwani (1989b),it can be readily shown that the elasticityof the Pa
poverty measure w.r.t. the mean of the distribution,holding the Lorenz
curve constant is given by:

Va = -zf(z)/P
0 < 0 (for a
=

a(l

- Pa-l/Pa)

=

< 0 (for

0)
a > 1)

(3)

where f(z) denotes the probabilitydensity of consumptionat the poverty
line z. This also has to be estimated;details can be found in Appendix 2.
Table 4 also gives the base period point elasticitieswith respect
to mean consumption. All povertymeasures are found to respond elastically
to higher mean consumptionholding the Lorenz curve constant. In each case,
the (absolute)elasticitiesare highest with respect to the lower poverty

- 17

-

line and in urban areas. For a given poverty line and sector, the growth
elasticityis highest for the preferredmeasure oi poverty and lowest for
the headcount index.

On the Role of Initial Conditions

The results in Table 4 suggest that, by the beginningof the
adjustmentperiod, aggregatepoverty in Indonesiawas ready to respond quite
elasticallyto any further economicgrowth. Initial conditionsof the
period may thus have been favorableto maintainingthe momentum of poverty
alleviation,in spite of lower growth. We now investigatethat conjecture
more closely.
The growth elasticityof poverty is a functionof the parametersof
17
the underlyingconsumptiondistribution.

Consider first the mean.

Differentiating(3) with respectto the mean is, it can be readilyverified
that:
e,

=

Po/
< 0 (for a = O)

_

(a

(4)

- la-i)aPa-l (for a 2 1)

,UPa
The last derivativeis not necessarilynegative for all a > 1, though it is
found to be so for a = 1, 2, for

these

data

(Table

4). This result can be

interpretedas an "accelerationeffect" of growth on poverty;a higher mean
implies a more elastic response (in absolutevalue) of poverty to further
growth. (And, conversely,at low average consumption,higher growth rates
will be needed to achieve the same proportionatepoverty alleviation
impact).

- 18 -

Under plausible assumptionsabout how the Lorenz curves have
shifted over time, it can also be shown that, for these data, the (absolute)
elasticityof the FGT class of poverty measureswith respect to the mean
(again holding the Lorenz curve constant)is a monotonic decreasingfunction
1 8 differentiating(3) with
of the initialGini measure of inequality;

respect to the Gini coefficientG oae finds that (analogouslyto (4)):

OIIO/ G > 0

8-G '

(E a
-

ea

-

(for a = 0)

Ea-1)aPal
QGP

>0

(for a > 1)

(5)

where Ea denotes the elasticityof the Pa poverty measure to Lorenz
dominatingchanges in the Gini coefficient.
It can thus be argued that a past history of fairly equitable
growth (resultingin a relatively"high"mean and "low" inequality)resulted
in a 'high' elasticityof poverty to future growth in Indonesia. The growth
experienceover the 1970s and early 1980s would appear to have created a
sound foundationfor maintainingpoverty alleviationfrom the more modest
growth of the mid-1980s. This finding also has implicationsfor Indonesia's
poverty alleviationprospectsafter 1987, as will be consideredfurther in
Section 6.

5. An AlternativePoverty Assessment.

The methods we have used so far will have overestimatedpoverty
alleviationin Indonesiaif growth rates ir.mean real consumptionhave been
underestimated. The SUSENAS does imply a considerablyhigher growth rate of
real consumptionper capita than do the national accounts;15 percent over

-

19

-

the three years for the former as opposed to 5 percent for the latter. The
reason for this large discrepancyis quite unclear,and there does not
appear to be any sound basis for decidingwhich is closer to the truth.
However, the results of Tables 3 and 4 also allow us to make an
alternativeassessmentof how poverty changed during the period, avoiding
the potentiallycontentiousissues concerningthe comparabilityof the
levels of SUSENAS expendituresacross the two dates. For this purposewe
can use the SUSENAS to assess how overall inequalitychanged during the
period. This assumes that the samplesat both dates were representativeof
the populations,though this seems plausible (see Appendix 1 for further
discussion). From Table 3 we can then concludethat there was no
deteriorationin overall inequality;indeed equity improvedslightlyand in
a way which would have reduced poverty. This use of the SUSENAS data
requirescomparisonsof consumptionor income relativitiesacross dates, not
absolute levels. Since inequalitydid not increase,any positive growth in
per capita consumptionwould (as a rule) have reduced poverty (and, indeed,
since inequalityfell slightly,a modest contractionin mean consumption
would have been possiblewithout an increase in poverty). From the point
elasticityestimatesin Table 4, the national accounts'growth rate for
consumptionper capita implies that the headcount index at the lower poverty
line (for example)would have fallen by about 10 percent over the period (to
a first-orderapproximation). The preferredpoverty measure, on the other
hand, would have declinedby about 17 percent. While these figures are a
good deal less than those impliedby the SUSENAS,they still indicate a far
from negligibleimpact on poverty. The above calculationsalso ignore the
desirableeffect of lower inequalityon aggregatepoverty. The qualitative
conclusionthat poverty declined is robust to relaxingthe requirementthat
even inflationadjustedSUSENAS expendituresare level comparableover time.

-

20

-

6. Implicationsfor Future Poverty AlleviationProspects

Naturallywe cannot say anythingwith similar confidenceabout what
has happened to poverty in Indonesiasince early 1987 (the last available
SUSENAS survey),or what is likely to happen in the future. But some
aspects of the methodologywe have used here can throw light on the issue.
Our 1987 results also allow us to estimatethe elasticitiesof poverty to
any future distributionally
neutral growth in mean consumption,in a similar
manner to those we have used in the 1984-87 analyses. We find that the 1987
growth elasticitiesare also high; and, indeed,higher than those for 1984.
For example, our estimates are -4.1 for the 1987 elasticityof the poverty
gap measure with respect to mean consumptionusing the lower poverty line
(as compared to -2.9 for 1984 from Table 4); similarly,for our preferred
measure we obtain an elasticityof -4.8 (-3.4 for 1984). Furthermore,from
national accounts for 1987 and 1988 it appears that growth rates of real
consumptionper capita have increased. Both these factors - the higher
growth elasticitiesof poverty and the higher growth rates - imply
increasingpoverty alleviationthrough any distributionallyneutral growth
since early 1987.
Sustainablefuture consumptiongrowth in Indonesiawill clearly
require that investmentis revived,after the cuts of the mid 1980s. But it
is possible for the poverty alleviationbenefitsof even substantial
consumptiongrowth to be lost entirelythrough a sufficientcontemporaneous
deteriorationin overall equity. As we have seen, there was a modest
decreasein inequalityover 1984-1987. Is this likely to have abated since
then?

- 21 -

The sharp increasein rice prices during 1987 and 1988 (associated
with poor harvests and a reluctanceby the governmentto import rice) is
probably the main recent event which could mitigate continuedpoverty
alleviation. The effect on aggregatepoverty of higher rice prices is not
obvious. Ravallion and van de Walle (1988) found for Java in 1981 that even
if rice price increaseswere allowed to be passed on fully to the incomes of
the (typicallypoor) rice producers,all distributionallysensitivepoverty
measures would respond adversel'to higher rice prices. The effects are,
however, ambiguous for other mfasures of poverty which are not sensitiveto
the welfare of the poorest of the poor, such as the headcount index. Still,
the latter measure is known to be inadequate,and the preferredmeasures
favor a presumptionthat higher rice prices will (ceterisparibus) increase
poverty. It remains to be seen whether adverse effects on the poor of
changing relativeprices have been sufficientin size, or persistentenough,
to seriously jeopardizecontinuedpoverty alleviationthrough growth in
Indonesia.
A further issue beggingwhich has not been addr*ssedhere is
whether there exists a subset of the poor who will be unalie to escape
poverty in the future, solely by their participationin Inionesia's
seemingly robust aggregate economicgrowth. The results of this study
suggest that many of the poor have been able to do so in the 19809. But the
data studied here cannot tell us about any persistentlypoor personswho may
have benefittedlittle from growth, and are not so large in numbers or so
easily interviewedas to show up in sample averages;that would require
repeatedand painstakingobservationof a single sample. For this group,
direct policy interrentionmay remain their only hope.

- 22 -

7. Conclusions

The measurementof poverty at one point in time is fraughtwith
difficultv,and the comparisonof two points in time adds even further
problems. We have followedpast work on poverty in Indonesiaand elsewhere
in basing poverty measures on distributionsof household consumptionper
person. We have adjusted for inflationusing the ConsumerPrice Index,
though we have modified the underlyingexpenditureweights to accord more
closely to spendingpatterns of the poor. Recognizingthe uncertainties
involved in poverty measurement,an effort has been made to use a wider
range of both povertymeasures and poverty lines. Here we have drawn on
insightsfrom the recent literatureon poverty analysisusing dominance
conditionsto establishpartial orderingsof distributions.
From our comparisonsof income and consumptiondistributionsover
time we conclude that aggregatepoverty in Indonesiadecreased over the
period 1984 to 1987. This holds for both urban and rural areas. In a
neighborhoodof commonly assumed poverty lines for Indonesia,the magnitude
of the decline in poverty over the period is impressive;22 percent of the
population are identifiedas poor in 1987 using a consumptionpoverty line
for which 33 percent were poor in 1984. Such numbers can, however, be quite
sensitive to how poverty is measured, particularlythe choice of a poverty
line: for example, 46 percent are deemed to be poor in 1987 using a poverty
line for which 56 percent were poor in 1984. But, we find that the
aualitativeconclusionthat poverty declinedover the period holds for a
very broad class of poverty measures, and a wide range of poverty lines.
Indeed, it would hold even if one knew nothing about the poverty line, and
so allowed it to be anywherebetween the lowest and highest consumption

- 23 -

levelsl Underestimationin the rate of inflationover the period would
need, in our view, to be implausiblyhigh to alter the conclusionthat
poverty decreased.
A significant(though less dramatic)decline in poverty in both
urban and rural areas is also indicatedif one uses the share of total
consumptiondevoted to non-food goods as the indicatorof welfare, thus
avoiding the problems of adjustingfor inflation. The proportionof the
populationwho devoted three-quartersor more of their total consumptionto
food fell from 32 percent to 28 percent over the period. Furthermore,our
qualitativeconclusionsare robust to relaxingthe assumptionthat the
household level consumptiondata are even level comparableacross dates;
poverty is also found to have decreased (thoughagain by a less dramatic
margin than CPI adjustedconsumptionsindicate)if one bases the comparisons
of mean consumptionssolely on national accounts,only using the household
level data tapes to compare relativeconsumptionsat any one date.
Though the caloric intake data is less than ideal (with some
underestimationbeing likely at both dates), there is strong evidencethat
the extent of undernutritionalso fell significantly. This holds for both
urban and rural sectors,over a very wide range of alternativemeasures of
undernutritionand it holds for any (unknown)interpersonaldistributionof
nutritional requirements,providedthat this did not also change over the
period. The shift in the caloric intake distributionwas far from
negligibleat its lower end; we find, for example, that 27 percent of the
1987 populationdid not attain a caloric intake which 37 percent bad failed
to reach in 1984.
In summary,while legitimatedoubts can be raised about the
quantitativemagnitudes involved,the qualitativeconclusionis plain:

- 24

-

poverty and undernutritionin Indonesiacontinued to decline during the
difficult 1980s.
We have investigatedthe "proximatecauses" of Indonesia'ssuccess
using two methods of decomposition,one based on sectors,the other based on
distributionalparameters. The sectoraldecompositionof the change in
aggregate poverty indicatesthat gains to the rural sector were very
important,particularlyfor the poorest of the poor. Gains to the urban
sector and populationshifts between sectors (from rural to urban)
contributedto poverty alleviation,but were quantitativelyless important
than the direct gains to the rural poor. In terms of the parametersof the
aggregate consumptiondistributions,we found that both increases in average
real consumption (holdingrelativeinequalitiesconstant),and a modest
improvementin overall equity, contributedto poverty alleviationduring the
period. The former factor was quantitativelymore important,though less so
for the preferred (distributionally
sensitive)poverty measure.
A tadk for future researchis to understandmore deeply how this
success was achieved. Some possibleclues have been identifiedhere. The
gains to the rural farm sector were crucial and so policy adjustmentsaimed
at protectingthat sector probablyhad an importantrole. Certain
ingredientsof the macroeconomicpolicy responseprobablyhelped here, such
as the devaluations. Indonesia'srecent economichistory had a role too; by
the mid 1980s, past growth at relativelylow inequalityhad created a
situationwhereby relativelylarge effects on aggregatepoverty could be
generatedby seemingly small shifts in the distributionof consumption.
Thus, Indonesia'srecent econorichistory had created initial conditionsfor
the adjustmentwhich were quite favorableto maintainingthe country's
momentum in poverty alleviation,providedthat at least modest growth in
private consumptionper capita could be maintained.

- 25

-

Appendix 1: The Distributions

The HouseholdConsumptionand Caloric Intake Data

Past poverty assessmentsfor Indonesiahave generallybeen based on
the distributionsof householdconsumptionper person from the SUSENAS
expendituresurveysdone by the Central Bureau of Statistics. We follow
this method, though noting that it has both advantagesand disadvantages.
The followingpoints should be particularlynoted:
(i) The SUSENAS is widely recognizedas a householdconsumption
survey of relativelyhigh qualitywhich has used sound and
consistentsurvey methods over many years. The 1984 and 1987
surveysappear to be fully comparablein terms of the questions
asked, and the methods of samplingand interviewing. The
estimatesobtainedwill depend in part on the date of interview,
particularlyin rural areas where past work has found evidence
19
of seasonalityin consumption(as well as incomes).

The 1984

and 1987 SUSENAS surveys used here were done at approximately
the same time of year (Februaryand January respectively)in
comparableagriculturalyears,20 and so this is unlikely to be a
problem for comparisonsbetween those years.
(ii) The SUSENAShas the advantageover expendituresurveys fer a
rumber of developingcountriesthat it is a consumptionbased
survey i.e., it does not just ask how much was spent on variuus
goods and services,but also asks how much was actually consumed
from various sources,includingof course, market expenditure
but also includingconsumptionfrom own productionand transfers
(van de Walle, 1988).

-

26

-

(iii) Normalizinghousehold consumptionby the number of persons makes
no allowance for the likely variationin consumptionneeds
between differentpersons,accordingto (for example) age or
gender. There is a large literatureon the constructionof
equivalencescales which allow the normalizationto take account
of such differencesin household composition,though it remains
that there is no single ideal method of dealingwith this
21
issue.

Future work on measuringpoverty in Indonesiamight

fruitfullyaddress this issue, but for present purposeswe shall
follow past practice.

(iv) The use of consumptionrather than income is generallydesirable
in th's setting, since incomes fluctuatemore and householdscan
save. It can be argued that consumptionexpendituresfor a
single date will provide a better measure of a household's
standardof living than current income. However, it is also of
interestto examine the effects on incomes,as this is where
many of the immediateconsequencesof macroeconomicadjustment
are felt.22
(v) The SUSENAS aims to provide a random sample of the entire
population. However, it is probablydifficultto survey certain
types of households,and a disproportionate
number of these may
be poor; individualswithout fixed abode are an obvious example.
Though we do not have any evidence to support the view, it is
not implausiblethat the itineranturban poor in large cities
such as Jakarta are not well representedin the SUSENAS. If
there has been an increasein the numbers of such persons during
the adjustmentperiod then this will bias downward our

- 27

-

assessmentof poverty in 1987, relativeto 1984. The sampla was
stratifiedspatially. All our calculationhave used the local
expansion factors (inversesamplingrates) in aggregation.
(vi) The SUSENAS is the only availablesourc