00074918.2014.980377
Bulletin of Indonesian Economic Studies
ISSN: 0007-4918 (Print) 1472-7234 (Online) Journal homepage: http://www.tandfonline.com/loi/cbie20
Revisiting the Impact of Consumption Growth
and Inequality on Poverty in Indonesia during
Decentralisation
Riyana Miranti, Alan Duncan & Rebecca Cassells
To cite this article: Riyana Miranti, Alan Duncan & Rebecca Cassells (2014) Revisiting
the Impact of Consumption Growth and Inequality on Poverty in Indonesia during
Decentralisation, Bulletin of Indonesian Economic Studies, 50:3, 461-482, DOI:
10.1080/00074918.2014.980377
To link to this article: http://dx.doi.org/10.1080/00074918.2014.980377
Published online: 03 Dec 2014.
Submit your article to this journal
Article views: 454
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=cbie20
Download by: [Universitas Maritim Raja Ali Haji]
Date: 17 January 2016, At: 23:33
Bulletin of Indonesian Economic Studies, Vol. 50, No. 3, 2014: 461–82
REVISITING THE IMPACT OF CONSUMPTION
GROWTH AND INEQUALITY ON POVERTY IN
INDONESIA DURING DECENTRALISATION
Riyana Miranti*
Alan Duncan*
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
University of Canberra
Curtin University
Rebecca Cassells*
Curtin University
This article analyses the consumption growth elasticity and inequality elasticity of
poverty in Indonesia, with a particular focus on the decentralisation period. Using
provincial panel data, we show that the effectiveness of growth in alleviating poverty across provinces was greater during decentralisation—that is, between 2002
and 2010—than at any other point since 1984. The growth elasticity of poverty since
2002 is estimated to have been –2.46, which means that a 10% increase in average consumption per capita would have reduced the poverty rate by almost 25%.
However, we also ind that rising income inequality negated a quarter to a third of
the 5.7percentagepoint reduction in the headcount poverty rate. This increasing
inequality has contributed to a lower level of propoor growth than that maintained
in Indonesia before decentralisation.
Keywords: economic development, consumption growth, poverty, inequality, decentralisation
JEL classiication: D63, I30, O1, O4
INTRODUCTION
Shortly after the end of the New Order era in 1998, Indonesia entered a new
development phase in which policies and powers shifted from centralised to
decentralised governance. This process of decentralisation formally commenced
in 2001, marked by legislation that saw greater power given to municipal and
district governments. This legislation included Law 22/1999 on Regional Governance and Law 25/1999 on the Fiscal Balance between Central and Regional
* This article is based mainly on section 3 of the authors’ OECD working paper of 2013,
‘Trends in Poverty and Inequality in Decentralising Indonesia’. The authors thank Yogi
Vidyattama and Erick Hansnata, the other authors of that paper. They also thank Michael
Forster, Ana LlenaNozal, and other country delegates of the OECD for their funding, assistance, and feedback. Sonny Harmadi, Evi Nurvidya Ariin, Asep Suryahadi, and Jan
Priebe provided useful comments, as did the two anonymous referees. Those who gave
advice bear no responsibility for any errors or deiciencies.
ISSN 00074918 print/ISSN 14727234 online/14/00046122
http://dx.doi.org/10.1080/00074918.2014.980377
© 2014 Indonesia Project ANU
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
462
Riyana Miranti, Alan Duncan, and Rebecca Cassells
Governments, which are considered to be the foundations of a rapid process of
what has been called ‘big bang’ decentralisation (Hofman and Kaiser 2002). The
process faced both unsettled political conditions and a slow economic recovery;
Hill (2007) argues that the economy had only begun to recover by the beginning
of 2003. The two decentralisation laws were improved upon by Law 32/2004 and
Law 33/2004, which provided more clarity about the roles and responsibilities of
the different levels of government and interlinkages between central, provincial,
and district governments (Brodjonegoro 2009, Holtzappel 2009).
The dramatic changes in Indonesia’s political and economic environments over
the past decade, and the arguments that exist around the positive and negative outcomes of decentralisation, have highlighted the importance of examining movements in social and economic patterns since 2001—particularly trends in poverty
and inequality. In terms of poverty reduction efforts, decentralisation, although
not directly used as a sole strategy to alleviate poverty, is expected to improve
service delivery and provide better access to the poor by empowering credible
local governments that are well informed about the needs of their constituents.
Poverty alleviation strategies at the local level can be embedded into a number
of areas of responsibility that are associated with poverty—such as education, or
health support and welfare programs. Sumarto, Suryahadi, and Ariianto (2004)
argue the importance of civil society in decentralisation, in that it may create an
opportunity to closely monitor governance and thus give the poor a chance to be
heard, which will in turn be likely to facilitate more effective program targeting.
Decentralisation is also expected to promote higher economic growth and per
capita income, and therefore increase the potential to reduce poverty. Thornton
(2006) highlights several reasons that support this argument. First, local governments are in a better position to take account of local conditions when providing amenities and infrastructure. Second, competition among local governments
promotes incentives for investment, such as lowering investment tax rates. Third,
under revenue constraints local governments have an incentive to innovate the
production and supply of public goods and services for their communities.
Nevertheless, previous studies (such as Mahi 2010) have conjectured that
decentralisation in Indonesia has not improved household welfare signiicantly.
In addition, Hartono and Irawan (2008), for example, concluded that inequality
has not decreased, possibly because of a lack of policy coordination between central and local governments, with local government focusing on generating local
income rather than contributing to national programs of poverty alleviation.
This article is an extension of Miranti’s (2010) study, which examines the impact
of growth and change in inequality in Indonesia in 1984–2002. We have extended
the dataset used in the original study, ensuring that the time series data are consistent and comparable. While this article does not directly attempt to quantify
the impact of decentralisation on poverty and inequality, it examines in detail
what happened to both during this period and analyses the links between them.
MACROECONOMIC AND EMPLOYMENT INDICATORS
This section provides a background on the macroeconomic and employment indicators across development episodes in Indonesia, in order to compare the decentralisation period with preceding periods. Table 1 presents Indonesia’s economic
Revisiting the Impact of Consumption Growth and Inequality on Poverty
463
TABLE 1 Economic Growth by Development Episode, 1990–2010
(average % p. a.)
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
(1)
1990–96
GDP
GDP per capita
Manufacturing GDP
Agriculture GDP
Mining GDP
Services GDP
Mean consumption
per capita
(2)
1997–98
(3)
1999–2002
7.2
5.3
9.9
3.9
5.2
8.8
–13.1
–14.3
–11.4
–1.3
–2.8
–3.8
4.0
2.5
4.2
1.9
3.1
2.5
1.3
–17.0
3.3
(4)
2001–4
4.8
3.0
5.7
3.4
–1.6
4.5
(5)
2005–10
(6)
2001–10
5.7
4.1
3.9
3.7
2.4
6.3
5.4
3.7
4.6
3.5
1.1
5.6
2.4a
Sources: Authors’ calculations based on data from Miranti (2010) and CEIC Asia Database.
Note: Full column headings are as follows: (1) 1990–96 (prior to the Asian inancial crisis); (2) 1997–
98, (crisis period); (3) 1999–2002 (early recovery period); (4) 2001–4 (earlystage decentralisation); (5)
2005–10 (full implementation); and (6) 2001–10 (entire decentralisation period). The table shows average annual economic growth based on the compound rate and calculated using constant prices.
Data on mean consumption per capita cover 2002–10. Owing to the nature of Susenas consumption
data, which are available only every three years up to 2005, the growth of this indicator cannot be
broken down into early and full implementation periods.
a
growth since 1990 in several development episodes.1 Economic growth declined
by 13.1% during the crisis and then rebounded at 4.0% per year during the early
recovery period until 2002. This period overlapped with the irst stages of decentralisation, which we divide into two: (a) earlystage decentralisation (2001–4) and
(b) full implementation (2005–10). We deine full implementation as commencing after the laws implementing major funding reforms—Dana Alokasi Umum
(General Allocation Fund) and Dana Alokasi Khusus (Speciic Purpose Fund)—
took effect in 2004. This period also covers the 2008 global inancial crisis, during
which Indonesia’s economy fared relatively well. Its resilience was due in part
to the impact of increases in national spending, related to the 2009 presidential
election campaign that prompted increases in domestic demand. It is also related
to the fact that the ratio of Indonesia’s exports relative to the size of its economy
is small compared with those of neighbouring countries such as Singapore, Thailand, and Malaysia (Basri and Rahardja 2011).
The average rate of economic growth was lower during both early decentralisation and full implementation than it was before the crisis—particularly in the
manufacturing and services sectors, as relected in employment statistics (table 2).
Average employment growth in these sectors was lower during decentralisation
than before the crisis, with the manufacturing sector recording negative growth.
However, the share of employment in the manufacturing or the services sector of
total employment was higher than it was before the crisis. Employment elasticity
1. The data before 1990 can be seen in Miranti’s (2010) study.
464
Riyana Miranti, Alan Duncan, and Rebecca Cassells
TABLE 2 Structural Transformation of Employment, by Sector, 1990–2010
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
(1)
(2)
(3)
(4)
1990–96 1997–98 1999–2002 2001–4
Agriculture, forestry, & isheries
Employment
growth (% p. a.)
–1.9
6.4
Avg share of
employment (%)
49.6
41.9
Manufacturing
Employment
growth (% p. a.)
5.8
–12.9
Avg share of
employment (%)
11.6
11.9
Services
Employment
growth (% p. a.)
5.9
–0.4
Avg share of
employment (%)
34.1
39.8
(5)
(6)
(7)
2005–10 2001–10 Elasticity
1.9
0.7
–0.2
0.5
44.1
44.4
40.9
42.3
1.7
–2.9
3.5
1.5
13.1
12.6
12.4
12.5
–1.2
2.3
4.8
3.3
37.6
37.2
40.2
39.0
0.15
0.37
0.66
Sources: Authors’ calculations based on data from Miranti (2010) and CEIC Asia Database.
Note: See the note to table 1 for full headings for columns 1–6. Data on employment in mining, quarrying, electricity, gas and water, and construction are not presented. Employment elasticity is calculated
as the ratio of employment growth in a sector per year to the ratio of GDP in that particular sector per
year.
(the ratio of yearly employment growth to the ratio of yearly GDP growth) in the
services sector in this period was relatively high, at 0.66, relecting a growing sector, whereas it was lower in the manufacturing sector, at 0.37.2 Table 2 also shows
that growth in the agricultural sector increased on average by 0.5% per year from
the start of decentralisation to 2010—most likely a relection of the sector’s slow
expansion.
RECENT POVERTY AND INEQUALITY TRENDS
Recent studies of poverty in Indonesia—including those of Scherer and Scherer
(2011), Miranti (2010), Miranti et al. (2013), and Yusuf et al. (2014)—argue that
Indonesia’s performance in reducing poverty rates prior to the 1997–98 Asian
inancial crisis was impressive. Soeharto’s governments had adopted universal
policies that aimed to beneit most, including the poor and those in the rural sector (Huppi and Ravallion 1991).
The poverty rate increased during the crisis, reaching 24% in 1998, its highest
level since 1984 (igure 1). Poverty was still at a high 18% in 2001, when decentralisation commenced. Economic growth was slower during decentralisation
than before the crisis, while household survey data suggest that growth in
2. This employment elasticity was lower than it was during 1990–96 (0.58, as calculated in
Miranti 2007).
Revisiting the Impact of Consumption Growth and Inequality on Poverty
465
FIGURE 1 Trend in Poverty Rates, 1976–2010
(%)
45
ESD
FI
40
35
Total (r)
30
Rural (r)
Urban
20
15
Rural
10
Urban (r)
Total
5
10
08
20
06
20
04
20
02
20
00
20
20
98
96
19
94
19
92
19
90
19
88
19
86
19
19
84
82
19
80
19
78
19
19
76
0
19
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
25
Source: Data from Susenas, various years.
Note: ESD = earlystage decentralisation. FI = full implementation. (r) = revised. There is a break in the
series from 1996 because BPS revised its oficial poverty rates, due to changes in the methodology. The
oficial poverty rates calculate the proportion of the Indonesian population who live under the poverty line, which is deined as whether a person can fulil the cost of basic needs in terms of explicit food
items covering a 2,100calorie intake per day, represented by 52 commodities and basic nonfood items
covering clothing, housing, education, and health, represented by 51 commodities in urban areas and
47 commodities in rural areas. Calorie intake is estimated through household consumption patterns.
consumption was slower during 2002–10 than it was in the early recovery period.
Was the slower economic growth after 2002 still ‘propoor’, as Timmer (2004)
labelled Indonesia’s earlier growth experience?3
Figure 1 shows a decreasing trend in the national poverty rate between 2001
and 2005. The rate increased in 2006, in part because of the reduction in fuel subsidies in 2005 in conjunction with increases in the price of rice and other commodities. In 2010, the poverty rate was 13.3%, as a result of a lower reduction in
average poverty by 3.7% per year—slower than the 5% annual decrease during
1990–96). Figure 1 also demonstrates that poverty rates in rural areas have long
been higher than in urban areas. This gap was accentuated by the fact that most
nonagricultural employment was created in urban areas (Suryahadi et al. 2011).
Further, we ind that provincial poverty rates decreased between 2001 and 2010 in
line with the national trend (table 3), except in Aceh and DKI Jakarta.
While poverty rates decreased, consumption inequality increased, as Miranti
et al. (2013) and Yusuf et al. (2014) have discussed. Figure 2 shows trends in
3. The deinitions of propoor growth vary, covering both absolute and relative deinitions.
This article uses the absolute deinition, in which the poor beneit from the overall growth
of income in the economy,
466
Riyana Miranti, Alan Duncan, and Rebecca Cassells
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
TABLE 3 Annualised Change in Provincial Poverty Rates, 2001–10
(poverty headcount, %)
Province
2001
%
2010 change
Banten
Jambi
South Kalimantan
West Kalimantan
Bangka Belitung
East Kalimantan
Central Kalimantan
Bali
West Sumatra
North Maluku
Southeast Sulawesi
DI Yogyakarta
East Nusa Tenggara
East Java
West Nusa Tenggara
Central Sulawesi
17.2
19.7
11.9
19.2
13.3
14.0
11.7
7.9
15.2
14.0
25.2
24.5
33.0
21.6
30.4
25.3
7.2
8.3
5.2
9.0
6.5
7.7
6.8
4.9
9.5
9.4
17.1
16.8
23.0
15.3
21.6
18.1
–9.3
–9.1
–8.8
–8.1
–7.6
–6.5
–5.9
–5.2
–5.1
–4.3
–4.2
–4.1
–3.9
–3.8
–3.8
–3.7
Province
2001
%
2010 change
South Sulawesi
Indonesia
West Java
Central Java
Lampung
Gorontalo
Maluku
Bengkulu
Riau
North Sulawesi
Papua
South Sumatra
North Sumatra
Aceh
DKI Jakarta
16.5
18.4
15.3
22.1
24.9
29.7
34.8
21.7
10.1
10.7
41.8
16.1
11.7
19.2
3.1
11.9
13.3
11.3
16.6
18.9
23.2
27.7
18.3
8.5
9.1
36.4
15.5
11.3
21.0
3.5
–3.6
–3.5
–3.4
–3.1
–3.0
–2.7
–2.5
–1.9
–1.8
–1.8
–1.5
–0.4
–0.4
1.0
1.1
Source: Authors’ calculations based on data from Susenas, various years.
inequality, using household consumption data as the basis for calculation.4 Overall inequality increased by ive percentage points between 2002 and 2010.5 Figure
2 also shows that inequality is higher in urban areas, and closely aligned to overall
trends, while rural inequality is consistently lower by around nine percentage
points. This most likely relects the large increases in urban populations in recent
years (Mishra 2009): in 2010, 53% of Indonesia’s population was in urban areas
and this proportion is expected to reach 65% by 2025 (Bappenas 2011). Table 4
shows the change in inequality; all provinces in the table experienced an increase
in inequality between 2002 and 2010, while Gorontalo, a province established in
2000, experienced the most rapid increase.
THE IMPACT OF CONSUMPTION GROWTH AND INEQUALITY ON
POVERTY DURING DECENTRALISATION
This section explores the direction and strength of the associations between poverty,
inequality, and growth during 1984–2010. Miranti (2010) examined the impact of
changes in consumption growth and changes in inequality on headcount poverty
4. The consumptionbased Gini coeficient is an imperfect approximation of income inequality (Nugraha and Lewis 2013), and Susenas has the wellknown problem that it does
not accurately capture consumptive expenditure of the rich (Yusuf, Sumner, and Rum
2014).
5. Miranti et al. (2013) discussed the potential causes of this increasing inequality.
Revisiting the Impact of Consumption Growth and Inequality on Poverty
467
FIGURE 2 Gini Coeficients (Total, Urban, and Rural), 1996–2010
0.40
Urban
0.38
0.36
0.34
Total
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
0.32
0.30
Rural
0.28
0.26
0.24
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Sources: Data from Susenas, various years, for total inequality, and data from Yusuf, Sumner, and Rum
(2014), for urban and rural inequality.
Note: Based on consumption expenditure.
in Indonesia during 1984–2002.6 Pritchett (2011) compared the poverty elasticity
of growth, as the ratio of the percentage reduction in the poverty headcount rate,
with the percentage increase in GDP per capita during 1976–96 and 2000–8, but
without controls for provincial differences in poverty and growth.7 Using provincial data to expand Miranti’s work, we incrementally examine the consumption
growth elasticity of poverty (GEP) during a fourth development episode —decentralisation (2002–10)—again taking into account changes in inequality. To what
extent did the change in the degree of inequality offset the alleviating impact of
consumption growth on poverty? Was growth during this period propoor?
One of the key elements in the analysis is the high degree of heterogeneity of
economic circumstances across provinces in Indonesia (see, for example, Miranti
2011, and Hill and Vidyattama 2014). For an effective assessment of the under
lying impact of consumption growth and inequality on poverty within provinces,
it is essential to control for such local conditions. We do so by using econometric methods that exploit the longitudinal nature of provincial data on headcount
poverty derived from successive rounds of Indonesia’s National Socioeconomic
Survey (Survei Sosio Ekonomi Nasional [Susenas]).8
6. The second liberalisation period was characterised by slower and more cautious liberalisation than the irst (see Miranti 2010 for a more detailed discussion).
7. Pritchett (2011) inds an average elasticity of –1.15 during 1976–96 and a calculated GEP
of –0.70 for 2000–8.
8. See Priebe’s (2014) study for a review of the history of Susenas and oficial poverty measurement in Indonesia.
468
Riyana Miranti, Alan Duncan, and Rebecca Cassells
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
TABLE 4 Annualised Change in Gini Coeficients, 2002–10
Province
2002
%
2010 change
Gorontalo
Southeast Sulawesi
West Nusa Tenggara
Bengkulu
Lampung
North Sulawesi
South Sulawesi
Central Sulawesi
East Nusa Tenggara
Banten
South Kalimantan
West Java
Bali
West Sumatra
0.24
0.27
0.27
0.25
0.25
0.27
0.30
0.28
0.29
0.33
0.29
0.29
0.30
0.27
0.43
0.42
0.40
0.37
0.36
0.37
0.40
0.37
0.38
0.42
0.37
0.36
0.37
0.33
9.8
6.9
6.3
5.8
5.2
4.6
4.1
3.8
3.8
3.4
3.3
3.1
3.0
2.9
Province
2002
%
2010 change
West Kalimantan
Central Kalimantan
East Kalimantan
North Sumatera
BangkaBelitung
Central Java
South Sumatra
Indonesia
Jambi
Riau
DKI Jakarta
DI Yogyakarta
East Java
0.30
0.25
0.30
0.29
0.25
0.28
0.29
0.33
0.26
0.29
0.32
0.37
0.31
0.37
0.30
0.37
0.35
0.30
0.34
0.34
0.38
0.30
0.33
0.36
0.41
0.34
2.9
2.8
2.7
2.7
2.7
2.5
2.1
1.9
1.9
1.6
1.5
1.5
1.2
Source: Authors’ calculations based on data from Susenas, various years.
Note: Changes are calculated for those provinces for which data are available for both 2002 and 2010.
Data
We use Susenas data from 1984 to 2010 on provincial headcount poverty, monthly
mean consumption per capita, and provincial inequality. For provincial headcount poverty data before 1996, we draw on the poverty series used in Miranti’s
(2010) study. These allow for consistent comparisons with the revised poverty
rates published since 1996 by Badan Pusat Statistik (BPS), Indonesia’s central statistics agency, because Miranti (2010) used the BPS methodology of 2003 to re
estimate BPS poverty igures from 1984 to 1993.
We determined that the provincial level would be an appropriate geographical
unit for constructing a consistent paneldata source for empirical analysis. We use
11 rounds of Susenas consumption data to assemble the provincial panel used in
our estimations: every three years from 1984 to 1996, 2002, 2005, and then annually from 2007 to 2010. Our poverty igures and Gini coeficients are based on
Susenas consumption data. A provincial series is therefore only available every
third year from 1984 to 2005, using the Susenas consumption module, and annually from 2007, using the panel data from the Susenas modules.
In some years, Susenas data were not collected in provinces experiencing conlict, such as Aceh, Maluku, and Papua. This created a small number of missing
observations, leading to an unbalanced panel of provincial data. A second problem is the expansion of the number of provinces, from 26 in 2001 to 33 in 2003. For
consistency, in the empirical estimation, we reallocated the data for the new provinces back into the original provincial boundaries of 2001. We combined the data
for Bangka–Belitung with South Sumatra, the Riau Islands with Riau, Banten with
West Java, Gorontalo with North Sulawesi, West Sulawesi with South Sulawesi,
Maluku Utara with Maluku, and West Papua with Papua. The end result is a
workable dataset with 308 observations, covering 26 provinces.
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
Revisiting the Impact of Consumption Growth and Inequality on Poverty
469
We use mean per capita consumption (in expenditure terms) from Susenas as
a proxy of household income rather than per capita GDP data from national or
regional accounts. This follows previous literature in this ield (see Deaton 2001,
Ravallion and Chen 1997, Ravallion and Chen 2003, Adams 2004, and Miranti
2010) and is justiied for four reasons: (a) there is only a weak correlation between
provincial headcount poverty and economic growth in both the national and the
regional accounts; (b) it has yet to be determined whether increased average living standards translated into poverty reduction (a trickledown effect); (c) mean
consumption per capita is suggested to relect the welfare level more accurately
than income from the national accounts (Ravallion 1995), because of its effectiveness in capuring the life cycle or permanent income and is therefore suitable for
poverty analysis and (d) our regressions require consistent time series that complement those used by Miranti (2007, 2010).
Neither mean per capita consumption nor per capita GDP is free from measurement errors; both may underestimate or overestimate household income. Bhalla
(2002), for example, has argued that using the survey mean as a growth proxy
can greatly underestimate the GEP in developing countries. Adams (2004) found
the opposite. Ravallion (2001) tried to correct this problem by using the growth
rate from the national accounts as an instrumental variable for the growth rate in
the survey mean; but the growth rate from the national accounts may not be the
best instrument, since it may be correlated with the error terms in the regression.
We therefore deine growth as the percentage change in mean consumption per
capita. (For comparison, appendix table A1 contains the results using regional
GDP per capita.)
We use headcount poverty rates, or the proportion of poor people in the population, and Gini coeficients to represent, respectively, provincial poverty and
inequality (that is, as statistical measures of the dispersion of income distribution). Both are simpler to understand than other poverty and inequality measures,
and both variables are oficially published by BPS (except for poverty data before
1996) and calculated from Susenas. To allow for comparisons over time, we use
mean consumption (expenditure) per capita data in 1984 rupiah, using the ratio of
provincial poverty lines to the 1984 provincial poverty line as a delator.9
In line with the indings in the literature (such as Friedman 2001, 2005), and for
reasons discussed in Miranti’s (2010) study, consumption growth and inequality
act together to inluence the provincial headcount poverty rate, with prior expectations of a negative relationship between poverty and mean consumption and a
positive relationship between poverty and inequality (as measured using the Gini
coeficient). Simple scatterplots of the (bivariate) association between poverty and
either consumption (igures 3a and 3b) or inequality (igures 4a and 4b) for all
periods or in each of Indonesia’s development episodes provide indicative support for these relationships. Without a controlled variable, the igures also reveal
some variation in the strengths of such relationships over time—particularly for
inequality, for which the associations seem weaker during decentralisation.
Two caveats apply when seeking to draw conclusive inferences about the direction and strength of the associations between poverty, consumption growth, and
inequality using the simple representations in igures 3a, 3b, 4a, and 4b. First, it
9. See Miranti’s (2007) study for this conversion methodology.
Riyana Miranti, Alan Duncan, and Rebecca Cassells
470
FIGURE 3a Mean per Capita Consumption and Provincial Headcount Poverty Rates
1984–2010
Headcount poverty rate (%)
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
60
40
20
0
10
20
30
40
50
Mean per capita consumption (Rp ’000 per month)
60
Source: Authors’ calculations based on data from Susenas, 1984–2010.
Note: Locally smoothed regressions are generated using the Lowess method, with a bandwidth of 0.95.
is important that such effects are simultaneously controlled for when estimating
growth and inequality elasticities of poverty. Not to do so would lead to a bias in
the estimated effects—for example, ignoring the marginal impact of inequality
on poverty would force the growth elasticity to absorb this additional inluence.
Second, the apparent association between poverty and consumption is affected
to a large degree by persistent differences in poverty, consumption, and inequality among provinces. Papua, for example, consistently records a higher level of
poverty and a lower level of mean consumption than Jakarta. Not to control for
such differences could also lead to bias in the apparent impact of growth and
inequality on poverty within each province.
Empirical Methodology
We use two general models to estimate consumption growth and inequality elasticities of poverty. Both models are derived from the basic model suggested by
Ravallion and Chen (1997) and applied by Miranti (2010):
P
ln Pi,t = γ 0 + γ 1 ln MEAN i,t + γ 2 lnGINI i,t + ∑ β ep d p + δ i + ε i,t
(1a)
p=1
T
ln Pi,t = γ 0 + γ 1 ln MEAN i,t + γ 2 lnGINI i,t + ∑ βt dt + δ i + ε i,t
(1b)
t=1
P
P
P
p=1
p=1
p=1
ln Pi,t = γ 0 + ∑ γ 1 p e p ln MEAN i,t + ∑ γ 2 p e p lnGINI i,t + ∑ β ep e p + δ i + ε i,t
(2a)
Revisiting the Impact of Consumption Growth and Inequality on Poverty
471
FIGURE 3b Mean per Capita Consumption and Provincial Headcount Poverty Rates,
by Development Period, 1984–2010
Headcount poverty rate (%)
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
60
40
First liberalisation
(1984–90)
Second liberalisation
(1991–96)
Recovery
(1999–2002)
Decentralisation
(2002–10)
20
0
60
40
20
0
0
20
40
60
0
20
40
60
Mean per capita consumption (Rp ’000 per month)
Source: Authors’ calculations based on data from Susenas, 1984–2010.
Note: Locally smoothed regressions are generated using the Lowess method, with a bandwidth of 0.95.
P
P
T
p=1
p=1
t=1
ln Pi,t = γ 0 + ∑ γ 1 p e p ln MEAN i,t + ∑ γ 2 p e p lnGINI i,t + ∑ βt dt + δ i + ε i,t
(2b)
where Pi,t represents headcount poverty in province i at time t (%); MEANi,t represents mean consumption per capita (rupiah per month, in 1984 prices); GINIi,t
is the Gini coeficient of province i at time t; t is the year index (t = {1984, 1987,
1990, 1993, 1996, 1999, 2002, 2005, 2007, 2008, 2009, 2010}); and dt is a dummy variable for each Susenas year from 1984 to 2010 (for example, d1984 = 1 if t = 1984 and
0 otherwise). The variable ep represents dummies for four distinct development
episodes in Indonesia: the irst liberalisation period (1984–90); the second liberalisation period (1991–96); the early recovery period (1999–2002); and decentralisation (2002–10). The irst three periods are consistent with those used in Miranti’s
(2010) study: e1 = 1 if t = {1984, 1987, 1990} and 0 otherwise; e2 = 1 if t = {1993, 1996}
and 0 otherwise; e3 = 1 if t = {1999, 2002} and 0 otherwise; e4 = 1 if t = {2005, 2007,
2008, 2009, 2010} and 0 otherwise; d i is the province ixed effect (unobserved heterogeneity); and ei,t is a whitenoise error term that includes errors in the poverty
measure.
In each case, the relation between poverty, consumption growth, and inequality
takes a logarithmic form for both dependent and independent variables, so that
the coeficients of each of the core explanatory variables are presented directly as
elasticities.10 Models (1a) and (2a) include distinct development episodes as time
10. We acknowledge that the possibility of reverse causality runs from poverty rates to
growth of consumption per capita. We conclude, however, that the likelihood is small, since
Riyana Miranti, Alan Duncan, and Rebecca Cassells
472
FIGURE 4a Gini Coeficient and Provincial Headcount Poverty Rates,
1984–2010
Headcount poverty rate (%)
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
60
40
20
0
0.20
0.25
0.30
0.35
Gini coefficient
0.40
0.45
Source: Authors’ calculations based on data from Susenas, 1984–2010.
Note: Locally smoothed regressions are generated using the Lowess method, with a bandwidth of 0.95.
effects while models (1b) and (2b) include year dummies as the time effects. We
control for time ixed effects to capture macroeconomic conditions in each development episode or Susenas consumption module year.
Under this choice of speciication, the coeficients attached to the variables
involving lnMEAN refer to a 1% change in monthly mean consumption per capita, and the coeficients of the variable lnGINI refer to a 1% change in inequality. The irst speciications (1a and 1b) assume constant growth and inequality
elasticities of poverty across the whole period covered by the data, whereas the
second speciications (2a and 2b) allow for a different elasticity to be estimated in
each development episode.
We use ixed effects methods here to capture provincial differences in poverty
in different development episodes. Ravallion and Chen (1997) and Adams (2004)
used irstdifferences estimation in their analyses, to control for provincial heterogeneity. We prefer a twoway ixed effects approach to control simultaneously
for both provincial heterogeneity and systematic national trends in poverty over
time. We adopt two other assumptions in these ixed effect methods. First, we
assume that, in any province, random errors are usually thought to be serially
independent—that is, not correlated with each other over time (see Wooldridge
we use the same Susenas year as the source of poverty rates and growth of mean consumption per capita. Further, causality runs only one way, from mean consumption per capita to
the headcount poverty index, as in Ravallion and Datt’s (1996, 1999, 2002) series of papers;
Ravallion and Chen’s (1997) study; and Meng, Gregory, and Wang’s (2005) study. The other
possible source of endogeneity has already been solved by the ixed effects and the year
dummies (owing to the nature of panel data).
Revisiting the Impact of Consumption Growth and Inequality on Poverty
473
FIGURE 4b Gini Coeficient and Provincial Headcount Poverty Rates,
by Development Period, 1984–2010
Headcount poverty rate (%)
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
60
First liberalisation
(1984–90)
40
Second liberalisation
(1991–96)
20
0
60
Recovery
(1999–2002)
40
Decentralisation
(2002–10)
20
0
0.20
0.25
0.30
0.35
0.40
0.20
0.25
0.30
0.35
0.40
Gini coefficient
Source: Authors’ calculations based on data from Susenas, 1984–2010.
Note: Locally smoothed regressions are generated using the Lowess method, with a bandwidth of 0.95.
2003). Second, for comparison, we modify this assumption by allowing the random errors to be correlated within provinces across years but uncorrelated among
provinces (see Bertrand, Dulo, and Mullainathan 2004 and Hoechle 2007 for further discussion).11 For the second assumption, we apply the clustering method.
Empirical Results
Tables 5 and 6 provide a series of regression results for the range of speciications
nested in equations (1a), (1b), (2a), and (2b). The irst panel of results in table 5
restricts growth and inequality to constant levels over the full period of analysis,
whereas the second panel in table 6 provides separate elasticity estimates for each
of Indonesia’s four main development episodes since 1984.
Both sets of results demonstrate the importance of controlling for provincial
differences when estimating consumption growth and the inequality elasticity of
poverty (IEP). The irst two columns of table 5 report estimates of the (constant)
GEP, without controlling for provincial ixed effects. When inequality is ignored,
the GEP is estimated to be –1.34 (column 1). The additional control of inequality
(column 2) adjusts the GEP to –1.37 (which means that a 10% increase in average
consumption per capita will reduce the poverty rate by almost 14%). The additional estimated IEP in column 2 is 0.26, but this is insigniicant even at the 10%
level.
11. We also test for crosssectional dependence regardless of whether the residuals from a
ixedeffects estimation of regression model are spatially independent, following Hoechle
(2007). The test proves that the random errors are not correlated among provinces and
clusters.
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
TABLE 5 Growth Elasticity Regression Results, Constant across Development Periods
(1)
Explanatory variable
Consumption and inequality
ln(mean consumption)
ln(GINI)
Development period
EPISODE1 (irst liberalisation)
EPISODE2 (second liberalisation)
EPISODE3 (recovery)
EPISODE4 (decentralisation)
Constant
Provincial ixed effects
Year effects
Clustering
R2 / Adjusted R2
(2)
(3)
(4)
(5)
Coeficient t-statistic Coeficient t-statistic Coeficient t-statistic Coeficient t-statistic Coeficient t-statistic
–1.34***
—
16.18***
No
No
No
0.58
–20.62***
25.09***
–1.37***
0.26
15.62***
No
No
No
0.58
–20.24***
1.63
–2.30***
0.81***
–19.36***
5.60***
–2.28***
0.86***
1.85*
–2.24**
–3.34***
21.46***
0.08*
–0.08**
–0.14***
–
23.37***
—
—
—
—
23.05***
Yes
No
No
0.90
19.94***
Yes
Yes
No
0.91
–19.49***
6.16***
–2.28***
0.86***
–8.62***
3.13***
19.18***
—
—
—
—
22.55***
2.14***
Yes
Yes
Yes
0.82
Source: Authors’ calculations based on data from Susenas, 1984–2010.
Note: Observations = 308. For regressions that include provincial and time ixed effects, the reference province is Jakarta and the reference period is 2010.
* p < 0.1; ** p < 0.05; *** p < 0.01.
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
TABLE 6 Growth Elasticity Regression Results (Varying across Development Periods)
(1)
Explanatory variable
(2)
(4)
(5)
Coeficient t-statistic Coeficient t-statistic Coeficient t-statistic Coeficient t-statistic Coeficient t-statistic
Consumption and inequality
EPISODE1 × ln(mean cons)
EPISODE2 × ln(mean cons)
EPISODE3 × ln(mean cons)
EPISODE4 × ln(mean cons)
EPISODE1 × ln(GINI)
EPISODE2 × ln(GINI)
EPISODE3 × ln(GINI)
EPISODE4 × ln(GINI)
–0.88***
–1.38***
–1.38***
–1.37***
0.48
0.68
0.36
0.77***
–6.08***
–9.11***
–8.88***
–12.81***
1.54
1.37
0.74
2.94***
Development period
EPISODE1 (irst liberalisation)
EPISODE2 (second liberalisation)
EPISODE3 (recovery)
EPISODE4 (decentralisation)
Constant
—
3.95
4.91*
3.42
10.30***
1.53
1.84*
1.52
5.82***
Provincial ixed effects
Year effects
Clustering
R2 / Adjusted R2
Prob > F for different episodes:
ln(mean cons)
ln(GINI)
(3)
–2.08***
–2.31***
–2.34***
–2.50***
0.52***
0.54*
0.75***
1.13***
—
2.06
1.48
1.95
22.40
–15.42***
–17.87***
–17.51***
–19.77***
2.71***
1.88*
2.66***
6.57***
–2.00***
–2.33***
–2.29***
–2.46***
0.50***
0.93***
0.92***
1.13***
1.55
1.08
1.65*
15.56***
—
—
—
—
23.94***
–15.49***
–19.19***
–18.25***
–20.13***
2.82***
3.49***
3.49***
6.87***
–2.00***
–2.33***
–2.29***
–2.46***
0.50
0.93***
0.92***
1.13**
18.14***
—
—
—
—
23.42***
–5.77***
–10.10***
–8.98***
–9.45***
1.34
3.40***
2.87***
2.56**
10.03***
–1.99***
–2.31***
–2.27***
–2.45***
–5.75***
–10.58***
–9.36***
–9.84***
—
—
—
—
24.29***
11.19***
No
No
No
0.63
Yes
No
No
0.90
Yes
Yes
No
0.92
Yes
Yes
Yes
0.85
Yes
Yes
Yes
0.84
0.03
0.85
0.00
0.03
0.00
0.02
0.09
0.46
0.07
0.86
3.40
Source: Authors’ calculations based on data from Susenas, 1984–2010.
Note: Observations = 308. For regressions that include provincial and time ixed effects, the reference province is Jakarta and 2010 is the reference period. Unrestricted estimates allow both consumption and inequality parameters to vary across periods (columns 3 and 4), whereas restricted estimates refer to inequality
parameters that are ixed across periods between 1984 and 2002 (column 5).
* p < 0.1; ** p < 0.05; *** p < 0.01.
476
Riyana Miranti, Alan Duncan, and Rebecca Cassells
TABLE 7 Summary of the GEP and IEP, by Period, 1984–2010
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
Unrestricted model
First liberalisation (1984–90)
Second liberalisation (1991–96)
Recovery (1999–2002)
Decentralisation (2002–10)
All periods (average)
Restricted model
GEP
IEP
GEP
IEP
–2.00
–2.33
–2.29
–2.46
–2.28
0.50
0.93
0.92
1.13
0.86
–1.99
–2.31
–2.27
–2.45
–2.28
0.86
0.86
0.86
0.86
0.86
Source: Tables 5 and 6.
Note: Unrestricted estimates allow both consumption and inequality parameters to vary across periods (columns 3 and 4 of table 6) whereas restricted estimates refer to inequality parameters that are
ixed across periods between 1984 and 2002 (column 5 of table 6).
Table 5 also shows that including provincial ixed effects and time effects in
our estimations has two implications. First, the estimated GEP strengthens substantially, to –2.30, when controlling only for provincial ixed effects (column 3),
or to –2.28, with the addition of time effects (columns 4 and 5). Second, the impact
of inequality on poverty increases when we account for provincial differences.
The IEP strengthens to 0.81 (column 3) and 0.86 (columns 4 and 5), respectively,
and becomes statistically signiicant. This is an important result, and emphasises
that the positive impact of growth on poverty across Indonesian provinces can be
diluted by high levels of consumption inequality.
Table 6 gives separate estimates of the GEP and IEP across Indonesia’s four
development episodes since 1984. Column 1 reports a series of growth and inequality elasticities of poverty for each of Indonesia’s main development phases,
but with no controls for systematic provincial differences. Again, results are biased
downwards on this basis. Nevertheless, they align broadly with those of Pritchett
(2011) and show a rising impact of growth on poverty as Indonesia progressed
through each phase of development. The last four columns of table 6 provide the
most reliable estimates of the GEP and IEP, with respective controls for provincial
ixed effects (column 2) and both provincial and time ixed effects—in column 3,
without using the clustering method, and in columns 4 and 5, using the clustering
method.
Two key issues emerge from these results. First, the effectiveness of growth in
alleviating poverty across provinces was greater during decentralisation than at
any other point since 1984. The GEP since 2002 is estimated to have been around
–2.46, which means that a 10% increase in average consumption per capita reduced
the poverty rate by almost 25%. Second, in relation to the offsetting impact of inequality on provincial poverty over time, the results in column 3 of table 6 show a
rising inluence of inequality on provincial poverty over time, with the strength of
this effect peaking during decentralisation at an IEP of 1.13 (suggesting that a 10%
increase in inequality would have increased headcount poverty rates by more
than 11%). If we apply clustering methods to allow for correlation in the random
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
Revisiting the Impact of Consumption Growth and Inequality on Poverty
477
errors over time and within provinces (column 4 of table 6), the inequality effects
become less signiicant. The inal series of estimates (column 5) further restrict the
IEP to a constant level over time, returning an estimated (constant) effect of 0.86.
The effects of other explanatory factors not separately included in the empirical
speciications may be absorbed into year ixed effects and provincial ixed effects.
The explanatory variables that have not been included in the estimation include,
in particular, relevant government policies or interventions such as various targeted poverty alleviation programs.
For comparison, we present another measure of income growth: regional gross
domestic product (RGDP) per capita (see appendix table A1). The results show
that the impact of growth of RGDP on elasticities of poverty is smaller than those
that use the consumption data. This is in line with Adams’s (2004) study, which
inds a weaker statistical relationship between poverty reduction and the growth
of income measured by the national accounts. This may be a limitation of the
account data for poverty analysis, since the output produced by a region may not
necessarily be associated with the welfare of that particular region (see columns
3 and 4 of appendix table A1). This result supports Miranti’s (2013) study, which
examines the determinants of regional poverty in Indonesia during 2006–11 and
inds that the GEP during this period using the RGDP per capita as a proxy for
income growth is estimated to be low, at –0.28.12
Appendix table A1 also shows that although the GEP was negative and signiicant during decentralisation, it was lower than the elasticities during the second
liberalisation period and the recovery period. None of those inequality elasticities
of poverty is statistically signiicant.
Quantifying the Consumption Growth and Inequality Effects on Poverty
Table 8 shows the quantiied impacts of growth and changes in inequality effects
on poverty change, combining the period in Miranti’s (2010) study and the decentralisation period (2002–10). The quantiied impacts represent the contribution of
growth and changes in inequality to changes in the poverty rate. The magnitudes
of the results presented here differ from those of Miranti’s (2010) study, owing
to the additional data included in the analysis in this article and the improved
methodology—including using midpoint or average consumption or inequality
during the period we investigate rather than using consumption or inequality at
the beginning of the period.
Our indings indicate that changes in inequality (between 1.43 to 1.88 percentage points) offset the negative impact of growth of consumption on changes
in poverty (between 5.69 to 5.71 percentage points). Although economic growth
was propoor during decentralisation, the increasing degree of inequality over the
same period reduced its impact. It may well be worth exploring further whether
this outcome represents an adverse impact from decentralisation for districts
within a province, and, if so, what mechanisms caused such a rise in inequality.
12. Miranti’s (2013) study adopts a slightly different speciication, by including explanatory variables such as interprovincial migration, intergenerational transfers, human capital, and living conditions.
Riyana Miranti, Alan Duncan, and Rebecca Cassells
478
TABLE 8 Contribution of Consumption Growth and Inequality to
Change in Poverty, by Period (percentage points)
Contribution to poverty change
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
Growth
Inequality
change
Total poverty
change
Unrestricted model
First liberalisation (1984–90)
Second liberalisation (1991–96)
Recovery (1999–2002)
Decentralisation (2002–10)
All periods (average)
–3.54
0.54
–4.84
–5.71
–13.55
–0.61
0.79
0.99
1.88
3.05
–4.15
1.33
–3.85
–3.83
–10.50
Restricted model
First liberalisation (1984–90)
Second liberalisation (1991–96)
Recovery (1999–2002)
Decentralisation (2002–10)
All periods (average)
–3.53
0.53
–4.80
–5.69
–13.49
–1.05
0.73
0.93
1.43
2.04
–4.58
1.26
–3.87
–4.26
–11.45
Source: Authors’ calculations.
Note: Unrestricted estimates allow both consumption and inequality parameters to vary across periods
(columns 3 and 4 of table 6) whereas restricted estimates refer to inequality parameters that are ixed
across periods between 1984 and 2002 (column 5 of table 6).
CONCLUSIONS
This article uses 11 rounds of Susenas consumption modules to calculate the
impact of consumption growth and inequality on poverty, focusing on the decentralisation period and taking into account unobserved heterogeneity among provinces. The results show that the GEP during decentralisation was negative and
signiicant, which means that an increase in average living standards in terms of
consumption per capita went hand in hand with poverty reduction. In contrast,
the IEP was positive and signiicant, which suggests that increasing inequality was
associated with an increasing poverty rate. From the indings of the unrestricted
model, there were more pronounced effects of income inequality on regional poverty rates during later development episodes up to the post2002 decentralisation.
The propoor impact of economic growth, using mean consumption per capita as
a proxy of economic growth during decentralisation (a reduction of around 5.7
percentage points in the headcount poverty rate), was offset to a greater extent by
rising income inequality as measured by the Gini coeficient (up from 0.33 in 2002
to 0.38 in 2010). In combination, the stronger negative impact of rising inequality
contributed to an increase of between 1.4 to 1.9 percentage points in the headcount
poverty rate, offsetting the reduction in the poverty rate by a quarter to onethird.
The impact of different episodes on poverty differs over time; the quantiied
impact of consumption growth and inequality on poverty was the greatest during decentralisation. In addition, the results also suggest that changes in inequality offset some of the negative impacts of consumption growth on changes in
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
Revisiting the Impact of Consumption Growth and Inequality on Poverty
479
poverty. Although consumption growth was propoor during decentralisation—
as in other development episodes (the irst and second liberalisation period and
the recovery period)—the offsetting effects of changes in inequality hampered the
impact of consumption growth during 2002–10.
The fact that increases in inequality countered propoor growth may have some
relevant policy implications. Indonesia may need to have more
ISSN: 0007-4918 (Print) 1472-7234 (Online) Journal homepage: http://www.tandfonline.com/loi/cbie20
Revisiting the Impact of Consumption Growth
and Inequality on Poverty in Indonesia during
Decentralisation
Riyana Miranti, Alan Duncan & Rebecca Cassells
To cite this article: Riyana Miranti, Alan Duncan & Rebecca Cassells (2014) Revisiting
the Impact of Consumption Growth and Inequality on Poverty in Indonesia during
Decentralisation, Bulletin of Indonesian Economic Studies, 50:3, 461-482, DOI:
10.1080/00074918.2014.980377
To link to this article: http://dx.doi.org/10.1080/00074918.2014.980377
Published online: 03 Dec 2014.
Submit your article to this journal
Article views: 454
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=cbie20
Download by: [Universitas Maritim Raja Ali Haji]
Date: 17 January 2016, At: 23:33
Bulletin of Indonesian Economic Studies, Vol. 50, No. 3, 2014: 461–82
REVISITING THE IMPACT OF CONSUMPTION
GROWTH AND INEQUALITY ON POVERTY IN
INDONESIA DURING DECENTRALISATION
Riyana Miranti*
Alan Duncan*
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
University of Canberra
Curtin University
Rebecca Cassells*
Curtin University
This article analyses the consumption growth elasticity and inequality elasticity of
poverty in Indonesia, with a particular focus on the decentralisation period. Using
provincial panel data, we show that the effectiveness of growth in alleviating poverty across provinces was greater during decentralisation—that is, between 2002
and 2010—than at any other point since 1984. The growth elasticity of poverty since
2002 is estimated to have been –2.46, which means that a 10% increase in average consumption per capita would have reduced the poverty rate by almost 25%.
However, we also ind that rising income inequality negated a quarter to a third of
the 5.7percentagepoint reduction in the headcount poverty rate. This increasing
inequality has contributed to a lower level of propoor growth than that maintained
in Indonesia before decentralisation.
Keywords: economic development, consumption growth, poverty, inequality, decentralisation
JEL classiication: D63, I30, O1, O4
INTRODUCTION
Shortly after the end of the New Order era in 1998, Indonesia entered a new
development phase in which policies and powers shifted from centralised to
decentralised governance. This process of decentralisation formally commenced
in 2001, marked by legislation that saw greater power given to municipal and
district governments. This legislation included Law 22/1999 on Regional Governance and Law 25/1999 on the Fiscal Balance between Central and Regional
* This article is based mainly on section 3 of the authors’ OECD working paper of 2013,
‘Trends in Poverty and Inequality in Decentralising Indonesia’. The authors thank Yogi
Vidyattama and Erick Hansnata, the other authors of that paper. They also thank Michael
Forster, Ana LlenaNozal, and other country delegates of the OECD for their funding, assistance, and feedback. Sonny Harmadi, Evi Nurvidya Ariin, Asep Suryahadi, and Jan
Priebe provided useful comments, as did the two anonymous referees. Those who gave
advice bear no responsibility for any errors or deiciencies.
ISSN 00074918 print/ISSN 14727234 online/14/00046122
http://dx.doi.org/10.1080/00074918.2014.980377
© 2014 Indonesia Project ANU
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
462
Riyana Miranti, Alan Duncan, and Rebecca Cassells
Governments, which are considered to be the foundations of a rapid process of
what has been called ‘big bang’ decentralisation (Hofman and Kaiser 2002). The
process faced both unsettled political conditions and a slow economic recovery;
Hill (2007) argues that the economy had only begun to recover by the beginning
of 2003. The two decentralisation laws were improved upon by Law 32/2004 and
Law 33/2004, which provided more clarity about the roles and responsibilities of
the different levels of government and interlinkages between central, provincial,
and district governments (Brodjonegoro 2009, Holtzappel 2009).
The dramatic changes in Indonesia’s political and economic environments over
the past decade, and the arguments that exist around the positive and negative outcomes of decentralisation, have highlighted the importance of examining movements in social and economic patterns since 2001—particularly trends in poverty
and inequality. In terms of poverty reduction efforts, decentralisation, although
not directly used as a sole strategy to alleviate poverty, is expected to improve
service delivery and provide better access to the poor by empowering credible
local governments that are well informed about the needs of their constituents.
Poverty alleviation strategies at the local level can be embedded into a number
of areas of responsibility that are associated with poverty—such as education, or
health support and welfare programs. Sumarto, Suryahadi, and Ariianto (2004)
argue the importance of civil society in decentralisation, in that it may create an
opportunity to closely monitor governance and thus give the poor a chance to be
heard, which will in turn be likely to facilitate more effective program targeting.
Decentralisation is also expected to promote higher economic growth and per
capita income, and therefore increase the potential to reduce poverty. Thornton
(2006) highlights several reasons that support this argument. First, local governments are in a better position to take account of local conditions when providing amenities and infrastructure. Second, competition among local governments
promotes incentives for investment, such as lowering investment tax rates. Third,
under revenue constraints local governments have an incentive to innovate the
production and supply of public goods and services for their communities.
Nevertheless, previous studies (such as Mahi 2010) have conjectured that
decentralisation in Indonesia has not improved household welfare signiicantly.
In addition, Hartono and Irawan (2008), for example, concluded that inequality
has not decreased, possibly because of a lack of policy coordination between central and local governments, with local government focusing on generating local
income rather than contributing to national programs of poverty alleviation.
This article is an extension of Miranti’s (2010) study, which examines the impact
of growth and change in inequality in Indonesia in 1984–2002. We have extended
the dataset used in the original study, ensuring that the time series data are consistent and comparable. While this article does not directly attempt to quantify
the impact of decentralisation on poverty and inequality, it examines in detail
what happened to both during this period and analyses the links between them.
MACROECONOMIC AND EMPLOYMENT INDICATORS
This section provides a background on the macroeconomic and employment indicators across development episodes in Indonesia, in order to compare the decentralisation period with preceding periods. Table 1 presents Indonesia’s economic
Revisiting the Impact of Consumption Growth and Inequality on Poverty
463
TABLE 1 Economic Growth by Development Episode, 1990–2010
(average % p. a.)
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
(1)
1990–96
GDP
GDP per capita
Manufacturing GDP
Agriculture GDP
Mining GDP
Services GDP
Mean consumption
per capita
(2)
1997–98
(3)
1999–2002
7.2
5.3
9.9
3.9
5.2
8.8
–13.1
–14.3
–11.4
–1.3
–2.8
–3.8
4.0
2.5
4.2
1.9
3.1
2.5
1.3
–17.0
3.3
(4)
2001–4
4.8
3.0
5.7
3.4
–1.6
4.5
(5)
2005–10
(6)
2001–10
5.7
4.1
3.9
3.7
2.4
6.3
5.4
3.7
4.6
3.5
1.1
5.6
2.4a
Sources: Authors’ calculations based on data from Miranti (2010) and CEIC Asia Database.
Note: Full column headings are as follows: (1) 1990–96 (prior to the Asian inancial crisis); (2) 1997–
98, (crisis period); (3) 1999–2002 (early recovery period); (4) 2001–4 (earlystage decentralisation); (5)
2005–10 (full implementation); and (6) 2001–10 (entire decentralisation period). The table shows average annual economic growth based on the compound rate and calculated using constant prices.
Data on mean consumption per capita cover 2002–10. Owing to the nature of Susenas consumption
data, which are available only every three years up to 2005, the growth of this indicator cannot be
broken down into early and full implementation periods.
a
growth since 1990 in several development episodes.1 Economic growth declined
by 13.1% during the crisis and then rebounded at 4.0% per year during the early
recovery period until 2002. This period overlapped with the irst stages of decentralisation, which we divide into two: (a) earlystage decentralisation (2001–4) and
(b) full implementation (2005–10). We deine full implementation as commencing after the laws implementing major funding reforms—Dana Alokasi Umum
(General Allocation Fund) and Dana Alokasi Khusus (Speciic Purpose Fund)—
took effect in 2004. This period also covers the 2008 global inancial crisis, during
which Indonesia’s economy fared relatively well. Its resilience was due in part
to the impact of increases in national spending, related to the 2009 presidential
election campaign that prompted increases in domestic demand. It is also related
to the fact that the ratio of Indonesia’s exports relative to the size of its economy
is small compared with those of neighbouring countries such as Singapore, Thailand, and Malaysia (Basri and Rahardja 2011).
The average rate of economic growth was lower during both early decentralisation and full implementation than it was before the crisis—particularly in the
manufacturing and services sectors, as relected in employment statistics (table 2).
Average employment growth in these sectors was lower during decentralisation
than before the crisis, with the manufacturing sector recording negative growth.
However, the share of employment in the manufacturing or the services sector of
total employment was higher than it was before the crisis. Employment elasticity
1. The data before 1990 can be seen in Miranti’s (2010) study.
464
Riyana Miranti, Alan Duncan, and Rebecca Cassells
TABLE 2 Structural Transformation of Employment, by Sector, 1990–2010
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
(1)
(2)
(3)
(4)
1990–96 1997–98 1999–2002 2001–4
Agriculture, forestry, & isheries
Employment
growth (% p. a.)
–1.9
6.4
Avg share of
employment (%)
49.6
41.9
Manufacturing
Employment
growth (% p. a.)
5.8
–12.9
Avg share of
employment (%)
11.6
11.9
Services
Employment
growth (% p. a.)
5.9
–0.4
Avg share of
employment (%)
34.1
39.8
(5)
(6)
(7)
2005–10 2001–10 Elasticity
1.9
0.7
–0.2
0.5
44.1
44.4
40.9
42.3
1.7
–2.9
3.5
1.5
13.1
12.6
12.4
12.5
–1.2
2.3
4.8
3.3
37.6
37.2
40.2
39.0
0.15
0.37
0.66
Sources: Authors’ calculations based on data from Miranti (2010) and CEIC Asia Database.
Note: See the note to table 1 for full headings for columns 1–6. Data on employment in mining, quarrying, electricity, gas and water, and construction are not presented. Employment elasticity is calculated
as the ratio of employment growth in a sector per year to the ratio of GDP in that particular sector per
year.
(the ratio of yearly employment growth to the ratio of yearly GDP growth) in the
services sector in this period was relatively high, at 0.66, relecting a growing sector, whereas it was lower in the manufacturing sector, at 0.37.2 Table 2 also shows
that growth in the agricultural sector increased on average by 0.5% per year from
the start of decentralisation to 2010—most likely a relection of the sector’s slow
expansion.
RECENT POVERTY AND INEQUALITY TRENDS
Recent studies of poverty in Indonesia—including those of Scherer and Scherer
(2011), Miranti (2010), Miranti et al. (2013), and Yusuf et al. (2014)—argue that
Indonesia’s performance in reducing poverty rates prior to the 1997–98 Asian
inancial crisis was impressive. Soeharto’s governments had adopted universal
policies that aimed to beneit most, including the poor and those in the rural sector (Huppi and Ravallion 1991).
The poverty rate increased during the crisis, reaching 24% in 1998, its highest
level since 1984 (igure 1). Poverty was still at a high 18% in 2001, when decentralisation commenced. Economic growth was slower during decentralisation
than before the crisis, while household survey data suggest that growth in
2. This employment elasticity was lower than it was during 1990–96 (0.58, as calculated in
Miranti 2007).
Revisiting the Impact of Consumption Growth and Inequality on Poverty
465
FIGURE 1 Trend in Poverty Rates, 1976–2010
(%)
45
ESD
FI
40
35
Total (r)
30
Rural (r)
Urban
20
15
Rural
10
Urban (r)
Total
5
10
08
20
06
20
04
20
02
20
00
20
20
98
96
19
94
19
92
19
90
19
88
19
86
19
19
84
82
19
80
19
78
19
19
76
0
19
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
25
Source: Data from Susenas, various years.
Note: ESD = earlystage decentralisation. FI = full implementation. (r) = revised. There is a break in the
series from 1996 because BPS revised its oficial poverty rates, due to changes in the methodology. The
oficial poverty rates calculate the proportion of the Indonesian population who live under the poverty line, which is deined as whether a person can fulil the cost of basic needs in terms of explicit food
items covering a 2,100calorie intake per day, represented by 52 commodities and basic nonfood items
covering clothing, housing, education, and health, represented by 51 commodities in urban areas and
47 commodities in rural areas. Calorie intake is estimated through household consumption patterns.
consumption was slower during 2002–10 than it was in the early recovery period.
Was the slower economic growth after 2002 still ‘propoor’, as Timmer (2004)
labelled Indonesia’s earlier growth experience?3
Figure 1 shows a decreasing trend in the national poverty rate between 2001
and 2005. The rate increased in 2006, in part because of the reduction in fuel subsidies in 2005 in conjunction with increases in the price of rice and other commodities. In 2010, the poverty rate was 13.3%, as a result of a lower reduction in
average poverty by 3.7% per year—slower than the 5% annual decrease during
1990–96). Figure 1 also demonstrates that poverty rates in rural areas have long
been higher than in urban areas. This gap was accentuated by the fact that most
nonagricultural employment was created in urban areas (Suryahadi et al. 2011).
Further, we ind that provincial poverty rates decreased between 2001 and 2010 in
line with the national trend (table 3), except in Aceh and DKI Jakarta.
While poverty rates decreased, consumption inequality increased, as Miranti
et al. (2013) and Yusuf et al. (2014) have discussed. Figure 2 shows trends in
3. The deinitions of propoor growth vary, covering both absolute and relative deinitions.
This article uses the absolute deinition, in which the poor beneit from the overall growth
of income in the economy,
466
Riyana Miranti, Alan Duncan, and Rebecca Cassells
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
TABLE 3 Annualised Change in Provincial Poverty Rates, 2001–10
(poverty headcount, %)
Province
2001
%
2010 change
Banten
Jambi
South Kalimantan
West Kalimantan
Bangka Belitung
East Kalimantan
Central Kalimantan
Bali
West Sumatra
North Maluku
Southeast Sulawesi
DI Yogyakarta
East Nusa Tenggara
East Java
West Nusa Tenggara
Central Sulawesi
17.2
19.7
11.9
19.2
13.3
14.0
11.7
7.9
15.2
14.0
25.2
24.5
33.0
21.6
30.4
25.3
7.2
8.3
5.2
9.0
6.5
7.7
6.8
4.9
9.5
9.4
17.1
16.8
23.0
15.3
21.6
18.1
–9.3
–9.1
–8.8
–8.1
–7.6
–6.5
–5.9
–5.2
–5.1
–4.3
–4.2
–4.1
–3.9
–3.8
–3.8
–3.7
Province
2001
%
2010 change
South Sulawesi
Indonesia
West Java
Central Java
Lampung
Gorontalo
Maluku
Bengkulu
Riau
North Sulawesi
Papua
South Sumatra
North Sumatra
Aceh
DKI Jakarta
16.5
18.4
15.3
22.1
24.9
29.7
34.8
21.7
10.1
10.7
41.8
16.1
11.7
19.2
3.1
11.9
13.3
11.3
16.6
18.9
23.2
27.7
18.3
8.5
9.1
36.4
15.5
11.3
21.0
3.5
–3.6
–3.5
–3.4
–3.1
–3.0
–2.7
–2.5
–1.9
–1.8
–1.8
–1.5
–0.4
–0.4
1.0
1.1
Source: Authors’ calculations based on data from Susenas, various years.
inequality, using household consumption data as the basis for calculation.4 Overall inequality increased by ive percentage points between 2002 and 2010.5 Figure
2 also shows that inequality is higher in urban areas, and closely aligned to overall
trends, while rural inequality is consistently lower by around nine percentage
points. This most likely relects the large increases in urban populations in recent
years (Mishra 2009): in 2010, 53% of Indonesia’s population was in urban areas
and this proportion is expected to reach 65% by 2025 (Bappenas 2011). Table 4
shows the change in inequality; all provinces in the table experienced an increase
in inequality between 2002 and 2010, while Gorontalo, a province established in
2000, experienced the most rapid increase.
THE IMPACT OF CONSUMPTION GROWTH AND INEQUALITY ON
POVERTY DURING DECENTRALISATION
This section explores the direction and strength of the associations between poverty,
inequality, and growth during 1984–2010. Miranti (2010) examined the impact of
changes in consumption growth and changes in inequality on headcount poverty
4. The consumptionbased Gini coeficient is an imperfect approximation of income inequality (Nugraha and Lewis 2013), and Susenas has the wellknown problem that it does
not accurately capture consumptive expenditure of the rich (Yusuf, Sumner, and Rum
2014).
5. Miranti et al. (2013) discussed the potential causes of this increasing inequality.
Revisiting the Impact of Consumption Growth and Inequality on Poverty
467
FIGURE 2 Gini Coeficients (Total, Urban, and Rural), 1996–2010
0.40
Urban
0.38
0.36
0.34
Total
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
0.32
0.30
Rural
0.28
0.26
0.24
1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Sources: Data from Susenas, various years, for total inequality, and data from Yusuf, Sumner, and Rum
(2014), for urban and rural inequality.
Note: Based on consumption expenditure.
in Indonesia during 1984–2002.6 Pritchett (2011) compared the poverty elasticity
of growth, as the ratio of the percentage reduction in the poverty headcount rate,
with the percentage increase in GDP per capita during 1976–96 and 2000–8, but
without controls for provincial differences in poverty and growth.7 Using provincial data to expand Miranti’s work, we incrementally examine the consumption
growth elasticity of poverty (GEP) during a fourth development episode —decentralisation (2002–10)—again taking into account changes in inequality. To what
extent did the change in the degree of inequality offset the alleviating impact of
consumption growth on poverty? Was growth during this period propoor?
One of the key elements in the analysis is the high degree of heterogeneity of
economic circumstances across provinces in Indonesia (see, for example, Miranti
2011, and Hill and Vidyattama 2014). For an effective assessment of the under
lying impact of consumption growth and inequality on poverty within provinces,
it is essential to control for such local conditions. We do so by using econometric methods that exploit the longitudinal nature of provincial data on headcount
poverty derived from successive rounds of Indonesia’s National Socioeconomic
Survey (Survei Sosio Ekonomi Nasional [Susenas]).8
6. The second liberalisation period was characterised by slower and more cautious liberalisation than the irst (see Miranti 2010 for a more detailed discussion).
7. Pritchett (2011) inds an average elasticity of –1.15 during 1976–96 and a calculated GEP
of –0.70 for 2000–8.
8. See Priebe’s (2014) study for a review of the history of Susenas and oficial poverty measurement in Indonesia.
468
Riyana Miranti, Alan Duncan, and Rebecca Cassells
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
TABLE 4 Annualised Change in Gini Coeficients, 2002–10
Province
2002
%
2010 change
Gorontalo
Southeast Sulawesi
West Nusa Tenggara
Bengkulu
Lampung
North Sulawesi
South Sulawesi
Central Sulawesi
East Nusa Tenggara
Banten
South Kalimantan
West Java
Bali
West Sumatra
0.24
0.27
0.27
0.25
0.25
0.27
0.30
0.28
0.29
0.33
0.29
0.29
0.30
0.27
0.43
0.42
0.40
0.37
0.36
0.37
0.40
0.37
0.38
0.42
0.37
0.36
0.37
0.33
9.8
6.9
6.3
5.8
5.2
4.6
4.1
3.8
3.8
3.4
3.3
3.1
3.0
2.9
Province
2002
%
2010 change
West Kalimantan
Central Kalimantan
East Kalimantan
North Sumatera
BangkaBelitung
Central Java
South Sumatra
Indonesia
Jambi
Riau
DKI Jakarta
DI Yogyakarta
East Java
0.30
0.25
0.30
0.29
0.25
0.28
0.29
0.33
0.26
0.29
0.32
0.37
0.31
0.37
0.30
0.37
0.35
0.30
0.34
0.34
0.38
0.30
0.33
0.36
0.41
0.34
2.9
2.8
2.7
2.7
2.7
2.5
2.1
1.9
1.9
1.6
1.5
1.5
1.2
Source: Authors’ calculations based on data from Susenas, various years.
Note: Changes are calculated for those provinces for which data are available for both 2002 and 2010.
Data
We use Susenas data from 1984 to 2010 on provincial headcount poverty, monthly
mean consumption per capita, and provincial inequality. For provincial headcount poverty data before 1996, we draw on the poverty series used in Miranti’s
(2010) study. These allow for consistent comparisons with the revised poverty
rates published since 1996 by Badan Pusat Statistik (BPS), Indonesia’s central statistics agency, because Miranti (2010) used the BPS methodology of 2003 to re
estimate BPS poverty igures from 1984 to 1993.
We determined that the provincial level would be an appropriate geographical
unit for constructing a consistent paneldata source for empirical analysis. We use
11 rounds of Susenas consumption data to assemble the provincial panel used in
our estimations: every three years from 1984 to 1996, 2002, 2005, and then annually from 2007 to 2010. Our poverty igures and Gini coeficients are based on
Susenas consumption data. A provincial series is therefore only available every
third year from 1984 to 2005, using the Susenas consumption module, and annually from 2007, using the panel data from the Susenas modules.
In some years, Susenas data were not collected in provinces experiencing conlict, such as Aceh, Maluku, and Papua. This created a small number of missing
observations, leading to an unbalanced panel of provincial data. A second problem is the expansion of the number of provinces, from 26 in 2001 to 33 in 2003. For
consistency, in the empirical estimation, we reallocated the data for the new provinces back into the original provincial boundaries of 2001. We combined the data
for Bangka–Belitung with South Sumatra, the Riau Islands with Riau, Banten with
West Java, Gorontalo with North Sulawesi, West Sulawesi with South Sulawesi,
Maluku Utara with Maluku, and West Papua with Papua. The end result is a
workable dataset with 308 observations, covering 26 provinces.
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
Revisiting the Impact of Consumption Growth and Inequality on Poverty
469
We use mean per capita consumption (in expenditure terms) from Susenas as
a proxy of household income rather than per capita GDP data from national or
regional accounts. This follows previous literature in this ield (see Deaton 2001,
Ravallion and Chen 1997, Ravallion and Chen 2003, Adams 2004, and Miranti
2010) and is justiied for four reasons: (a) there is only a weak correlation between
provincial headcount poverty and economic growth in both the national and the
regional accounts; (b) it has yet to be determined whether increased average living standards translated into poverty reduction (a trickledown effect); (c) mean
consumption per capita is suggested to relect the welfare level more accurately
than income from the national accounts (Ravallion 1995), because of its effectiveness in capuring the life cycle or permanent income and is therefore suitable for
poverty analysis and (d) our regressions require consistent time series that complement those used by Miranti (2007, 2010).
Neither mean per capita consumption nor per capita GDP is free from measurement errors; both may underestimate or overestimate household income. Bhalla
(2002), for example, has argued that using the survey mean as a growth proxy
can greatly underestimate the GEP in developing countries. Adams (2004) found
the opposite. Ravallion (2001) tried to correct this problem by using the growth
rate from the national accounts as an instrumental variable for the growth rate in
the survey mean; but the growth rate from the national accounts may not be the
best instrument, since it may be correlated with the error terms in the regression.
We therefore deine growth as the percentage change in mean consumption per
capita. (For comparison, appendix table A1 contains the results using regional
GDP per capita.)
We use headcount poverty rates, or the proportion of poor people in the population, and Gini coeficients to represent, respectively, provincial poverty and
inequality (that is, as statistical measures of the dispersion of income distribution). Both are simpler to understand than other poverty and inequality measures,
and both variables are oficially published by BPS (except for poverty data before
1996) and calculated from Susenas. To allow for comparisons over time, we use
mean consumption (expenditure) per capita data in 1984 rupiah, using the ratio of
provincial poverty lines to the 1984 provincial poverty line as a delator.9
In line with the indings in the literature (such as Friedman 2001, 2005), and for
reasons discussed in Miranti’s (2010) study, consumption growth and inequality
act together to inluence the provincial headcount poverty rate, with prior expectations of a negative relationship between poverty and mean consumption and a
positive relationship between poverty and inequality (as measured using the Gini
coeficient). Simple scatterplots of the (bivariate) association between poverty and
either consumption (igures 3a and 3b) or inequality (igures 4a and 4b) for all
periods or in each of Indonesia’s development episodes provide indicative support for these relationships. Without a controlled variable, the igures also reveal
some variation in the strengths of such relationships over time—particularly for
inequality, for which the associations seem weaker during decentralisation.
Two caveats apply when seeking to draw conclusive inferences about the direction and strength of the associations between poverty, consumption growth, and
inequality using the simple representations in igures 3a, 3b, 4a, and 4b. First, it
9. See Miranti’s (2007) study for this conversion methodology.
Riyana Miranti, Alan Duncan, and Rebecca Cassells
470
FIGURE 3a Mean per Capita Consumption and Provincial Headcount Poverty Rates
1984–2010
Headcount poverty rate (%)
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
60
40
20
0
10
20
30
40
50
Mean per capita consumption (Rp ’000 per month)
60
Source: Authors’ calculations based on data from Susenas, 1984–2010.
Note: Locally smoothed regressions are generated using the Lowess method, with a bandwidth of 0.95.
is important that such effects are simultaneously controlled for when estimating
growth and inequality elasticities of poverty. Not to do so would lead to a bias in
the estimated effects—for example, ignoring the marginal impact of inequality
on poverty would force the growth elasticity to absorb this additional inluence.
Second, the apparent association between poverty and consumption is affected
to a large degree by persistent differences in poverty, consumption, and inequality among provinces. Papua, for example, consistently records a higher level of
poverty and a lower level of mean consumption than Jakarta. Not to control for
such differences could also lead to bias in the apparent impact of growth and
inequality on poverty within each province.
Empirical Methodology
We use two general models to estimate consumption growth and inequality elasticities of poverty. Both models are derived from the basic model suggested by
Ravallion and Chen (1997) and applied by Miranti (2010):
P
ln Pi,t = γ 0 + γ 1 ln MEAN i,t + γ 2 lnGINI i,t + ∑ β ep d p + δ i + ε i,t
(1a)
p=1
T
ln Pi,t = γ 0 + γ 1 ln MEAN i,t + γ 2 lnGINI i,t + ∑ βt dt + δ i + ε i,t
(1b)
t=1
P
P
P
p=1
p=1
p=1
ln Pi,t = γ 0 + ∑ γ 1 p e p ln MEAN i,t + ∑ γ 2 p e p lnGINI i,t + ∑ β ep e p + δ i + ε i,t
(2a)
Revisiting the Impact of Consumption Growth and Inequality on Poverty
471
FIGURE 3b Mean per Capita Consumption and Provincial Headcount Poverty Rates,
by Development Period, 1984–2010
Headcount poverty rate (%)
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
60
40
First liberalisation
(1984–90)
Second liberalisation
(1991–96)
Recovery
(1999–2002)
Decentralisation
(2002–10)
20
0
60
40
20
0
0
20
40
60
0
20
40
60
Mean per capita consumption (Rp ’000 per month)
Source: Authors’ calculations based on data from Susenas, 1984–2010.
Note: Locally smoothed regressions are generated using the Lowess method, with a bandwidth of 0.95.
P
P
T
p=1
p=1
t=1
ln Pi,t = γ 0 + ∑ γ 1 p e p ln MEAN i,t + ∑ γ 2 p e p lnGINI i,t + ∑ βt dt + δ i + ε i,t
(2b)
where Pi,t represents headcount poverty in province i at time t (%); MEANi,t represents mean consumption per capita (rupiah per month, in 1984 prices); GINIi,t
is the Gini coeficient of province i at time t; t is the year index (t = {1984, 1987,
1990, 1993, 1996, 1999, 2002, 2005, 2007, 2008, 2009, 2010}); and dt is a dummy variable for each Susenas year from 1984 to 2010 (for example, d1984 = 1 if t = 1984 and
0 otherwise). The variable ep represents dummies for four distinct development
episodes in Indonesia: the irst liberalisation period (1984–90); the second liberalisation period (1991–96); the early recovery period (1999–2002); and decentralisation (2002–10). The irst three periods are consistent with those used in Miranti’s
(2010) study: e1 = 1 if t = {1984, 1987, 1990} and 0 otherwise; e2 = 1 if t = {1993, 1996}
and 0 otherwise; e3 = 1 if t = {1999, 2002} and 0 otherwise; e4 = 1 if t = {2005, 2007,
2008, 2009, 2010} and 0 otherwise; d i is the province ixed effect (unobserved heterogeneity); and ei,t is a whitenoise error term that includes errors in the poverty
measure.
In each case, the relation between poverty, consumption growth, and inequality
takes a logarithmic form for both dependent and independent variables, so that
the coeficients of each of the core explanatory variables are presented directly as
elasticities.10 Models (1a) and (2a) include distinct development episodes as time
10. We acknowledge that the possibility of reverse causality runs from poverty rates to
growth of consumption per capita. We conclude, however, that the likelihood is small, since
Riyana Miranti, Alan Duncan, and Rebecca Cassells
472
FIGURE 4a Gini Coeficient and Provincial Headcount Poverty Rates,
1984–2010
Headcount poverty rate (%)
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
60
40
20
0
0.20
0.25
0.30
0.35
Gini coefficient
0.40
0.45
Source: Authors’ calculations based on data from Susenas, 1984–2010.
Note: Locally smoothed regressions are generated using the Lowess method, with a bandwidth of 0.95.
effects while models (1b) and (2b) include year dummies as the time effects. We
control for time ixed effects to capture macroeconomic conditions in each development episode or Susenas consumption module year.
Under this choice of speciication, the coeficients attached to the variables
involving lnMEAN refer to a 1% change in monthly mean consumption per capita, and the coeficients of the variable lnGINI refer to a 1% change in inequality. The irst speciications (1a and 1b) assume constant growth and inequality
elasticities of poverty across the whole period covered by the data, whereas the
second speciications (2a and 2b) allow for a different elasticity to be estimated in
each development episode.
We use ixed effects methods here to capture provincial differences in poverty
in different development episodes. Ravallion and Chen (1997) and Adams (2004)
used irstdifferences estimation in their analyses, to control for provincial heterogeneity. We prefer a twoway ixed effects approach to control simultaneously
for both provincial heterogeneity and systematic national trends in poverty over
time. We adopt two other assumptions in these ixed effect methods. First, we
assume that, in any province, random errors are usually thought to be serially
independent—that is, not correlated with each other over time (see Wooldridge
we use the same Susenas year as the source of poverty rates and growth of mean consumption per capita. Further, causality runs only one way, from mean consumption per capita to
the headcount poverty index, as in Ravallion and Datt’s (1996, 1999, 2002) series of papers;
Ravallion and Chen’s (1997) study; and Meng, Gregory, and Wang’s (2005) study. The other
possible source of endogeneity has already been solved by the ixed effects and the year
dummies (owing to the nature of panel data).
Revisiting the Impact of Consumption Growth and Inequality on Poverty
473
FIGURE 4b Gini Coeficient and Provincial Headcount Poverty Rates,
by Development Period, 1984–2010
Headcount poverty rate (%)
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
60
First liberalisation
(1984–90)
40
Second liberalisation
(1991–96)
20
0
60
Recovery
(1999–2002)
40
Decentralisation
(2002–10)
20
0
0.20
0.25
0.30
0.35
0.40
0.20
0.25
0.30
0.35
0.40
Gini coefficient
Source: Authors’ calculations based on data from Susenas, 1984–2010.
Note: Locally smoothed regressions are generated using the Lowess method, with a bandwidth of 0.95.
2003). Second, for comparison, we modify this assumption by allowing the random errors to be correlated within provinces across years but uncorrelated among
provinces (see Bertrand, Dulo, and Mullainathan 2004 and Hoechle 2007 for further discussion).11 For the second assumption, we apply the clustering method.
Empirical Results
Tables 5 and 6 provide a series of regression results for the range of speciications
nested in equations (1a), (1b), (2a), and (2b). The irst panel of results in table 5
restricts growth and inequality to constant levels over the full period of analysis,
whereas the second panel in table 6 provides separate elasticity estimates for each
of Indonesia’s four main development episodes since 1984.
Both sets of results demonstrate the importance of controlling for provincial
differences when estimating consumption growth and the inequality elasticity of
poverty (IEP). The irst two columns of table 5 report estimates of the (constant)
GEP, without controlling for provincial ixed effects. When inequality is ignored,
the GEP is estimated to be –1.34 (column 1). The additional control of inequality
(column 2) adjusts the GEP to –1.37 (which means that a 10% increase in average
consumption per capita will reduce the poverty rate by almost 14%). The additional estimated IEP in column 2 is 0.26, but this is insigniicant even at the 10%
level.
11. We also test for crosssectional dependence regardless of whether the residuals from a
ixedeffects estimation of regression model are spatially independent, following Hoechle
(2007). The test proves that the random errors are not correlated among provinces and
clusters.
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
TABLE 5 Growth Elasticity Regression Results, Constant across Development Periods
(1)
Explanatory variable
Consumption and inequality
ln(mean consumption)
ln(GINI)
Development period
EPISODE1 (irst liberalisation)
EPISODE2 (second liberalisation)
EPISODE3 (recovery)
EPISODE4 (decentralisation)
Constant
Provincial ixed effects
Year effects
Clustering
R2 / Adjusted R2
(2)
(3)
(4)
(5)
Coeficient t-statistic Coeficient t-statistic Coeficient t-statistic Coeficient t-statistic Coeficient t-statistic
–1.34***
—
16.18***
No
No
No
0.58
–20.62***
25.09***
–1.37***
0.26
15.62***
No
No
No
0.58
–20.24***
1.63
–2.30***
0.81***
–19.36***
5.60***
–2.28***
0.86***
1.85*
–2.24**
–3.34***
21.46***
0.08*
–0.08**
–0.14***
–
23.37***
—
—
—
—
23.05***
Yes
No
No
0.90
19.94***
Yes
Yes
No
0.91
–19.49***
6.16***
–2.28***
0.86***
–8.62***
3.13***
19.18***
—
—
—
—
22.55***
2.14***
Yes
Yes
Yes
0.82
Source: Authors’ calculations based on data from Susenas, 1984–2010.
Note: Observations = 308. For regressions that include provincial and time ixed effects, the reference province is Jakarta and the reference period is 2010.
* p < 0.1; ** p < 0.05; *** p < 0.01.
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
TABLE 6 Growth Elasticity Regression Results (Varying across Development Periods)
(1)
Explanatory variable
(2)
(4)
(5)
Coeficient t-statistic Coeficient t-statistic Coeficient t-statistic Coeficient t-statistic Coeficient t-statistic
Consumption and inequality
EPISODE1 × ln(mean cons)
EPISODE2 × ln(mean cons)
EPISODE3 × ln(mean cons)
EPISODE4 × ln(mean cons)
EPISODE1 × ln(GINI)
EPISODE2 × ln(GINI)
EPISODE3 × ln(GINI)
EPISODE4 × ln(GINI)
–0.88***
–1.38***
–1.38***
–1.37***
0.48
0.68
0.36
0.77***
–6.08***
–9.11***
–8.88***
–12.81***
1.54
1.37
0.74
2.94***
Development period
EPISODE1 (irst liberalisation)
EPISODE2 (second liberalisation)
EPISODE3 (recovery)
EPISODE4 (decentralisation)
Constant
—
3.95
4.91*
3.42
10.30***
1.53
1.84*
1.52
5.82***
Provincial ixed effects
Year effects
Clustering
R2 / Adjusted R2
Prob > F for different episodes:
ln(mean cons)
ln(GINI)
(3)
–2.08***
–2.31***
–2.34***
–2.50***
0.52***
0.54*
0.75***
1.13***
—
2.06
1.48
1.95
22.40
–15.42***
–17.87***
–17.51***
–19.77***
2.71***
1.88*
2.66***
6.57***
–2.00***
–2.33***
–2.29***
–2.46***
0.50***
0.93***
0.92***
1.13***
1.55
1.08
1.65*
15.56***
—
—
—
—
23.94***
–15.49***
–19.19***
–18.25***
–20.13***
2.82***
3.49***
3.49***
6.87***
–2.00***
–2.33***
–2.29***
–2.46***
0.50
0.93***
0.92***
1.13**
18.14***
—
—
—
—
23.42***
–5.77***
–10.10***
–8.98***
–9.45***
1.34
3.40***
2.87***
2.56**
10.03***
–1.99***
–2.31***
–2.27***
–2.45***
–5.75***
–10.58***
–9.36***
–9.84***
—
—
—
—
24.29***
11.19***
No
No
No
0.63
Yes
No
No
0.90
Yes
Yes
No
0.92
Yes
Yes
Yes
0.85
Yes
Yes
Yes
0.84
0.03
0.85
0.00
0.03
0.00
0.02
0.09
0.46
0.07
0.86
3.40
Source: Authors’ calculations based on data from Susenas, 1984–2010.
Note: Observations = 308. For regressions that include provincial and time ixed effects, the reference province is Jakarta and 2010 is the reference period. Unrestricted estimates allow both consumption and inequality parameters to vary across periods (columns 3 and 4), whereas restricted estimates refer to inequality
parameters that are ixed across periods between 1984 and 2002 (column 5).
* p < 0.1; ** p < 0.05; *** p < 0.01.
476
Riyana Miranti, Alan Duncan, and Rebecca Cassells
TABLE 7 Summary of the GEP and IEP, by Period, 1984–2010
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
Unrestricted model
First liberalisation (1984–90)
Second liberalisation (1991–96)
Recovery (1999–2002)
Decentralisation (2002–10)
All periods (average)
Restricted model
GEP
IEP
GEP
IEP
–2.00
–2.33
–2.29
–2.46
–2.28
0.50
0.93
0.92
1.13
0.86
–1.99
–2.31
–2.27
–2.45
–2.28
0.86
0.86
0.86
0.86
0.86
Source: Tables 5 and 6.
Note: Unrestricted estimates allow both consumption and inequality parameters to vary across periods (columns 3 and 4 of table 6) whereas restricted estimates refer to inequality parameters that are
ixed across periods between 1984 and 2002 (column 5 of table 6).
Table 5 also shows that including provincial ixed effects and time effects in
our estimations has two implications. First, the estimated GEP strengthens substantially, to –2.30, when controlling only for provincial ixed effects (column 3),
or to –2.28, with the addition of time effects (columns 4 and 5). Second, the impact
of inequality on poverty increases when we account for provincial differences.
The IEP strengthens to 0.81 (column 3) and 0.86 (columns 4 and 5), respectively,
and becomes statistically signiicant. This is an important result, and emphasises
that the positive impact of growth on poverty across Indonesian provinces can be
diluted by high levels of consumption inequality.
Table 6 gives separate estimates of the GEP and IEP across Indonesia’s four
development episodes since 1984. Column 1 reports a series of growth and inequality elasticities of poverty for each of Indonesia’s main development phases,
but with no controls for systematic provincial differences. Again, results are biased
downwards on this basis. Nevertheless, they align broadly with those of Pritchett
(2011) and show a rising impact of growth on poverty as Indonesia progressed
through each phase of development. The last four columns of table 6 provide the
most reliable estimates of the GEP and IEP, with respective controls for provincial
ixed effects (column 2) and both provincial and time ixed effects—in column 3,
without using the clustering method, and in columns 4 and 5, using the clustering
method.
Two key issues emerge from these results. First, the effectiveness of growth in
alleviating poverty across provinces was greater during decentralisation than at
any other point since 1984. The GEP since 2002 is estimated to have been around
–2.46, which means that a 10% increase in average consumption per capita reduced
the poverty rate by almost 25%. Second, in relation to the offsetting impact of inequality on provincial poverty over time, the results in column 3 of table 6 show a
rising inluence of inequality on provincial poverty over time, with the strength of
this effect peaking during decentralisation at an IEP of 1.13 (suggesting that a 10%
increase in inequality would have increased headcount poverty rates by more
than 11%). If we apply clustering methods to allow for correlation in the random
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
Revisiting the Impact of Consumption Growth and Inequality on Poverty
477
errors over time and within provinces (column 4 of table 6), the inequality effects
become less signiicant. The inal series of estimates (column 5) further restrict the
IEP to a constant level over time, returning an estimated (constant) effect of 0.86.
The effects of other explanatory factors not separately included in the empirical
speciications may be absorbed into year ixed effects and provincial ixed effects.
The explanatory variables that have not been included in the estimation include,
in particular, relevant government policies or interventions such as various targeted poverty alleviation programs.
For comparison, we present another measure of income growth: regional gross
domestic product (RGDP) per capita (see appendix table A1). The results show
that the impact of growth of RGDP on elasticities of poverty is smaller than those
that use the consumption data. This is in line with Adams’s (2004) study, which
inds a weaker statistical relationship between poverty reduction and the growth
of income measured by the national accounts. This may be a limitation of the
account data for poverty analysis, since the output produced by a region may not
necessarily be associated with the welfare of that particular region (see columns
3 and 4 of appendix table A1). This result supports Miranti’s (2013) study, which
examines the determinants of regional poverty in Indonesia during 2006–11 and
inds that the GEP during this period using the RGDP per capita as a proxy for
income growth is estimated to be low, at –0.28.12
Appendix table A1 also shows that although the GEP was negative and signiicant during decentralisation, it was lower than the elasticities during the second
liberalisation period and the recovery period. None of those inequality elasticities
of poverty is statistically signiicant.
Quantifying the Consumption Growth and Inequality Effects on Poverty
Table 8 shows the quantiied impacts of growth and changes in inequality effects
on poverty change, combining the period in Miranti’s (2010) study and the decentralisation period (2002–10). The quantiied impacts represent the contribution of
growth and changes in inequality to changes in the poverty rate. The magnitudes
of the results presented here differ from those of Miranti’s (2010) study, owing
to the additional data included in the analysis in this article and the improved
methodology—including using midpoint or average consumption or inequality
during the period we investigate rather than using consumption or inequality at
the beginning of the period.
Our indings indicate that changes in inequality (between 1.43 to 1.88 percentage points) offset the negative impact of growth of consumption on changes
in poverty (between 5.69 to 5.71 percentage points). Although economic growth
was propoor during decentralisation, the increasing degree of inequality over the
same period reduced its impact. It may well be worth exploring further whether
this outcome represents an adverse impact from decentralisation for districts
within a province, and, if so, what mechanisms caused such a rise in inequality.
12. Miranti’s (2013) study adopts a slightly different speciication, by including explanatory variables such as interprovincial migration, intergenerational transfers, human capital, and living conditions.
Riyana Miranti, Alan Duncan, and Rebecca Cassells
478
TABLE 8 Contribution of Consumption Growth and Inequality to
Change in Poverty, by Period (percentage points)
Contribution to poverty change
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
Growth
Inequality
change
Total poverty
change
Unrestricted model
First liberalisation (1984–90)
Second liberalisation (1991–96)
Recovery (1999–2002)
Decentralisation (2002–10)
All periods (average)
–3.54
0.54
–4.84
–5.71
–13.55
–0.61
0.79
0.99
1.88
3.05
–4.15
1.33
–3.85
–3.83
–10.50
Restricted model
First liberalisation (1984–90)
Second liberalisation (1991–96)
Recovery (1999–2002)
Decentralisation (2002–10)
All periods (average)
–3.53
0.53
–4.80
–5.69
–13.49
–1.05
0.73
0.93
1.43
2.04
–4.58
1.26
–3.87
–4.26
–11.45
Source: Authors’ calculations.
Note: Unrestricted estimates allow both consumption and inequality parameters to vary across periods
(columns 3 and 4 of table 6) whereas restricted estimates refer to inequality parameters that are ixed
across periods between 1984 and 2002 (column 5 of table 6).
CONCLUSIONS
This article uses 11 rounds of Susenas consumption modules to calculate the
impact of consumption growth and inequality on poverty, focusing on the decentralisation period and taking into account unobserved heterogeneity among provinces. The results show that the GEP during decentralisation was negative and
signiicant, which means that an increase in average living standards in terms of
consumption per capita went hand in hand with poverty reduction. In contrast,
the IEP was positive and signiicant, which suggests that increasing inequality was
associated with an increasing poverty rate. From the indings of the unrestricted
model, there were more pronounced effects of income inequality on regional poverty rates during later development episodes up to the post2002 decentralisation.
The propoor impact of economic growth, using mean consumption per capita as
a proxy of economic growth during decentralisation (a reduction of around 5.7
percentage points in the headcount poverty rate), was offset to a greater extent by
rising income inequality as measured by the Gini coeficient (up from 0.33 in 2002
to 0.38 in 2010). In combination, the stronger negative impact of rising inequality
contributed to an increase of between 1.4 to 1.9 percentage points in the headcount
poverty rate, offsetting the reduction in the poverty rate by a quarter to onethird.
The impact of different episodes on poverty differs over time; the quantiied
impact of consumption growth and inequality on poverty was the greatest during decentralisation. In addition, the results also suggest that changes in inequality offset some of the negative impacts of consumption growth on changes in
Downloaded by [Universitas Maritim Raja Ali Haji] at 23:33 17 January 2016
Revisiting the Impact of Consumption Growth and Inequality on Poverty
479
poverty. Although consumption growth was propoor during decentralisation—
as in other development episodes (the irst and second liberalisation period and
the recovery period)—the offsetting effects of changes in inequality hampered the
impact of consumption growth during 2002–10.
The fact that increases in inequality countered propoor growth may have some
relevant policy implications. Indonesia may need to have more