Faktor-Faktor Yang Mempengaruhi Konsumsi Beras Di Indonesia.

DETERMINANTS OF RICE CONSUMPTION IN INDONESIA

CAHYA NAJMUDINROHMAN

GRADUATE SCHOOL
BOGOR AGRICULTURAL UNIVERSITY
BOGOR
2015

STATEMENT OF THESIS, SOURCE OF INFORMATION
AND COPYRIGHT*
I hereby declare that master thesis entitled “Determinants of Rice
Consumption in Indonesia” is my work under the direction of the advisory
committee and has not been submitted in any form to any other universities.
Sources of information derived or quoted from works published and unpublished
by other authors have been mentioned in the text and listed in the References at
the end of this master thesis.
I hereby assign the copyright of my master thesis to the Bogor Agricultural
University.
Bogor, September 2015
Cahya Najmudirohman

H451110581

* Copyright transfer due to the collaborative research work with other parties outside the Bogor
Agricultural University should be based on a related agreement.

SUMMARY
CAHYA NAJMUDINROHMAN. Determinants of Rice Consumption in
Indonesia. Supervised by HARIANTO and MUHAMMAD FIRDAUS.
Rice consumption in Indonesia is relatively high. In the supply side, both
domestic production and import of rice face challenges such as susceptible to
climate change and price instability. Coupled with potential benefit of more
diverse and balanced diet, rice consumption needs to be reduced. Thus the drivers
of the rice consumption need to be investigated to find potential channel to work
with. Using SUSENAS quarter I data, rice consumption model is estimated.
The result shows that rice remains normal good for low and middle
income group. It becomes inferior good for high income group, but the impact
decreases. The first two quartile have higher magnitude of own price elasticity
than the last two. It implies that any price spike has bigger impact on lower
income group, threatens their calorie intake due to lower rice consumption. The
result also shows that rice has limited substitutes, and their magnitudes of cross

price elasticities are relatively small. Female headed household consume less rice,
thus a campaign targeted to them to reduce rice consumption and promote more
diverse and balanced diet would be a viable option. Spatial factor also plays role
in the quantity of rice consumed, but it might be eroded by income increase.
Keywords: cross price elasticity, expenditure elasticity, own price elasticity, rice
consumption, SUSENAS

RINGKASAN
CAHYA NAJMUDINROHMAN. Faktor-Faktor yang Mempengaruhi Konsumsi
Beras di Indonesia. Dibimbing oleh HARIANTO dan MUHAMMAD FIRDAUS.
Konsumsi beras di Indonesia relatif tinggi. Di sisi penawaran, baik
produksi dalam negeri dan impor beras menghadapi tantangan seperti rentan
terhadap perubahan iklim dan ketidakstabilan harga. Dengan mempertimbangkan
potensi manfaat dari diet yang lebih beragam dan seimbang, konsumsi beras perlu
untuk dikurangi. Oleh karena itu, hal-hal yang mempengaruhi konsumsi beras
perlu diteliti guna menginvetigasi saluran yang potensial. Model konsumsi beras
di estimasi dengan menggunakan data SUSENAS 2013 kuartal I.
Metode estimasi yang digunakan ialah metode rata-rata terkecil dengan
mempertimbangkan hal-hal yang berpotensi menghasilkan estimasi yang bias.
Oleh karena itu, nilai-nilai pencilan dikeluarkan karena metode rata-rata terkecil

sangat sensitif terhadap pencilan. Selain itu, penimbang rumah tangga digunakan
untuk memperoleh simpangan baku yang tegar terhadap heteroskedastisitas yang
tidak terspesifikasikan yang ditemukan dalam proses estimasi.
Hasil penelitian menunjukkan bahwa beras termasuk barang normal bagi
kelompok berpenghasilan rendah dan menengah. Pada dua kuartil pertama
pengeluaran, elastisitas harga beras memiliki nilai mutlak yang lebih tinggi
daripada dua kuartil berikutnya. Hal ini menggambarkan bahwa lonjakan harga
memberikan dampak yang besar pada kelompok berpenghasilan rendah, dimana
akan mengacam asupan kalori akibat konsumsi beras yang menurun. Namun beras
menjadi barang inferior bagi kelompok berpenghasilan tinggi. Oleh karena itu,
masyarakat berpenghasilan menengah ke atas berpotensi untuk semakin
mengurangi konsumsi beras dan meningkatkan konsumsi makanan bernilai tinggi
seperti sayur, buah, dan produk hewani. Hal tersebut akan membantu pencapaian
kemandirian pangan melalui sisi permintaan beras.
Hasil penelitian juga menunjukkan bahwa beras memiliki barang substitusi
yang terbatas dengan besaran elastisitas harga silang yang relatif kecil. Rumah
tangga dengan kepala rumah tangga berjenis kelamin wanita mengkonsumsi beras
lebih sedikit, sehingga dapat dijadikan target kampanye dalam rangka mengurangi
konsumsi beras dan mempromosikan diet yang lebih beragam dan seimbang.
Faktor spasial juga memiliki peran dalam kuantitas beras yang dikonsumsi,

namun berpotensi untuk tergerus oleh kenaikan pendapatan.
Kata kunci:

elastisitas harga silang, elastisitas pengeluaran, elastisitas harga
sendiri, konsumsi beras, SUSENAS

© All Rights Reserved by Bogor Agricultural University, 2015
Copyright Reserved by Law
It is prohibited to quote part or all of this paper without including or citing the
source. Quotations are only for purposes of education, research, scientific
writing, preparation of reports, critics, or review an issue; and those are not
detrimental to the interests of the Bogor Agricultural University.
It is prohibited to announce and reproduce part or all of this paper in any form
without the permission of the Bogor Agricultural University.

DETERMINANTS OF RICE CONSUMPTION IN INDONESIA

CAHYA NAJMUDINROHMAN

Master thesis

as one of the requirements to obtain a degree of
Magister Sains
in an Agribusiness Study Program

GRADUATE SCHOOL
BOGOR AGRICULTURAL UNIVERSITY
BOGOR
2015

External Examiner:
Dr. Amzul Rifin SP. MA
Study Program Representative Examiner:
Dr. Ir. Burhanuddin, MM

Thesis Title
Name
Student ID

: Determinants of Rice Consumption in Indonesia
: Cahya Najmudinrohman

: H451110301

Approved by
Advisory Committee

Dr. Ir. Harianto, MS
Chairman

Prof. Dr. Muhammad Firdaus, SP, M.Si
Member

Agreed by
Head of Agribusiness Study Program

Prof. Dr. Ir. Rita Nurmalina, MS.

Examination Date: September 16, 2015

Dean of Graduate School


Dr. Ir. Dahrul Syah, MScAgr

Submission Date:

ACKNOWLEDGMENT
Praise to Allah the Almighty for all His blessings. I feel grateful to finally
finish my thesis with all the efforts and supports I received from all parties:
1. Dr. Ir. Harianto, MS and Prof. Dr. Muhammad Firdaus, SP, M.Si., as the
advisory committee from Bogor Agricultural University, Indonesia, where
this thesis would not have been possible without their guidance and support.
2. Prof Dr Matin Qaim and Dr. Stefan Schwarze, as thesis advisors from GeorgAugust-University of Göttingen, Germany, for the advice and support during
the thesis writing process in Germany.
3. Michael Euler, in providing the SUSENAS data used in this research.
4. Prof Dr. Rita Nurmalina, MS and Dr. Ir. Suharno MADev as the Head and
Secretary of Agribusiness Study Program Bogor Agricultural University
respectively.
5. The Ministry of Higher Education - Indonesia in providing financial support
during the study.
6. All of my friends in SIA and MSA program for all support during my study in
Göttingen, Germany and Bogor, Indonesia.

7. My deepest gratitude goes to my lovely parents and family who always
support me with their dedication. I dedicate this work to my beloved parents,
whom always provide their love and prayers for me.

Bogor, September 2015
Cahya Najmudinrohman

CONTENTS

LIST OF TABLES

ix

LIST OF FIGURES

ix

LIST OF APPENDICES

ix


LIST OF ABBREVIATIONS

x

1 INTRODUCTION
Background
Problem Statement
Research Objectives

1
1
4
4

2 LITERATURE REVIEW

5

3 FRAMEWORK

Theoritical Groundwork
Operational Framework

7
7
8

4 RESEARCH METHOD
Data Set
Sampling Method
Construction of Key Variables
Model
Estimation Method
Goodness of Fit Test of Model
Dealing with Missing Values
Dealing with Outliers
Dealing with Heteroscedastic Residual Variances

10
10

10
11
12
12
12
14
14
15

5 DESCRIPTION OF RICE CONSUMPTION IN INDONESIA

15

6 RESULTS AND DISSCUSSION

17

7 CONCLUSIONS AND RECOMMENDATIONS
Conclusions
Recommendations

21
21
21

REFERENCES

22

APPENDICES

27

BIOGRAPHY

50

ix

LIST OF TABLES
1
2
3
4
5
6
7

Review of studies of rice consumption
Hypotheses for t-test
Mean of household consumption by expenditure group (kg/week)
Mean of calorie and expenditure shares of rice by expenditure group
Prices of rice and related commodities by islands (Rp.)
Estimation result
Mean of household consumption by island (kg/week)

6
13
16
16
17
18
20

LIST OF FIGURES
1
2
3
4

Price of grains, January 2003-April 2014
Per capita supply of staple foods in Indonesia, 1961-2009
Derivation of Marshallian demand curve
Operational framework

2
3
7
9

LIST OF APPENDICES
1
2
3
4
5
6

Paired test of quantity of rice grain and rice overall
Regression results by expenditure quartile
Normality of the residual
Heteroscedasticity in the initial estimation without robust standard errors
Multicollinearity testing
Questionnaire of SUSENAS 2013 quarter I

27
28
33
33
34
35

x

LIST OF ABBREVIATIONS
BPS

: Badan Pusat Statistik / The Centre Board of Statistics

CB

: Census Block

FAO

: Food and Agricultural Organization

OLS

: Ordinary Least Square

PSU

: Primary Sampling Units

USDA

: United States Department of Agriculture

SAKERNAS : Survei Angkatan Kerja Nasional / The National Labour Force
Survey
SUSENAS

: Survei Sosial Ekonomi Nasional / Indonesia’s National SocioEconomic Survey

1 INTRODUCTION
Background
Rice is the main staple food in Indonesia. It provides more than half of
total energy intake for Indonesian households in average (Timmer, 2004).
Considering that the number of population in Indonesia reaches 246.86 million in
2012 and its positive population growth (World Bank, 2014), providing sufficient
rice for its growing population remain an important issue. Regarding this matter,
there is long debate whether emphasize on domestic production or rely on import
to meet the rice demand in Indonesia.
Soon after independence of Indonesia, self-sufficiency in rice has been
sought to decrease foreign political influence and to save foreign exchange
(Moon, 1998 p.194-195, Barker and Hayami, 1976 p.617). Apart from political
economy perspective, the uncertainty in rice availability was also a significant
issue, notably when rice was unavailable in the world market in 1974 (Monke and
Salam, 1986 p.238-239). Such condition of uncertainty is one of the justifications
for self-sufficiency (Cheng, 1987).
After that rice crisis period in 1970s, green revolution came into fruition.
High yielding varieties contributed to rice production growth boost throughout
Asia. Accelerated yield together with higher commercial orientation in major
exporting countries made world rice market became more reliable for importing
countries (Dawe, 2002 p.369). The world price also has become less volatile since
1980s and it is less likely the top exporting countries to form a successful cartel in
rice market, so that Indonesia can rely on world market to meet the rice
consumption (Dawe, 2008).
However, both domestic production and import of rice face challenges.
Rice production in Indonesia encounters decelerating yield growth due to
degraded irrigation infrastructures, less attention in research and development,
conversion of productive land to other functions (Simatupang and Timmer, 2008),
and overused chemical fertilizer (Osorio et al., 2011). Improvement in technology
via research and development as well as its dissemination is yet to be seen
(Simatupang and Timmer, 2008). Likewise, rice production is also susceptible to
climate variability such as El Niño (Naylor et al., 2001).
Meanwhile, even though the quantity traded has been much larger after
1980s, the world rice market is relatively small compared to maize and wheat; the
quantity traded are respectively 7%, 11%, and 20% of the world production of
each commodity on average (Childs and Baldwin, 2010). Besides thin, world rice
market also remains unstable due to some characteristics as described by Jayne
(1993), which are fragmented, concentrated geographically in production,
inelastic in demand, and sensitive to oil price change. The instability became more
apparent in food crisis 2008 in which rice received the biggest impact in price
(Figure 1).

1000
900
800
700
600
500
400
300
200
100
0
Jan
Jul
Jan
Jul
Jan
Jul
Jan
Jul
Jan
Jul
Jan
Jul
Jan
Jul
Jan
Jul
Jan
Jul
Jan
Jul
Jan
Jul
Jan
Jul

Price
(USD/Ton)

2

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Year
Rice

Figure 1

Wheat

Maize

Price of grains, January 2003-April 2014

Source: FAO (2014), used also by Piesse and Thirtle (2009)

Given the condition of domestic production and international market of
rice, Indonesia bears more risk if there are limited alternatives other than rice.
Contrarily, Indonesia could be better in coping with supply shock of rice if
Indonesia has diverse diet in staple. For example, when food crisis hit
international market of rice the hardest in 2008 (Figure 1), Indonesia could have
given rice up to other commodities such as maize or wheat which had less severity
of the price crisis. If drought caused by El Niño hampered domestic rice
production (Naylor et al., 2001), cassava which is more drought resistant could
have been consumed more as an alternative staple. Given the scarcity of water and
productive land, higher yield crops could play important role in providing
sufficient food. Annual yields of potato, cassava, and sweet potato in Indonesia
are respectively 15.41, 15.39, 11.69 tons/ha, higher than rice which is 4.54 tons/ha
(Siregar and Surayadi, 2006). Nevertheless, utilization of cassava and sweet
potato tend to decline in last several decades (Figure 2).
As shown in Figure 2, Indonesia faces an upward trend in per capita rice
utilization – food utilization is equal to its supply – in the last four decades,
meanwhile per capita rice consumption declines in some other Asian countries
namely Japan, Malaysia, Nepal, Singapore (Ito et al., 1989), Thailand (Ito et al.,
1989, Isvilanonda, 2006), and Taiwan (Ito et al., 1989, Huang and Bouis, 1996).
Figure 2 also shows a discrepancy in staple food utilization, where rice is
dominantly utilized. Maize per capita utilization looks increase, but the utilization
for feed is nearly a half of food utilization (FAO, 2014). The utilization of wheat
and potato rise, both remain small compared to rice.

3

140
120
100
Per capita
80
supply
(kg/year) 60
40
20
2009

2006

2003

2000

1997

1994

1991

1988

1985

1982

1979

1976

1973

1970

1967

1964

1961

0

Year

Figure 2

Rice

Cassava

Maize

Sweet Potatoes

Wheat

Potatoes

Per capita supply of staple foods in Indonesia, 1961-2009
Source: FAO (2014)

Commonly, rice known and consumed in Indonesia is white refined grain
type of rice. Without its bran, white rice contains as high as 793.4 cal/gr energy
(USDA, 2011), but lacks vitamins, minerals, dietary fibre, and other
phytochemicals (Slavin et al., 1999, Zuñiga et al., 2013). Poor households might
suffer further from insufficient intake of those nutrients, since the relatively higher
price of rice (Figure 1) could hinder their ability to purchase nutrient rich foods
such are vegetables, fruits, and animal products. Those conditions could increase
the risk of diet-related disease. For example there was 50 percent of child
suffering iron deficiency anaemia in rural area of the Central Java province,
moreover it was more than 70 percent during the financial crisis 1998 (Block,
2004, Block et al., 2004), and iron deficiency anaemia is linked with long term
adverse effect on cognitive development of children (Alloway, 2013, Algarín et
al., 2013). Besides, high white rice consumption also correlates to insulin
resistance and hyperglycaemia, known as type 2 diabetes mellitus (Zuñiga et al.,
2013).
Integrating other staple might not fully answer those problems, but might
provide some alleviation. Integrating orange fleshed sweet potato into diet which
is evidently improves vitamin A status (van Jaarsveld et al., 2005) could play
pivotal role to combat micronutrient deficiency such as vitamin A which is also
prevalent in Indonesia (Berger et al., 2008, Suharno et al., 1993). Consuming
cheaper staple, such as cassava, could increase the budget for nutrient rich food
such as vegetable. By consuming more cassava, cassava leaf which is commonly
consumed as vegetable in Indonesia could be more utilized to provide ß-carotene,
iron, zinc, and sulphur (Wobeto et al., 2006). Furthermore, pairing some staple
food could be beneficial because each staple food is poor in some nutrient and
rich in other nutrient, as example rice lacks thiamine which maize has five times

4

higher of it, while rice has about three times pantothenic acid than yellow maize
(USDA, 2011).
Thus, this paper argues that the debate should shift away from selfsufficiency issue towards reducing rice consumption and promote more diverse
and balanced diet. More diverse and balanced diet enables households to have
better coping mechanism from supply and price shocks of each commodity. While
the effect on nutrient and health is not as sound as overall food diversity, diversity
in staple food could give potential benefit. To reduce rice consumption, the
current consumption behaviour of rice needs to be investigated.
Problem Statement
Developing countries face typically the same problem of population
growth and other problems related. Those problems also remain relevant to
Indonesian current condition. The last official census recording the population of
Indonesia took place in 2010 and it showed that there were 237,424,363 people
coupled with 1.49 growth rate (BPS, 2014). The dynamics of population growth
combined with the high level of per capita rice consumption, relatively stagnant
yields of rice, and thin international market of rice lead to an argument that it is
also important to decrease the consumption of rice. Therefore, the main problem
of this research is examines what drives consumption of rice and the magnitude of
those drivers.
The first hypothesis is that rice price affects the rice consumption
negatively, but the magnitude is low as a basic food. The second hypothesis is that
income has a sound negative impact, where the diet tends to shift toward more
expensive as income increases. The third hypothesis is that there are some food
complement and substitute for rice. The fourth is that education and female
gender of household head have negative impact on rice consumption. The fifth is
there are differences on rice consumption related to spatial factor.
Research Objectives
The challenge of high consumption of rice caused by high level per capita
consumption and huge population are the sum of Indonesian general food matter.
Increasing production alone is not enough to achieve and maintain self-sufficiency
of rice if it can’t meet the rise of its consumption. Thus, it is also crucial to reduce
the rice consumption to fulfil that goal. Figuring out what determines such high
level of rice consumption is a key to control it. Furthermore, measuring the
income elasticity and cross elasticity are the objectives of this research, in which
social, demographic and spatial difference are used as control variables.

5

2 LITERATURE REVIEW
As the main source of energy intake, rice consumption in Indonesia has
been a focus among researchers. Studies on Indonesian rice consumption of differ
in model specification and quantity measurement of rice. With few exceptions like
Timmer and Alderman (1979) and Timmer (1981), most of the Indonesian rice
consumption studies use budget share as quantity measurement. Timmer and
Alderman (1979) and Timmer (1981) used a straightforward measurement which
is kilogram of consumed rice and double log model specification. Budget share
would be useful for commodity grouping, but direct quantity measurement would
be more sensible for a specific commodity because it would help to take out the
quality difference from the estimation (McKelvey, 2011). Timmer and Alderman
(1979) also showed that disaggregating income level in the analysis is necessary
because they have different sensitivity to economic changes. One of the key
findings by Timmer and Alderman (1979) is that the first two quartiles of income
classes have 6.445 and 3.971 respective rice income elasticities, indicating rice is
almost a luxury good for them, and rice remains normal good even for the highest
income group. On the other hand, an influential work of Deaton (1990) changed
double-logarithmic demand functions into functions that relate budget shares to
the logarithms of prices and incomes to construct a demand system. One of the
findings is -0.424 own price elasticity of rice. He also found that the price of rice
has consistently higher impact on the quantity of other commodities compared to
the impacts of changes of prices of other commodities to rice quantity.
Rice is not only staple food in Indonesia but also in other Asia countries,
thus the parameters of rice consumption might be comparable. Isvilanonda and
Kongrith (2008) employs Almost Ideal Demand System (AIDS) model to
examine expenditure and price elasticity of demand for rice consumption in
Thailand households using cross-sectional data of a socio-economic survey
conducted in 2002. Some of the results stated that among the first and second
quartiles of income groups, the expenditure elasticity of rice quantity are 0.057
and 0.256 respectively, implying rice as a normal good in those income groups. In
contrast, the top income class of 25 percent had -0.436 elasticity, indicating rice is
an inferior good among richer households. Chern et al. (2002) examined income
elasticity of rice demand in Japan, using Linear Almost Ideal Demand System
(LA/AIDS) model and monthly basis cross-sectional household data named
Annual Report on the Family Income and Expenditure Survey (FIES) in 1997.
The results stated that expenditure elasticity is 1.76, indicated that rice is normal
good in Japan. Estimated cross price elasticity between rice and both fish and
meat carried negative sign so that fish and meat are complements to rice.
Rice also gains attention in the other countries in which rice is not the
main staple food. In this context, they usually face considerable amount of nonconsuming households, so that they usually employs logistic regression. Bamidele
et al (2010) using logistic model found that the major factors that significantly
influence household demand in Nigeria were the income of the head of household,
household size and the educational status of the heads of household. The study
also found that price per unit kilogram of rice was not a significant factor. BatresMarquez (2009) also used logistic regression to examined relationships among

6

economic, social, and demographic factors that affect rice consumption in the
United States. Data come from the Continuing Survey of Food Intakes by
Individuals (1994-1996) and the National Health and Nutrition Examination
Survey (2001-2002). The result is race/ethnicity and education are determinants of
the probability of consuming rice, and more so than low-income status.
Table 1

Review of studies of rice consumption

Author

Title

Data set

Model

Key finding

Timmer and
Alderman
(1979)

Estimating
consumption
parameters for food
policy analysis

SUSENAS
1976

Double-log:
log of quantity
as a function
of the log of
prices and
incomes

Rice is
almost a
luxury good
for the first
two of
quartiles of
income

Deaton
(1990)

Price elasticities
from survey data:
extensions and
Indonesian results

SUSENAS
1981

budget shares
as a function
of the
logarithms of
prices and
incomes

Asymmetric
cross price
elasticity

Isvilanonda
and
Kongrith
(2008)

Rice consumption
in Thailand: The
slackening demand

Thailand
Socioeconomic
survey data
2002

AIDS

Rice is an
inferior good
among richer
households

Chern et al.
(2002)

Analysis of food
consumption
behavior by
Japanese
households

FIES 1997

LA/AIDS

Rice is
normal good
in Japan

Bamidele et
al. (2010)

Economic analysis
of rice consumption
patterns in Nigeria

Primary data
2007 (survey)

logistic
regression

Income,
household
size, and
education are
significant in
the odds of
consuming rice

BatresMarquez
(2009)

Rice consumption
in the United
States: recent
evidence from food
consumption
surveys

CSFII 19941996 and
NHANES
2001-2002

logistic
regression

Ethnicity and
education are
significant in
the odds of
consuming rice

7

3 FRAMEWORK
Theoritical Groundwork
This chapter does not imply that the paper would impose most of
properties of demand in the model, but to address that the correlation of variables
has a theoretical ground. The current neoclassical demand theory is evolved from
marginalist theory of value by Carl Menger, William Stanley Jevons, and Léon
Walras (Moscati, 2007 p.362, Stigler, 1937). The marginalist explains that value
of commodity depends on marginal utility, which is increment of utility acquired
from consuming an increment of a commodity which is already held in some
number of quantities (Jevons, 1871). In this context, utility is property of an object
which gives happiness, advantage, benefit, pleasure, good, or prevents the
occurrence of unhappiness, mischief, pain (Bentham, 1879).

Figure 3

Derivation of Marshallian demand curve
Source: Salvatore 2008

8

Facing various commodities and constrained by a level of budget,
consumer has a combination of those commodities which gives equal utility,
where the combination of two commodities in two dimensional spaces can be
drawn namely indifference curve (Salvatore, 2008 fifth edition). When price of
one commodity (normal goods) decrease, ceteris paribus, consumer could afford
more quantity of the commodity (budget line shifts unevenly toward the
commodity) thus the indifference curve shifts. By capturing the impacts of several
shifts of the indifference, demand schedule shall be obtained (Figure 3).
However, the price effect captures both substitution and income effect.
The demand in which the price effect is described as above (uncompensated) is
known as Marshallian demand, while the (compensated) price effect in the
Hicksian demand differentiate them (Alston and Larson, 1993). This study
follows the Marshallian, so that the cross price elasticity needs to be treated
cautiously. If the rice expenditure is relatively high in the budget share, the
income effect is expected high as well. Thus, the substitution effect might be
undermined by the income effect. It might be the case for the lowest expenditure
group where they spend considerable share of their budget on rice.
Operational Framework
Harvey et al (2001) consider consumption as a dynamic process that must
be analyzed in relation to a range of macro and micro social changes (including
economic, spatial, temporal and cultural change). Economic factor is captured by
variable of expenditure, price of rice, cereals, meats, eggs, milk, vegetables,
noodle, oils, and fruits. Spatial factor is represented by island where the household
live. Social factor is represented by education of household head. Demographic
factor is represented by household size, age and gender of the head of household.
Educational and spatial variable are probably capture the taste of households
(Timmer and Alderman, 1979). Therefore, this study is designed to examine
determinants of rice consumption by testing the expected factors. From the
estimation, the sign and magnitude of each variable would give hints for
appropriate recommendation based on actual household consumption provided by
the data set.

9

Rice as the main staple food in Indonesia

PROBLEMS
Production: yield growth decelerate, susceptible to climate change
Import: thin market, sensitive to oil price change
Potential benefit of diverse diet: more resistant to shock, nutritional benefit
and preventable chronic disease

Analysis of Determinants of Rice Consumption in Indonesia
is necessary to be performed

Economic factors:
Expenditure, price of
rice, cereals, meats,
eggs, milk,
vegetables, noodle,
oils, fruits

Social and spatial
factors:

Demographic
factors:

Education level of
household head,
island

Household size, age
and gender of the
head of household

Rice consumption

Policy Recommendation
Figure 4

Operational framework

10

4 RESEARCH METHOD
Data Set
National socio-economics survey of Indonesia (known as SUSENAS) is
utilized in this paper. SUSENAS has been conducted by Centre Board of Statistics
(BPS) since 1965 (BPS, 2014). However, the data sets are mostly not panel.
SUSENAS sample was initially a small sample but it has been expanded since
(Walle, 1988). Sample of 2013 quarter 1 SUSENAS used in this study covers
75,000 households across 497 regencies/municipalities, represent all geographical
area of Indonesia (BPS, 2014).
The data set of SUSENAS records detailed household’s consumption of
215 food commodities classified into 14 groups. Both quantity and value of those
commodities are recorded within recall period of seven days prior to enumeration.
It recalls the actual consumption, which may differ from the purchase considering
at least these following factors. First, the purchasing habit may vary across
households as the household income (e.g. wage) may be received differently by
each household; monthly, two-weekly, weekly, or even daily. Within the recall
period, a household may purchase a bulk commodity for two weeks stock. In this
case, interviewer has to make sure that he records the quantity consumed for a
week instead of quantity purchased for two week stock. Second, food consumed is
not only from purchase but also from own production or give away. The quality
and value of those sources are recorded by SUSENAS in which the value of nonpurchase source is imputed from the local market price (Walle, 1988).
Consumption from own production remains important especially for 50.2 percent
of population living in rural area (BPS, 2014). Whereas food gift by neighbour or
relatives is common practice as part of Indonesian culture.
To use the data set for a consumer behaviour analysis specifically on rice
commodity, the commodity grouping needs be considered in detail. Indonesian
language differentiates terms for uncooked and cooked rice; beras and nasi
respectively. Rice (beras) consumption is recorded in group of Cereals, listed as
the first group of food in the data set. Consumption of any kind of cooked rice
(nasi) is recorded in the different group named Prepared Foods listed as the 13th
group of food in the data set. Overlooking rice consumption in the latter food
group will underestimates the quantity of rice consumption, especially on the
households (e.g. working couple) who buy prepared foods more than cook.
Sampling Method
A stratified – based on rural and urban classification, in which geographic
and socioeconomic conditions are expected to differ – multistage cluster sampling
was applied in each district/municipality to collect the SUSENAS 2013 data. In
the first stage, a number of primary sampling units (PSU) were chosen out of the
sampling frame by probability proportional to size principle, so that the
probability of each PSU being selected is proportional to its size (number of
households). Then the selected PSUs are randomly allocated into four quarters.

11

The sampling frame of this stage is list of PSUs from population survey 2010 in
which information on number of household (SP2010-RBL1 list), dominating
census block (normal, elite, or slum area), remoteness, rural or urban
classification of the PSU are included. PSU is a group of several census blocks
(CB), and CB itself is a group of several – approximately a hundred –
households).
There are two methods in second stage: (i) select two CBs from each
selected PSU for SUSENAS quarter II, III and I which is selected for The
National Labour Force Survey (SAKERNAS) as well, then the two CBs are
randomly allocated to SUSENAS and SAKERNAS, one for each, (ii) select a CB
by probability proportional to size (number of households) for SUSENAS quarter
IV and I which is exclusive for SUSENAS. The sampling frame of this stage is
list of CBs from each PSU which has been selected from the previous stage.
In the third stage, ten households from each census block were
systematically selected based on updated lists of households (SP2010-C1) using
VSEN11-P list. List of head of household were constructed from ExtractSP2010-C1 for variables of name, address, and education of the head of
household, then a field update was conducted. Sampling frame of this stage is list
of households in each CB gathered from population census 2010 (SP2010-C1)
which has been updated shortly before the survey. Institutional households
(orphanages, hospitals, barracks, prisons etc.) were excluded from the list of
households.
Regarding to the non-random sampling, the possibility of each
individual/household to be selected as a sample depends on the criteria used (each
chance may differs) instead of random (equal chance). SUSENAS has provided
design/sampling weight, so that the sample stays representative. Weight was
calculated from the inverse of the product of the probabilities of the sample being
selected at each stage. Distribution of the weight variable was examined as well.
Probability of the PSU being selected (sampling factor) is the same for each
quarter. Sampling factor of CB is the same for SUSENAS and SAKERNAS.
Construction of Key Variables
Income and price elasticities are the key parameters in the consumption
model upheld by consumer theory (Timmer and Alderman, 1979). However, those
variables are often not completely available in survey data. Thus, Income is
frequently replaced by expenditure in order to illustrate the living standard in
transforming economy (Qaim et al., 1997). In Indonesian context, one of the
plausible reasons is that reported income tends to be understated in such survey,
while expenditure is more precisely reported. SUSENAS also records expenditure
on the likes of insurance, education, housing, gold and jewellery which are
considered as investments in Indonesia. Covering those items, expenditure is quite
proximate to income. Furthermore, income may vary significantly over time in the
transforming economy setting, thus using expenditure in a spot survey is expected
to be more precise in illustrating the standard of living (Qaim et al. 1997).
Price is also barely available in survey data including SUSENAS.
Although SUSENAS surveyor imputes local market price for subsistence and gift

12

consumption, this information is sporadic. As previously mentioned, SUSENAS
sampling is conducted in clusters, thus unit value – obtained by dividing total
expenditure on commodities by its quantity – could be employed as a proxy for
price which tend to be similar in the same cluster (Deaton, 1980, McKelvey,
2010). It would roughly capture the spatial variance of prices (Deaton, 1980)
which prominent in Indonesian context where thousands of islands are scattered
and the transport infrastructures are not equally developed.
However, unit values capture both price and quality difference of
consumed good across households. To tackle this problem, household specific
unit value is replaced by mean unit value for each census block as proxy of price
of each commodity or group of commodities (similar to Skoufias, 2012 where
median unit values per each of the urban and rural were employed) and quantity
of rice is employed as dependent variable instead of budget share which is widely
used in demand analysis (McKelvey, 2010).
Model
Despite the neoclassical demand is derived from utility function, demand
equation could be specified directly in which quantity is a function of income (or
expenditure) and prices, without refer to the utility function, i.e. double-log
demand equation which is widely used as well (Clements, 1987). Considering that
the quantity is utilized instead of budget share as explained previously, this study
follow the direct approach. Thus, the model is specified as follows.
ln qrice = β0 + β1 ln X + ∑

2j ln

pj + ∑

3j ln

Dj + ∑

4j ln

Sj

j=1,…,n

where qrice is the quantity of rice consumed, X is expenditure per capita, pj is the
prices of rice and other food commodities, Dj is demographic factor, and Sj is
spatial factor, while βi is coefficient as the estimator of the model.
Estimation Method
Given the model above, the linear estimation is employed in this paper. If
the assumptions hold, least square gives the best linear unbiased estimator. Thus
least square shall be utilized while potential source of bias shall be taken into
account and the assumption shall be addressed.
Goodness of Fit Test of Model
a. R2
Based on Thomas (1997), “coefficient of determination is a proportion of
the total variation in dependent variable that can be attributed to variations in all
the explanatory variables acting together”. In the other words, R2 is used to
determine whether the variables in the model can explain the variation that occurs
in the independent variable. The higher the R2 indicates the better fit of the model.

13

b. t-Test
The t-test is a statistical test that is used to measure whether the parameters
or variables of the equation are individually significant or not, it is also known as
a partial test of significance because the significance of each variable can be
observed in the model. Generally, there are two types of alternative hypotheses
and both types have different decision rule.
Table 2

Hypotheses for t-test
One-sided alternative

Two-sided alternative

H0 : βk = 0
H1 : βk > 0 or H1 : βk < 0

H0 : βk = 0
H1 : βk ≠ 0

Source: Wooldridge (2006)

If the t-statistic is less than in the t-table (t-statistic < t-tableα/2, (n-k)) on the
real level of α, then H0 is accepted. Accepting H0 (βk = 0) indicates that the
variables tested did not significantly affect the dependent variable. Conversely, if
the t-statistic obtained on the real level of α is greater than in the t -table (tstatistic > t-tableα/2, (n-k)), then H0 is rejected. Rejection of the H0 (βk = 0) implies
that the variables tested significantly affect the dependent variable.
c. Heteroscedasticity
Heteroscedasticity occurs due to the variance of the error term not being
constant. Conversely, the homoscedasticity assumption states that the variance of
error of dependent variables should be constant, which is one of the assumptions
to make best linear unbiased estimators or BLUE (Wooldridge, 2006). The impact
arising from heteroscedasticity is non-constant variance, causing the value of the
variance to be larger than estimated. The high variance causes the hypothesis test
(F-test and t-test) to become less precise, with the confidence intervals becoming
larger due to large standard errors, and further resulting in an improper
conclusion. One of the formal tests of the heteroscedasticity is Breusch-Pagan
test.
d. Normality
The test of normality is conducted to examine whether the residual or error
term is close to normal distribution or not. A normality test of the error term is
conducted by using the Kernel density test. The Kernel density test is one of nonparametric analysis method. However, normality is relatively not as restrictive on
large sample size as shown by “Fahrmeir et al. (2013 p.124) that the tests and
confidence intervals, derived under the assumption of normality, remain valid for
large sample size even with non-normal errors”.
e. Multicollinearity
Multicolinearity occurs when there is correlation among the independent
variables, between two of them or more, in a multiple regression. The presence of
multicollinearity usually can be detected when the sign of the coefficient is not as
expected, and have high R2 but in the result of many individual-tests are not

14

significant. Another method to test the multicolinearity is Variance Inflation
Factor (VIF). The rule of thumb is that multicollinearity exists if the VIF value is
more than five.
(̂ )

The presence of perfect multicollinearity leads to the inability to determine
the least squares coefficient, as well as causing the variance and the covariance
values of the coefficients to become infinite.

Dealing with Missing Values
a. Quantity of Consumed Rice
In the data set, missing value in rice consumption presents as few as 1,164
percent of households. As the main source of calorie intake, rice is consumed by
almost all households in Indonesia. In this case, it is not convincing to assume that
those missing values are zero consumptions indicating preference as a choice to
not consume (Walle, 1988). Thus, those missing values are excluded instead of
using a method dealing with censored distribution.
b. Prices
Price of a commodity in a certain area is expected to be the same. The
closer the households locate, the higher the likelihood of facing the same price is.
In the same province, prices are likely to vary between rural and urban area due to
the spatial and socioeconomic factors. Thus, missing values of prices are replaced
by mean value of the commodity in each rural and urban area.
c. Education
In the data set, the percentage missing values in education of household’s
head is 7.01 percent. In education categories, there is 46.15 percent household
whose head has elementary school as his highest completed education, 15.74
percent middle school, 22.60 percent high school, 1.99 percent diploma, 5.78
bachelor, 0.73 percent master or doctor. Given the proportion and the Indonesian
sociocultural context, missing value is possibly the same as no education. Thus in
this study thus missing values are assumed as no education, the missing values are
replaced by zero values.
Dealing with Outliers
Least squares estimators are very sensitive to the presence of outliers in
which can affect estimation as well as inference (Fahrmeir et al., 2013 p.105
p.160). Using estimation method which is not susceptible to outliers (robust
regression) would be desirable, however heteroscedasticity which will be
discussed in the following part of this paper is needed to be addressed as a
priority. Thus robust regression would be used to detect the outliers, and then the
outliers are excluded in the main analysis.

15

Dealing with Heteroscedastic Residual Variances
Heteroscedasticity yields biased estimators (Trenkler, 1984). However, the
theory about the severity and the type of it is almost nonexistent and the validity
of formal tests highly depends on whether the model is correctly specified, i.e.
Breusch–Pagan test assumes multiplicative variances with exactly defined
predictors (Fahrmeir, et al. 2013 p.186). Formal tests should not be the only tool
to detect heteroscedasticity, thus residual plots need to be part of it since they are
in many cases the only way to diagnose the specific type of heteroscedasticity or
to discover which of the variables affect the variances of the residuals (Fahrmeir
et al., 2013 p.186).
Thus in this paper, heteroscedasticity is encountered by examining the
residual plots produced by WLS estimation using each predictor as estimation
weight, then selecting the weight from the predictor which has the highest impact
in reducing the magnitude of the heteroscedastic residual variances. Following
those steps, none of the predictors in the model gives obvious impact. Thus, this
paper use more generic solution using sampling weights to obtain standard errors
that are robust to such unspecified heteroscedasticity. Besides, households from
different income group might have different elasticity of income and prices, thus
the estimation is done in each expenditure groups.

5 DESCRIPTION OF RICE CONSUMPTION IN INDONESIA
As mentioned previously, the quantity of rice consumption would have
been undervalued if prepared rice had not been included. Given the structure of
commodity grouping in the SUSENAS data, there are some food items which are
not so clear whether they are made out of rice or other commodity, such as cakes
and snacks. However, their quantities are expected to be relatively negligible
compared to other obvious rice based prepared food such as plain or fried rice.
The SUSENAS quarter 1 data show that typical household in Indonesia consumes
7.0970 kg of rice in a week (Table 4), including 6.6250 kg of uncooked rice grain
(beras), and the difference of both numbers is statistically significant. On the
individual level, the average of the rice consumption is 1.9025 kg in a week.
Assuming the consumption does not have significant seasonal fluctuation, the
individual consumption is proximate to 99.2035 kg per capita per year.

16

Table 3

Mean of household consumption by expenditure group (kg/week)
Quartile of per capita total expenditure
Commodity
Overall
1
2
3
4
Rice
8.058640
7.458152
6.9158270 5.982665
7.097278
Roots
4.082572
2.293414
0.9469490 1.520720
2.421175
Meats
0.732101
0.820666
1.8332090 1.244944
1.007032
Cereals
1.569261
0.834532
0.7040790 0.679723
0.952018
Eggs
0.693285
0.788658
0.9299817 1.072429
0.884327
Vegetable
4.485042
5.118447
5.5519810 5.978528
5.264798
Fruits
2.134889
2.334012
2.6747510 3.465321
2.708186
Instant noodle 0.320906

0.363425

0.4038715

0.439922

0.383834

Table 3 shows that the quantity of consumed rice and other food
commodities vary across expenditure group. It indicates that the higher the group
of expenditure, the lower the level of rice consumption. However, it does not give
any explanation on the variability of rice consumption within each group of
expenditure. Thus, it would be explored further in the result of the model
estimation. This chapter focuses on the brief overview of the rice consumption
and its importance in Indonesia (Table 4).
Table 4

Mean of calorie and expenditure shares of rice by expenditure group
Quartile of per capita total expenditure
Variable
Overall
1
2
3
4
Rice calorie share
0.5729
0.5273
0.4914
0.4490
0.5097
Rice expenditure share
0.0454
0.0327
0.0249
0.0140
0.0291

In accordance to Timmer (2004), Table 5 shows that rice remains the main
source of calorie intake in Indonesia, especially for the low expenditure group.
The first quartile of expenditure group relies on rice to provide about 57.29
percent of calorie intake in average. The shares consistently decreasing along the
increasing expenditure group, however it remains relatively high even in the
fourth quartile. Meanwhile, the expenditure on rice contributes by 4.54 percent to
total expenditure of the first quartile of expenditure in average. The proportion is
decreasing into 1.40 percent in the last quartile of expenditure.
Beside expenditure, the variability of prices – both own price and potential
cross price – is expected to explain the variability of rice consumption. SUSENAS
as a cross-section data does not capture the price variability over time, however it
capture the spatial variability of prices. Insinuation on spatial price variability is
shown in Table 5 where islands are used to depict the spatial factor.

17

Table 5

Prices of rice and related commodities by islands (Rp.)
Mean of Price by Islands

Commodity
Rice
Roots
Cereals
Meats
Eggs
Milk
Vegetables
Noodle
Oils
Fruits

Sumatra

Java

Bali-Nusa

Borneo

Sulawesi

8992.17
6058.37
9329.59
32930.08
15283.56
57190.50
5620.22
20883.38
12383.90
9313.52

7771.57
4373.31
8149.46
29865.79
14700.18
51700.74
4219.35
18904.09
11565.67
8309.79

8462.63
4798.36
6592.61
41072.91
15261.98
58275.96
4842.51
23579.26
14680.11
8524.41

8874.43
5730.59
8807.04
32210.18
16747.97
54365.60
7269.80
22095.51
13340.99
10393.19

7379.58
5012.21
7070.97
35715.38
15249.73
50037.86
4402.58
21552.44
11741.99
7416.49

MalukuPapua
10391.13
7895.64
10376.52
40480.82
22701.19
54241.33
8789.15
29162.33
18947.11
11380.58

In general, Maluku and Papua have the highest prices. The development
inequity makes the transportation cost higher to distribute goods to some areas,
such as Papua, while area having good transportation infrastructure and or is close
to the producer may have lower prices, such as Java. Even more, the development
inequity may presents within each island itself.

6 RESULTS AND DISSCUSSION
Variance of consumed rice quantity might be correlated to the changes of
the likes of expenditure, spatial indicator, education and gender of consumer. The
rice consumption varies in each quartile of per capita total expenditure where the
lower the quartile, the higher the mean of rice quantity in kg/week. People who
are able to afford higher expense obviously earn higher income. They can
consume more diverse food both fancy staple food (i.e. pasta and potato) and
more nutrient dense type of food (i.e. meat, dairy, fruit and vegetable).
The result of the estimation (Table 6) shows that the R2 of the model
varies between 0.564, 0.597, 0.615, 0.616, and 0.642. It means that minimum 56.4
percent of the variance of rice quantity across the households can be explained by
the variance of the independent variables, which are expenditure, price of rice,
price of roots, price of fruits, price of other cereals, price of meats, price of eggs,
price of milk, price of vegetables, price of noodle, price of oil, household size, age
of household head, gender of household head, education of household head, and
spatial factor represented by island. The highest explanatory power is 0.642,
which means that 64.2 percent of the variance of rice quantity across the
households can be explained by the variance of the independent variables.

18

Table 6

Estimation result

Variable

Ln quantity of rice by quartile of per capita total
expenditure (PCTE)

Ln quantity
of rice

1

2

3

4

ln PCTE
ln price of rice

0.211***
-0.277***

0.146***
-0.154***

0.0608**
-0.0490*

-0.0336***
-0.0937***

0.0530***
-0.132***

ln price of roots
ln price of cereals
ln pri