J Geograph Syst (2006) 8: 253268

DOI 10.1007/s10109-005-0016-3

O R I GI N A L A R T I C L E

Suahasil Nazara Georey J. D. Hewings Michael Sonis

An exploratory analysis of hierarchical spatial

interaction: the case of regional income shares
in Indonesia

Received: 28 February 2005 / Accepted: 2 November 2005 / Published online: 14 January 2006
Springer-Verlag 2006

Abstract Regional interaction is generally understood as implying movement

between regions at the same level of an hierarchy. This paper extends the
notion to include an hierarchical system, thus facilitating the consideration of
vertical interaction in the analysis of regional interaction. Obviously, vertical
interaction is not altogether a new concept. One could nd this concept in
many analyses related to national-local or federal-state relationships. This
paper treats hierarchy in a strict sense; spatial units are interacting one with
another within, but not between, presumed super regions. A particular
example drawn in this paper is the province-region relationships for the case
of Indonesia. In this framework, provinces form a region, and regions form
the nation. The DendrinosSonis model is then used as the basis for measuring the hierarchical spatial interaction in Indonesia. The analysis will explore the degree to which complementarity and competitive interaction
revealed at one level in the hierarchy persist at lower or higher levels.
Keywords Hierarchical interaction DendrinosSonis model Indonesia
JEL Classication C5 O1 R1
S. Nazara
Demographic Institute, Department of Economics,
University of Indonesia,
Kampus FEUI, 16424, Depok, Indonesia
G. J. D. Hewings (&) M. Sonis
Regional Economics Applications Laboratory, University of Illinois,
607 S. Mathews, #318, Urbana, IL 61801-3671, USA
E-mail: hewings@uiuc.edu
M. Sonis
Department of Geography, Bar Ilan University, Ramat Gan, Israel

254

S. Nazara et al.

Introduction
Regional interaction happens in two important dimensions: horizontal and
vertical. Horizontal interaction takes place among regions at the same level
of an hierarchical structure, for example, interaction among states, among
provinces, etc. Vertical interaction, on the other hand, is among regions at a
dierent level of the hierarchy. A typical example of this interaction is that
between central and local economies, or federal and state governments.
Clearly, these vertical and horizontal relationships can take place at the same
time. Integrating this combination will be the focus of this paper, providing
an analytical framework to examine regional interaction that takes into
account both vertical and horizontal relationships.
Interaction can be of two distinct forms. First, two interacting regions
may engage in a complementary relationship. In the economic growth
sphere, this would mean that economic growth in one region is positively
related to the growth in the other region. Alternatively, two interacting regions may be involved in a competitive relationship. Again, in the sphere of
economic growth, the economic growth of one region results in a decline in
the other region.1 In the model to be developed here, interaction may be
considered as the resultant of a complex set of ows that might include
capital and other nancial ows such as remittances, goods and services and
government expenditures. Since many of these ows are not documented
individually or in terms of a strict origin-destination ow, the analyst is
forced to consider surrogates and to view competitive forces at a more macro
(aggregated) level than might otherwise be desirable.
The Dendrinos-Sonis (1988, 1990) model will be employed as the basic
methodology. The model, originally developed to explain the dynamics of
population changes, has been applied to income variables in several different contexts, for example Hewings et al. (1996), Nazara et al. (2001)
and Magalhaes et al. (2001). The present paper diers from previous
applications since several layers of hierarchical regional structure are
considered. In that sense, the paper provides a framework for a multilayered economic analysis of interaction (for alternatives, see Funck 2000;
Johansson 2000).
The model will be applied to Indonesian data, for the 19751999 period.
The analysis will use the gross domestic regional product (GDRP) at the
provincial level at 1993 constant prices.2 The uneven distribution of domestic
regional product is a signicant, sustained feature in Indonesian regional
development. Using an interregional inputoutput structure, Sonis et al.
(1997) revealed that the western part (Sumatra, Java and Bali) is in

This way of understanding regional interaction is dierent from those dening regional
interaction that is agent-based. An example of agent-based denition of regional interaction
is provided in Poot (2000, 205) where he denes regional competition, which is one type of
interactions, as actions of economic agents that are taken to enhance the standard of living
of their own territories, such as regions, cities, or countries.
2
The term product and income should have dierent meaning at the regional level.
However, we ignore such distinction in this paper. Hence, the two terms will be used
interchangeably.

An exploratory analysis of hierarchical spatial interaction

255

competition with the poorer eastern part. This paper will be dierent from
Sonis et al. (1997) in two respects. First, we will examine the nature of
interaction using a long-term time-series, spanning over a quarter of century,
rather than employing a one-point snapshot that would characterize analysis
based on interregional inputoutput data. Secondly, this paper will employ a
modeling framework where province and region are interrelated at the same
time vertically as well as horizontally. This is in contrast to Sonis et al. (1997)
where interaction was assumed to take place among regions horizontally, at
a single point in time.
Notwithstanding the uneven distribution of regional income, some
studies have asserted the existence of convergence at the provincial level
(Garcia and Soelistianingsih 1998; Nazara 1999). Convergence studies typically assume that provinces operate independently of each other. The
present paper, however, tries to explore the nature of interaction among
provinces and regions. The immediate connection between convergence and
interaction is not directly clear; this paper is an attempt to shed some light
upon the relationship.
This paper is organized as follows. Part two will elaborate the DendrinosSonis model, the basic framework for the interaction analysis. Part three
of this paper will extend the standard model to t the hierarchical regional
structure. Part four presents the context for the Indonesian economy during
the study period. Part ve analyses the estimation results for the 19751999
period in Indonesia. Some reections and further considerations complete
the paper.

The basic DendrinosSonis model

The model was rst introduced by Dendrinos and Sonis (1988, 1990).
Initially proposed to handle the dynamics of population changes, this
model has also been applied to economic analysis in various countries. For
example, Hewings et al. (1996) explored applications to economic interactions in the US Midwest economy, while Magalhaes et al. (2001) compared these results with an application in the Northeast Brazilian
economies. In this section, we will elaborate the model as it is used for the
analysis of interaction among regions at the same level of hierarchy. We
will extend the methodology further for the hierarchical regional interaction in the following section.
Let yi(t) denote the relative income of province i, i.e., its share in the
total national income at time t. Further, assume that there are n provinces
in the economy. Thus, the distribution of the relative income can be written
as
Y t y1 t; :::; yi t; :::; yn t

i 1; :::; n t 1; :::; T :

It can be seen here that we are dealing with a discrete system of distributional
dynamics. The relative discrete socio-spatial dynamics is given by:

256

S. Nazara et al.

B F yt
C
B i
C
yi t 1 B P
C
@ n
A
Fj yt

i; j 1; :::; n;

t 1; :::; T ;

j1

P
where 0<yi(0)<1, Fi[y(t)]>0, and i yi 0 1: Although the function Fi(.)
can take any arbitrary form as long as it satises the positive value property,
the expression Fi[y(t)] presents the locational and temporal comparative
advantages enjoyed by the population at (i, t) (Dendrinos and Sonis 1990,
26; Sonis and Hewings 2000). We also need to determine a numeraire or
reference region, a reason for which will become clear as we move on with
the model. Assume that the rst province is considered as the numeraire,
then we can always state another regions observations in terms of this
numeraire, that is:
Gj y0

Fj y0

F1 y0

for all j 2; 3; :::; n:

With this specication, Eq. 1 can be expressed in the following system of

equations
8
1
y1 t 1 P
n
<
1
Gj yt

where j 2; 3; :::; n:
2
j2
:
yj t 1 y1 t 1Gj yt

The numeraire serves two functions in the model. First, it ensures that the
shares of all regions sum up to one. This notion is important since regions
can only divide a xed total; the national income has to be mutually divided
among existing regions. One may think of this model as the working
framework of Richardsons (1973) competitive model of growth in terms of
the regional income share. Secondly, the numeraire is also a way to represent
regional interaction because the growth of a specic region is specically
expressed in terms of others. A single region is not standing alone in the
system of regions; it interacts with others, the nature of which is represented
by the existence of the numeraire.
As mentioned earlier, the function Fi(.) can take any arbitrary form as
long as it satises the positive value property. In this paper, we will assume a
multiplicative specication of Gj[y(0)] as suggested by Dendrinos and Sonis
(1988), that is:
a

Gj y0
Aj Pk yktjk

where j 2; 3; :::; n; k 1; :::; n:

Coecient Aj>0 represents the locational advantages of provinces j=2,

3, ..., n. The log-linear modication yields:
ln yj t 1  ln y1 t 1 ln Aj

n
X
k1

ajk ln yk t;

An exploratory analysis of hierarchical spatial interaction

257

where j=2, 3, ..., n, k=1, ..., n. The coecient ajk thus implies ajk
@ ln Gj y0
=@ ln ykt which is an elasticity term. It is the percentage change of
income, i.e., the percentage growth in share in region j relative to that in
region 1, the numeraire, with respect to a 1% change of income in region k.
The coecient ajk is central to the competition and complementarity
analysis, both in terms of its sign as well as its magnitude. A positive value
would indicate complementarity growth in shares between the two regions j
and k. That is, every 1% income growth in the share in region k would correspond to an ajk percent income growth in the share in region j. On the other
hand, a negative value of ajk would indicate a competitive relationship between
the two regions; if the share in one region grows, the others share will decline.
Given the assumed functional form, note that Eq. 4 is completely linear in
parameters. As it is a system, then Seemingly Unrelated Regressions technique may be employed. Indeed, since the regressors on the right hand side
are the same for a province, then OLS may be used. Still, we will use SUR for
the estimation of Eq. 4 to make the results comparable with the other versions of the DendrinosSonis model, which will be presented in the next
section.3
As a nal note to this section, this model deals with spatial interaction
without the need of a so-called spatial weight matrix, an a priori structure of
regional interrelationship imposed on the regional system. In a sense, the
current model arrives at the spatial interaction structure by letting the data
speak for themselves. This suits certain cases, such as the Indonesian case in
this paper; as it is an archipelago country, the determination of spatial
weights that are typically based on contiguity criteria presents a daunting
logistical and denitional problem. As an island country, many provincial
borders may be separated by sea, rather than physically contiguous on
land. In some places, like East Java and Bali provinces, ground and air
transportation are available to facilitate movements of people and goods.
However, take the case of provinces in Java and Kalimantan. Administratively, there are no common borders between these provinces as the sea
separating the two is quite large. In retrospect, this is exactly the problem of
excluding Hawaii from the rest of US contiguous states.4

Interaction in a hierarchical structure: a methodological framework

As asserted in the introduction to this paper, the vertical and horizontal
interregional interaction schemes take place at the same time. Vertical
interaction takes place between two or more spatial units of dierent levels,
and horizontal interaction occurs among spatial units of the same hierarchical

3
SUR technique on Eq. 4 will yield the same estimates as those obtained with the
ordinary least square (OLS) estimator. The reason is simply the fact that each of n - 1
equation in Eq. 4 has exactly the same set of explanatory variables (Judge et al. 1988). The
use of SUR here is justied by its greater eciency than OLS.
4
For further exposition on the spatial econometrics technique involving the use of
spatial weight matrix, look at Anselin (1988).

258

S. Nazara et al.

Fig. 1 Hierarchical regional structure

layer. Therefore, it is necessary to build a methodological framework where

the two are taken into account at the same time. This is exactly what this
section will try to accomplish.
Before the modeling framework is presented, a review will be provided of
the kind of hierarchical structure that will be considered. Figure 1 provides
the typical hierarchical regional structure of an economy, the one that will be
adopted in this paper. The economy has two layers in the hierarchy, namely
regions and provinces. Assume that there are two regions, R and S, in the
economy. The dotted line between R and S means that the two are interacting; this is the horizontal interaction. The solid line from the nation to
the two regions means that R and S engage in a vertical relationship with the
upper level. Further assume that there are three provinces within region R,
and two provinces within region S. Just as each region interacts with the
nation, each province has a vertical interaction with its region and it is in a
horizontal interaction with other fellow provinces within the region. Further,
interaction between Province 1 in R and Province 4 in S is not direct. Such an
across-region interaction is conducted through the appropriate regions. Such
a structure is also known as a strict hierarchy. Obviously the hierarchical
structure can be extended to lower administrative units such as districts
within each province, sub-districts within each sub-district, and so on.5
Recall the DendrinosSonis model as discussed in the previous section.
One may note that the regions dened there, i.e., i=1, ..., n are of the
same level. When provinces are grouped within dierent regions, then the

5
That fact raises another issue to address, which is beyond the scope of this paper, i.e.,
the appropriateness of the spatial categorization. This question carries many dierent
concerns, among them are the appropriate level of structure for the analysis, should vertically hierarchical structure be dened in accordance with the term nodal or functional
regions or administrative regions, etc.

An exploratory analysis of hierarchical spatial interaction

259

economic growth of a particular province is also dependent on the economic

growth of other regions in the country. Employing the DendrinosSonis
model as is shown in Eq. 4, the regional interaction within an hierarchical
structure can be written as follows:
ln yj t 1  ln y1 t 1 ln Aj

n
X

ajk ln yk t

k1

n
X

ajl ln yl t;

l1

where now l=1, 2, ..., m denotes regional indices, and the other variables are
dened as earlier.
Note that the economic growth of region j is still represented in terms of
the numeraire region 1. Coecient ajk as before denotes the percentage
change of income growth in region j (in terms of the numeraire) as a result of
a percentage change of income growth in province k. In the same spirit,
coecient ajl would denote the percentage change of income growth in
region j (in terms of the numeraire) as a result of a percentage change of
income growth in region l. The interpretation of the sign of the as will be
similar to the previous denitions. Seemingly Unrelated Regressions technique is the proper estimation technique since, in contrast to Eq. 4, Eq. 5 has
income of regions on the right hand side. Hence, the regressors for each
province are no longer the same and OLS is no longer appropriate.
Note that Eq. 5 is specied in a very general way so as not to restrict any
symmetrical relationship between two provinces within the same region.
Asymmetry in provincial relationships is not an implausible feature of
regional interaction. Of course, that does mean that symmetry is impossible.
When a region j nds region k as its complement (or competitor), it is always
possible for k to have a mutual feeling toward j. This is the case of symmetric
interaction. What can be asserted here, however, is the possibility of an
asymmetrical relationship. This is the case where region j nds k as its
complement while k nds j as its competitor. More specically, economic
growth in k will bring a positive eect to the economic growth in j, but
economic growth in the latter region negatively aects that of the former.
How is such a relationship possible? One could argue that this kind of situation may be triggered by the existence of an imbalance in the transactions
between the two regions (Nazara et al. 2001). For instance, imagine that
there exists a massive ow of production input, i.e., capital, labor, etc., from
one region to the other. The sending region may be negatively aected by the
outow while the receiving region may be beneting from this phenomenon.
The asymmetric relationship may also take place when one region is used as
nothing but a pure market, such as a hinterland serving as a market for the
core regions products.
Indonesian application: context and model structure
Indonesia is an archipelago country, comprising more than 13,000 islands,
currently divided into 32 provinces. The number of provinces has been in
ux recently, in particular since the launch of the new reform movement in
1998 and the decentralization scheme adopted in 1999. More and more

260

S. Nazara et al.

Fig. 2 Indonesian map of regions

localities are asking to form their own, new provinces. For the purpose of
this paper, we will use the 26-province system.6 The convenience of this
aggregation stems from the fact that it is in accordance with the available
published data by the Indonesian Central Bureau of Statistics. The GDRP
data are at the provincial level from 19752000 in 1993 constant prices. The
data are available by types of expenditures and by the industrial origin.
There is a general consensus in Indonesia that the quality of regional data
has been improving over time.
The provinces can be categorized into several regions. In this paper, we
are going to use a ve-region classication.7 They are Sumatra (eight
provinces), Java and Bali (six provinces), Kalimantan (four provinces),
Sulawesi (four provinces), and the Eastern Islands (four provinces). As is
immediately clear from Fig. 2, this regionalization conforms to the geography of the country. As a country with major islands, it is straightforward to
consider that provinces within the major island will interact more one with
another. In some sense, the major islands provide the natural break for the
regionalization.
The case for the adoption of a strict hierarchy for Indonesia is not only
supported by the geographical perspective, but is also sustained from the
policy-making point of view. The above regional classication is the one used
by the Development Planning Agency (Bappenas) in conducting the National
Consultative Planning Meeting (Konasbang) in Indonesia. Hierarchically, a
series of regional consultative planning meetings are also conducted within

6
This is the old 27-province system minus East Timor, who became an independent
country in 1999.
7
There are several other regionalizations that are common, for instance, Java and
outer islands. Another commonly used regionalization is the western and eastern parts
of Indonesia. The former refers to Java and Sumatra (and sometimes also includes Bali),
and the latter is the rest of the country.

An exploratory analysis of hierarchical spatial interaction

261

Table 1 Intermediate input transaction by regions (in percentage)

Sending region

Sumatra
Java and Bali
Kalimantan
Sulawesi
Eastern Islands

Destination region
Sumatra

Java and Bali

Kalimantan

Sulawesi

Eastern
Islands

89.9
2.2
0.3
0.5
0.3

9.3
96.0
13.6
5.4
12.7

0.6
1.0
84.9
2.2
3.1

0.1
0.3
0.9
90.5
4.8

0.0
0.5
0.3
1.3
79.2

each region. A major aim of such consultative meeting is to synchronize the

public and direct (domestic and foreign) investment plans in the country.
More than that, the strict hierarchy case is also economically reasonable.
Refering to Table 1, one may note the dominance of self-contained interaction at the regional level in Indonesia. Derived from the intermediate input
transaction from the 1995 Indonesian interregional inputoutput table, the
table shows that the vast majority of input transactions are within-region,
rather than between-regions.
The distribution of income among these ve regions can be seen in Fig. 3.
Java is the dominant economy having almost 60% of national income.
Adding Sumatra, to form the western part of Indonesia, the proportion
increases to about 80%. The hegemony of Java in the Indonesian regional
structure is no surprise. Several other studies have also pointed this out (see,
among others, Sonis et al. 1997). Partly, it is a result of the centralistic
pattern of development during the last three decades. High dependency on
the central government, which is located in Java, results in the one-way ow

70

Percentage

60
50
40
30
20
10
0
1975

1990

1995
Year

Sumatra
Sulawesi

Java & Bali

Eastern Island

Fig. 3 Regional share of income in Indonesia, 19752000

Kalimantan

262

S. Nazara et al.

of all resources to the region. Therefore, it should not come as a surprise that
there is almost no signicant change in the proportional distribution during
the last 25 years. One may appropriately predict that the pattern shown in
Fig. 3 will continue into the future should there be no signicant policy
intervention.
Despite the long-term regional share of income shown above, some
studies do nd evidence of regional convergence in Indonesia. Garcia and
Soelistianingsih (1998) nd evidence of r and b convergence for the period of
19751993. The former, r convergence, refers to a decrease in the dispersion
of provincial per capita GDRP over time. The latter, b convergence, refers to
the test of absolute and conditional convergence in the annual growth rate of
regional GDP. Several time periods, i.e., 19751993, 19801993, and 1983
1993, were tried and convergence is conrmed for all three periods. In the
same spirit, Nazara (1999) took suggestions made by Mankiw et al. (1992)
on the correct specication of the Solow growth model, and using the xed
eects model for pooled provincial and time series data, also conrms the
evidence of convergence among provincial per capita income in Indonesia.8
The above discussion brings to the surface the necessity to take into
account regional considerations, in addition to the standard provincial
framework. We have shown that the regional perspective is supported by the
geography, policy-making process, interregional trade, and lastly by convergence phenomenon. In the latter, we contrast the provincial and regional
perspectives. In retrospect, convergence studies typically assume that each
province operates independently one with another. It is not clear what kind
of regional interaction is implied by regional convergence or divergence. It is
the intention of this paper to shed some light on this subject.

The hierarchical system: regional-provincial interaction

Two alternative models, as shown in Eqs. 4 and 5, will be estimated. The
analysis here will focus on the signs of the interaction. As elaborated earlier, a
positive sign for the regression coecients means that the growth in the share
of the explanatory region will have a positive eect on the share of the region
in the dependent variable. A negative coecient will correspondingly bring
negative eects on the share of the region in the dependent variable. Each cell
in Table 1 represents the sign impact of the growth of the shares of the
column province to that of the row province. Each row would represent Eq. 4
for each region j. For the numeraires, we use the following: Lampung (18) for
Sumatra, Jogya (24) for Java, South Kalimantan (33) for Kalimantan,
Southeast Sulawesi (44) for Sulawesi, and East Nusa Tenggara (52) for
8
Some other studies have also conrmed the existence of regional convergence within a
country, e.g., Barro and Sala-i-Martin (1995) for the case of US states and Japanese prefectures. For the case of developing countries look at Cashin and Sahay (1996) for the
convergence among 20 states in India. From the theoretical point of view, the convergence
among sub-nations is sparked by the relatively homogenous conditions of exogenous
variables. In terms of the Solow growth model, the exogenous variables may be the saving
rates, population or labor supply growth, depreciation, or rate of technical progress.

An exploratory analysis of hierarchical spatial interaction

263

Eastern Islands. The numbers in parentheses are the provincial numbers.9

The whole system, 21 equations for each province with each 25 observations,
is estimated using the SUR technique.
Table 2 presents the signs of interaction without regional eect, i.e.,
following specication (4), and Table 3 presents the signs of interaction with
regional eects, i.e., specication (5). The identical set of numeraires is used
in producing both Tables 2 and 3. Insignicant signs are denoted with 0.
The gray shades in Table 3 highlight the dierences with Table 2. Gray
cells with 0 in Table 3 means the cell was statistically signicant in Table 2.
There are only a few gray 0 cells in Table 3. Altogether there are 18 gray 0
cells, or about 15% of the total interaction signs in Table 2. In addition to
that, gray cells with + or  signs means that the sign of that cell was the
opposite in Table 2. There are only two cases found in the whole Table 3.
The fact that there are only relatively small dierences in terms of sign
changes attests to some degree of robustness in the estimation. While as a
consequence one may infer that Tables 2 and 3 are qualitatively comparable,
recall that in Table 3 we have incorporated the regional eect that may be at
play in Indonesian sub-national interaction.
The regional eects are shown in the right columns in Table 3. For
example, the North Sumatra province in the Sumatra region interacts with
other provinces within the region, and also with other regions in the country.10 To illustrate, the interpretation of results for Aceh province in Sumatra
is shown by row signs. Acehs GDP share growth is positively related to the
share growth of other provinces in Sumatra island except Bengkulu, and is
positively related to the share growth of Java, Kalimantan, and Sulawesi
regions; but is negatively impacted by growth in the shares in the Eastern
Islands region.
Provinces in Sumatra are largely complementary with each other.
A negative impact on other provinces, however, is apparent from Bengkulu
(17). This is the case where the province receives a positive impact of the
share growth of other provinces and regions; but not vice versa. It is also a
case where the impact of a region is asymmetric. In relation to other regions,
provinces in this region are also positively impacted by growth in shares in
Java and Kalimantan. Java, Sumatra and Kalimantan form the western part
of Indonesia. This is the area where ows of goods and people are the largest.
Share growth in the eastern regions, Sulawesi and the Eastern Islands, also
have positive impacts on the Sumatra provinces, although a competitive
relationship is statistically signicant for Aceh.
In Java, the capital city Jakarta seems to have signicant positive interaction with its own region and West Java provinces, and a negative impact
with Central Java. Share growth of other regions also turns out to impact the
capital city negatively. That puts Jakarta in a competitive fashion with most
of the Indonesian regions and provinces. In retrospect, Java provinces are
primarily in a competitive mode one with another, as well as with other
9
We use the provincial numbering system to make table reading easier. Note that this
numbering system is dierent from one ocially used by the Central Bureau of Statistics.
10
Note again that it is assumed that provinces of dierent regions do not interact
directly. Recall the strictly hierarchical structure as presented in Fig. 1.

All signs are signicant at 5%. 0 denotes statistically insignicant signs

Table 2 Qualitative (sign) analysis without regional eect

264
S. Nazara et al.

All signs are signicant at 5%. Shaded 0 cells mean the sign was statistically signicant in Table 2. Shaded + or  means the
opposite signicant sign in Table 2

Table 3 Qualitative (sign) analysis with regional eect

An exploratory analysis of hierarchical spatial interaction

265

266

S. Nazara et al.

regions. As a region, Javas growth is favorable to provinces in Sumatra and

actually Sulawesi, although for the latter it is not statistically signicant.
However, growth in Java will negatively impact Central and East Kalimantan, and Papua provinces. This is another conrmation of Javas economic hegemony in the Indonesian regional system.
Interaction among provinces in Kalimantan is mixed. West and Central
Kalimantan province seems to benet from share growth of other provinces
in the region. East Kalimantan, however, seems to be dierent. These
provinces are also negatively impacted by share growth in other regions in
the country. However, this region gives positive impact to most provinces in
Sumatra, Central Java and Central Sulawesi. A negative impact is brought
by this region to the eastern provinces of Maluku and Papua.
In Sulawesi, share growth in North and Central Sulawesi province seems
to have positive impact on provinces in the region. South Sulawesi, being the
richest province in the region, seems to bring a negative impact to North and
Central Sulawesi. Moreover, Central Sulawesi seems to benet from share
growth in Sumatra, Kalimantan and Eastern Islands. As a region, Sulawesi
is in a complementary mode with several of provinces in Sumatra, but primarily in competition with Kalimantan and the Eastern Islands.
The poorest province in the Eastern Islands, i.e., East Nusa Tenggara
(52), is denitely in competition with all other provinces. West Nusa
Tenggara also seems to be in competition with the rest of the provinces in the
region. Maluku is the only province that complements the others. Maluku
and Papua, however, are in competition with Kalimantan and Sulawesi. In
the case of Papua, the competition is with the rest of the nation. As a region,
Eastern Islands is in competitive mode with most of provinces in Indonesia,
except a few provinces in Sumatra.
It is important to comment about the interpretation of non-signicant
signs. While these signs econometrically denote coecients of zero values,
from the interregional-trade point of view obviously the no-interaction
argument is not sutainable. Such an argument is even weaker among adjacent or close regions as is suggested by the rst law of geography: all points
on the map are interrelated but close points are more so than distant ones.
Therefore, we would argue that non-signicant coecients merely denote
unclearly signicant patterns of relationship rather than absence of interaction. Non-signicant coecients do not dismiss the possibility of interregional interaction; they simple denote that no convincing positive or negative
relationships are apparent using this methodology.

Concluding remarks
This paper has demonstrated a methodological framework of regional
interaction analysis in an hierarchical regional structure, interpreted as the
existence of several layers of vertically-related spatial structure. The methodology extends the standard DendrinosSonis model that has been applied
to several cases. We use the strict hierarchical structure as the working
framework where the available observations for the estimation process are

An exploratory analysis of hierarchical spatial interaction

267

limited, and so derived the working specication from the structure. Applied
to the provincial-regional interaction in Indonesia, the model has reveals the
importance of taking into account both vertical and horizontal eects in
regional interaction analysis. Ignoring regional eects in the Indonesian case
leads to substantially dierent congurations of interaction phenomena.
This is especially true for the most-connected provinces and regions such as
those in Sumatra and Java. Other regions play important roles in these
regions interaction scheme. However, it seems that other regions eect is
not so much an important issue when one deals with the eastern part of
Indonesia. This paper has conrmed the hegemony of the western part in the
Indonesian regional economic structure. It was shown earlier that there is a
pattern where provinces in the western part, particularly in Sumatra, complement each other while at the same time engage in a competitive fashion
with the eastern part of Indonesia.
The interpretation of statistical signicance in the interaction coecient
remains a potentially contentious issue. We argue that statistical insignicance does not signify no interaction between two localities, but rather denotes unclear and indeterministic interaction patterns. As has already been
noted, the methodology developed in this paper deals with the regional
interaction in an indicative fashion, i.e., it shows the pattern but says nothing
about the determinants. Clearly the latter are important elements for a
complete understanding of regional interaction system as well as policy
making process. Further studies should address these issues; their interpretation may assist in developing interregional computer general equilibrium
(CGE) models that are sensitive to the nature of spacetime interactions
within an hierarchical system of regions. The work initiated by Resosudarmo
et al. (1999) oered one approach that combined a more exible spatial
system with macro and micro regions specied within a multi-region CGE
model for Indonesia to explore alternative decentralization policies. However, for the most part, the methodology of DendrinosSonis has not been
embraced by spatial CGE analysis; the application would seem to oer
signicant benets.
References
Anselin L (1988) Spatial econometrics: methods and models. Kluwer, Dordrecht
Barro R, Sala-i-Martin X (1995) Economic growth. McGraw-Hill, New York
Cashin P, Sahay R (1996) Regional economic growth and convergence in India. Finance
Dev 33(1):4952
Dendrinos D, Sonis M (1988) Nonlinear discrete relative population dynamics of the US
regions. Appl Math Comput 25:265285
Dendrinos D, Sonis M (1990) Chaos and socio-spatial dynamics. Springer, Berlin Heidelberg New york
Funck RH (2000) Hard and soft determinants of interregional competition. In: Batey PWJ,
Friedrich P (eds) Regional competition. Springer, Berlin Heidelberg New York,
pp 6685
Garcia JG, Soelistianingsih L (1998) Why do dierences in provincial incomes persist in
Indonesia? April 1998. Bull Indones Econ Stud 34(1):95120
Hewings GJD, Sonis M, Cuello FA, Mansouri F (1996) The role of regional interaction in
regional growth: competition and complementarity in the US regional system. Aust J
Reg Stud 2:133149

268

S. Nazara et al.

Johansson B (2000) Regional competition: endogenous and policy-supported processes. In:

Batey PWJ, Friedrich P (eds) Regional competition. Springer, Berlin Heidelberg New
York, pp 3465
Judge GG et al (1988) Introduction to the theory and practice of econometrics, 2nd edn.
Wiley, New York
Magalhaes A, Sonis M, Hewings GJD (2001) Regional competition and complementarity
reected in relative regional dynamics and growth of GSP: a comparative analysis of the
Northeast of Brazil and the Midwest States of the US. In: Joaquim JM Guilhoto,
Georey JD Hewings (eds) Structure and structural change in the Brazilian Economy.
Aldershot, Ashgate
Mankiw NG, Romer D, Weil DN (1992) A contribution to the empirics of economic
growth. Q J Econ 107:407437
Nazara S (1999) Provincial Convergence Within the Neoclassical Growth Theory in
Indonesia. In: Paper presented at 21st conference of northeast section of regional science
association international (RSAI), held by Cornell University and SUNY Binghamton,
Ithaca, May 78
Nazara N, Hewings GJDD, Sonis M (2001) Interregional competition and complementarity
in Indonesia. Discussion Paper, 01-T-02, Regional Economics Applications Laboratory,
University of Illinois, Urbana
Poot J (2000) Reections on local and economy-wide eects of territorial competition. In:
Batey PWJ, Friedrich P (eds) Regional competition. Springer, Berlin Heidelberg New
York, pp 205230
Richardson HW (1973) Regional growth theory. Macmillan, Bristol
Resosudarmo BP, Wuryanto LE, Hewings GJD, Saunders L (1999) Decentralization and
income distribution in the interregional Indonesian economy. In: Hewings GJD, Sonis
M, Madden M, Kimura Y (eds) Understanding and interpreting economic structure,
advances in spatial sciences. Springer, Berlin Heidelberg Germany
Sonis M, Hewings GJD, Jiemin G, Edison H (1997) Interpreting spatial economic structure:
feedback loops in the Indonesian interregional economy, 1980, 1985. Reg Sci Urban
Econ 27:325342
Sonis M, Hewings GJD (2000) Regional competition and complementarity: comparative
advantages/disadvantages and increasing/diminishing returns in discrete relative spatial
dynamics. In: Batey PWJ, Friedrich P (eds) Regional competition. Springer, Berlin
Heidelberg Newyork, pp 139158

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.