Spatial disparities across the regions of Turkey: an exploratory spatial ...

Ann Reg Sci (2010) 45:379–400 DOI 10.1007/s00168-009-0313-8 ORIGINAL PAPER

Spatial disparities across the regions of Turkey: an exploratory spatial data analysis Fatih Celebioglu · Sandy Dall’erba

Received: 22 August 2008 / Accepted: 18 June 2009 / Published online: 5 July 2009 © Springer-Verlag 2009

Abstract The aim of this paper is to perform an exploratory spatial data analysis on the growth and development level of the 76 Turkish regions over 1995–2001. While our choropleth maps indicate that the Western part of the country is significantly more developed than the East, the tools of spatial statistics reveal the presence of spatial dependence across provinces. The presence of heterogeneity is reflected in the distribution of Local Indicators of Spatial Association statistics. Overall, our results shed new light on the distribution of growth across Turkish regions and its relation with public investments and human capital, two indicators of development. JEL Classification

O18 · R11 · R58

1 Introduction Turkey is formally composed of several provinces used as administrative units. The definition of regions is only used for geographic classification purposes (for example Marmara, Aegean, Southeastern areas) and to cluster provinces according to their level of economic development. For instance, the provinces located in the Southeastern and Eastern Anatolia areas are known to be lagging behind in economic and social terms.

F. Celebioglu (B) Department of Economics, Faculty of Economics and Administrative Sciences, University of Dumlupinar, 43020 Kutahya, Turkey e-mail: [email protected] S. Dall’erba Department of Geography and Regional Development, University of Arizona, Harvill Building, Box #2, Tucson, AZ 85721-0076, USA e-mail: [email protected]

123

380

F. Celebioglu, S. Dall’erba

A couple of reasons have been highlighted in the past to justify the East–West divide that has marked the Turkish regional economies for a couple of decades (Ates et al. 2000; Balkir 1995; Gezici and Hewings 2004). They are, among others, inequalities in salaries (Elveren and Galbraith 2008), the dependence on agriculture and weakness of industrial sector (Akgungor 2003; Özaslan et al. 2006), the divide in the education level (Ozturk 2002; Tansel and Gungor 2000), the migratory flows from the east to the west (Keles 1985; Kirdar and Saracoglu 2007), ethnic terrorism (PKK terrorist organization, especially after the 1970s) (Feridun and Sezgin 2008), populist and misguided policies applied by governments (TUGIK Report 2008), and the lack of private investment in the east (Deliktas et al. 2008). Balkir (1995) proposed to classify the Turkish regional disparities into three categories: demographic disparities, including migration and urbanization, economic disparities, that include several of the components mentioned above, and the disparities in infrastructures, including the provision of public services. However, it is very difficult to assess the extent to which the phenomena above are the reason or the consequence for the divide observed within Turkey.1 What is more certain is that history has had an impact on these regional imbalances. They go back as far as the ages of the Ottoman Empire when the geographical location of Western Anatolia, especially the coastal areas like Izmir, Istanbul, and their hinterlands, gave it an important role in the external trade of the country. Since then, trade and industry have always been more developed there than in East Anatolia. With the foundation of the Republic of Turkey in 1923, the attention of the successive governments has shifted to Central Anatolia where the capital, Ankara, has been established. In order to lessen regional disparities, the Turkish authorities have implemented successive Five-Year Industrial Plans that promote investments in public infrastructures and private investments in Anatolia. A few decades later, the State Planning Organization was formally established to supervise the country’s regional policies. They were included for the first time in the Five-year National Development Plan of 1963–1967. Shortly after, the Second Plan (1968–1972) introduced the term “backward regions” and state-of-the-art planning techniques at the province level. Back then, the main aim of Turkish regional policies was already to ensure the development of the least favored eastern regions. The Third Plan period (1973–1977) included policies based on the intersectoral linkages between regions. This was the first time the concept of Priority Provinces for Development (PPD) was used. It consisted in focusing on industrial investments in the least favored areas. After 1980, the influence of export base theory (Arslan and Wijnbergen 1993) led to the decentralization of industrial activities from the metropolitan cities (Istanbul, Ankara, Izmir etc.) and thus to the industrial expansion of the provinces adjacent to the metropolitan regions (Gezici and Hewings 2004). During the 1980s, Turkey adopted a regional policy strategy grounded on two main components: public investments in infrastructures and financial incentives (such as tax break, lump-sum payments) for the private sector to locate in the backward areas. In this sense, the Turkish approach

1 We want to thank an anonymous referee for bringing this point to our attention.

123

Spatial disparities across the regions of Turkey

381

did not differ much from what was commonly done in the EU (see Dall’erba and Le Gallo 2008; Molle 2007) and in the US (Drabenstott 2006; Vadali 2008) during the same period. Turkish regional imbalances have been the subject of a vast amount of literature over the years. As Ertugal (2005) indicates, previous contributions conclude that re-distribution policies did not necessarily achieve the aim of increasing cohesion within the country. Atalik (1990) investigates regional income disparities over 1975–1985 and finds that inequalities have increased. Ates et al. (2000) try to assess the performance of the Southeastern Anatolia Project by measuring sigma and beta-convergence between the recipient regions and the rest of Turkey (excluding Marmara). They discover the convergence, among these (Eastern) regions. Berber et al. (2000) also conclude to the absence of beta and sigma convergence over 1975–1997. A couple of authors have decided to get a deeper insight into the Turkish regional dynamics by focusing on the distribution of income within each part of the country. Based on the polarization East versus West highlighted by Senesen (2002) and Dogruel and Dogruel (2003) discover that sigma-convergence took place over 1987–1999, but among the developed regions only. Gezici and Hewings (2003) split the country into three groups and use the Theil index to discover that interregional inequalities have increased while intra-regional inequalities have declined over 1980–1997. Their approach is innovative because they are the first ones to use spatial econometrics to assess beta-convergence at the regional level. This technique allows them to measure that growth spreads from one region to its neighbors. A year later, Gezici and Hewings (2004) confirmed that the absence of convergence is not only true at the provincial level, but for the functional regions as well. In addition, they conclude that the regions that received most of the regional development funds did not grow faster than the more developed regions of the West. Further works that reach the same results include Karaca (2004) and Yildirim and Ocal (2006). Overall, the level of internal imbalances in Turkey is much greater than the one experienced in most developed countries, but it is also less pronounced than what can be seen in most developing countries. When it comes to assessing the effect of regional development funds on growth, we believe that further work is needed. Indeed, it is actually hard to estimate whether regional disparities would have been even greater in the absence of funding in the poor areas. Following the spirit of the literature cited above, the aim of this paper is to investigate inequalities across the 76 Turkish regions over 1995–2001 by means of an exploratory spatial data analysis (ESDA). It is a set of techniques used to describe and visualize spatial distributions, identify atypical locations or spatial outliers, discover patterns of spatial association, clusters or hot spots, and suggest spatial regimes or other forms of spatial heterogeneity (Anselin 1988, 1999; Haining 1990). Several ESDA have been performed on the issue of regional inequalities. For instance, Dall’erba (2005), Ezcurra et al. (2007a,b), Le Gallo and Ertur (2003), Battisti and Di Vaio (2008) focus on the EU regions while Rey and Montouri (1999), Rey (2001) and Voss et al. (2006) focus on the US states. Ying (2000) investigates the case of China, Magalhaes et al. (2005) are interested in the Brazilian regions, Manfred et al. (2001) and Jensen et al. (2006) care about Austria and Chile, respectively, while van Oort and Atzema (2004) focus on the municipalities in the Netherlands. The only ESDA ever performed on the Turkish case is Gezici and Hewings (2003). However, the authors only focus on

123

382


the growth and income distributions, not on the factors at the origin of it, and use the former 67 provinces that composed the country before the 1990 definition of new administrative boundaries. As a result, this paper intends to fill this gap by analyzing regional income and growth with an extended time period and regional sample, and by focusing on two factors affecting regional inequalities (human capital and public investments). ESDA offers the opportunity to compare the differences between the eastern and western provinces by means of choropleth maps, box plots, and scatter plots and measure the extent of spatial autocorrelation. In the absence of input-output tables at the regional level in Turkey, ESDA is a second-best solution when it comes to modeling spillover effects across regions. This paper is organized as follows: the next section describes our data and their distribution. Section 3 introduces first the spatial weight matrix which captures the relationships between the regions of our sample and upon which the rest of ESDA relies. The discussion then turns to the measurement of global and local spatial autocorrelation by means of the Moran’s I, the Moran’s scatter plot and Local Indicators of Spatial Association (LISA) statistics. The last section will summarize our findings and provide some concluding remarks.

2 Data analysis Our dataset comes from the Turkish Statistical Institute2 and the State Planning Organization.3 They represent for each region the level of per capita income in 1995, the growth rate of per capita income over 1995–2001, the level of public investment divided by the GDP of the region in 1995 and the percentage of a region’s population with a university degree in 1995. All data are expressed in 1987 constant prices. The time frame we use (1995–2001) is limited by data availability. Indeed, data before and after that period simply do not exist at the regional level. As a result, even if Turkey currently counts 81 provinces, we are obliged to work with the 76 provinces that correspond to that period.4 While the period we have selected is limited by the data, it is worth providing a clearer picture of the economic situation of Turkey during this period.5 The two most recent financial crises experienced by the country took place in 1994 and 2001. Because dependence on the financial sector varies by Turkish province, it is important that the reader keeps in mind that the growth process we are reviewing in this paper may have been affected by these events.6 Other events that may influence our results is Turkey’s 1996 application for membership to the European Union and the implementation of the Seventh Development Plan (until 2000) to promote internal cohesion. 2 http://www.turkstat.gov.tr, Turkish Statistical Institute. 3 http://www.spo.gov.tr, State Planning Organization. 4 A map of the 76 provinces we use in this paper appears in the appendix. 5 We thank an anonymous referee for this advice. 6 Even if many other studies have also concluded to a duality in the growth process across Turkish provinces

(see the references in the introduction).

123


383

0.547 - 0.574 0.574 - 0.59 0.59 - 0.602 0.602 - 0.683

Fig. 1 Growth rate for period 1995–2001 in Turkey 17.369 - 18.04 18.04 - 18.462 18.462 - 18.795 18.795 - 19.78

Fig. 2 Log of GDP per capita (1995) in Turkey

However, note that the tools we are using in this paper do not allow us to draw any causation between these events and growth. 2.1 Mapping the distributions We start our analysis with the quartile maps of the distribution of our variables for each province. Figure 1 displays the distribution of the regional growth rate of per capita GDP relative to Turkey’s average over 1995–2001. The darker areas indicate a greater level of relative growth. It appears from this map that the distribution of growth is pretty random, which is an idea that will need to be assessed in the next section. Figure 2 displays the distribution of regional per capita GDP levels in 1995 relative to Turkey’s average. A clear core-periphery (or east/west) pattern appears in this map, with the core composed of the richest regions, whereas the peripheral regions are also the poorest ones. This confirms the findings of the various studies mentioned in the introduction above. In the Western part of the country, the coastal areas and the

123

384

F. Celebioglu, S. Dall’erba 442493.115 - 892661.151 892661.151 - 1490197.921 1490197.921 - 2670017.232 2670017.232 - 9016173.415

Fig. 3 Total public investment divided by province level GDP in Turkey (1995)

1.079 - 1.955 1.955 - 2.533 2.533 - 3.334 3.334 - 8.464

Fig. 4 University degree as % of population in Turkey (1995)

province of the capital city (district number 25 in the appendix map) are clearly better off than the rest of the country. This is because trade, industry, and tourism are developed in these areas. Figure 3 shows the distribution of public investments divided by the GDP of each province in 1995 relative to Turkey’s average. The east-versus-west pattern is not necessarily clear in this map. In the West, public investments seem randomly distributed, while they seem more clustered among the Eastern and Southeastern Anatolia provinces. When focusing on Figs. 2 and 3 together, it seems that public investments are not enough to counterbalance the low development of the Eastern provinces. As mentioned by (TUGIK Report 2008) the gap may come from private investors who favor Western Anatolia over Eastern and Southeastern Anatolia. Figure 4 may give us more insights into the East–West disparities mentioned so far. Indeed, as can be seen on this quartile map, the percentage of the population with a university degree (in 1995 and relative to Turkey’s average) is much greater in the

123

Spatial disparities across the regions of Turkey Table 1 Net migration rates of provinces in Turkey (2000 year)

Region or Province name

Istanbul

Net migration rate (%0) 46.09

Tekirdag, Edirne, Kirklareli

43.41

Balikesir, Çanakkale

11.56

Izmir

39.88

Aydin, Denizli, Mugla

35.83

Manisa, Afyon, Kütahya, Usak

−5.76

Bursa, Eskisehir, Bilecik

38.73

Kocaeli, Sakarya, Düzce, Bolu, Yalova

−9.47

Ankara

25.59

Konya, Karaman Antalya, Isparta, Burdur Adana, Mersin

The regional statistics on http:// www.turkstat.gov.tr

385

0.01 47.24 −6.94

Hatay, Kahramanmaras, Osmaniye

−30.23

Kirikkale, Aksaray, Nigde, Nevsehir, Kirsehir

−19.39

Kayseri, Sivas, Yozgat

−28.58

Zonguldak, Karabük, Bartin

−69.07

Kastamonu, Çankiri, Sinop

−39.83

Samsun, Tokat, Çorum, Amasya

−46.65

Trabzon, Ordu, Giresun, Rize, Artvin, Gümüshane

−26.11

Erzurum, Erzincan, Bayburt

−43.51

Agri, Kars, Igdir, Ardahan

−57.3

Malatya, Elazig, Bingöl, Tunceli

−27.06

Van, Mus, Bitlis, Hakkari

−39.49

Gaziantep, Adiyaman, Kilis

−22.91

Sanliurfa, Diyarbakir

−39.45

Mardin, Batman, Sirnak, Siirt

−46.78

West than in the East. The provinces of East Anatolia are known for their low literacy rate and for losing their most skilled workers to the rich provinces of West Anatolia provinces.7 Table 1 below indicates the net migration rate for the provinces of Turkey. Negative values indicate outmigration, while the positive values reflect in-migration. The results clearly indicate that the provinces which experience a high percentage of net out-migration rates are located in the eastern part of Turkey and along the Black sea. It is interesting to note from the table above that Kocaeli, Sakarya, Düzce, Bolu, and Yalova have negative net migration rate while they are among the rich regions.8 The reason is not due to economic factors. In August 1999, the Marmara Region 7 See the regional statistics on http://www.turkstat.gov.tr. 8 We thank an anonymous referee for raising this point.

123

386


Fig. 5 Growth rate for period of 1995–2001 in Turkey

which includes Kocaeli, Sakarya, and Yalova experienced an earthquake scaled 7.4 on Richter’s scale. Three months later, it was Düzce and Bolu’s turn to experience a second earthquake (this time rated 7.2). Following these tragedies, many inhabitants became homeless, many factories collapsed, and the cities’ unemployment rates soared. 2.2 Box plots The box plot is another tool of ESDA. Designed by Tukey (1977), box plots display five interesting pieces of information about a dataset: the lowest value, the lower quartile of the distribution (25% of the cumulative distribution, noted Q1), the median (Q2), the upper quartile (75% of the cumulative distribution, noted Q3), and the highest value. The median value is represented by the line in the center of the rectangular box. Because we are working with 76 provinces, the median represents the value of the 38th province in the ranked distribution (from the lowest to the highest value) of the variable under study. The second advantage of a box plot is to display the outliers which are defined as the values above or below a given multiple (arbitrarily set to 1.5 or 3 by Geoda) of the difference between the first and third quartiles. For instance, a lower outlier corresponds to a value below [Q1 − 1.5∗ (Q3 − Q1)] and an upper outlier is defined as a value above [Q3 + 1.5∗ (Q3 − Q1)]. The thin line on the upper part of box plots is called the hinge, here corresponding to the default criteria of 1.5 times the difference between the first and third quartiles (Thompson 2003). This tool has been commonly used in exploratory data analysis (see, for instance, Chambers et al. 1983; Leinhardt and Leinhardt 1980). The box plots of our variables appear in Figs. 6, 5, 7 and 8. They show that Bolu and Zonguldak are the only (upper) outliers in the distribution of provincial growth

123


387

Fig. 6 Log of Per Capita GDP in Turkey (1995)

Fig. 7 Total pub. invest. divided by province level GDP in Turkey (1995)

rate while only Kocaeli is the province with the highest value of per capita GDP in 1995, but it is not an outlier. There are four upper outliers in the distribution of total public investment (Ankara, Sivas, Hakkari, and Igdir) as well as in the distribution of the share of the population with a university degree (Ankara, Istanbul, Izmir, and Antalya). Canakkale is the province with the lowest value of growth rate in the period of 1995–2001. There are two lowest values (Agri and Mus) in the distribution of log of

123

388



per capita GDP in 1995. While Erzincan and Tunceli are the provinces with the lowest values in the distribution of total public investment, three other eastern provinces (Agri, Mus, and Sirnak) have the lowest values in the distribution of the share of the population with a university degree. In addition, Agri and Mus have also the lowest values per capita GDP for each year of period 1995–2001. Quartile maps and box plots are useful tools to get some insights into the distribution of a variable. However, they do not formally test whether the spatial distribution of a variable is random. For instance, the distribution of the per capita income and education level across Turkish provinces is marked by two distinct clusters (East versus West) as can be seen from Figs. 2 and 4 above. This observation needs to be tested by the formal tools of Exploratory Spatial Data Analysis. It starts with the definition of a spatial weight matrix and continues with the measurement of spatial autocorrelation. 3 Exploratory spatial data analysis (ESDA) 3.1 Spatial weight matrix A spatial weight matrix is the necessary tool to impose a neighborhood structure on a spatial dataset. As usual in the spatial statistics literature, neighbors are defined by a binary relationship (0 for non-neighbors, 1 for neighbors). All our work is performed under GeoDa. We have used two basic approaches for defining neighborhood: contiguity (shared borders) and distance. Contiguity-based weights matrices include rook and queen. Areas are neighbors under the rook criterion if they share a common border, not vertices. Distance-based weight matrices include distance bands and k nearest neighbors. Based on these two concepts, we decided to create four weight matrices to investigate the distribution of our variables of interest: a rook contiguity matrix, k_4

123


389

nearest neighbor matrix, k_7 nearest neighbor matrix, and W-90 miles matrix which defines as neighbors all the provinces located within a great circle distance with a cutoff of 90 miles. Due to space constraints, we present the k_7 nearest neighbor matrix only, below: ⎧ ⎨ wi j (k) = 0 if i = j ∗ w (k) = 1 if di j ≤ Di (k) and wi j (k) = wi j (k)/ j wi j (k) for k = 7 ⎩ ij wi j (k) = 0 if di j > Di (k)

(1)

where di j is great circle distance between centroids of region i and j and Di (k) is the seventh order smallest distance between regions i and j such that each region i has exactly 7 neighbors. Now that the weight matrix has been defined, we estimate a couple of spatial statistics that will shed some light on the spatial distribution of our variables. The most common of them is Moran’s I which is a measure of global spatial autocorrelation (Anselin 1988). 3.2 Moran’s I for global spatial autocorrelation Spatial autocorrelation refers to the correlation of a variable with itself in space. It can be positive (when high values correlate with high neighboring values or when low values correlate with low neighboring values) or negative (spatial outliers for high–low or low–high values). Note that positive spatial autocorrelation can be associated with a small negative value (e.g., −0.01) since the mean in finite samples is not centered on 1. Spatial autocorrelation analysis includes tests and visualization of both global (test for clustering) and local (test for clusters) Moran’s I statistic (Anselin et al. 2006). Global spatial autocorrelation is a measure of overall clustering and it is measured here by Moran’s I. It captures the extent of overall clustering that exists in a dataset. It is assessed by means of a test of a null hypothesis of random location. Rejection of this null hypothesis suggests a spatial pattern or spatial structure, which provides more insights into data distribution ,that what a quartile map or box plot does. For each variable, it measures the degree of linear association between its value at one location and the spatially weighted average of neighboring values (Anselin et al. 2007; Anselin 1995) and is formalized as follows: n It =

i=1

n

∗ j=1 wi j (k)x it x jt n i=1 j=1 x it x jt

n

(2)

∗

where wi j is the (row-standardized) degree of connection between the spatial units i and j and xi j is the variable of interest in region i at year t (measured as a deviation from the mean value for that year). Values of I larger (smaller) than the expected value E(I ) = −1/(n − 1) indicate positive (negative) spatial autocorrelation. In our study, this value is (−0.0133). There are different ways to draw inference here. The approach we use is a permutation approach with 999 permutations. It means that 999 re-sampled datasets were automatically created for which the I statistics are computed. The value

123

390


Table 2 Moran’s I and P-Value Variables

K_4

K_7

Rook

W-90 miles

Growth rate (1995–2001)

0.069 (0.137)

0.045 (0.129)

0.087 (0.076)

0.071 (0.141)

Log of per capita GDP (1995)

0.636 (0.001)

0.647 (0.001)

0.639 (0.001)

0.652 (0.001)

Public investment/GDP (1995)

0.089 (0.081)

0.105 (0.025)

0.103 (0.067)

0.075 (0.154)

University degree as % of population (1995)

0.400 (0.001)

0.384 (0.001)

0.391 (0.001)

0.396 (0.001)

P-values are into brackets Fig. 9 Standard deviation of per capita GDP

Fig. 10 Moran’s I values of per capita GDP

obtained for the actual dataset has then been compared to the empirical distribution obtained from these re-sampled datasets. The results of Moran’s I are presented in Table 2 below. All the results indicate a positive spatial autocorrelation, i.e., the value of a variable in one location depends positively on the value of the same variable in neighboring locations. For instance, when the per capita income in one province increases by 1%, the one of its neighbors increases by slightly more than 0.6%. Three out of our four variables of interest are significant (at 5%) with the k_7 nearest neighbor matrix. For this reason, this is the weight matrix we will use in the rest of our study. In fact, Turkey’s regional income divergence shows signs of decreasing spatial dependence from 1995 to 2001, as shown in Figs. 9 and 10 below. Therefore, the question of whether decreasing spatial effects are one of the reasons for regional

123


391

Fig. 11 Growth rate for period of 1995–2001 in Turkey

Fig. 12 Log of per capita GDP in Turkey (1995)

divergence needs to be explored. While this is not the point of this paper, a spatial econometric approach should shed some lights on this issue and should be considered by policy-makers when promoting convergence. 3.3 Moran’s scatter plot for global and local spatial autocorrelation The Moran scatter plot often complements Moran’s I because it provides an easy way to categorize the nature of spatial autocorrelation into four types: low–low (noted LL), low–high (LH), high–low (HL), and high–high (HH). The x-axis captures the value of a variable compared to the average value of the sample. For instance, all the points on the right-hand side of the figure mean (the vertical axis in the middle) that in the corresponding provinces, the value of the variable under study was above the sample’s

123

392


Fig. 13 Total pub. invest. divided by province level GDP in Turkey (1995)


average. On the other hand, the y-axis captures the average value of the same variable in the neighboring locations (with the neighbors being defined by the weight matrix). For instance, all the points below the mean (the horizontal axis in the middle of the figure) represent provinces of which neighbors display, on average, a lower value than the sample’s mean. The result of this approach is a figure with four windows which reflect the correlation between the relative (to the mean) value of a variable in one location and the relative value of the same variable in neighboring locations. For instance, the quadrant HH means a high value in the studied area and a high value in the neighboring areas. Regions located in quadrants I and III refer to positive spatial autocorrelation, i.e., the spatial clustering of similar values, whereas quadrants II and IV represent negative spatial autocorrelation, i.e., the spatial clustering of dissimilar values. Note also that

123


393

the link between a scatter plot and Moran’s I is reflected by a line of which slope is the value of Moran’s I statistic.9 Figures 11, 12, 13 and 14 below display the Moran scatter plots of our variables of interest. For both the per capita income and education level, positive spatial autocorrelation is reflected by the value of Moran’s I and the fact that most of the provinces are located in quadrants HH and LL with HH displaying a cluster a Western provinces while LL shows a cluster of Eastern provinces. Once again, it reflects the dualistic structure of Turkey’s provinces. The other two variables (growth and public investment) are more spread around their mean and across the four quadrants. The slope indicates that spatial autocorrelation is less high (actually not significant for growth) for these two variables than for the two previous ones. Table 3 indicates the name of the regions according to their distribution in the Moran scatterplot quadrants. Positive spatial autocorrelation is reflected by the fact that most provinces are in the high–high and low–low quadrants. More precisely, for the per capita GDP and university degree variables, the Western provinces are mostly High–high areas, while the Eastern ones are low-low. Obviously, the low–high and high–low quadrants contain fewer provinces. 3.4 LISA statistics for local spatial autocorrelation Local Indicators of Spatial Association statistics measure, by definition, the presence of spatial autocorrelation for each of the location of our sample. It captures the presence or absence of significant spatial clusters or outliers for each location. Combined with the classification into four types defined in the Moran scatter plot above, LISA indicates significant local clusters (high–high or low–low) or local spatial outliers (high–low or low–high). The average of the Local Moran statistics is proportional to the Global Moran’s I value (Anselin 1995; Anselin et al. 2007). Anselin (1995) formulated the local Moran’s statistics for each region i and year t as the follows: xi wi j x j with m 0 = xi2 /n (3) Ii = m0 j

i

where wi j is the elements of the row-standardized weights matrix W and xi (x j ) is the observation in region i ( j). The significant results (at 5%) of the LISA statistics are presented in Table 3. Their significance level is based on a randomization approach with 999 permutations of the neighboring provinces for each observation. The randomization approach is used in the context of a numeric permutation approach to describe the computation of pseudo significance levels for global and local spatial autocorrelation statistics. In order to determine how likely it would be to observe the actual spatial distribution at hand, the actual values are randomly 9 Numerous articles have used the Moran’s scatterplot to capture spatial autocorrelation in the distribution of their variables. See, among many others, Dall’erba (2005), Rey and Montouri (1999) and Ezcurra et al. (2007a).

123

LL

LH

HL

123

Public Investment divided by GDP (1995)

Univ. degree as % of population (1995)

Log of province level per capita GDP (1995)

Adiyaman, Agri, Ardahan, Batman, Afyon, Aksaray, Bartin, Cankiri, Ankara, Antalya, Aydin, Balikesir, Isparta, Yozgat Bayburt, Bingol, Bitlis, Diyarbakir, Bilecik, Burdur, Bursa, Canakkale, Erzincan, Erzurum, Giresun, Denizli, Edirne, Eskisehir, Icel, Gumushane, Hakkari, Igdir, Izmir, Karaman, Kastamonu, K.Maras, Kars, Kirsehir, Malatya, Kirklareli, Konya, Kocaeli, Mardin, Mus, Ordu, S.Urfa, Siirt, Kutahya, Manisa, Mugla, Nigde, Sinop, Sirnak, Sivas, Tokat, Sakarya, Tekirdag, Usak, Istanbul, Tunceli, Van Zonguldak, Adana, Bolu Adiyaman, Agri, Aksaray, Ardahan, Afyon, Bartin, Cankiri, Gumushane, Ankara, Antalya, Aydin, Balikesir, Karaman, Kastamonu, Kirikkale, Artvin, Batman, Bayburt, Bingol, Bilecik, Burdur, Bursa, Canakkale, Konya, Kutahya, Manisa, Nigde, Bitlis, Corum, Diyarbakir, Denizli, Edirne, Eskisehir, Isparta, Sakarya, Usak, Zonguldak Erzurum, Giresun, Hakkari, Hatay, Izmir, Kirklareli, Kocaeli, Mugla, Igdir, K.Maras, Kars, Kirsehir, Tekirdag, Istanbul, Bolu Mardin, Mus, Nevsehir, Ordu, Rize, Samsun, S.Urfa, Siirt, Sinop, Sirnak, Sivas, Tokat, Tunceli, Van, Yozgat, G.Antep Adiyaman, Amasya, Artvin, Bartin, Agri, Ankara, Ardahan, Bitlis, Afyon, Aksaray, Aydin, Balikesir, Eskisehir, Hakkari, Igdir, Kars, Bingol, Burdur, Canakkale, Corum, Batman, Bilecik, Cankiri, Giresun, Kastamonu, Konya, Malatya, Kocaeli, Mus, Sirnak, Van, Bolu Denizli, Edirne, Elazig, Erzincan, Mardin, Sakarya, Samsun, Siirt, Hatay, K.Maras, Karaman, Tokat, Zonguldak Kirikkake, Kirklareli, Kirsehir, Kutahya, Manisa, Nevsehir, Nigde, Ordu, Rize, Tekirdag, Trabzon, Tunceli, Usak, G.Antep, Adana

Antalya, Bayburt, Bursa, Diyarbakir, Erzurum, Gumushane, Icel, Isparta, Izmir, Kayseri, Mugla, S.Urfa, Sinop, Sivas, Yozgat, Istanbul

Amasya, Elazig, Erzincan, Icel, Kayseri, Malatya, Trabzon, Adana

Amasya, Artvin, Corum, Elazig, Hatay, Kayseri, Kirikkale, Nevsehir, Rize, Samsun, Trabzon, G.Antep

Adiyaman, Amasya, Ankara, Afyon, Batman, Denizli, Edirne, Aksaray, Antalya, Aydin, Balikesir, Growth rate Agri, Bayburt, Bingol, Bitlis, Ardahan, Artvin, Bartin, Bilecik, Isparta, K.Maras, Kirklareli, Burdur, Bursa, Canakkale, Corum, for period of Eskisehir, Giresun, Gumushane, Cankiri, Diyarbakir, Erzincan, Kirsehir, Manisa, Mugla, Nevsehir, Elazig, Icel, Izmir, Karaman, 1995–2001 Hakkari, Igdir, Kastamonu, Erzurum, Hatay, Kars, Malatya, Nigde, Sinop, Tokat Kirikkale, Konya, Kutahya, Kayseri, Kocaeli, Mus, Ordu, Rize, Mardin, Sakarya, Siirt, Sirnak, Samsun, Sanliurfa, Tekirdag, Usak, Sivas, Tunceli, Van, Zonguldak, Trabzon, Adana Yozgat, Istanbul, G.Antep Bolu

HH

Table 3 Distribution of spatial autocorrelation

394 F. Celebioglu, S. Dall’erba


395

Table 4 Provinces with significant LISA statistics at 5% (with spatial weight matrix k_7 nearest neighbors) G 95-01 LN 95 Tot.Pb.In Univ.Deg.

G 95-01 LN 95 Tot.Pb.In Univ.Deg.

Adıyaman

Karaman

Afyon

Kars

A˘grı

LL

HH

LL

Aksaray

HH

LL

Kayseri

Amasya Ankara

LL

Kastamonu Kırıkkale

LH

Kırklareli

Antalya

HL

HH

HH

Kır¸sehir

Ardahan

LL

Artvin

HL

HH

LL

Kocaeli

Aydın

HH

HH

Kütahya Malatya

HH

Konya

LH HH

Balıkesir

LL

HH

HH

Bartın

LH

LH

LH

Manisa

HH

LH

LL

LL

Mardin

LL

LL

Batman

HL

Bayburt

LL

Mu˘gla

HH

HH

Bilecik

HH

HH

Mu¸s

LL

LL

Bingöl

LL

LL

Bitlis

LL

LL

Nev¸sehir ˙gde NI˘

HH

Ordu

HH

Sakarya

Burdur Bursa

LL

Çanakkale Çankırı

HH

Rize

HH LH

Sanlıurfa ¸

Denizli

Siirt

Edirne

HL

Elazı˘g

LL

LL

HH

HH

HL

Eski¸sehir

HH

HH LL

Sırnak ¸

LL

LL HL

HH

HH

Trabzon

LH

Tunceli

HL LL

U¸sak

Hakkari

LL

HH

LL

Hatay icel/mersin LL LL

HH

HH

Van Yozgat ˙Istanbul

HL

I˘gdır

Kahramanmara¸s

LL

Tokat

Gümü¸shane

Isparta ˙Izmir

LL

Sinop

Tekirda˘g LL

Giresun

LH

Sivas

Erzincan Erzurum

HH

Samsun

Çorum Diyarbakır

HL LH

LL

Zonguldak

HH

Gaziantep

LL

HH

LL

HL LL

HH LH

Adana Bolu

HH

123

396

F. Celebioglu, S. Dall’erba non-significant High-High Low-Low Low-High High-Low

Fig. 15 Cluster map (Growth rate 1995–2001) non-significant High-High Low-Low Low-High

Fig. 16 Cluster map (Log of per capita GDP 1995)

reshuffled over space 999 times. This table point out that some Eastern provinces (Agri, Ardahan, Batman, Bingol, Bitlis, Diyarbakir, Hakkari, Igdir, Kars, Mardin, Mus, Siirt, Sirnak, and Van) display LL-type autocorrelation for both their per capita GDP and university degree, but also HH-type autocorrelation for public investment (in Agri, Ardahan, Hakkari, Igdir, Kars, and Van). Once again, this result reflects the will of the authorities to counterbalance poverty in the East. Following the results displayed in Table 4, we also provide the LISA maps (Figs. 15, 16, 17, and 18) as a visual representation of these results. 4 Conclusions The aim of this paper has been to perform an exploratory analysis of the economic disparities across 76 Turkish provinces. We have investigated the spatial distribution

123


397

non-significant High-High Low-Low High-Low

Fig. 17 Cluster map (Total public investment 1995) non-significant High-High High-Low

Fig. 18 Cluster map (University degree 1995)

of growth over 1995–2001, of the per capita GDP, total public investments, and the share of the population with a university degree in 1995 across these provinces. First, our quartile maps have revealed the gap between East and West when it comes to per capita GDP and education levels. Second, the Box plots showed that West Anatolia and the coastal area provinces are upper outliers in the distribution of almost all our variables. Some provinces of East Anatolia (Hakkari and Igdir) appear as outliers only in the distribution of public investments, thus reflecting the efforts of the government to counterbalance internal disparities. When we measure spatial autocorrelation by means of Moran’s I, our results indicate positive (and significant) global autocorrelation for all our variables except growth, and thus indicating the geographical location of a province influences its level of income, public investment, and education. These results are corroborated by the corresponding Moran’s Scatterplots that display most of the eastern provinces in the low–low quadrant and the western ones in the high–high quadrant. Finally, LISA statistics confirm the significant presence of local spatial autocorrelation and highlight spatial

123

398


heterogeneity in the form of two distinct spatial clusters of high and low values of per capita income. Overall, these results confirm the dualistic structure of Turkey’s economic geography, as many previous studies had showed. However, our results also show that this form of spatial heterogeneity goes along with the presence of spatial autocorrelation among provinces. In addition, this paper has explored for the first time the relationship between the spatial distribution of economic development and two of the factors at the origin of it (human capital and public investments). Based on our results, we recommend fighting internal imbalances by promoting investments in education and the training of unemployed in the poorest areas. We also believe that developing the social and economic conditions in the East should be on the government’s priority list so that migration to the West and eventually ethnic terrorism in the East will be reduced to a minimum. In addition, if future results indicate that spatial autocorrelation continues to decrease over time, as in our Fig. 10, we recommend policies that support a sector rather than a province in particular. Because Eastern Turkey’s provinces tend to display the same sectoral composition, this should guarantee to promote the development of several provinces at once. Finally, given the exploratory nature of the work above, future research will consist in using a spatial econometric approach in order to formally model the interactions between the growth of the Turkish provinces, their socio-economic characteristics, and the regional development measures to the least-developed areas. Appendix

2

29

28

43

27

4

3

1

6

5

25

31

19

12

33

13

21

15

24

23

73 74

52

60

53

67 66

38 34

22 14

72 68

54 39

16

10

59

55

46

32

20

11

58

41 40

71

69

57

56

45

42 30

17

9

8

7

70

44 26

18

47 37

35

61

51 50

64 62

75

65 63

76

48 36

49

The regions are the following: Edirne (1), Kirklareli (2), Tekirdag (3), Istanbul (4), Kocaeli (5), Sakarya (6), Canakkale (7), Balikesir (8), Bursa (9), Izmir (10), Manisa (11), Kutaya (12), Aydin (13), Mugla (14), Denizli (15), Usak (16), Bilecik (17), Bolu (18), Eskisehir (19), Afyon (20), Isparta (21), Burdur (22), Antalya (23), Konya (24), Ankara (25), Cankiri (26), Zonguldak (27), Bartin (28), Kastamonu (28), Corum (30), Kirikkale (31), Kirsehir (32), Aksaray (33), Nigde (34), Karaman (35), Mersin (36), Adana (37), Kayseri (38), Nevsehir (39), Yozgat (40), Tokat (41), Amasya (42), Sinop (43), Samsun (44), Ordu (45), Sivas (46), K.Maras (47), G.Antep (48), Hatay (49), S.Urfa (50), Adiyaman (51), Malatya (52), Elazig (53), Tunceli (54), Erzincan (55),

123


399

Giresun (56), Trabzon (57), Gumushane (58), Bayburt (59), Bingol (60), Diyarbakir (61), Mardin (62), Sirnak (63), Batman (64), Siirt (65), Bitlis (66), Mus (67), Erzurum (68), Rize (69), Artvin (70), Ardahan (71), Kars (72), Igdir (73), Agri (74), Van (75), Hakkari (76).

References Akgungor S (2003) Exploring regional specializations in turkey’s manufacturing industry. Regional studies association international conference, Pisa, Italy, April 12–15 Anselin L (1988) Spatial econometrics: methods and models. Kluwer, Dordrecht Anselin L (1995) Local indicator of spatial association–LISA. Geogr Anal 27:93–115 Anselin L (1999) Interactive techniques and exploratory spatial data analysis. In: Longley PA, Goodchild MF, Maguire DJ, Rhind DW (eds) Geographic information system: principles, techniques, management and applications. Wiley, New York Anselin L, Sridharan S, Gholston S (2007) Using exploratory spatial data analysis to leverage social indicator databases: the discovery of interesting patterns. Soc Indic Res 82:287–309 Anselin L, Syabri I, Kho Y (2006) GeoDa: an introduction to spatial data analysis. Geogr Anal 38:5–22 Arslan I, Wijnbergen S (1993) Export incentives, exchange rate policy and export growth in turkey. Rev Econ Stat 75(1):128–133 Atalik G (1990) Some effects of regional differentiation on integration in the European community. Papers Reg Sci Assoc 69:11–19 Ates S, Erk N, Direkci T (2000) Convergence and growth within GAP (South Eastern Anatolia Project) Region and Overall Turkey’s Regions. http://www.econturk.org/sanli8.pdf, International METU Conference in Economics IV, September 13–16, Ankara Balkir C (1995) Less development regions and regional development policies in Turkey. European Urban and Regional Studies, SAGE Publications, 1995/2; pp. 253–264 Battisti M, Di Vaio G (2008) A spatially filtered mixture of β-convergence regressions for EU regions, 1980–2002. Emp Econ 34:105–121 Berber M, Yamak R, Artan S (2000) Turkiye’de yakinlasma hipotezinin bolgeler bazinda gecerliligi uzerine ampirik bir calisma: 1975–1997, Ulusal Bolge Bilimi Kongresi—Bildiri Kitabi 9:51–59, Trabzon Chambers J, William C, Beat K, Tukey P (1983) Graphical methods for data analysis. Wadsworth, Belmont Dall’erba S (2005) Distribution of regional income and regional funds in Europe 1989–1999: an exploratory spatial data analysis. Ann Reg Sci 39:121–148 Dall’erba, Le Gallo J (2008) Regional convergence and the impact of structural funds over 1989–1999: a spatial econometric analysis. Papers in regional science (forthcoming) Deliktas E, Önder AÖ, Karadag M (2008) The spillover effects of public capital on the Turkish private manufacturing industries in the Geographical Regions. Ann Reg Sci. doi:10.1007/s00168-008-0220-4 Dogruel F, Dogruel S (2003) Türkiye’de bölgesel gelir farkliliklari ve büyüme. In: Iktisat Uzerine Yazilar—Korkut Boratav’a Armagan, Iletisim Yayinlari, Istanbul Drabenstott M (2006) Rethinking federal policy for regional economic development. Econ Rev 91(1): 115–142 Elveren AY, Galbraith JK (2008) Pay inequality in Turkey in the Neo-Liberal Era: 1980–2001. University of Texas Inequality Project Working Paper No. 49, April 21 Ertugal E (2005) Strategies for regional development: challenges facing turkey on the road to EU membership. Turkish Policy Quarterly 4(3), Fall Ezcurra R, Iraizoz B, Pascual P, Rapún M (2007a) Spatial disparities in the European agriculture: a regional analysis. Applied Economics, pp 1–16 Ezcurra R, Pascual P, Rapún M (2007b) Spatial disparities in the European union: an analysis of regional polarization. Ann Reg Sci 41:401–429 Feridun M, Sezgin S (2008) Regional underdevelopment and terrorism: the case of south eastern Turkey. Defence Peace Econ 19(3):225–233 Gezici F, Hewings G (2003) Spatial analysis of regional inequalities in Turkey. In: 43rd congress of european regional science, 27–30 August 2003, Jyvaskyla, Finland Gezici F, Hewings G (2004) Regional convergence and the economic performance of peripheral areas in Turkey. RURDS 16(2):113–132

123

400


Haining R (1990) Spatial data analysis in the social and environmental sciences. Cambridge University Press, Cambridge Jensen L, Yang TC, Matthews SA (2006) Spatial inequality in Chile: stasis and change. Population Research Institute, The Pennsylvania State University Karaca O (2004) Türkiye’de Bölgeler Arasi Gelir Farkliliklari: Yakinsama Var mi? Türkiye Ekonomi Kurumu (Turkish Economy Association), http://www.tek.org.tr Keles R (1985) The effects of external migration on regional development in Turkey. In: Hudson R, Lewis J (eds) Uneven development in southern Europe: studies of accumulation, class, migration and the state. Methuen Publishing, London, pp 54–75 Kirdar M, Saracoglu S (2007) Migration and regional convergence: an Empirical investigation for Turkey. http://mpra.ub.uni-muenchen.de/2648/, MPRA Paper No. 2648 Le Gallo J, Ertur C (2003) Exploratory spatial data analysis of the distribution of regional per capita GDP in Europe, 1980–1995. Papers Reg Sci 82(2):175–201 Leinhardt G, Leinhardt S (1980) Exploratory data analysis: new tools for the analysis of empirical data. Rev Res Educ 8:85–157 Magalhaes A, Hewings G, Azzoni CR (2005) Spatial dependence and regional convergence in Brazil. Investigaciones Regionales 6:5–20 Manfred FM, Fröhlich J, Gassler H, Varga A (2001) The role of space in the creation of new technological knowledge in Austria: an exploratory spatial data analysis. In: Manfred FM, Fröhlich J (eds) Knowledge, complexity and innovation systems. Springer, Berlin, pp 124–145 Molle W (2007) European cohesion policy. Routledge, London Özaslan M, Dinçer B, Özgür H (2006) Regional disparities and territorial indicators in Turkey: Socio-economic development index (SEDI). The department of planning and regional development, university of thessaly and the Greek section of ERSA, 46th Congress of the European Regional Science Association (ERSA): Enlargment, Southern Europe, and the Mediterranean, Volos, Greece, 30 August to 3 September Ozturk I (2002) Economic and social issues of east and southeast Turkey: policy implications. http://mpra. ub.uni-muenchen.de/1582/, MPRA Paper No. 1582 Rey SJ (2001) Spatial empirics for economic growth and convergence. Geogr Anal 33:194–214 Rey SJ, Montouri BD (1999) US regional income convergence: a spatial econometric perspective. Reg Stud 33:143–156 Senesen U (2002) Türkiye’de Bölgesel Gelir Dagiliminda Son Gelismeler. 10. Ulusal Bölge Bilimi-Bölge Planlama Kongresi, 17-18 Ekim 2002, ITÜ Mimarlik Fakültesi, Istanbul State Planning Organization (SPO), First five-year national development plan (1963–1967), http://www. spo.gov.tr State Planning Organization (SPO), Second five-year national development plan (1968–1972), http://www. spo.gov.tr State Planning Organization (SPO), Third five-year national development plan (1973–1977), http://www. spo.gov.tr Tansel A, Gungor AD (2000) Provincial inequalities in school enrollments in Turkey. Economic Research Forum Working, Paper No. 2003 Thompson JR (2003) John Tukey (1915–2000): Deconstructing Statistics. In: Barry AB, Edward JW (eds) Proceedings of the Sixth US Army conference on applied statistics, ACAS00/ThompsonJames/ThompsonJames.pdf, Accessed 21 August 2008 TUGIK (Turkish Young Businessmen’s Confederation) Report (2008) (prepared by Prof. Mustafa Sonmez) Do˘gu ve Güneydo˘gu Anadolu’nun Geli¸simi ˙Için Çözüm Önerileri. http://www.tugik.org.tr/ tuganasayfa.wc Tukey JW (1977) Exploratory data analysis. Addison-Wesley, Reading Vadali S (2008) Toll roads and economic development: exploring effects on property values. Ann Reg Sci 42:591–620 Van Oort FG, Atzema OALC (2004) On the conceptualization of agglomeration economies: the case of new firm formation in the dutch ICT sector. Ann Reg Sci 38:263–290 Voss PR, Long DD, Hammer RB, Friedman S (2006) County child poverty rates in the US: a spatial regression approach. Popul Res Policy Rev 25:369–391 Yildirim J, Ocal N (2006) Income inequality and economic convergence in Turkey. Trans Stud Rev 13(3):559–568 Ying LG (2000) Measuring the spillover effects: some Chinese evidence. Papers Reg Sci 79:75–89

123