1 Regional Inequality and Regional Convergence Clubs -
Analyzing Reunited Germany
Master thesis, MSc International Economics & Business University of Groningen, Faculty of Economics and Business
DATE 9/01/2018
CAROLINE SCHULZ Student number: 2596113 Email: c.s.schulz@student.rug.nl
2 Regional Inequality and Regional Convergence Clubs -
Analyzing Reunited Germany
Abstract
This research investigates on regional inequalities and convergence in Germany. Using a panel data set consisting of 402 regions and covering the years 2000 - 2014, the club clustering by Phillips and Sul (2009) is applied and convergence clubs are created based on the idea of Bartkowska and Riedl (2012). Following the idea of inclusive growth the analysis includes economic as well as socioeconomic factors aiming to check on their relevance for the creation of clusters and regional disparities in Germany. Overall convergence of regions as well as convergence between East and West regions in Germany can be rejected. The results of the multinomial logistic regression approach reveal persistent disparities between regions located in East and West Germany, as well as between urban and rural regions.
3 CONTENT
1. INTRODUCTION ... 5
2. THEORETICAL BACKGROUND ... 6
2.1 Regional Inequality ... 6
2.2 Regional (Economic) Inequality and Total Factor Productivity ... 7
2.3 Inclusive Growth ... 8 2.4 Club Convergence ... 8 2.4.1 Economic Factors ... 9 2.4.2 Socioeconomic Factors ... 10 2.5 Exogenous Factors/Effects ... 11 2.5.1 Agglomeration Effects ... 11
2.5.2 Specific Characteristics of Germany ... 12
3. METHODOLOGY ... 14
3.1 Data and Panel Construction... 14
3.2 Club Construction ... 14
3.2.1 Log T Regression Test ... 15
3.2.2 Regional Convergence Clubs ... 15
3.3 Variables ... 16
3.4 Multinomial Logistic Regression ... 17
3.5 Baseline Model ... 17
4. ANALYSIS AND FINDINGS ... 19
4.1 Coefficient of Variation Analysis ... 19
4.2 Cluster Analysis ... 21
4.2.1 Description/Descriptive Statistics of Clusters ... 21
4.2.2 Results of Cluster Analysis ... 22
4.3 Regression Analysis ... 23
5. DISCUSSION ... 27
5.1 Patterns of Regional Inequality ... 27
5.2 Regional Inequality Policies ... 29
6. CONCLUSION... 30
REFERENCES ... 32
APPENDIX ... 36
Appendix 1: State and Region Allocation ... 36
4
Appendix 3: Club Distribution ... 40
Appendix 3a: Distribution Club 1 ... 41
Appendix 3b: Distribution Club 2... 42
Appendix 3c: Distribution Club 3 ... 43
Appendix 3d: Distribution Club 4... 44
Appendix 3e: Distribution Club 5 ... 45
Appendix 4: Distribution of Unemployment Rates in Germany ... 46
Appendix 5: Coefficient of Variation Results ... 47
Appendix 6: Overview and Explanation of Variables ... 48
Appendix 7: Descriptive Statistics of Variables per Club ... 50
5 1. INTRODUCTION
Since the fall of the Berlin Wall in 1989 and the subsequent reunification of Germany in 1990 politicians and authorities have strived to reunite former East and West Germany not only politically, but also according to economic and social standards. In 1990 promises were made, also captured in the German constitution, for regional equality and equality across the united country to be achieved, referring for instance to the improvement of economic growth and living standards in the poorer regions (Berentsen, 2006). Since then high amounts of funds were spent, meant to help closing the existing gaps (Frick & Goebel, 2008).
The question arises whether the measures taken have succeeded and former Eastern regions converged to Western German levels or whether there are still inequalities to be found in reunited Germany.
Multiple recent studies have found evidence for persisting social and economic inequalities between and also within countries of the European Union (e.g. Barrios & Strobl, 2009; Bartkowska & Riedl, 2012; Fiaschi, Gianmoena, & Parenti, 2017; Von Lyncker & Thoennessen, 2017), some of them particularly addressing Germany (e.g. Berentsen, 2006; Burda & Severgnini, 2017; Granato, Haas, Hamann, & Niebuhr, 2015; Kubis, Brachert, & Titze, 2012; Maseland, 2014; Uhlig, 2008). The distribution of income per capita was found not to follow a common growth path but rather to establish cluster patterns (Bartkowska & Riedl, 2012). This holds for economies across different continents as well as integrated markets like economies in Western Europe (Corrado, Martin, & Weeks, 2005).
In this research we want to investigate on the development, the current state and the existence of regional inequalities in Germany, and potential patterns the inequality could be assigned to. We examine whether there is still a recognizable and statistically significant East West pattern to be found for German regions, or evidence for alternative patterns that emerged in the course of time after reunification. The underlying research question is thus
“Is there regional inequality in reunited Germany, and what are possible patterns and underlying factors that can explain regional inequality?
We do so and aim to contribute to existing research by applying and testing for the club convergence hypothesis for German regions. The club convergence hypothesis refers to economies similar in their structural characteristics but diverse in initial conditions moving to different steady state equilibria (Bartkowska & Riedl, 2012). Economies moving to the same steady state equilibrium build together a convergence club (Galor, 1996). We will apply the club clustering algorithm by Phillips and Sul (2007) to check for the existence and distribution patterns of different convergence clubs in Germany.
6 we perform a multinomial logistic regression analysis, which focuses on economic as well as socioeconomic indicators.
This paper is structured as follows: Firstly, relevant concepts and findings from academic literature will be discussed. Afterwards, the method together with the applied model for the analysis and its composition will be explained, which is followed by the econometric analysis using the panel data set on German regions. Finally the results of the analysis will be presented and discussed, and conclusions be drawn pointing out the most important findings of this research.
2. THEORETICAL BACKGROUND 2.1 Regional Inequality
Regional inequalities refer to disparities between regions regarding different dimensions, for instance their economic performance, growth and living standards (Rodríguez-Pose & Tselios, 2015). Multiple scholars, among others Maguire (2016), have observed that e.g. GDP per capita dispersion across regions within one country can be larger than the dispersion across regions of different countries. This raises the question of how regional inequalities actually evolve and persist over time.
Kuznets (1955) was one of the first to conduct research on the development of regional inequalities. He suggested that regional inequalities are likely to appear in a non-linear way arguing that the industrialization process firstly leads to a rise followed by a fall of income inequalities (Kuznets, 1955). Following this reasoning, Kim and Margo (2003) apply the idea of regional inequalities on US regions. Their findings suggest that industrialization has led to an increase in regional income disparities within the US, as manufacturing was concentrated in the Northern parts and the specialization in agricultural activities in the Southern parts of the US (Kim & Margo, 2003).
7 capital (Granato et al., 2015; Rodriguez-Pose & Vilalta-Bufi, 2005; Südekum, 2008), innovation (Lee & Rodriguez-Pose, 2013), labour productivity (Combes, Lafourcade, Thisse, & Toutain, 2011), total factor productivity (Beugelsdijk, Klasing, & Milionis, 2017; Burda & Severgnini, 2017), income, consumption and wealth (Brzozowski, Gervais, Klein, & Suzuki, 2010; Heathcote, Perri, & Violante, 2010). Following the idea of inclusive growth, that is explained in a later section, this research aims to combine economic as well as socioeconomic factors in the analysis of regional inequalities in Germany.
2.2 Regional (Economic) Inequality and Total Factor Productivity
Recent research suggests large disparities in the level of economic development within countries (Beugelsdijk et al., 2017). The study of Gennaioli, La Porta, Lopez-de Silanes and Shleifer (2013), covering data for regions across the world, reveals that on average the income per worker is more than four times higher in a country’s richest regions compared to the same country’s poorest regions. Although a lot of research has been done on regional economic and income differences, and governments strive to diminish regional inequalities, regional income differences are persistent (Beugelsdijk et al., 2017).
Economic growth can enhance divergence across different regions (Barrios & Strobl, 2009). In their research Barrios and Strobl (2009) analyze the development of regional inequalities and conclude that regional inequalities are likely to increase on average with higher levels of national GDP per capita, and decrease after a certain relative level of national GDP per capita has been achieved. The European Commission (2004) could confirm these theories in their observations of former Soviet Union member states of the European Union: countries like Poland, the Czech Republic, Slovakia and Hungary, have experienced a fast national catching-up in GDP per capita since 1995, accompanied by increasing regional inequalities within these countries. Similarly, regions belonging to former East Germany were meant to level standards of former West Germany, which will be discussed in more detail in section 2.5.2.
Total Factor Productivity
8 On the country level TFP (rather than traditional factor endowments) is said to explain even more than 90% of growth rate disparities (Easterly & Levine, 2001). Thus, not only physical factors should be considered but also the efficiency with which they are being used, when analyzing regional economic development and its implications for regional inequality.
However, research confirmed the systematic depreciation of physical, human and match capital for formerly planned economies, like those of the Eastern bloc, including the East German economy (Blanchard & Kremer, 1997; Burda & Hunt, 2001; Roland & Verdier, 1999), making it problematic to use TFP as a measure for economic development and growth, and an instrument for comparison (in this research) (Burda & Severgnini, 2014, 2017). Therefore, Burda and Severgnini (2017) propose alternative measures and suggest to apply the available data for measures like trend regression estimates or moving averages. Keeping in mind the mentioned relevance of not only physical factors, but also the efficiency of their use, as well as the earlier discussed importance of socioeconomic indicators, this research will apply the concept of inclusive growth and its determinants for the analysis of regional inequalities in Germany to create a coherent picture of the latest developments and situation.
2.3 Inclusive Growth
Literature defines inclusive growth as broad-based economic growth across sectors accessing a large part of a respective country’s labour force (Ianchovichina & Lundstrom, 2009). The European Union aims to achieve inclusive growth by supporting high employment, which should lead to social and territorial cohesion (European Commission, 2010). This implies that people contribute to economic growth and also profit from its benefits (Ianchovichina & Lundstrom, 2009). Thus, inclusive growth policies should both, support the economy and convergence of per capita income as well as foster social welfare (Rodríguez-Pose & Tselios, 2015). Not only Germany but also the European Commission aims to diminish economic differences between EU regions in order to enhance social cohesion, and provides a budget for regions with lower levels of development (Beugelsdijk et al., 2017; European Commission, 2010). It is important that both aspects, the economic and the social, are considered as “economic convergence does not necessarily imply social convergence, and vice versa” (p.31; Rodríguez-Pose & Tselios, 2015). Following this argumentation (as well as Burda and Severgnini (2017)) factors related to economic and social welfare are included in the analysis of regional inequalities.
2.4 Club Convergence
9 such that they would move to the same steady state equilibrium (Bartkowska & Riedl, 2012). Economies that move towards the same steady state equilibrium build a convergence club (Galor, 1996).
Research has found different initial conditions and variables being responsible for an economy to belong to a certain club. Azariadis and Drazen (1990) state that an economy’s initial level of income and human capital, which is also adapted by Durlauf and Johnson (1995), is responsible for an economy to be part of a club. Multiple authors agree that the distribution of income per capita across economies highly influences the club formation and affiliation of an economy (e.g. Basile, 2009; Fiaschi & Lavezzi, 2007; Fotopoulos, 2008; Phillips & Sul, 2009). Corrado et al. (2005) confirm income per capita being a relevant variable in the club formation when considering integrated markets, such as Western Europe. Latest research proves the club convergence hypothesis to hold for European countries with initial conditions being relevant for the resulting income distribution (Von Lyncker & Thoennessen, 2017).
2.4.1 Economic Factors
Lately, research found and agrees that economic factors such as the distribution of income per capita does not follow a common growth path but instead creates cluster patterns (Basile, 2009; Fiaschi & Lavezzi, 2007; Fotopoulos, 2008; Phillips & Sul, 2009). This implies that income is not distributed equally but differences between regions evolve. The cluster patterns and the associated income disparities have not only been found when comparing economies of different continents, but also for integrated economies like the Western European (Corrado et al., 2005). Therefore, when assessing whether growth is inclusive with the majority of people benefitting from it, or follows rather the unequal distribution and cluster patterns that were described previously, multiple economic aspects are included in the analysis of this research. Firstly, per capita income, which can be measured by the Gross Value Added per worker (Bartkowska & Riedl, 2012), is taken into account in the analysis of regional disparities. Multiple studies use income per capita to analyze whether economies achieved convergence or rather build cluster patterns moving to different state equilibria (e.g. Bartkowska & Riedl, 2012; Corrado et al., 2005). Following their approach this research uses data on income per capita to test overall convergence and clustering patterns for German regions.
Secondly, as the club convergence hypothesis underlines their importance, the role of initial conditions should be investigated. This research uses initial GDP, measured by the difference between the GDP in 2000 and the GDP in the respective years included in the analysis, to look for the influence of initial economic conditions on regional inequalities and exhibition of cluster patterns (clubs).
10 socioeconomic (explained in the next section), is included, presuming that the past years have an essential influence on the recent situation. In terms of economic factors the growing importance of the service sector in modern economies, and thus the industrial structure of a region, is considered. As in the research of Bartkowska and Riedl (2012) Gross Value Added in the service sector represents the industrial structure of a region.
For reasons of data availability as well as multicollinearity between variables the selection of economic indicators is limited in this research. Nonetheless, multiple indicators covering social and socioeconomic aspects are included in the analysis and will be presented in the next section.
2.4.2 Socioeconomic Factors
Social welfare and socioeconomic factors are important aspects to consider when analyzing regional inequalities. As mentioned earlier, recent studies on regional disparities tend to move away from the earlier sole focus on economic factors and more towards the inclusion of factors covering the well-being of humans and the population as a whole (e.g. Brzozowski et al., 2010; Granato et al., 2015; Heathcote et al., 2010; Rodriguez-Pose & Vilalta-Bufi, 2005; Südekum, 2008). The German Socio-Economic Panel (SOEP) is a longitudinal study launched in 1984 aiming to grasp information on changes and development of living conditions and social welfare (Wagner, Frick, & Schupp, 2007). The study involves multiple modules covering among others the topics population and demography; education, training and qualification; labour market and occupational dynamics (Wagner et al., 2007) that are picked up in this research.
Population and demography
This category refers to the structure and density of the population. One important aspect to consider when analyzing regional economic performance and development is the age structure of the population. Generally, people aged between 15 and 65 are considered to be part of the labour force and thus be able to contribute to the economic performance of a region (Bundesinstitut für Bau- Stadt- und Raumforschung (BBSR), 2013). People aged above 65 are regarded as pensioners, and belong together with children, to the group op dependants (Bloom & Canning, 2003) that cannot add to economic growth. Thus, a lower number of elderly people relative to the labour force leads to an increase in the economic performance (Bloom & Canning, 2003), implying that a higher share of elderly negatively is considered to have a rather negative impact.
11 Garretsen, & Van Marrewijk, 2009). However, population density could also be an indicator of agglomeration, and thus whether a region is considered to be urban or rural, and related effects of this structural characteristic. The effects that a classification as urban or rural have on regions and their economy will be explained further in section 2.5.
Education, training and qualification
Education and training are crucial factors to prevent the expansion of socio-economic differences in the European Union (Dieckhoff, 2007). In a knowledge-based economy it is of utmost importance to continuously develop human resources, their skills and apply life-long learning (European Commission, 2000). Disparities in human capital endowments have been found to highly influence regional economic performance in Europe (Rodriguez-Pose & Vilalta-Bufi, 2005). Human capital positively affects productivity and growth levels (Mankiw, Romer, & Weil, 1992). This is further illustrated by the Lisbon agenda, which states that intangible inputs might increase the regional efficiency level by generating more advantageous economic environments for regional firms (Dettori et al., 2012). In the past few decades labour market conditions have changed with a declining demand for low-skilled labour, and a shift and increasing demand for knowledge-intensive jobs (Maurin & Thesmar, 2004). At the regional level a higher availability of well-educated labour forces represents an advantage for the localization of innovative firms, thus promoting local productivity (Rauch, 1993). Literature confirms the crucial role of human capital for economic performance and development: Dettori et al. (2012) found that intangible productive factors, to which human capital belongs to, support economic growth and social coherence. Thus, human capital is considered to be an important factor for the analysis and explanation of regional inequalities, and qualification of the labour force is included as a structural characteristic in the analysis. Nowadays, a large mobility of the labour force has been recognized, which influences the availability of human capital and the labour force participation rate, and thus impacts a region’s performance (Granato et al., 2015; Niebuhr, Granato, Haas, & Hamann, 2012; Rodriguez-Pose & Vilalta-Bufi, 2005). Considering these aspects, not only the qualification of labour force and their availability, represented by the labour force participation rate, but also the commuter and migration balance are included in the analysis (Niebuhr et al., 2012).
2.5 Exogenous Factors/Effects
This section will present and explain additional factors that are considered to be relevant in the analysis of regional inequalities.
2.5.1 Agglomeration Effects
12 geographically close, or agglomerated, gives rise to positive spillovers, which can function regionally as well as city-specific, but always under the term of close geographical proximity (Brakman et al., 2009).
The importance of human capital has already been explained in the previous section. However, within agglomeration economies additional effects arise with hu man capital becoming important in a quantitative and qualitative manner. First, the effect of supply of labour, which is larger in agglomerated economies – a higher concentrated population attracts more people, thus human capital is accumulated (Lucas, 1988). On a qualitative level, urban agglomerations tend to be more favourable as they often offer a strong industrial base, an advanced service sector as well as a larger share of highly educated/qualified people (Maseland, 2014).
Moreover, knowledge and intellectual capital are currently key drivers for innovation and subsequently positively influence economic growth and employment (Guastella & Timpano, 2015; Roth & Thum, 2010). Since increased knowledge can spillover and can be used to produce higher quality products, services, and process innovations, these are key to future innovative capacity of a region (Guastella & Timpano, 2015). Urban agglomerations tend to attract these qualities, thereby pushing economic growth more compared to less urbanized regions.
Concerning the physical geographical location, Cuadrado-Roura (2001) found that the accessibility of a region in crucial for the diffusion of these innovations, technological progress, and investments. Moreover, (ESPON, 2009) states that the access to, and quality of, the infrastructure is another important aspect positively influencing economic growth. Significant differences in terms of accessibility between urban and rural areas become clear. All of these factors are subject to positive spillover effects on the condition that regions are more agglomerated or closer to each other. Thus, urban and metropolitan regions profit more from these positive spillovers, which in turn causes regions to follow different growth paths and, hence, could lead to regional inequality due to less favourable conditions and resulting poorer performance of rural regions.
2.5.2 Specific Characteristics of Germany
13 necessity of these transfers in order to counter the large scale migration from East to West Germany and improve the local labour markets. Finally, all measures and policies by the German government were aimed to improve economic conditions, catalyze growth and in the end regional convergence (Ragnitz, 2014). Numerous obstacles had and possibly still have to be overcome to succeed, or at least, partly succeed and reach the goal of regional equality in general and between East and West regions specifically. The picture painted by Uhlig (2007) and Demary and Röhl (2009) does not provide an optimistic view. Compared to Western regions, Eastern regions suffer from issues like persistent outward migration, especially those younger than 30; lower fertility rates; hence a net loss of population; lower GDP per capita; higher unemployment, lower wages and labor productivity. On another note, legal systems and regulations are equal, unemployment benefits are more or less similar, and the costs of living are lower in Eastern regions.
Next to these described historical less favourable conditions East Germany faces issues related to urbanization and related metropolitan areas. The previous section pointed out the economic advantages of urban regions compared to rural regions, and as possible consequence resulting regional inequalities between urban and rural regions. The German government has created metropolitan areas. Within these metropolitan areas economic ties are strengthened by cooperation and could lead to improved labour markets and employment opportunities, a better business environment, and above all the creation of positive spillovers. East Germany has only two metropolitan areas (see figure below) and could thereby have a disadvantage, which could possibly make (some) East German regions lag behind in terms of economic performance and thus lead to the creation of regional inequalities.
14 As illustrated, it is important to keep this specific East West history of Germany and factors relating to it in mind when analyzing regional inequalities as the history still might have an impact on the situation nowadays. Multiple researcher account for differences between East and West in their research, e.g. Berentsen (2006), Borsi & Metiu (2015), Kubis et al. (2012), Maseland (2014), Vollmer, Holzmann, Ketterer, & Klasen (2013).
Thus, East as well as additional factors will be included as a variable in the analysis to check for potential East West patterns and disparities resulting from the German separation.
3. METHODOLOGY
3.1 Data and Panel Construction
Data for this research was extracted from the database INKAR, “Indikatoren und Karten zur Raum- und Stadtentwicklung” (indicators and maps for spatial and urban development). This database was established by the Federal Institute for Research on Building, Urban Affairs and Spatial Development including data from the multiple federal statistical offices (Statistische Ämter des Bundes und der Länder).
Data from INKAR is available on different spatial levels. The “Kreise” level relates to administrative city district and counties, and mainly corresponds to the NUTS3 classification of the European NUTS classification system (Bundesinstitut für Bau- Stadt- und Raumforschung (BBSR), 2013). As we investigate inequalities and disparities on a regional level, the German regions on the “Kreise” level are our unit of analysis and thus data on the “Kreise” level was retrieved from INKAR.
The panel covers the period from 2000 to 2014, containing 402 regions and 6030 observations. After detailed observation some outliers were recognized and removed leading to a total of 398 regions and 5970 observations being included in the analysis. It contains multiple indicators to be used as explanatory variables for the regression analysis that will be explained in more detail in section 3.3.
For some indicators data were incomplete due to non-availability of the data for the entire period. Missing values were replaced using the multiple imputation method in Stata. Multiple imputations describe missing values being replaced by multiple sets of plausible values (StataCorp, 2013). Thus the multiple imputation method by Stata that fills in the missing information which then can be used for further analysis (StataCorp,1985, 2013).
3.2 Club Construction
15 Bartkowska and Riedl (2012), and Von Lyncker and Thoennessen (2017), the variable considered is GVA per worker representing per capita income.
The club clustering algorithm consists of two steps, a log t test and subsequently the formation of convergence clubs.
3.2.1 Log T Regression Test
Firstly, the log t test was applied to the entire sample to analyze the transition behavior and growth path of the different regions. The idea behind this test is that regions with a similar development can be grouped together.
Using panel data, the log t test is based on an innovative decomposition of the variable of interest (Bartkowska & Riedl, 2012; Simo-Kengne, 2016), as shown below:
(1)
where yit represents the log of income per capita that is proxied by the GVA per worker, the region specific characteristic component, the common factor and the error term (Bartkowska & Riedl, 2012; Simo-Kengne, 2016). The development of a specific region log yit is illustrated by the common factor and two region-specific factors and .
This equation was further modified by Bartkowska and Riedl (2012).1
Additionally, Phillips and Sul (2007) developed a log t regression test to test the null hypothesis of overall convergence:
against
The null hypothesis of overall convergence is rejected for our sample, which implies that regions are moving towards different steady state equilibria (Bartkowska & Riedl, 2012; Galor, 1996). “Rejection of the null hypothesis of convergence for the whole panel cannot rule out the existence of convergence in subgroups of the panel.” (p.4; Du, 2016). Following this statement, we look for regions that are converging using the club clustering algorithm of Phillips and Sul (2007).
3.2.2 Regional Convergence Clubs
The club clustering algorithm by Phillips and Sul (2007) is used to identify groups of regions moving to the same steady state equilibria (Bartkowska & Riedl, 2012; Du, 2016). The convergence clubs can be formed based on the results of the log t test (Bartkowska & Riedl, 2012). Following the steps by Du (2016) we applied the club clustering algorithm of Phillips and Sul (2007) to our sample. This included the implementation of log t tests in order to recognize and merge regions that fulfil the convergence hypothesis jointly (Du, 2016), thus moving towards the same steady state equilibrium (Bartkowska & Riedl, 2012).
1
16 The aim of this research is to analyze the latest situation and state of regional inequality in Germany. Therefore only data of the most recent periods is included in the club clustering algorithm. Thus, clubs are formed based on the GVA per worker data for the years 2010 – 2014 consulting and following the steps described in the manuscript of Du (2017).15 clubs have been identified and four regions that could not be allocated to any of the convergence clubs.
Additional t tests were performed to discover which pairs of adjacent clubs could eventually be merged together. This exercise led to the reduction of clubs to a total of nine clubs and one non-convergent club. The number of clubs was further reduced to five clubs based on the t test results in order to perform the regression analyses with a limited number of clubs and a clearer club composition. An overview of the regions and their respective club allocation is provided in Appendix 1 and 2. The non-convergent regions were put together as a group named Club 6. Because of the large heterogeneity in Club 6 it can be considered as an outlier and is excluded from the regression analysis. Table 1 displays the club classification and the number of regions assigned to each club.
Clubs Number of regions
Club 1 84
Club 2 76
Club 3 50
Club 4 121
Club 5 67
Table 1. Convergence Club Classification 3.3 Variables
The selection of variables for the model meant to explain the regional disparities and potential patterns across Germany was inspired by related scientific articles such as Bartkowska & Riedl (2012), Berentsen (2006), Beugelsdijk et al. (2017), Maseland (2014) and Von Lyncker & Thoennessen (2017).
As this research aims to explain regional inequalities and disparities the dependent variable should be a measure of variation. The existence of multiple clubs indicates that variation and inequality are present, and the probability of a region to belong to a certain club retrieved from the club clustering algorithm represents the dependent variable of our model.
The explanatory variables meant to explain the club affiliation are variables related to economic and socioeconomic indicators as well as initial conditions and structural characteristics.
The table below shows the explanatory variables of the baseline and the extended models. All variables cover the period 2000-20142. A detailed description and explanation of all variables used in the analysis is provided in Appendix 6.
2
17
Variable Name Description
Labour force participation rate
LFPR Labour force participation rate per 100 inhabitants of working age
Employees with Academic Qualification
EMPL_AQ Share of employed with academic qualification relative to all
employed in % East (Dummy
variable)
EAST 0 = region located in West Germany
1 = region located in East Germany Urban (Dummy
variable)
URBAN 0 = rural area
1 = urban area Initial GDP per
capita
GDPCIN Difference to initial GDP per capita (2000)
GDP density GDPDENSA GDP per km²
GVA Service Sector GVATS Percentage share of Gross Value Added of the service sector
Share of older population
POP65 Share of population aged 65 and older relative to inhabitants in %
Commuter balance COM_TOT Commuter balance (net commuters) per 100 employed at place of
living
Total net migration MIGR_TOT Total migration balance per 1.000 inhabitants
Population density POP_DENS Inhabitants per km²
Table 2. Explanatory Variables Baseline and Extended Models
With the dummy variables we can check on structural characteristics and patterns for specific regions. For East West differences an East German dummy is used and for urban rural an urban dummy.
3.4 Multinomial Logistic Regression
For this research a multinomial logistic regression approach is chosen to analyze the significance of economic and socioeconomic factors for membership in a club. This method transforms logistic regressions to problems with multiple discrete outcomes (Greene, 2012). The discrete outcomes in this research are the different clubs a region can be assigned to, thus ranging from club 1 to club 6.
The multinomial logistic regression model consists of a categorically distributed dependent variable and multiple independent variables (Greene, 2012). It can be used to estimate the outcome of the dependent variable, thus in this case the probability of a region to belong to a certain club (Greene, 2012). In a multinomial logistic model it is possible to interpret the so-called score directly as a probability value, which indicates the probability of an observation belonging to a club based on the measured characteristics of the independent variables of this region (Carter Hill, Griffiths, & Lim, 2011).
3.5 Baseline Model
18 theoretical background provided in Chapter 2, but also influenced by statistical issues of multicollinearity.
As pointed out in section 3.4, the dependent variable describes the probability of a region to belong to a certain club. It is regressed on the labour force participation rate, human capital measured by the share of employees with academic qualification, the East and Urban dummy variables, initial GDP measured by the difference between the GDP in 2000 and the respective year, and GDP density measured by GDP per km². Furthermore, the error term is included in the regression.
The dummy variables are included to investigate on differences and patterns of Eastern and Western, as well as urban and rural regions respectively.
The idea and approach of this research is to emphasize more the importance of socioeconomic factors for regional equality and convergence. Therefore, additional models were created, including more variables related to socioeconomic conditions. Two additional models were created in order to check for multiple factors and at the same time avoid a too high correlation between the explanatory variables.
Model 2 adds the indicators elderly population, measured by population over 65, as well as the net commuter balance, measuring the difference between incoming and outgoing commuters of a respective region.
Model 2:
In Model 3, population density is added, among other factors. As GDP density is a variable that is self calculated including population density, it needed to be excluded from this model. Further, model 3 includes the service sector, measured by its GVA, and both indicators related to mobility: next to commuter balance the migration balance is included, measuring the difference between people moving to and away from a region.
Model 3:
19 4. ANALYSIS AND FINDINGS
4.1 Coefficient of Variation Analysis
With the log t regression test in section 3.2.1 the null hypothesis of overall convergence could be rejected for our sample indicating that the development across the country differs and regions in Germany move towards different steady state equilibria. Before analyzing regional inequality in Germany and the underlying “reasons” in more detail we conducted tests to investigate on the distribution and on potential patterns for GDP per capita, GDP per employee and GVA per employee in Germany. For these tests the coefficient of variation was chosen, being a measure that covers the degree of variation of the underlying data, which provides insights into an even or rather uneven distribution of GDP and GVA in Germany, as well as the size of the variable.
The coefficient of variation tests were performed on the whole German sample as well as on composed subsamples, which were East and West regions, rural and urban regions, North and South regions, in order to compare them and investigate on potential patterns.
East and West regions
Table 3. Coefficient of Variation
The results for East and West regions, displayed in Table 3, suggest that the coefficient of variation (CoV) and thus the degree of variation of GDP per capita, GDP per employee and GVA per employee is different in the subsamples. Regarding GDP per capita, the CoV values are significantly higher for the Western regions and very similar to the results for whole Germany. While the values for West Germany and whole Germany increase, Eastern German values decrease, implying different developments in East and West Germany between 2000 and 2013. Looking at the minimum and maximum values, one can recognize that the maximum and mean values are higher for the Western regions - the maximum value is even more than 2.5 times higher - indicating higher income and welfare in the Western regions. The minimum values though are slightly lower for the Western regions.
20 The values for GDP per employee differ less compared to the results for GDP per capita. The East German CoV increases, while the West and whole German CoV decrease, suggesting lower variation. These results could possibly suggest convergence regarding GDP per employee in Germany and between East and West regions. However, although the differences between minimum and maximum values decreased, there are still disparities between East and West regions - the West maximum is significantly higher - that persisted over time. The results for GVA per employee are similar. Convergence takes place regarding the CoV values, however, the West maximum values for 2013 are still almost 50% higher compared to the East values implying that the richer regions are located in West Germany.
Concluding, the CoV values between the subsamples differ and all Western CoV values are higher compared to the Eastern values meaning heterogeneity is higher in the Western regions. Only, when indicators are measured per employee, the CoV values of Eastern and Western regions seem to converge. A possible explanation for this phenomenon could be the higher number of unemployed in the Eastern regions compared to Western regions (see Appendix 4), however, this issue will not be further investigated in this research. Nevertheless, the apparently persisting disparities between East and West regions will be kept in mind and paid attention to by using an East dummy variable in the analysis.
Rural and Urban regions
The results for rural and urban regions, provided in Appendix 5, show a higher CoV of GDP per capita for urban regions implying higher variation in urban and lower variation in rural regions. Both types of regions experience increasing variation over time. Looking at the minimum and maximum values, one can recognize that these are higher for the urban regions; the maximum values are even more than 50% higher for the urban regions. The results for GDP per employee and GVA per employee are similar, showing higher values for urban regions as well. The urban CoV values are always higher compared to the rural values implying that heterogeneity is higher in the urban regions. For all indicators CoV values of urban regions are more similar to whole German. The GDP per capita values increase suggesting higher heterogeneity of GDP per capita in both types of regions, while the per employee CoV values decrease, implying lower heterogeneity of GDP and GVA per employee over the years. Thus, as one might have expected, per capita income and GVA are on average higher in the urban regions, and disparities between urban and rural regions exist. This together with possible patterns of rural and urban regions will be further considered by making use of an urban dummy variable in the regression analysis.
North and South Regions
21 time period. Therefore, no further attention is paid to distinguish North and South regions in the analysis.
4.2 Cluster Analysis
4.2.1 Description/Descriptive Statistics of Clusters
The table below shows the club composition based on the results from the club clustering algorithm including the log t tests and small adjustments, and the descriptive statistics for GVA per worker for each club respectively. In general a descending order of the GVA per worker values is recognizable, implying that lower clubs consist of regions experiencing lower levels of GVA per worker in the past years. Looking at the mean and median values one can recognize that Club 1 represents the club with the highest performance while Club 5 consists of regions with lower levels of GVA per worker (Club 6 was marked as outlier and is therefore excluded from the analysis).
Clubs No. of regions Mean Std. Dev. Min. Max. Median
Club 1 84 65.49 9.54 44.6 88.7 65.05
Club 2 76 57.24 4.97 42.4 66.2 58.3
Club 3 50 55.95 3.52 48.1 71.3 55.95
Club 4 121 53.78 7.12 43.1 89.5 52
Club 5 67 49.39 7.24 42 86.1 47.6
Table 4. Convergence club classification and GVA per worker descriptive statistics
Appendix 4 shows the club distribution on a map. Darker colours on the colour scale represent higher clubs. Looking at the map multiple patterns can be recognized.
Firstly, one can see that larger cities and their surrounding regions often belong to the highest clubs, such as the cities of Hamburg, the southern agglomeration of Berlin, Munich, Nuremberg, Mannheim, Heidelberg and Frankfurt, as well as regions of the Rhine-Ruhr metropolitan area with cities like Düsseldorf and Essen, and in the East the southern agglomeration of Dresden. This observation, together with the findings from the previous section, suggests that urban areas, especially the larger cities and their surroundings, are more favourable to be located in higher clubs.
Looking at Germany as a whole it seems that regions belonging to the higher clubs are rather located in the West. A cluster of regions belonging to Club 1 and 2 can be found in the South and South-West of Germany. East Germany has only very few regions allocated to Club 1 and 2, and it seems that multiple of the regions belonging to the lowest club, Club 5, are lined up at the borderlines, e.g. at the border with Poland and the Czech Republic, and at the former borderline with West Germany. This observation corresponds with the observations on potential East West patterns from section 4.1, and gives the impression that there is still a significant difference in GDP and income levels between East and West regions.
22 A third pattern that will not be further investigated is that there seems to be a line in the shape of a parable consisting of Club 1 regions located in South and South-West Germany. The line connects the cities of Munich and Frankfurt including regions in the North of Munich, Heidelberg, Karlsruhe, Mannheim, Ludwigshafen, Mainz and Wiesbaden. However, as the observations from section 4.1 led to the conclusion that there are no significant differences between North and South this proposition of more favourable regions in the South will not be further investigated.
4.2.2 Results of Cluster Analysis
After having looked at the club composition and the club distribution on the map, the focus now is on the descriptive statistics of the explanatory variables to be included in the regression analysis (Appendix 7), to check on potential trends and patterns of these variables to be derived for the clubs.
The two dummy variables in the analysis are East and Urban. East takes the value 1 in case the respective region is located in East Germany, otherwise it is 0. Looking at the descriptive statistics for East, provided in Appendix 7, one recognizes that the median is 0 for all clubs. This indicates that at least half of the regions per club are located in West Germany. This can partly be explained by the fact that the sample consists of more regions located in West Germany (see Appendix 1). The mean values of the clubs, representing the arithmetic average (Carter Hill et al., 2011), follows an ascending order. While the mean for Club 1 is approximately 0.06 and rather small, the value is multiple times higher for Club 5 (0.45). This implies that the share of regions located in East Germany increases for lower clubs, and East regions are rather assigned to lower clubs, which confirms the proposition made in the previous section.
The descriptive statistics for the urban dummy variable reveal another trend. The urban dummy variable takes the value 1 when regions are classified as urban, 0 when rural. While the median value is 1 for the first three clubs, it becomes 0 for Club 4 and 5 implying that these latter clubs consists of more rural than urban regions. The decreasing mean value confirms this trend. Lower clubs have lower mean values implying that the share of rural regions follows a growing trend, the lower the club. The table below confirms and illustrates the findings described.
Club 1 Club 2 Club 3 Club 4 Club 5
East 5 9 6 27 30
West 79 67 44 94 37
Urban 62 41 27 54 16
Rural 22 35 23 67 51
23 A lot of the remaining explanatory variables also show some sort of patterns or order how they are distributed among the clubs, with more favourable variables like labour force participation rate, GDP density or employees with academic qualification showing higher values for the higher clubs. Details on the descriptive statistics are provided in Appendix 7. However, these propositions should be treated with care/caution as they are just observations that have been made based on the descriptive statistics. The regression analysis, which is presented in the next section, provides findings and results based on statistical analysis and evaluation.
4.3 Regression Analysis
The models composed for the multinomial regression analysis analyze the meaning and importance of economic as well as socioeconomic indicators for a region’s club membership. The multinomial logistic regression approach that was chosen allows us to evaluate the significance of the indicators and the outcome of the dependent variable representing the probability of a region to belong to a specific club (Greene, 2012).
In all models Club 1 was chosen as the baseline club, which means that the results for the other clubs should be interpreted relative to Club 1. In our case this means the probability of belonging to the respective club compared to belonging to Club 1.
Baseline Model
The results for the baseline model are presented in the table below.
(1) (2) (3) (4) (5)
VARIABLES Logit coeff Logit coeff Logit coeff Logit coeff Logit coeff
LFPR -0.0725*** -0.116*** -0.114*** -0.108*** (0.0116) (0.0130) (0.0112) (0.0130) EMPL_AQ -0.0678*** -0.0963*** -0.130*** -0.189*** (0.0202) (0.0276) (0.0277) (0.0302) EAST 0.822*** 0.987*** 1.833*** 2.955*** (0.179) (0.210) (0.193) (0.207) URBAN -0.435*** -0.269** -0.476*** -0.952*** (0.110) (0.130) (0.116) (0.137) GDPCIN 0.0159*** 0.0302*** 0.0400*** 0.0544*** (0.00479) (0.00560) (0.00461) (0.00581) GDPDENSA -0.00909*** -0.0173*** -0.0141*** -0.0195*** (0.00202) (0.00287) (0.00244) (0.00336) Constant 6.668*** 9.869*** 10.80*** 10.09*** (0.961) (1.100) (0.964) (1.086) Observations 5,970 5,970 5,970 5,970 5,970
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 6 Regression Baseline Model
24 value for the respective indicator, the higher is the probability that the region stays in the club it was assigned to relative to Club 1 (the base outcome). The reverse applies for negative coefficients. A negative coefficient implies that a higher score on the respective factor decreases the probability of staying in this club compared to Club 1.
The coefficients of the first two variables, labour force participation rate representing employment, and employees with academic qualification are negative. Higher employment and a higher share of employees with academic qualification, both increase the probability of being in Club 1. Additionally, a higher GDP density is associated with a higher propensity of being in Club 1, though the coefficient is relatively small and therefore has a minor impact compared to the impact of labour force participation rate and employees with academic qualification.
Surprisingly the coefficient for the factor initial GDP, measured by the difference to the GDP of the year 2000, is positive implying that a higher initial GDP leads to less probability of being in Club 1. However, the coefficient is rather small and thus this factor only has a minor impact.
East and Urban are the two dummy variables included in the regression. The coefficient for East is positive and considerably increasing for lower clubs, being more than three times higher for Club 5 compared to Club 2. Thus, one can conclude that East regions have a lower probability of being in Club 1, and further, that the probability of East regions to stay in the respective club increases with the clubs becoming lower. This implies that East regions are rather clustered in the lowest clubs, which was also shown in Table 5. The negative coefficient for urban makes clear that regions that are urban have higher probabilities of being in Club 1, or generally assigned to higher clubs.
Except for the constant and labour force participation rate coefficients, all coefficients experience the highest (for negative coefficients the lowest) value for Club 5, which implies that the impacts are higher for this club.
Extended Models
25 Model 2
(1) (2) (3) (4) (5)
VARIABLES Logit coeff Logit coeff Logit coeff Logit coeff Logit coeff
LFPR -0.133*** -0.141*** -0.194*** -0.207*** (0.0130) (0.0143) (0.0124) (0.0149) EMPL_AQ -0.0362* -0.0806*** -0.0905*** -0.143*** (0.0206) (0.0284) (0.0283) (0.0310) EAST 0.418** 0.715*** 1.228*** 2.361*** (0.187) (0.219) (0.201) (0.216) URBAN -0.585*** -0.327** -0.624*** -1.170*** (0.113) (0.133) (0.121) (0.142) GDPCIN 0.0228*** 0.0303*** 0.0479*** 0.0702*** (0.00483) (0.00554) (0.00468) (0.00599) GDPDENSA -0.00407* -0.0179*** -0.0149*** -0.0126*** (0.00218) (0.00326) (0.00272) (0.00358) POP65 0.222*** 0.110*** 0.299*** 0.339*** (0.0224) (0.0247) (0.0215) (0.0249) COM_TOT -0.0179*** -0.00465** -0.0126*** -0.0237*** (0.00175) (0.00208) (0.00173) (0.00204) Constant 6.514*** 9.541*** 10.73*** 10.33*** (0.973) (1.095) (0.964) (1.093) Observations 5,970 5,970 5,970 5,970 5,970
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 7 Regression Model 2
The coefficients of the variables that have been taken over from the baseline models, labour force participation rate, employees with academic qualification, East, urban, initial GDP per capita and GDP density, keep their signs and thus can be interpreted in the same way. The value of some of these coefficients changes slightly, implying higher or lower impact on t he propensity to be in the same club or in Club 1.
The coefficient of the newly added variable elderly population is positive implying that regions with a higher share of people above the age of 65, have a lower probability of being in Club 1. Thus, a higher share of elderly rather leads to being allocated to a lower club. The variable commuter balance describes the balance between incoming and outgoing commuters. It has negative coefficients implying that a higher net commuter score, thus the higher the number of incoming commuters exceeding the number of outgoing commuters, increases the probability of being allocated to Club 1.
26 Model 3
(1) (2) (3) (4) (5)
VARIABLES Logit coeff Logit coeff Logit coeff Logit coeff Logit coeff
LFPR -0.0616*** -0.105*** -0.0877*** -0.0581*** (0.0118) (0.0132) (0.0114) (0.0138) EMPL_AQ -0.0926*** -0.125*** -0.156*** -0.195*** (0.0199) (0.0267) (0.0259) (0.0296) EAST 0.755*** 0.909*** 1.592*** 2.357*** (0.182) (0.211) (0.191) (0.212) URBAN -0.536*** -0.256** -0.510*** -0.960*** (0.111) (0.130) (0.115) (0.143) GDPCIN 0.0207*** 0.0274*** 0.0486*** 0.0883*** (0.00487) (0.00554) (0.00473) (0.00635) GVATS 0.0189*** 0.0212*** 0.0562*** 0.117*** (0.00468) (0.00527) (0.00454) (0.00608) COM_TOT -0.0169*** -0.00659*** -0.00925*** -0.0184*** (0.00170) (0.00201) (0.00168) (0.00203) MIGR_TOT -0.0463*** -0.0502*** -0.0681*** -0.127*** (0.00960) (0.0108) (0.00914) (0.0111) POP_DENS 0.000223** -0.000379*** -0.000384*** -0.00109*** (9.00e-05) (0.000123) (0.000105) (0.000175) Constant 4.529*** 7.913*** 5.337*** -1.607 (1.062) (1.199) (1.048) (1.270) Observations 5,970 5,970 5,970 5,970 5,970
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 8 Regression Model 3
The coefficients of the variables that have been used in the previous models keep the same signs, and thus their interpretation is confirmed. Surprisingly, the coefficients for GVA of the service sectors are positive implying that a higher GVA achieved in the service sector decreases the probability to be in Club 1. The migration balance, measuring the net of inflows and outflows of inhabitants, is negative implying that a positive migration balance increases the probability of a region to be assigned to Club 1. Lastly, the coefficient of population density experiences a switch of the sign: it is positive for Club 2 implying that higher population density decreases the probability of being in Club 1 compared to being in Club 2. However, for Clubs 3 to 5 the coefficient becomes negative, which means a higher population density leads to a higher probability of being in Club 1 relative to the respective club. As the coefficients are very small the impact of this variable is considered to be rather minor.
27 Variables with
positive coefficients
East, initial GDP, population above 65, GVA in the service sector and population density for Club 2
Variables with negative coefficients
Labour force participation rate, employees with academic qualification, urban, GDP density, commuter balance, migration balance and largely for population density
The impact of these results will be further discussed in Chapter 5.
5. DISCUSSION
5.1 Patterns of Regional Inequality
The regression analysis has shown that although the government has implemented policies in the past and provided budget and funding for regions lagging behind overall convergence has been rejected and regional inequality persists in Germany following different patterns. The existence of the clubs implies that regions cluster and move towards different steady state equilibria.
The results of the regressions showed clearly the significant impact of two variables: urban and East, whose coefficients have been the two highest for all regression results (excluding the constant coefficient). This implies large disparities between urban and rural areas, and all efforts by the government to support the catching-up process of regions in East Germany and diminish inequalities have not succeeded yet. The size of the coefficient is a clear sign that the achievement of overall convergence is still linked to large efforts.
Looking again at Table 5, which is again provided below, to better illustrate and explai n the results, one clearly recognized that rural and East regions rather cluster in the lower club making clear that they are lagging behind the more favourable West and urban regions. Out of the 77 East regions, 57, which represent almost 75% of the East regions, are assigned to Club 4 and 5. Regarding the West regions only about 40% are categorized to Club 4 and Club 5. This imbalance is also illustrated in Appendix 3a-e showing the distribution of clubs per club respectively. Thus, a huge gap between East and West regions can be recognized.
Club 1 Club 2 Club 3 Club 4 Club 5 Total
East 5 9 6 27 30 77
West 79 67 44 94 37 321
Urban 62 41 27 54 16 200
Rural 22 35 23 67 51 198
Total 84 76 50 121 67 398
28 A similar picture can be drawn for urban and rural regions. About 60% of the rural regions has been categorized to Club 4 and Club 5, while only 35% of urban regions are assigned to these clubs.
This raises the question of whether there could possibly be a link between urban and East regions both being marked by weak economic performances. Looking at the map provided in the figure below one can recognize some patterns. The orange-colored areas are urban areas, which are characterized by higher population densities. The green-colored areas are rural areas and are characterized by lower population densities. It is evident that most urban regions, and thus related positive spillovers, are located in Western Germany. The only urban regions in Eastern Germany are found in and around the few larger cities.
Figure 2- City districts and counties. Urban-rural division. Source: INKAR, BBSR Bonn (2015).
29 5.2 Regional Inequality Policies
Not only the club composition with East regions clustering in the lower clubs but also the analysis of the descriptive statistics and CoV test indicate that the gap between economic as well as socioeconomic factors between East and West remains a problem as it creates regional inequalities. Thus, looking at inclusive growth determinants one can conclude that convergence between East and West has not been achieved yet. More than a quarter century after reunification of East and West this creates doubts on how effective government policies and structural transfers are in stimulating growth and development in regions lagging behind. Shortly after reunification the solidarity surcharge has been introduced and as explained earlier, the net transfers made until 2005 are worth €1.600 billion (Ragnitz et al., 2009). Still nowadays €100 billion resulting from this solidarity surcharge are transferred to East Germany meant to support the development and create convergence between East and West regions.
Another means to support financial equalization on the state level is called the Länderfinanzausgleich and includes the redistribution of funds (Frick & Goebel, 2008). The idea behind these transfers is that richer states support poorer states with financial means. Since the integration of East states in this existing system of financial equalization in 1995 a clear East West patterns emerged showing that especially the states in South-West Germany, such as Bavaria, Hesse and Baden-Württemberg, are donor states while all East states profit from additional financial gains (see figure below).
Figure 3 – Financial equalization across regions (Frick & Goebel, 2008)
Data source: Bundesfinanzministerum, Statistisches Bundesasmt, authors‘ calculations
30 East West transfers are the most efficient means to increase regional equality between regions or whether urban rural development policies and related transfers could have a higher impact and lead to successful convergence of regions in East and West Germany. Looking at the current efforts and outcomes, it seems that instruments on the state level make less sense nowadays and concepts on the regional level would do better, including policies that are tailor-made for specific locations paired with strategies on local growth.
6. CONCLUSION
The results of this research imply that overall convergence of German regions was rejected. Regional inequality still matters and disparities are persistent in unified Germany. The presence of multiple clubs is an indicator for existing income differences between German regions. The descriptive statistics, the Coefficient of Variation tests as well as the regression results showed that regional inequality follows two specific patterns: the first is that East regions are less favourable compared to West regions and rather allocated to lower clubs. The growing East coefficient of the regression analysis is an indicator of lower probability of th e East regions to be located in higher convergence clubs. Although, the German reunification took place more than 25 years ago and multiple instruments and systems were implemented by the governments and its policies to equalize living conditions and economic performance to West standards, the efforts seem not (yet) to have resulted in the desired outcomes.
Further, the analysis revealed striking differences between urban and rural regions. Similar to the East West pattern, observations have shown that also rural regions face a clear disadvantage, and the majority of them is allocated to the lowest convergence clubs. The advantage of urban towards rural region can be summarized under the term agglomeration effects. The analysis of the results revealed a possible link between East-West and urban rural regions that was also found by Maseland (2014). According to his research East-West patterns should rather be considered urban-rural patterns as East Germany contains a higher share of rural regions, leading to regional disparities with West Germany consisting of more urban regions.
31 above 65, GVA in the service sector and population density (only for Club 2) have the opposite effect and thus decrease the propensity of being assigned to Club 1.
Overall the results are considered to be in line with underlying theories and this research to be a contribution in the field of regional inequality, convergence clubs and the development and convergence of East and West Germany.
However, this research faces some limitations. The time period and the availability of data put some limitations to the analysis of data, including the imputations of one variable, which might have slightly affected the results. The capital transfers and equalization measures have been mentioned in the discussion part but were not quantified and not part of the analysis. Lastly, a possible impact of the financial crisis of 2007/2008 was neglected, assuming that it had similar effects on all regions and thus did not influence the relative comparison.
Future research on this topic could try to capture both of these effects: equalization instruments and systems could be included in the analysis quantifying related measures like capital transfers. Including the effects of the financial crisis is a more challenging issue and future research would have to develop a way in which these could be taken into account. Finally, it might be interesting to consider different or also longer time spans and see whether the analysis reveals different results and insights on regional inequalities and convergence in Germany.
Acknowledgements:
32 REFERENCES
Azariadis, C., & Drazen, A. (1990). Threshold externalities in economic development. Quarterly Journal of Economic Development, 105(2), 501–526.
Barrios, S., & Strobl, E. (2009). The dynamics of regional inequalities. Regional Science and Urban Economics, 39, 575–591.
Bartkowska, M., & Riedl, A. (2012). Regional convergence clubs in Europe: Identification and conditioning factors. Economic Modelling, 29(1), 22–31.
Basile, R. (2009). Productivity polarization across regions in Europe: the role of nonlinearities and spatial dependence. International Regional Science Review, 31, 92–115.
Berentsen, W. H. (2006). Changing regional inequalities in United Germany. Eurasian Geography and Economics, 47(4), 462–477.
Beugelsdijk, S., Klasing, M. J., & Milionis, P. (2017). Regional economic development in Europe: the role of total factor productivity. Regional Studies, 1–16.
Blanchard, O., & Kremer, M. (1997). Disorganization. The Quarterly Journal of Economics, 112(4), 1091–1126.
Bloom, D. E., & Canning, D. (2003). Contraception and the Celtic Tiger. Economic and Social Review, 34, 229–247.
Borsi, M. T., & Metiu, N. (2015). Discussion paper. Empirical Economics, 48(2), 657–681.
Brakman, S., Garretsen, H., & Van Marrewijk, C. (2009). An Introduction to Geographical Economics (Second Edi). Cambridge: Cambridge University Press.
Brzozowski, M., Gervais, M., Klein, P., & Suzuki, M. (2010). Consumption, income, and wealth inequality in Canada. Review of Economic Dynamics, 13, 52–75.
Bundesinstitut für Bau- Stadt- und Raumforschung (BBSR). (2013). Europäische Raumbeobachtung – Raumabgrenzungen NUTS-Regionen. Retrieved from
http://www.bbsr.bund.de/BBSR/DE/Raumbeobachtung/Raumabgrenzungen/Siedlungsstrukturell eRegionstypenEuropa/NUTSRegionen/NUTS_2_Regionen.html
Burda, M. C., & Hunt, J. (2001). From reunification to economic integration: productivity and the labor market in Eastern Germany. Brookings Papers on Economic Activity, 2, 1–71.
Burda, M. C., & Severgnini, B. (2014). Solow residuals without capital stocks. Journal of Development Economics, 109, 154–171.
Burda, M. C., & Severgnini, B. (2017). Total factor productivity convergence in German states since reunification: Evidence and explanations. Journal of Comparative Economics, 1–20.
Carter Hill, R., Griffiths, W. E., & Lim, G. C. (2011). Prinicples of Econometrics (Fourth Edi). John Wiley & Sons, Inc.
Combes, P.-P., Lafourcade, M., Thisse, J.-F., & Toutain, J.-C. (2011). The rise and fall of spatial inequalities in Franc: A long-run perspective. Explorations in Economic History, 48, 243–271. Corrado, L., Martin, R., & Weeks, M. (2005). Identifying and Interpreting Regional Convergence
